This item is only available as the following downloads:
1 ASTHMA RELATED MORBIDITY AND MORTALITY IN PATIENTS EXPOSED TO INHALED LONG ACTING BETA 2 ADRENOCEPTOR AGONIST BRONCHODILATORS By AYAD K. ALI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PA RTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012
2 2012 Ayad K. Ali
3 To the 300 million individuals in the world who are afflicted with asthma
4 ACKNOWLEDGMENTS Gratitude is expressed to my wife; family; academic advisor; dissertation committee members ; the faculty, staff and graduate students at the D epartment of P harmaceutical O utcomes and P olicy; the Graduate School Editorial staff; the staff at the I nternational C enter; the staff at t he G raduate S tudent and F amily H ousing department; the faculty who taught and administered the core and elective courses pertinent to the program at the University of Florida ; and the staff administering the Fulbright Program at the United States Departmen and Cultural Affairs
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 BACKGROUND ................................ ................................ ................................ ...... 14 Definition of Asthma ................................ ................................ ................................ 14 Natural History of Asthma ................................ ................................ ....................... 15 Asthma and Chronic Obstruct ive Pulmonary Disease ................................ ............ 16 Epidemiology of Asthma ................................ ................................ ......................... 19 Burden of Asthma ................................ ................................ ............................. 20 Risk Factors ................................ ................................ ................................ ..... 21 Asthma Pharmacotherapy ................................ ................................ ...................... 24 Asthma Management Guidelines ................................ ................................ ............ 25 Inhaled Long Acting Beta 2 Adrenoceptor Agonists ................................ ............... 27 Pharmacology of LABA Products ................................ ................................ ..... 28 LABA Safety Controversy ................................ ................................ ................. 30 LABA safety in experimental studies context ................................ ............. 31 LABA safety in observational studies context ................................ ............ 34 Heterogeneity of Response to LABA ................................ ................................ 37 2 PHARMACOEPIDEMIOLOGICAL PERSPECTIVE ................................ ................ 44 Methodological Cha llenges in Observational Research ................................ .......... 44 Confounding ................................ ................................ ................................ ..... 44 Confounding by indication ................................ ................................ .......... 45 Time dependent confounding ................................ ................................ .... 46 Residual confounding ................................ ................................ ................ 46 Effect Modification ................................ ................................ ...................... 47 Selection Bias ................................ ................................ ................................ ... 49 Informative censoring ................................ ................................ ................. 50 Depletion of susceptibles ................................ ................................ ........... 52 Measurement Bias ................................ ................................ ........................... 53 Exposure misclassification ................................ ................................ ......... 55 Diagnostic misclassification ................................ ................................ ....... 59 Outcome misclassification ................................ ................................ .......... 60 Modeling Techniques to Control Bias and Confounding ................................ ......... 61 Multiple Regression Analysis ................................ ................................ ............ 62
6 Conventional Cox proportional hazards regression ................................ ... 63 Cox PHREG with time dependent co variates ................................ ............ 65 Marginal structural models ................................ ................................ ......... 67 Exposure Propensity Scores Technique. ................................ .......................... 71 Instrumental Variable Analysis Technique. ................................ ....................... 73 3 STUDY AIMS AND SIGNIFICANCE ................................ ................................ ....... 89 Study Objective ................................ ................................ ................................ ....... 89 Research Questions, Specific Aims, and Hypotheses ................................ ............ 89 Research Question No. 1 ................................ ................................ ................. 89 Specific aim no. 1 ................................ ................................ ....................... 89 Hypothesis no. 1 ................................ ................................ ........................ 89 Research Question No. 2 ................................ ................................ ................. 90 Specific aim no. 2 ................................ ................................ ....................... 90 Hypothesis no. 2 ................................ ................................ ........................ 91 Research Question No. 3 ................................ ................................ ................. 91 Specific aim no. 3 ................................ ................................ ....................... 91 Hypothesis no. 3 ................................ ................................ ........................ 92 Rationale and Significance ................................ ................................ ..................... 92 4 METHODS ................................ ................................ ................................ .............. 94 Research Ethics ................................ ................................ ................................ ...... 94 Study Type and Design ................................ ................................ ........................... 94 Data Source ................................ ................................ ................................ ............ 94 The General Practice Research Database (GPRD) ................................ ......... 95 The Office for National Statistics (ONS) Mortalit y Data ................................ .... 97 The Index of Multiple Deprivation (IMD) Scores ................................ ............... 98 Study Population ................................ ................................ ................................ ..... 98 Inclusion Criteria ................................ ................................ ............................... 99 Exclusion Criteria ................................ ................................ ........................... 100 Study Duration and Selection of Comparison Groups ................................ .......... 102 Exposure Measurement ................................ ................................ ................. 103 Cohort definition for specific aim no. 1 (morbidity outcome) ..................... 104 Cohort definition for specific aim no. 2 (mortality outcome) ..................... 105 Cohort definition for specific aim no. 3 (subgroup morbidity outcome) ..... 107 Outcome Measurement ................................ ................................ .................. 109 Mortality outcome ................................ ................................ ..................... 109 Morbidity outcomes ................................ ................................ .................. 110 Covariate Measurement ................................ ................................ ................. 111 Patient characteristics ................................ ................................ .............. 111 Practice characteristics ................................ ................................ ............ 112 Asthma severity ................................ ................................ ....................... 113 Concurrent asthma drug prescriptions ................................ ..................... 115 Other concurrent pr escriptions ................................ ................................ 116 Concomitant immunizations ................................ ................................ ..... 117
7 Comorbid conditions ................................ ................................ ................ 118 Other factors ................................ ................................ ............................ 11 9 Sample Size and Power Calculations ................................ ................................ ... 121 Statistical Analysis Procedures ................................ ................................ ............. 121 Descriptive Statistics ................................ ................................ ...................... 122 Inferential Statistics ................................ ................................ ........................ 122 Conventional Cox PHREG model ................................ ............................ 123 Time dependent covariate Cox PHREG model ................................ ........ 123 Marginal structural model ................................ ................................ ......... 125 Sensitivity Analyses ................................ ................................ ........................ 126 5 RESULTS ................................ ................................ ................................ ............. 137 Descriptive Statistics ................................ ................................ ............................. 137 Exclusion Cohort ................................ ................................ ............................ 137 Original Cohort ................................ ................................ ............................... 139 Step Down Therapy Cohort ................................ ................................ ............ 148 Original high dose ICS initiators ................................ ............................... 149 Original medium dose ICS initiators ................................ ......................... 154 Inferential Statistics ................................ ................................ ............................... 160 Asthma related Morbidity ................................ ................................ ................ 161 Prescriptions for oral corticosteroids ................................ ........................ 161 As thma related A&E visits ................................ ................................ ........ 164 Asthma related Mortality ................................ ................................ ................. 166 Comparison of Models ................................ ................................ ................... 167 Sensitivity Analyses ................................ ................................ .............................. 177 6 DISCUSSION AND CONCLUSIONS ................................ ................................ .... 222 Discussion ................................ ................................ ................................ ............ 222 Asthma related mortality ................................ ................................ ................. 225 Asthma related morbidity ................................ ................................ ................ 227 Limitations ................................ ................................ ................................ ............. 229 Conclusions ................................ ................................ ................................ .......... 231 Future Work ................................ ................................ ................................ .......... 232 LIST OF REFERENCES ................................ ................................ ............................. 234 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 255
8 LIST OF TABLES Table page 1 1 Comparison of the stepwise approach for the treatment of a sthma in adults across different asthma management guidelines ................................ ............... 39 1 2 Availability of long acting beta agonist products in t he US and the UK .............. 41 4 1 Estimated equipotent daily doses of ICS for a sthmatic adults .......................... 127 4 2 Dosage strengths o f ICS and ICS/LABA formulations available in the UK ....... 127 4 3 Variable definitions ................................ ................................ ........................... 128 5 1 Patient characteristics for original study coho rt stratified by exposure type ...... 180 5 2 Patient characteristics for step down therapy cohort stratified by approach type among origin al high ICS/LABA dose initiators ................................ .......... 184 5 3 Patient characteristics for step down the rapy cohort stratified by approach type among original m edium ICS/LABA dose initiators ................................ .... 188 5 4 Incidence rates of morbidity outcomes among initiators of study drugs ............ 192 5 5 Incidence rates of prescriptions for short courses of oral corticosteroids for asthma exacerbations among initiators of study drugs ................................ ..... 193 5 6 Inciden ce rates of mortality outcomes among initiators of study drugs ............. 194 5 7 Characteristics of ast hmatics died of asthma in ONS Mortality database ......... 195 5 8 Distribution of average time to event among exposure groups ........................ 196 5 9 Hazard ratios of oral corticosteroid prescriptions among original cohort .......... 197 5 10 Hazard ratios of oral corticosteroid prescriptions among step down cohort with high dose ICS/LABA initiators ................................ ................................ ... 199 5 11 Hazard ratios of oral corticos teroid prescriptions among step down cohort with medium dose ICS/LABA initiators ................................ ............................. 200 5 12 Hazard ratios of asthma related A&E department visits among original cohort and step down cohort with medium dose ICS/LABA initiators .......................... 201 5 13 Hazard ratios of asthma deaths among original practices unlinked to ONS mortality database ................................ ................................ ............................ 202 5 14 Hazard ratios of all cause deaths among original cohort ................................ .. 203
9 LIST OF FIGURES Figure page 1 1 Chemical structure of inhaled beta agonist b ronchodilators commonly prescribed for asthma in the UK ................................ ................................ ......... 41 1 2 Chemical structure of inhaled corticosteroids commonly prescribed for asthma in th e UK ................................ ................................ ............................... 42 1 3 Illustration of the relationship between inhaled corticosteroid dosage and benefit/risk balance ................................ ................................ ............................. 43 2 1 I llustration of confounding ................................ ................................ .................. 75 2 2 Illustration of the relationship between study sample and target population in pharmacoepidemiologic studies ................................ ................................ ......... 76 2 3 Illustration of informative censori ng in pharmacoepidemiology .......................... 76 2 4 ................................ .............. 77 2 5 Illustration of measurement bi as t ypes in pharmacoepidemiology ...................... 78 2 6 Illustration of confounder misclassif ication in pharmacoepidemiology ................ 79 2 7 Illustration of exposure misclassification in randomized controlled trials ........... 80 2 8 Sample of a prescription in the UK general practice ................................ ........... 81 2 9 Il lustration of immortal time bias in pharmacoepidemiology ............................... 82 2 10 Pharmacoepidemiologic approaches to account for bias and confounding ........ 83 2 11 Data structure for Cox proportional hazards model s ................................ .......... 84 2 12 Illustration of association versus causation in pharmacoepidemiology ............... 85 2 13 Illustration of attaining quasi randomization by marginal structural models. ....... 86 2 14 Characteristics of an instrumental variable ................................ ......................... 88 3 1 Study profile ................................ ................................ ................................ ........ 93 4 1 Study profile for morbidity outcome (Study Aim No. 1) ................................ ..... 133 4 2 Study profile f or mortality outcome (Study Aim No. 2) ................................ ...... 134 4 3 Study profile for mortality outcome among prevalent users .............................. 135
10 4 4 Study profile f or subgroup analysis for morbidity outcome (Study Aim No. 3) .. 136 5 1 Cohort sample disposition ................................ ................................ ................ 204 5 2 Distribution of exposure initiators across UK countries. ................................ .... 207 5 3 Prescribing trend of study exposures ................................ ............................... 208 5 4 Distribution of prescriptions in the step d own therapy cohort with original high ICS/LABA dose initiators ................................ ................................ .................. 209 5 5 Distribution of prescriptions in the step down therapy cohort with original medium ICS/LABA dose initiators ................................ ................................ ..... 210 5 6 Prescribing trend of study exposures in step down therapy coh ort .................. 211 5 7 Product limit surviva l estimates of prescribing oral cortico steroids among original cohort of ICS, LABA and ICS/LABA initiators ................................ ...... 212 5 8 Product limit survival estimates of prescribing oral corticosteroids amon g high ICS/LABA dose initiators ................................ ................................ ........... 213 5 9 Product limit survival estimates of prescribing oral corticosteroids among medi um ICS/LABA dose initiators ................................ ................................ ..... 214 5 10 Product limit survival estimates of asthma related A&E departments visits among original coh ort of ICS, LABA and ICS/LABA initiators .......................... 215 5 11 Product limit survival estimates of asthma related deaths among original coh ort of ICS LABA and ICS/LABA initiators in practices unlinked to the Office of National Statistics mortality database ................................ ................. 216 5 12 Product limit survival estimates of all cause deat hs among original cohort of ICS, LABA and ICS/LABA initiators ................................ ................................ 217 5 13 Hazard ratios of asthma related morbidity outcomes stratified by comparison groups and regression models ................................ ................................ ......... 218 5 14 Hazard ratios of asthma related and all cause mortality outcomes stratified by comparison groups and regression models ................................ ................. 219 5 15 Distri bution of stabilized weights estimated by marginal structural models across study follow up year ................................ ................................ .............. 220 5 16 Distribution of unstabilized weights estimated by marginal structural models across st udy follow up year ................................ ................................ .............. 221
11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ASTHMA RELATE D MORBIDITY AND MORTALITY IN PATIENTS EXPOSED TO INHALED LONG ACTING BETA 2 ADRENOCEPTOR AGONIST BRONCHODILATORS By Ayad K. Ali December 2012 Chair: Abraham G. Hartzema Major: Pharmaceutical Sciences Bronchial asthma is a chronic inflammatory disease of the respiratory system, with increasing prevalence and severity in society. Inhaled long acting beta agonist (LABA) bronchodilators are considered part of the regular maintenance therapy in asthmatic patients. Although the purpose of asthma treatment is t o control the disease with medications that have satisfactory safety profile, there has been mounting concern about the safety of inhaled LABA bronchodilators as monotherapy, including increased asthma deaths and poor asthma outcomes. The role of inhaled c orticosteroids (ICS) in asthma management is crucial, and regular inflammation preventer therapy with ICS is associated with lower asthma mortality and morbidity rates. However, ICS have safety concerns as well. While it is recommended to use inhaled LABA bronchodilators in combination with ICS to attain and maintain acceptable outcomes, increased risks of asthma deaths and exacerbations are reported with the combination therapy. The aim of th is study is to use advanced and novel approaches in causal infere nce to evaluate asthma related outcomes in patients exposed to inhaled LABA bronchodilators as monotherapy, and ICS combination therapy. Among the 51,103 adults with asthma who
12 met the inclusion criteria and followed for 12 months after receiving first pre scription of study drugs from January 4, 1993 to August 20, 2010, about 92% initiated ICS monotherapy, 1% initiated LABA monotherapy, and 7% initiated ICS/LABA combination therapy. Among the ICS/LABA combination therapy initiators, 78% were in single devic e formulations and 22% were in separate devices. Among the 3,226 asthmatics who initiated ICS/LABA combination therapy, 14.2% continued on original strength ICS/LABA combination regimen, 26.5% reduced the dose of ICS to 50% while continuing LABA as combina tion regimen, and 59.3% discontinued LABA and continued ICS in original strength as monotherapy. The findings suggest presence of time dependent confounding of asthma exacerbations requiring prescriptions of oral corticosteroids by inhaled LABA products, b ut absence of time dependent confounding of accident and emergency department visits for asthma attacks or asthma related deaths or all cause deaths Compared with ICS monotherapy, LABA monotherapy is associated with 1 0 % increased risks of asthma exacerbat ions requiring short courses of oral corticosteroids (HR, 1.1 0 ; 95%CI, 1.0 7 1. 18 ). Initiators of ICS/LABA combination therapy are respectively 62 % and 50 % less likely to receive prescriptions of oral corticosteroids for asthma exacerbations than initiators of ICS (HR, 0.38; 95%CI, 0.12 0.66) or LABA monotherapies (HR, 0.50; 95%CI, 0.14 0.78). Compared to continuing original combination therapy regimens, step down therapy approaches by LABA discontinuation or ICS dose reduction are associated with significan t reduction in asthma morbidity. Nevertheless, within step down therapy approaches, LABA stoppers are associated with worsened asthma ( short courses of oral corticosteroids ) than ICS/LABA dose reducers when ICS monotherapy is in medium strength (HR, 1.24;
13 95%CI, 1.07 5.01 ). When ICS monotherapy is in high strength, withdrawing LABA is associated with better asthma control than continuing LABA as reduced ICS/LABA regimen (HR, 0.35; 95%CI, 0.06 0.51) Patients who survived for a minimum of 12 months after ini tiating LABA monotherapy are 25% more likely to die from asthma than patients who initiated ICS monotherapy (HR, 1.25; 95%CI, 1.11 3.01). There were no differences in asthma related deaths, all cause deaths, and asthma related visits to accident and emerge ncy departments between exposure groups. In conclusion, inhaled LABA should not be prescribed as monotherapy to adults with asthma, and should be used as an add on to ICS as maintenance therapy. For patients with less severe asthma, discontinuing LABA appe ars to worsen disease control, and such step down therapy approach appears more beneficial in patients with more severe disease state
14 CHAPTER 1 BACKGROUND Definition of Asthma Asthma is a chronic pulmonary disorder that was recognized since ancient times. The symptoms of asthma and an apparatus used to relieve them were first described in cuneiform clay tablets found in ancient Mesopotamia modern day Iraq (Cserhti, 2004) The term asthma an open mouth (Cserhti, 2004). Asthma is a chronic inflammatory disorder of the airways that is characterized by chronic bronchial inflammation, hyperresponsiveness, and reversible bronchocon striction (GINA, 2009). These pathophysiological features cause the distinctive signs and symptoms of asthma: wheezing, shortness of breath, chest tightness, coughing, and tachypnea (BTS, 2009). These signs and symptom s are crucial in the diagnosis, treatm ent, and monitoring of the disease. With the progress in understanding disease pathophysiology the definition of asthma has changed over time Reversible airway obstruction and hyperresponsiveness were the predominant characteristics of asthma definition in the 1950s and 1960s (Guilbert & Krawiec, 2003). The concept of preventing bronchoconstriction was first introduced in the 1970s (Guilbert & Krawiec, 2003). In the 1980s, inflammation was recognized as a predominant disease process in asthma (Nadel, 1985 ; Guilbert & Krawiec, 2003 ). Acute and chronic inflammation occurs in patients w ith varying severity of asthma which causes bronchial infiltration with inflammatory mediators, e.g. eosinophils, mast cells, and interleukins; airway remodeling manifested by hyp erplasia and hypertrophy of bronchial smooth muscles; and mucus plugging of the bronchial lumen. The effectiveness of bronchodilators is minimized when extensive mucus
15 plugging occurs in severe asthma (Hopp & Townley, 2006). Medications with anti inflam matory effects, such as inhaled corticosteroids, oral corticosteroids, and leukotriene modifiers are increasingly becoming the mainstay of asthma therapy. Bronchial hyperresponsiveness in asthmatic patients occurs due to increased sensitivity of the airway s to irritants, such as pollens and tobacco smoke Reversible airway obstruction (bronchoconstriction) is a common corollary to airways inflammation and hyperresponsiveness. Natural History of Asthma The development and consequences of untreated asthma are described in light of measures of di sease assessment and monitoring that are recommended by the National Heart Lung and Blood Institute (NHLBI) guidelines for the diagnosis and management of asthma (NHLBI, 2007) The degree of bronchoconstriction is propo rtional to the duration and severity of asthma. Adults with asthma have significantly greater loss in pulmonary function than non asthmatics. The daytime and nighttime symptoms of shortness of breath, wheezing, and coughing induce functional impairment in terms of disturbance of sleep and missing school and work days. The progressive decline in lung functions raises the risks of unpleasant events in terms of exacerbations Therapeutic interventions improve disease symptoms and minimize functional impairment and the risks of untoward events. However, such interventions have no influence on the underlying pathogenesis of asthma. For instance, countering bronchopulmonary inflammation with inhaled corticosteroids ( ICS ) improve s symptoms and reduce s exacerbations but, symptoms are rebound ed after ICS withdrawal (NHLBI, 2007)
16 Asthma and Chronic Obstructive Pulmonary Disease Chronic obstructive pulmonary disease (COPD) is a respiratory disorder that includes chronic bronchitis and emphysema. Both asthma and COP D a ccount for the majority of the obs tructive respiratory disorders (Ryu & Scanlon, 2001) In the US, about 24 million adults have COPD (12 million physician confirmed and 12 million under diagnosed ) ( CDC, 2011 b ). Globally, COPD is projected to be the fifth l eading burden o f illness es in 2020 as measured by disability adjusted life years ( Lopez & Murray, 1998 ). Also, mortality rates from COPD are projected to increase over the next eight years ( Murray & Lopez, 1997 ). In early 1960s, sugg ested that both asthma and COPD should be considered as two different representations of the same disease ( Orie, Sluiter, DeVries, Trammeling, & Wiktop 1961). This is because both disorders share similar pathophysiological characteristi cs (e.g. chronic in flammation, airway obstruction, and airway remodeling) and genetic and environmental risk factors (e.g. atopic tendency, airway hyperresponsiveness and tobacco smoke) However, the current consensus in pulmonary medicine weighs against th is hypothesis an d both conditions are considered distinct disorders that have distinct clinical presentation, prognosis, and therapeutic management guidelines ( Bleecker, 2004; Barnes, 2006 ; GINA, 2009; GOLD, 2010 ). Although both asthma and COPD have underlying i n flammatio n in the airways, the inflammatory mechanisms are distinct (GOLD, 2010). T he inflammatory reaction in asthma involves CD4+ T lymphocytes and eosinophils ; while in COPD CD8+ T lymphocytes, macrophages and neutroph ils ( Rang, Dale, Ritter, & Flower 2007;
17 G OLD, 2010). This distinction is important regarding therapeutic effectiveness of anti inflammatory medications, e.g. ICS where ICS is more effective in mitigating asthma type (eosinophilic) inflammation than COPD type (neutrophilic) inflammation (Buist, 2 003). The consequential a irflow limitation is reversible in asthma, while it is not fully reversible in COPD (GOLD, 2010) a feature that shows episodic breathlessness in asthmatic patients and continuous breathlessness in COPD patients The pulmonary path ological expressions that attribute to the reversibility of airflow limitation are different in both disorders. In asthma, bronchoconstriction is mainly due to inflammatory cytokines that mediate airway smooth muscle contractions; whereas in COPD the bronc hoconstriction arises from structural remodeling of the airways and surrounding lung parenchyma ( Barnes, 2000; Ichinose, 2009 ) However, underlying inflammation and airflow limitation in asthma can be similar to that of COPD when asthmatics are chronically exposed to noxious substances, including cigarette smoke (GOLD, 2010). Equally, patients with COPD can present inflammatory profiles similar to that in asthmatic patients (Chanez et al 1997) The most important risk factor for COPD is smoking. It is est imated that 15% of smokers have COPD, which corresponds to about 16 million Americans (Ryu & Scanlon, 2001); cigarette smoking accounts for 75% of COPD cases in the US (CDC, 2011 b ). Occupational exposure accounts for 15% of the cases; and other factors, in cluding uncontrolled asthma contributes to additional 10% COPD cases (CDC, 2011 b ). Controlling asthma plays a role in COPD prevention, where u ncontrolled asthma advances to progressive airflow limitation and COPD like clinical presentation ( Lange,
18 Parner, Vestbo, Schnohr, & Jensen 1998 ; CDC, 2011 b ). A longitudinal study showed that asthmatic s had about 12 times higher risk to develop COPD than non asthmatic counterparts after contr olling for smoking status (HR 12 .5 95%CI 6.84 22.8) ( Silva, Sherrill, Guerr a, & Barbee 2004). Coexisting a sthma and COPD Due to inherent challenges in the differential diagnosis of asthma and COPD both disorders can coexist Since different therapeutic guidelines are available for both conditions patients with coexisting asth ma and COPD present additional challen ges to healthcare team s in terms of suiting appropriate disease management strategies Generally p atients with more than one obstructive respiratory disease e.g. asthma and COPD are older than 50 years of age (Guerra, 2005) and asthma c omorbidity with COPD is associated with increased risks for morbidity and mortality (Guerra, 2005). A longitudinal study showed that 4.3% of Americans have physician confirmed concurrent asthma and COPD (compared to 7.3% physician confi rmed asthma and 11.1% physician confirmed COPD ) ( Sherrill, Guerra, Bobadilla, & Barbee 2003) In the United Kingdom ( UK ) the annual prevalence of asthma defined by general practitioner ( GP ) cons ultations is about 9%; for COPD, it is about 11%. H owever, asthma accounts for 1.17% of respiratory disease related mortality, compared to 23.4% cases due to COPD. The figure s for coexisting asthma and COPD are not conclusive. Asthma accounts for about 10% of hospital admissions for respiratory diseases in the UK. COPD accounts for about 17% of hospital admissions. About 4% and 21% of all respiratory bed days of inpatient stay are due to asthma and COPD, respectively (BTS, 2006)
19 In the United States ( US ) about 24 million adults have COPD and about 24 million adul ts and children have asthma. Asthma and COPD respectively account fo r 1.6% and 53. 7 % of respiratory related mortality Thirteen percent of respiratory hospital attendances are attributed to a sthma, while 20 % of respiratory hospitalizations are due to COPD On average, length of hospital stays for asthma is 3.2 days, and for COPD is 4.4 days ( respectively corresponding to 60% and 83% of total length of stay for all respiratory admissions). In the US, 30% and 46% of all GP respiratory visits were for asthma a nd COPD, respectively (NHLBI, 200 9 ) Morbidity, mortality, and healthcare utilization figures for coexisting asthma and COPD are not well established. Since such patients are usually excluded from clinical trials, pharmaceutical outcomes research in terms of medication effectiveness and safety in this population is desired. Asthma as a r isk f actor for COPD Asthma is an unde restimated risk factor for COPD. In absence of comorbid conditions, asthma does not affect expectancy ; however, untreate d and undertreated asthma can progress to C OPD and thereby diminishes patient survival ( Lange, Parner, Vestbo, Schnohr, & Jensen 1998; Silva, Sherrill, Guerra, & Barbee 2004; CDC, 2011 a ) I t is recommended to consider COPD as a po ssible prognosis in asth matic patients with refractory response to treatment and deteriorated clinical presentation (Decramer & Selroos, 2005). Epidemiology of Asthma Despite the tangible progress in scientific and clinical research, the availability of novel pharmaceutical and b iological interventions, and the development in medication delivery device technology, asthma became a comm on medical problem in the world
20 with growing prevalence, severity, and consequent individual, societal, and economic burdens (Long, 2011; CDC, 2011 a ) Burden of Asthma According to 2011 estimates of the World Health Organization (WHO), 235 million individuals in the world endured asthma (WHO, 2011 ). In the US about 1 in 12 American s (about 25 million; 8% of the US population) suffered from asthma in 2 009, which corresponded to 3,447 deaths in 2007; about 12 million asthma attacks in 2008; and approximately $56 billion in medical expenses and lost work days (CDC, 2011 a ). In the UK about 1 in 12 Brits (about 5.4 million; 9% of the UK population) had ast hma in 2010, with 1,204 deaths in 2008; over 79,794 hospital admissions in 2008 09; about 1.1 million lost work days; and about $1.64 billion a year in medical costs. Approximately 864 thousand asthmatics received asthma medications in 2010 in Northern Ire land, Scotland, and Wales (Asthma UK, 2010). Asthma accounted for about 38% of allergic disorder general practice visits in the UK; and about 57% (37.8 million) of the during 2001 were for asthma medications. During the same year, asthma accounted for 87% of all allergic condition hospitalizations in the UK ( Gupta, Sheikh, Strachan, & Anderson 2004). The prevalence of asthma in both countries is steadily increasing by about 1% annually (CDC, 2011 a ; Asthma UK, 2010 ). In the UK, m edications used for indications in the respiratory system accounted for 7.4% of all prescriptions in 2004, corresponding to £971 million. Inhaled bronchodilators accounted for the majority of respiratory drugs (49%), corresponding to £254 million; inhaled corticosteroids accounted for (26%), which tallie d to £411 million. In 2004, 23.8
21 million GP consultations were for respiratory diseases, corresponding to about £501 million. About 13% of hospital admiss ions were for respiratory diseases ( including lung cancer ) which accounted for about £1496 million (BTS, 2006). Although asthma is an incurable disease, appropriate disease management attribute s to symptom control and acceptable quality of life. About 75% of asthma hospitalizations and 90% of asthma deaths in the UK are deemed preventable (Asthma UK, 2010 ). Risk Factors There are many characteristics associated with an increased likelihood of asthma occurrence which include environmental factors as well a s intrinsic factors that are inherent to the ind ividual Atopy is a form of allergy in which hypersensitivity reaction s could be distant from the region of contact with the allergenic substance. Atopic conditions may include eczema, allergic dermatitis, ha y fever, allergic rhinitis, allergic development of persistent asthma; some studies showed that childhood atopy is associated with the development of adulthood asthma (Guilbe rt & Krawiec, 2003) Similarly, encountering asthma symptoms early in life is associated with higher risk of severe asthma development later in life (Guilbert & Krawiec, 2003). However, recent genetic studies showed limited correspondence between the genes that control immunoglobulin E (IgE) production and the genes that control asthma predispos ition; such findings suggest that atopy could be a corollary of asthma rather than a triggering factor (Zhang, Moffatt & Cookson, 2012). Family history of asthma is a well recognized risk factor for having asthma. Individuals with moderate to high family history of asthma
22 are 2 4 times more likely to develop asthma than individuals with average familial risks ( Liu, Valdez, Yoon, Crocker, Moonsinghe, & Khoury 2009). Current research in disease genetics revealed the association of multiple genes with the development and progression of asthma (Zhang, Moffatt & Cookson, 2012). These findings will bring future challenges in understanding the functions of these genes in r elation to the pathogenesis and pharmacogen etics of asthma. risk of asthma development and worsening exacerbations in both children and adults (Visness et al 2010; Fitzpatr ick, Joks & Silverberg, 201 2 ). However, the exact biochemical and pathological mechanisms correlating both conditions remain not well understood, and research in this area is in demand (van Huisstede & Braunstahl, 2010; DHHS, 2011). Furthermore, studies i nvestigating asthma treatment outcomes in obese asthmatics are encouraged in the field of outcomes research. The incidence and severity of childhood asthma is more common in boys than in girls; however, the status reverses after puberty, where the conditio n becomes more common and severer in women than in men, particularly after the age of 20 40 years (Guilbert & Krawiec, 2003). These chronological differences in disease occurrence and severity between genders are attributed to p ulmonary physiological chang es and hormonal homeostasis (Guilbert & Krawiec, 2003). Exposure to cigarette smoking active or passive exposure is a well recognized risk factor for man y respiratory disorders, e.g. lung ca ncer, COPD and a sthma. Cigarette smoke plays a role in both the d evelopment of asthma and the deterioration of existing asthma which increases airway sensitization and hyperactivity (Guilbert &
23 Krawiec, 2003) Likewise m aternal exposure to cigarette smoke is associated with increased risk of asthma like symptoms in ne wborns, and the development of childhood asthma (Guilbert & Krawiec, 2003). In addition to exposure to cigarette smoke, indoor and outdoor air pollution with aeroallergen s, e.g. house dust mites, molds, cockroaches, noxious fumes, cold and humid temperatur es, and some household pet s are associated with increased likelihood of atopy and subsequent asthma (Guilbert & Krawiec, 2003) A study showed that asthmatic patients have high IgE levels in response to sensitization to specific allergens derived from spec ies of house dust mites ( Dermatophagoides farina, Dermatophagoides pteronyssinus, and Blomia tropicalis ); however, the sensitization had no e ffect on asthma symptoms or disease control by asthma medications ( Albano & Ramos, 2011 ) Recurrent childhood v ira l infections that affect the lower respiratory tract, e.g. respiratory syncytial viral infection s are more p robable risks for the development of childhood asthma than infections that affect the upper respiratory tract (Guilbert & Krawiec, 2003) However, u pper respiratory tract infections by human rhinoviruses are risk factors for worsening asthma symptoms in children (Miller et al 201 2 ) The role of cholecalciferol (vitamin D ) and its deficiency in the occurrence and progression of asthma was investigate d in observational studies and clinical trials. A recent cross sectional study showed that asthmatics with lower serum vitamin D levels are associated with increased airway smooth muscle mass (hypertrophy) which affects bronchodilator responsiveness (Gupt a et al 2011). The findings suggest a beneficial role for vitamin D in the progression of asthma ; however current consensus regarding the recommendation of vitamin D supplementation as a prophylaxis against or an
24 adjuvant therapy for asthma is postponed until the findings from ongoing clinical trials are published ( Paul, Brehm, Alcorn, Holguin, Aujla, & Celedn 201 2 ) Asthma Pharmacotherapy The goal of asthma therapy is to maximize asthma control by effective, safe and cost effective therapeutic interve ntions. Asthma control is achieved by eradication of daytime, nocturnal, and exercise indu ced symptoms; prevention of exacerbations; and attaining and maintaining normal lung function ( Shaw, Haldar & Pavord 2007). Pharmacologic approaches for the managem ent of asthma involve two categories: reliever medications (bronchodilators) and controller medications (anti inflammatory). Bronchodilators include beta 2 adrenoceptor agonists, muscarinic receptor antagonists, and methylxanthines. Anti inflammatory agent s include corticosteroids, leukotriene modifiers, mast cell stabilizers (cromolyn sodium and nedocromil), and monoclonal immunoglobulin E antibody (omalizumab). Some anti inflammatory agents exert bronchodilator effects and vice versa, rendering this class ification rather not mutually exclusive ( Rang, Dale, Ritter, & Flower 2007). Inhaled beta 2 adrenoceptor agonists and inhaled corticosteroids are considered the main bronchodilator and anti inflammatory agents, respectively (GINA, 2009). Based on the phar macokinetic properties of individual agents, inhaled beta 2 adrenocpetor agonist bronchodilators are subdivided into two categories: short acting beta agonists (SABA), e.g. salbutamol (albuterol) (effect lasts 3 5 hours); and long acting beta agonists (LAB A), which include formoterol and salmeterol (effect lasts 8 12 hours). Inhaled corticosteroids (ICS) are further subdivided by their relative potency into standard (low) strength, and m edium high strength.
25 From therapeutic perspective, asthma medications c an be classified as quick reliever medications (rescue therapy ) and long term controller medications ( preventer therapy ) Quick relievers are used to relieve the symptoms of acute asthma and invo lve short acting beta agonists and muscarinic receptor antag onists. Long term controllers are used to control chronic asthma and involve inhaled corticosteroids, long acting beta agonists, leukotriene modifiers, methylxanthi nes, and mast cell stabilizers. Targeting inflammation is the mainstay of asthma pharmacolog ical treatment and findings from effectiveness and safety research showed no sign ificant differences between individual agents within controller classes (Jonas et al 2011) Yet, c ompared to current asthma medications, ICS are considered the most effecti ve therapeutic modality ( Suissa, Ernst & Kezouh, 2002; Jonas et al 2011; Nair, 2011). However, some patients have refractory response s to standard and increased doses of ICS, and novel pharmacological inter ventions are under development which can be use d in such patients (Colice, 2011). Asthma Management Guidelines Current therapeutic guidelines proposed by the Global Initiative for Asthma (GINA) (GINA, 2009) the British Thoracic Society (BTS) (BTS, 2009) and the US National Heart, Lung, and Blood Inst itute (NHLBI) (NHLBI, 2007) for the management of asthma in adults and children comprise of defined therapeutic steps. The GINA and BTS guidelines consist of five steps, while the US guideline encompasses six steps, where steps 3 and 4 are merged into the third step in the earlier guidelines ( T able 1 1 ). These guidelines recommend individualization of asthma therapy based on patient profile, and are not intended as a substitute for clinical judgment at the practice level, where clinical
26 information in a par ticular case might require deviation from these guidelines. According to the BTS guidelines (BTS, 2009), mild intermittent asthma is treated with an inhaled SABA as a reliever therapy alone (step I). If no symptom control is achieved, the addition of an IC S as a preventer therapy is required (step II). If symptoms are still not controlled, the addition of an inhaled LABA to the earlier step is suggested (step III). If the disease still uncontrolled, either the potency of the ICS is increased with continuous LABA, or the potency of the ICS is increased with LABA discontinuation (step IV). The addition of another asthma medication (e.g. a leukotriene modifier or a methylxanthine ) is considered in step IV as well. Likewise, other anti inflammatory alternatives might be considered in steps II IV. In severer cases, a supplement of the lowest effective dose of an oral corticosteroid is added to the previous step f or continuous control (step V). Other s exposu re to systemic corticosteroids. Although most patients achieve asthma control in this stepwise approach, many are still uncontrolled (Holgate, Price, & Valovirta, 2006; Long, 2011 ). The anti IgE omalizumab is considered as an add on therapy in pati ents with severe persistent allergic asthma who are exposed to high strength ICS and inhaled LABA (Step IV) (BTS, 2009; EMC, 2011). Patients move up and down between the steps depending upon the severity of their disease state (GINA, 2009). Variable patien t response could occur between steps, and alternative treatments might be selected within each step before stepping up therapy. Prescribers should consider the following factors prior to stepping up treatment: patient education including a personal asthma action plan; medication
27 adherence; inhaler technique; environmental control and triggers; comorbidities; asthma sy mptoms; functional limitation s, including lost work/school days, limitations of daily activities, sleep disturbances, and poor quality of life ; spirometry and peak expiratory flow rates; and ut ilization rates of inhaled rescue medications. Patients are stepped down in their treatment protocol when asthma is well controlled for at least three months (NHLBI, 2007; GINA, 2009). The risks for exacer bations, and asthma related hospitalizations and accident and emergency ( A & E ) department visits in the coming year can be assessed by evaluating these risks in the past year (NHLBI, 2007). Inhaled Long Acting Beta 2 Adrenoceptor Agonists Beta 2 adrenocepto r agonist bronchodilators play a fundamental role in the acute and maintenance management of asthma. SABA inhalers are used as rescue bronchodilators to relieve intermittent episodes of bronchospasm and breathlessness. LABA inhalers on the other hand, are used as maintenance therap y in addition to ICS ( Cazzola, Calzetta & Matera, 2011 ) The introduction of LABA inhalers considered a major advance in bronchodilator therapy. Currently, there are two LABA agents approved for use in the US and the UK, formoter ol and salmeterol. Both agents are available as single ingredient inhalers and as singl e device ICS combined inhalers. Table 1 2 lists LABA products that are approved for marketing by the US Food and Drug Administration (FDA), and the UK Medicines Healthca re Products Regulatory Agency (MHRA) (EMC, 2011; FDA, 2011 a ). Figures 1 1 and 1 2 respectively show the chemical structure of the approved L ABA and ICS products. In addition, new LABA products are development and others are under development which as ultra LABA agents with longer half lives including carmoterol,
28 indacaterol, LAS100977, olodaterol, PF 610355, and vilanterol (Cazzola, Calzetta & Matera, 2011) Among these agents, indacaterol (Onbrez Breezhaler Novartis ) was approved for marketing in the UK on November 30, 2009 as a treatment for COPD, but not for asthma ( EMC, 2011 ) is currently under review by the FDA (Novartis Briefing Document, 201 1) On the other hand, vilanterol in combination with fluticasone (Relova ir, GlaxoSmithKline) as a treatment for COPD and asthma is in the submission process in both the European Union ( EU ) and the US ( Hill, 2012 ) Similarly, Boehringer Ingelheim, the developer of olodaterol is currently intending to se ek FDA approval as a bron chodilator treatment for COPD (Garde, 2012). As the guidelines show, LABA inhalers are not indicated for the relief of acute exacerbations, rather they should be reserved as an adjunct therapy with ICS for the prevention of nocturnal and exercise induced s ymptoms of chronic and persistent asthma. Inhaled LABA as a treatment modality was introduced in asthma management guidelines because studies showed that patients who are not well controlled on ICS monotherapy had a better response when LABA inhalers were added (Bateman et al 2004; Greenstone et al 2005; Masoli, Weatherall, Holt, & Beasley 2005 ; Barnes, 2007 ). Pharmaco logy of LABA Products The h uman beta adren ergic receptors (adrenoceptors) are comprised of 413 amino acids and physiologically divided i 1 2 3 (Johnson, 1998). 1 receptors are found in 2 receptors are dispersed in the re spiratory tract smooth muscles; and 3 receptors are found in the adipose tissue ( Johnson, 1998 ). 1 receptors increase s cardiac output and oxygen consumption, which is
29 antagonized by beta blockers and used as antihypertensive and antiarrhythmic medications ; 2 agonists are use d as bronchodilators in obstructive pulmonary conditions e.g. asthma ; and 3 agonists increase l ipolysis and used as anti obesity agents (de Souza & Burkey, 2001) Inhaled LABA agents are administered to the bronchial epithelium and smooth muscle s through oral inhalation. These agents 2 receptor affinity than inhaled SABA br onchodilators which translates in to higher binding capacity to the receptor (Johnson, 1998) Inhaled SABA molecules are hydrophilic, which exert bronchodilation at the extracellular level with a rapid onset (5 15 minutes) and short duration of action (3 6 hours) On the other hand, inhaled LABA molecul es are lipophilic, which diffuse into the cell with relatively slower onset (1 5 20 minutes) and longer duration of action (8 12 hours) The duration of airway dilation for 2 adrenoceptor agonists is in the following order: formoterol > salmeterol > albuterol > terbutaline (Johnson, 1998 ) Undoubtedly, the bioavailability and subsequent effective ness of inhaled bronchodilators and anti inflammatory agents ar e contingent upon the type of inhalation device used in drug delivery (Virchow et al 2008) and proper inhaler technique ( Ovchinikova, Smith & Bosnic Anticevich, 2011 ). When locally administered into the bronchi via oral inhalation formoterol is deposit ed on bronchial epithelium and stored in the plasma membrane of bronchial smooth muscle cells, and slowly released to the extracellular compartment t o interact with 2 receptors to exert bronchodilation ; the plasma membrane storage size is proportional to the dru g dose that is deposited on bronchi al epithelium ; therefore, the
30 duration of bronchodilation for formote rol is dose dependent (Johnson, 1998) Salmeterol on t he other hand, has a side chain (Figure 1 1 ) that interacts with the 2 receptor within the plasma membrane of bronchial smooth muscle cells. This anchoring mechanism prevents salmeterol molecule from detachment from the receptor wh ile the active side of the molecule remains freely interacting with the active receptor site located on the outer layer of the cell membrane which continually attaches and detaches from the active site with a prolonged duration of action (Johnson, 1998 ; O ) LABA Safety Controversy In March 2, 2006, the FDA issued a Black Box Warning regarding all LABA products warning patients and healthcare professionals about an increased risk of asthma related death in patients exposed to LABA products (FD A, 2010 a ). In February a Medication Guide to all patients who are using any LABA product (FDA, 2010 b ). The trategy (REMS) and class labeling changes to all LABA products is largely based on the results of two clinical studies: The Serevent Nationwide Surveillance Study (SNS) ( Castle, Fuller, Hall, & Palmer 1993) and The Salmeterol Multicenter Asthma Research T rial (SMART) ( Nelson et al 2006). The 2007 NHLBI asthma management guidelines recommend and in tandem the addition of LABA to a low dose ICS rather than increasing the dose of ICS at step III of the stepwise therape utic approach (NHLBI, 2007). A lthough therapeutic guidelines and the FDA recommend ICS/LABA combination
31 therapy, there is no evidence that ICS protect the patient against LABA induced worsened asthma outcomes. Consequently, the FDA in the second quarter of 2011 required the manufacturers of LABA products to conduct five long term, large scale randomized, double blind, controlled clinical trials to further investigate the safety of ICS/LABA combination therapy in comparison with ICS monotherapy, with expecte d fi ndings to be published in 2017 (FDA, 2011 b ). Additionally, LABA products became the priority of drug safety research by the European Medicines Agency (EMA) (EMA, 2010). The controversy about the safety of regular use of inhaled LABA in asthmatic patien ts is significant and the answer for it is eagerly needed. LABA s afety in e xperimental s tudies c ontext Drug e fficacy is the ability of a drug to produce the expected therapeutic effect under standardized or experimental conditions (ideal effect) e.g. clin ical trials. The LABA safety issue is closely related to its efficacy ( or effectiveness in observational context) and can be examined as the reciprocal of its efficacy or effectiveness. The SNS is a randomized, controlled clinical trial (RCT) of 25,180 as thmatic patients who received either salmeterol (50 mcg twice daily) or albuterol (200 mcg four times daily) and followed for 16 weeks. The study was conducted in the UK by salmeterol manufacturer in the request of the MHRA to assess the safety of LABA in asthmatic patients. The majority of the patients were adults (about 6% were under 18 years old). The study showed a statistically non significant increase in the risk of asthma related mortality in salmeterol group (RR=3.00; p value=0.105) with no associat ion with asthma related hospitalization ( Castle, Fuller, Hall, & Palmer 1993). T he increased mortality was attributed to baseline disease severity, rather than salmeterol ( Castle,
32 Fuller, Hall, & Palmer ) When the SNS study was inco nclusive, the FDA requested salmeterol manufacturer to conduct another study to further assess the s afety matter. The SMART was incepted in 6,163 sites in the US as a RCT which enrolled 26,355 asthmatic patients who were randomly assigned to inhaled salmet erol (42 mcg twice daily) or identical placebo and followed for 28 weeks. About 12% of the patients were children (12 18 years old). The study was early terminated by the FDA for increased life threatening events, including asthma related mortality rates i n salmeterol group (RR=4.37; 95%CI=1.24 15.3) (Nelson et al 2006 ). Although both studies aimed at assessing the safety of LABA monotherapy, the participants in both studies were on ICS at baseline (69% in SNS; 47% in SMART). We cannot conclude from eithe r study if the addition of ICS to salmeterol has a positive or negative effect on asthma outcomes. Nevertheless, the FDA included in the Black Box Warning a recommendation to use LABA in combination with ICS and to discontinue LABA upon achieving asthma co ntrol (Product Label: www.serevent.com ); however, discontinuing LABA after accomplishing control with ICS/LABA combination therapy could result in a failure in asthma control and rebound symptom manifestations (Sears, 2011). Despite the inconclusive result s of the SNS and the SMART studies, the FDA, at that time might be concerned with the over utilization of LABA products in the population, and interested in reducing exposure to LABA. The utilization of inhaled LABA products is not consistent with the prop osed guidelines. Although the guidelines recommend using ICS as the preferred therapy, most of the patients are started on LABA. In the US, about 62% of asthmatic patients who received combined ICS/LABA LABA products ( Ye, Gutierrez,
33 Zarotsky, Nelson, & Blanchette 2009). Similarly, combination therapy is used more than indicated by the guidelines in other countries (Bisgaard & Szefler, 2006). After the SMART study, the FDA and individual investigators pr esented meta analyses of large RCTs to evaluate the safety of LABA inhalers with and without concomitant ICS in asthmatic patients. The findings revealed an increased risk for asthma related mortality and morbidity in LABA groups regardless of concomitant use of ICS ( Salpeter, S, Buckley, Ormiston, & Salpeter, E. 2006; Levenson, 2008; Salpeter, Wall & Buckley 2010). However two additional meta analyses contended against e risk of asthma related mortality and morbidity, whether alone or in combination with ICS (Jaeschke et al 2008; Nelson et al 2010). A meta analysis of Novartis sponsored clinical trials in 2,452 asthmatic adults (>18 years) compared formoterol with pl acebo showed no significan t increase in the risk of serious asthma exacerbations in formoterol users (OR=1.3; 95%CI=0.4 3.7); however, about 71% of the patients were on concomitant ICS use at baseline ( Kemp, Armstrong, Wan, Alagappan, Ohlssen, & Pascoe 20 11 ). Another meta analysis of randomized trials showed that ICS/LABA combination therapy is significantly protective against asthma exacerbations when compared with ICS monotherapy that has similar dose of the ICS in the combination group (RR=0.80; 95%CI=0 .73 0.89) However, the statistical significance disappear ed when ICS dose in the monotherapy group was higher than the counterpart in the combination group (RR=0.88; 95%CI=0.76 1.01) (Gibson, Powell & Ducharme, 2007) suggesting a plateau phenomenon in d ose response relationship between ICS and asthma outcomes, where benefits from dose increment reaches a plateau on the
34 expense of an increase in adverse reactions (Figure 1 3 ) In addition, one meta analysis showed statistically not significan t differences between fluticasone/salmeterol and budesonide/formoterol single device combination inhalers in terms of serious asthma related morbidity (Lasserson, Ferrara & Casali, 2011) These studies are limited in their generalizability, uncertainty of the pooled e stimates when the nu mber of events is small, and the application of weighted fixed effects models, which overlooked the fact that individual trials are heterogeneous and the assigned weights are random entities ( Shuster, Jones, & Salmon, 2007 ; Shuster, 201 0 ). LABA s afety in o bservational s tudies c ontext Drug effectiveness is the ability of a drug to demonstrate the therapeutic effect under real conditions of prescription and use (pragmatic effect). Randomized clinical trials are considered the gold standard for the assessment of treatment effect s Randomization ensures equal distribution of measured and unmeasured confounders across exposure groups, and when masking and placebo control are properly applied, bias from selection and measurement errors can be e liminated. However, these designs aim at testing drug efficacy in experimental conditions that do not reflect the variable conditions of clinical practice prescribing behavior, and patient adherence T herefore, findings from such studies are less generali zable and detection of rare and serious adverse outcomes is limited. On the other hand, findings from observational designs are more generalizable to clinical practice and target patient population, albeit lack of randomization predisposes these studies t o different types of biases ( Laupacis & Mamdani, 2004 ) However, properly designed observational studies complement the
35 findings from randomized trials and many such studies have been conducted to evaluate the effectiveness and safety of inhaled LABA prod ucts in large population of asthmatic patients Observational studies in the form of retrospective database analyses are rep orted with conflicting findings regarding LABA safety An e arlier study did not show an increased risk among asthmatic patients expo sed to salmeterol in comparison with theophylline ( OR=0.90; 95%CI=0.13 5.0 ) and ipratropium ( OR=0.12; 95%CI=0.02 0.71 ) (Meier & Jick, 1997 ) ; however, another study showed an association between salmeterol and asthma related morbidity (emergency department and hospital attendance), albeit statistically not significant compared to theophylline ( Lanes Lanza & Wentworth, 1998). In a case control study, salmeterol w as significantly associated with near fatal asthma events, the analysis however was not adjusted for asthma severity (OR=2.32; 95%CI=1.05 5.16) (Williams et al 1998). Nevertheless, the significance was removed after subgroup analyses are conducted within patient groups who were prescribed oral corticosteroids at the near fatal event date (OR=2.29; 95%CI=0.39 13.53) and those who were hospitalized for asthma during the 12 months prior to the near fatal event (OR=1.42; 95%CI=0.49 4.10) ( Williams et al 1998). A case control analysis of the GPRD data showed that chronic LABA users are found to be 3.2 times more likely to die from asthmatic attacks than nonusers (RR=3.2; 95%CI=0.7 14.1); chronic user s is defined by receiving more than eight LABA prescriptions during the 12 months prior to the occurrence of asthma related death ( Lanes, Garcia Rodr g uez & Huerta, 2002 ). Conversely, patients exposed to salmeterol within 12 months prior to experiencing asthma related death are found in another case
36 control study to be less likely to experience death than unexposed counterparts (OR=0.95; 95%CI=0.70 1.29) ( An derson et al 2005). Inhaled LABA monotherapy and ICS/LABA combination therapy were compared with inhaled SABA therapy in asthmatic patients. The study found a reduced risk of emergency department visits, and an increased risk of asthma related hospitaliz ations and intubations for LABA and ICS/LABA ( Guo, Tsai, Kelton, Bian, & Wigle 2011). Similar to a meta analysis (Lasserson, Ferrara & Casali, 2011), a cohort study showed that budeson ide/formoterol is less likely to be associated with asthma related mor bidity than fluticasone/salmeterol yet statistically not significant ( Blais, Beauchesne & Forget, 2009 ). In another study, a sthmatic patients in the GPRD who were initiators of ICS are followed until the addition of LABA or continuation of ICS ( Thomas, v on Ziegenweidt, Lee, & Price 2009). The findings were conflicting in terms of asthma morbidity outcomes. Compared to ICS users, p atients exposed to ICS/LABA are found to have lower hazards of rescue SABA utilization; yet, have higher hazards of oral corti costeroid use and asthma related hospitalizations ( Thomas, von Ziegenweidt, Lee, & Price 2009 ). An other GPRD study compared LABA monotherapy users with inhaled SABA users, and found the earlier group is more likely to die of asthmatic attack than the latt er (RR=2.5; 95%CI=1.6 3.8); likewise, for hospitalization rates (RR=4.9; 95%CI=3.1 7.8), and GP visits for exacerbations (RR=3.2; 95%CI=3.1 3.4) ( de Vries, Setakis, Zhang, & van Staa 2010). Similar analyses conducted to compare ICS/LABA combination therap y with inhaled SABA, which showed an increased risk of asthma related death (RR=2.7; 95%CI=1.9 3.9) and asthma related hospitalization rates (RR=3.4;
37 95%CI=2.3 5.0) with high dose ICS/LABA combination, and reduced risk of asthma death (RR=0.9; 95%CI=0.4 2. 1) with standard dose ICS/LABA combination ( de Vries, Setakis, Zhang, & van Staa 2010). Another study adjusted for ICS dose showed ICS/LABA combination therapy more beneficial in terms of reduced asthma exacerbations than ICS monotherapy (HR=0.65; 95%CI=0 .47 0.90) ( Wells, Peterson, Ahmedani, Severson, Gleason Comstock, & Williams 2012 ). These findings reflect the presence of residual confounding by time dependent confounders, especially confounding by disease severity. Despite identifying relevant surroga te variables for asthma severity, these studies failed to account for the time dependent nature of these variables, which predict study outcome (asth ma mortality and morbidity), determine future exposure (bronchodila tor and anti inflammatory type), an d aff ected by previous exposure. Heterogeneity of Response to LABA The findings from experimental and observational research suggest a bimodal response to LABA inhalers in asthmatic patients, a beneficial response by patients who are not well controlled on ICS alone; and a deleterious response by some patients regardless of concomitant use of ICS. Although the association of LABA with serious asthma outcomes seems paradoxical to pharmacodynamic characteristics of such agents that are full agonist at the beta 2 r eceptors, the following hypothetical mechanisms are proposed: LABA bronchodilators have a steroid sparing effect, which masks the underlying inflammatory process of asthma ( McIvor, Pizzichini, Turner, Hussack, Hargreave, & Sears 1998); LABA bronchodilator s improve asthma symptoms, which give mistaken sense of disease control and potential cease of anti
38 chronic LABA exposure reduce s the number of cell surface receptors due to receptor cell internalization (receptor desensitization) and shrinks the net number of cellular receptors (receptor down regulation) to prevent receptor overstimulation with beta agonists ( Simons, Gerstner & Cheang 1997; Tan, Grove, McLean, Gnosspelius, Hall, & Lipworth 1997 ; Johnson, 1998 ); increased bronchial hyperactivity in response to regular beta agonists exposure ( ); and genotypic variation in beta 2 receptors ( Palmer, Lipworth, Lee, Ismail, Macgregor, & Mukhopadhyay 2006 ) The response of LABA monotherapy is found to be variable across different beta 2 receptor polymorphisms (Bleecker et al 2010); however, pharmacogenetic studies showed that the addition of an ICS to a LABA is associated with improved asthma outcomes reg ardless of genotypic differences in the beta 2 receptors (Wechsler et al 2009; Bleecker et al 2006 ) On the other hand, individuals identified as non W hite are more at risk of exacerbations than W hites (NHLBI, 2007), and worsened asthma outcomes are ob served in African Americans exposed to inhaled LABA (Nelson et al 2006) who often have Arginine/Arginine genotype polymorphism for the beta 2 receptor gene and suboptimal asthma control on ICS monotherapy even though maximal doses are used (Glassroth, 2 006)
39 Table 1 1. Comparison of the stepwise approach for the treatment of asthma in adults (>12 years old) across different asthma management gui delines Asthma type Step Asthma management guidelines GINA, 2009 BTS, 2009 NHLBI, 2007 Intermitt ent I SABA SABA SABA Persistent asthma II SABA + L ICS SABA + LTM SABA + L ICS SABA + L ICS SABA + LTM SABA + MCS SABA + Xanthine III SABA + L ICS + LABA SABA + M ICS SABA + H ICS SABA + L ICS + LTM SABA + L ICS + Xanthine SABA + L ICS + LAB A SABA + M ICS + LABA SABA + M ICS SABA + M ICS + LTM SABA + M ICS + Xanthines SABA + L ICS + LABA SABA + M ICS SABA + L ICS + LTM SABA + L ICS + Xanthine IV SABA + M ICS + LABA SABA + H ICS + LABA SABA + M ICS + LTM SABA + H ICS + LTM SABA + M ICS + Xanthine SABA + H ICS + Xanthine SABA + H ICS + LABA SABA + H ICS + LABA + LTM SABA + H ICS SABA + H ICS + LABA + Xanthine SABA + H ICS + LABA SABA + H ICS + LABA + Anti IgE SABA + M ICS + LABA SABA + M ICS + LTM SABA + M ICS + Xanthine V SABA + M ICS + LABA + Oral CS § SABA + H ICS + LABA + Oral CS § SABA + M ICS + LABA + Anti IgE SABA + H ICS + LABA + Anti IgE SABA + H ICS + LABA + Oral CS § SABA + H ICS + LABA + Anti IgE VI n/a n/a SABA + H ICS + LABA + Anti IgE + Oral CS § GINA, Global initiative for asthma B TS, British thoracic society NHLBI, National he art, lung, and blood institute
40 SABA; Short acting beta agonist ICS, Inhaled corticosteroids LABA, Long acting beta agonist LTM, Leukotriene modifiers MCS, Mast cell stabilizers Anti IgE Anti immunog lobulin E (e.g. omalizumab) CS, Corticosteroids L, Low strength M, Medium strength H, High strength n/a Not applicable Step up treatment after considering patient education, including personal action plan; adherence; inhaler technique; env ironmental control and triggers; comorbidities; symptoms; spirometry and peak expiratory flow rates; and utilization rate of SABA rescue drugs. Step down treatment after checking if asthma is well controlled during the past 3 months. Consider patient refer ral to asthma specialist when reaching step 3 § Lowest effective dose of oral corticosteroids that produce control. Other treatments should be considered at the last steps t o minimize the exposure to systemic corticosteroids.
41 Table 1 2. Availability of lo ng acting beta agonist products in the US and the UK as of June 2011 Product Ingredient Marketing authorization date FDA MHRA Foradil Formoterol February 16, 2001 September 12, 1997 Serevent Salmeterol February 4, 1994 October 14, 1996 Symbicort Budesonide/Formoterol July 21, 2006 May 15, 2001 Dulera Mometasone/Formoterol June 22, 2010 NA Advair HFA Fluticasone/Salmeterol June 8, 2006 NA Advair Diskus Fluticasone/Salmeterol August 24, 2000 NA Seretide Fluticasone/Salmeterol NA February 1, 199 9 Product brand names are the property of their respective manufacturers Other g eneric products are not listed NA, Not avai lable in the respective country A ) B ) Figure 1 1. Chemical structure of inhale d beta agonist bronchodilators commonly prescribed for asthma in the UK A) Long acting. B) Short acting. (Source: http://www.drugbank.ca/drugs Last accessed July 23, 2012). Side Chain Formoterol Salmeterol Albuterol (Salbutamol) Terbutal ine
42 Figure 1 2 Chemical structure of inhaled corticosteroids commonly prescribed for asthma in the UK. (Source: http://www.drugbank.ca/drugs Last accessed July 23, 2012). Beclometasone Budesonide Ciclesonide Fluticasone Mometasone
43 Figure 1 3. Illustration of the relationship between inhaled corticosteroid dosage and benefit/risk balance 0 0 Dose Switching Time Risk Benefit Low Medium High ICS Dosage Categories 1 1
44 CHAPTER 2 PHARMACOEPIDEMIOLOGI CAL PERSPECTIVE Methodological Challenges in O bservational Research In observational pharmacoepidemiology, exposure to dru gs is not at random and probabilities of exposures are not the same across different populations; certain people with covariate structures have higher tendencies to be exposed to a specific treatment than others. The non experimental nature of observationa l designs presents challenges in accurately and reliably estimating the association between exposures and outcomes. Three sources of bias can distort this association in observational designs: confounding, selection bias, and measurement bias ( Rothman, Gre enland & Lash, 2008 ) Confounding In clinical practice, medications are not prescribed (and therefore, not utilized) at random ; p rescribing by physicians take into considerations multiple factors, s uch as patient, disease diagnostic and disease prognos ti c characteristics Such factors play the role of exposure determinants, which lead to the two most common types of confounding in pharmacoepidemiology: confounding by indication, and time dependent confounding ( Schneeweiss, 2008 ) Furthermore, failure to a ccount for all confounding characteristics in an observational study leads to residual confounding, which is the third common type of confounding in observational pharmacoepidemiology ( Glymour & Greenland, 2008 ) Confounding is Latin in origin ( confudere ), to mix together Epidemiolo gical confounding is defined as the distortion of exposure outcome relationship by some other variable, i.e. confounder (Schneeweiss, 2008) The distortion can lead to overestimation, underestimation, or reversal o f exposure effect
45 (Schneeweiss, 2008) A confounder is associated with the exposure and the outcome accounts for some or all of the observed exposure effect and not in the exposure out come causal pathway (Figure 2 1, A ). Here, association means that t he risk of the outcome is different among people with the confounder compared to those without and the distribution of the confounder is different among people with exposure compared to those without (or those with other comparator exposure category), i.e. the distribution of the confounder is unbalanced between exposure groups Confounding by indication Confounding by indication is defined as a distortion in the association between the drug and the outcome, which occurs when a drug (or a class of drugs) is preferentially prescribed to patients with specific baseline characteristics, such as those with worse disease state at baseline (confounding by disease severity), or those with preexisting comorbidity at baseline (channeling bias) (Bgaud, 2000). In case of confounding by disease severit y the observed exposure effect could be mixed with an effect of severer disease state ( Hudson & Suissa, 201 0 ) Both types of confounding by indication are common in chronic disease pharmacoepidemiology ( Psaty & Siscovick, 2010 ) ; however, confounding by disease severity may be the most important type in respiratory disease studies, including asthma; where inhale d medications are preferentially prescribed to patients with variable baseline disease severity states.
46 Time dependent confounding T ime dependent confounding is defined as an alteration in the association between the exposure and t he outcome as a result of a variable that may vary over time and concurrently act s as a confounde r and an intermediary (Figure 2 1, B) (Kattan & Cowen, 2009). In pharmacoepidemiology, drug effects are time dep endent, and are affected by time dependent confounders that themselves are affected by previous drug exposure, and affect subsequent drug e xposure and outcome. T his phenomenon is common in usual care real world settings including pulmonary medicine and conclusions drawn from studies that fail to account for time dependent confounding could be misleading. Residual confounding Databases ma y contain an array of variables; however, even the most complete and detailed database fails to include all potentiall y important confounding variables. The presence of confounding despite adjustment is referred to as residual confounding. F ailure to account for all confounding variables can rise from unmeasured variables (Glymour & Greenland, 2008) or measurement error s in the variables (Stram, Huberman & Wu, 2002) Further, g rouping a confounder that is on the continuous scale of measurement, e.g. age into few categories may result in inadequate confounding control in the observed exposure outcome relations hip ( Brenner & Blettner, 1997 ) Unmeasured confounders are either measurable but unmeasured in the main study or immeasurable ( unknown or difficult to measure) (Schneeweiss, 2008) Examples of variables that are frequently unavailable (unmeasured) in databases that ar e used for pharmacoepidemiologic research include : behavioral information
47 (smoking, alcohol drinking, nutrition and eating habits, exercise, substance abuse, and therapy adherence measures); laboratory information ( blood tests, lung function tests, and oth er functionality tests ) ; and exposure information (over the counter products, nutraceuticals and herbal remedies, and immunization history) In randomized, controlled clinical trials, randomization equally (or near equally) balances confounders across expo sure groups. Pharmacoepidemiologists aim at evaluating the causality of associations between dr ugs and outcomes; l ack of randomization in observational studies casts additional challenges in evaluating causal associations. M oreover, depending on the therap eutic effectiveness of a medication in an individual patient, subsequent medication use may change ; pati ents who tolerate the unintended side effects or experience the intended therapeutic effects of a medication are mo re likely to continue using it in the future compared to those who experience side effects or those who do not perceive the beneficial effects of a medication Confounding is one of the most important sources o f bias in observational studies, and any observed association in such studies is co nfounded to some degree ( Jepsen, Johnson, Gillman, & Srensen 2004) However, well designed and well conducted rigorous observational studies have increasingly important clinica l regulatory, and public health impacts. Effect Modification Observing exposu re effect in a subset of patients defined by particular characteristics is effect modification In another word, the average causal effect of the exposure on the outcome is different across levels of the effect modifying factor. Unlike confounding, effect modification is not a source of bias; rather it may be of public health
48 importance e.g. identifying exposure groups at risk Both confounding and effect modification can exist together ( incorrectly estimate s average causal effect co nfounded, ), or separately ( correctly estimate average causal effect not confounded, ; confounded, ) Stratified analysis is used to elucidate the association between exposure and outcome across levels of the factor. E.g. the association between obesity and asthma prevalence is different between men and women ( Loerbroks, Apfelbacher, Amelang, & Strmer 2008). In some epidemiology references the effect measure modification to signify th e point that identifying the presence of effect modification is contingent upon the type of effect measure used additive vs. multiplicative ( Greenland Rothman & Lash 2008 ) An e ffect modifier may change the direction or the magnitude of exposure effect W hen the direction of exposure effect is the same in all subsets of the modifier but the magnitude is strengthened or weakened between subsets, quantitative effect modification occur s When the aver age causal effects are in opposite directions in subsets of the effect modifier (i.e. the exposure and the outcome are associated in the presence of the effect modifier but not associated in the absence of the effect modifier, and vice versa ), qualitative effect modification occurs. In case of qualitative effect modification, both additive and multiplicative effect modifications are present (Thompson, 1991) I n the absence of qualitative effect modification effect modification may be present in one scale o f measurement than the other ( Greenland, Rothman & Lash, 2008 ) Therefore, the concept of effect modification in pharmacoepidemiology should always be examined in light of the chosen scale of effect measure, e.g. risk ratio (multiplicative), risk differen ce (additive).
49 Additive effect measure modification: Multiplicative effect measure modification: Selection Bias Selection bias is a distortion in the association between exposure a nd outcome that arises fro m systematic errors in the way study sample was selected This type of bias can arise in both randomized and observational studies. In randomized studies, randomization prevents confounding and selection bias when patient selection takes place after random ization to exposure arms ; intention to treat (ITT) and last observation carried forward (LOCF) analyses are valid in this situation However selection bias may be imposed when patient selection occurs after assigning exposures ( post randomization process ) In cohort studies, t he outcome of interest can be over or under represented in the preferentially selected sample of patients who have higher or lower likelihood of presenting the outcome ( Bgaud, 2000 ) ; t herefore, the sample will be unrepresentative of the target population to which study results are extrapolated (Figure 2 2 ) There are many forms of selection bias, including: info rmative censoring ; missing data bias ; surviv or bias (depletion of susceptibles) ; admission (referral) bias; diagnostic bias; incidence prevalence bias; volunteer (self selection) bias; and healthy user
50 effect the latter is considered a type of confounding more than of selection bias ( Hernn, Hernndez Daz & Robins, 2004 ) In retrospective database analyses informative censor ing and depletion of susceptibles are probably the most important types of selection bias. Informative censoring In the context of survival analysis, censoring refers to the termination of observations when they are not followed long enough to observe the outcome of interest ( Bgaud, 2000 ) ; the conditions for such termination are usually defined by study investigator When censoring occurs due to reasons that are not under the control of the investigator random censoring is said to happen ( Allison, 2010 ) Within this framework, informative censoring is likely to occur ( Allison, 2010 ) (Figure 2 3 ) If the censored individuals are biased subsample of the uncensored individuals who have similar covariate distribution (i.e. censored individuals have systematica lly higher or lower risks of observing the outcome than the uncensored counterparts) the censoring mechanism could be due to the exposure or the outcome i.e. informative Informative censoring can lead to biased estimates of the association between the e xposure and the outcome, which is likely in longitudinal studies with time dependent confounding (van der Laan & Robins, 2003). Selection bias due to random censoring can be prevented by including the variables that affect censoring and event times in the multivariate regression model, e.g. study starting time (index date) in case of random study entry and termination times (Allison, 2010) Calculating the inverse probability of censoring weighted (IPCW) estimator as part of the marginal structural models t echnique can account for
51 informative censoring ; however the procedure is inefficient when censoring is due to competing outcomes (van Wonderen et al 2009) outcome of interest to remove susceptible patients from the risk set pool ( Rothman & Greenland, 2008 ) If the rate of competing outcome is higher in the exposed, the HR will be overestimated. Sensitivity analyses on the other hand can be employed to see how sensitive the estimates are to informative censoring (Allison, 201 0). For example, the investigator may assume that censored individuals are at high risk of observing the outcome (outcome occur s immediately after censoring), or the reverse (censored individuals have longer time to event than any other individual in the s ample). When association estimates from sensitivity analyses bracket the estimate s from original analysis (i.e. act as a confidence limits) conclusions are not affected by treating censoring reason (e.g. other death, COPD) as non informative. In the GPRD, patient transfer out of the general practice or disenrollment f rom the healthcare insurance plan are unlikely related to the exposure or the outcome; t his is because healthcare is universal in the UK and such transfer or disenrollment could be due to pati ent m igration outside the country as an example However, drugs administered in hospitals are not recorded in the unlinked GPRD, and if admission to hospital and death are among the censoring conditions set by the investigator, selection bias can occur if the outcome of interest is time to hospitalization This is because censored patients are usually sicker than uncensored patients. The risk of asthma hospitalization is associated with the risk of asthma death ( Omachi, 2009), and in a study assessing the e ffect of ICS on asthma hospit alization in comparison to LABA, if ICS is effective in delaying or preventing asthma mortality, ICS will be associated with
52 increased risk for asthma hospitalization compared to LABA when asthma death is used as a censorship c riterion. This is bec ause ICS patients will survive longer than the LABA patients and be h ospitalized. Depletion of susceptibles There are many terms for this type of selection bias including survivor bias (Glymour & Greenland, 2008), ( Hernn, 2010), and ( Arrighi & Hertz Picciotto, 1994 ). Depletion of susceptibles is the most commonly used term in pharmacoepidemiology. Depletion of susceptibles is defined as the gradual exclusion of a subgroup of patients wh o are susceptible to the outcome of interest from one exposure group, which leads to subsequent selection of another subgroup of patients who are less suscept ible to the outcome from another exposure group (provided that exposure groups are comparable with regard to other factors other than exposure categories) This phenomenon is common in studies using prevalent users instead of incident users ( Ray, 2003; Szklo & Nieto, 2004; Danaei, Tavakkoli & Hernn, 2012 ). Figure 2 4 illustrates the depletion of susc eptibles concept in a retrospective cohort design assessing the association of a drug with an adverse drug reaction. At earlier time of the study (Period 1), patients at risk of experiencing the outcome are expected to develop the outcome and thereby exclu ded from the follow up. This will eventually leave a subgroup of patients who have low risk of experiencing the outcome at later time (Periods 2 3). If the drug of interest causes the outcome of interest (i.e. effective in comparat ive effectiveness researc h (CER) or less safe in a drug safety study), susceptible patients will differentially be excluded from the drug exposure group (Drug A, gray shaded blocks) than the other comparator group (Drug
53 B). The overtime depletion of susceptibles from Drug A group leads to the selection of susceptibles (who were less susceptible in prior periods) from Drug B group. Overtime (Periods 3 6), Drug A will appear inferior to Drug B (less effective in CER, or protective/safe in a drug safety study) when it is not in earli er periods. This phenomenon leads to the crossing of survival curves at a point of time due to the depletion of susceptibles and differential selection of less susceptibles overtime (HR>1 Periods 1 3; HR<1 Periods 4 6). Therefore, caution should be exercis ed in interpreting varying period specific HR where they are prone to selecti on bias due to this phenomenon; as a solution, adjusted survival curves utilizing inverse probability of treatment weighted (IPTW) estimator can be compared in terms of survival history of the entire sample when everybody exposed to Drug A, and the survival history of the entire sample when everybody exposed to Drug B ( Cole & Hernn, 2004; Hernn, 2010 ) Prevalent users invoke selection bias in two mechanisms : depletion of susceptibles, and adjusting for confounders that are affected by past exposures (Danaei, Tavakkoli & Hernn, 2012) Furthermore, p revention of selection bias induced by prevalent users is possible at the design stage of the study, where fol low up is restricted to those patients who were not exposed to the drug of interest for a period of time, then became exposed after the end of the specific period, when they will be considered exposure initiators after being unexposed (Danaei, Tavakkoli & Hernn, 2012) Measurement Bias Measurement (information) bias is a distortion in the association between the exposure and the outcome tha t arises from systematic errors in the way variables of
54 interest are measured for the comparison groups (Bgaud, 2000 ) In retrospective database analysis, information ( unobserved construct ; E, O, Z ) regarding true exposures, outcomes, and other variables are recorded in form of indicators ( observed measures ) to mimic these true experiences ( Hernn & Cole, 2 009 ) ; thus for example, drug exposure ascertainment in a database is not the reflection of true drug utilization, rather a reflection of a measured drug utilization. When these observed measures do not accurately reflect unobserved constructs, measurement error or misclassification (!e, !o, !z) is said to happen ( Grimes & Schulz, 2002 ) Measurement bias in pharmacoepidemiology is classified into four types (Figure 2 5 ) : independent nondifferential misclassification; independent differential misclassificati on; dependent nondifferential misclassification; and dependent differential misclassification (Hernn & Cole, 2009) Independent nondifferential misclassification denotes to the independence between the measurement errors in the e xposure and that in the ou tcome; however exposure misclassification is n ot affected by true outcome value (i.e. exposure misclassification is nondifferential across levels of the o utcome ) and likewise in outcome misclassification (Figure 2 5, A) Independent differential misclass ification refers to the independence between the measurement errors in the exposure and that in the outcome; however, the true value of the outcome affects exposure classification (i.e. exposure misclassification is differential across levels of the ou t com e) (Figure 2 5, B), and similarly in outcome misclassification (Figure 2 5, C). Dependent misclassification happens when measuremen t errors in the exposure and that in the outcome are dependent on each other e.g. because they share same mechanism of abstr acting information such as recall bias in pregnancy outcome Like
55 independent misclassification, dependent misclassification can be non differential (Figure 2 5, D) and differential ( exposure misclassification, Figure 2 5, E and outcome misclassification, F igure 2 5, F) Confounder misclassi fication is depicted in Figure 2 6 which is similar to the framework of residual confounding. This type of m isclassification may incline investigators to erroneously conclude the presence of effect modification by the co nfounder, when none exists ( Hernn & Cole, 2009 ) Nondifferential misclassification drives estimates of association between exposure and outcome towards the null hypothesis; differential misclassification on the other hand drives estimates either towards o r away from the null values ( Rothman, Greenland & Lash, 2008 ) In addition to measurement errors in abstracting the information, g rouping categorical or continuous variables into fewer categories can revert nondifferential misclassification to differentia l misclassification ( Rothman, Greenland & Lash, 2008 ); and association estimates can be driven away from the null estimates even when the misclassification is nondifferential. This may occur in multi category exposure s and when exposure or outcome miscla ssification depends on misclassification in other variables in the dataset e.g. confounders ( Rothman, Greenland & Lash, 2008 ) Exposure misclassification Characterization of exposure in observational designs is a challenging task. There is a myriad of r easons for imposing measurement bias in exposure characterization when utilizing retrospective database analyses. In pharmacoepidemiology, measures of drug exposure are not necessarily accurate reflections of true drug exposure. In databases, prescribing a actually received the medication from the phar macy and has actually taken it with
56 instructions. Patients may take or receive too little (e.g. incorrect inhaler techn ique) or too much of the correct drug (e.g. too high dose for the indication and interactions ) or they may not take or receive the drug prescribed ( e.g. produc t not affordable to the patient or the health care insurance system ; product is inconvenient to use, e.g. inhaler coordination; instructions not understood or remembere d or even agreed upon by the patient; experiencing undesirable effects ; failure to experience perceived benefits; product unavailability; or health beliefs and cultural factors). Thus, such patients may be classified as exposed to the drug of interest, when they may be not exposed or exposed to a variable amount of the drug. Similarly, exposure misclassification can occur in randomized, controlled trials (Figure 2 7 ) In open label and non placebo controlled trials in which patients are noncompliant the randomized exposure will be a misclassifi ed measure of the true exposure i.e. the randomized drug is different from the actually received dr ug (Figure 2 7, A) ; however, the association between the randomized exposure and the outcome of interest can be interpreted as an average causal effect when blinding and placebo controlled randomized trials are employed (Figure 2 7, B) (Hernn & Cole, 2009 ). When patients do not adhere to the randomized exposure, they may not be similar to those who adhered in terms of some characteristics, e.g. disease severity. This unmeasured confounding can lead to biased estimates when the aim of the study is to quantify the effect of the actually received exposure unless accounted for utilizing IPTW for example (Toh & Hernn, 2008) Likewise, biased estimates can be attributed to selection bias that can be invoked when patient censoring or loss t o follow up occurs after randomization (Toh & Hernn, 2008). On the other hand, if the study
57 aims at estimating the effect of the randomized exposure (i.e. the outcome when having the intention of treating with the drug), ITT analysis gives valid causal es timates of the ITT e ffect of the randomized drug ; however, it is a misclassified est imate of the actual exposure The magnitude of the ITT effect is proportional to the degree of compliance with randomization ( the arrow ), which gives conservative estimates that correspond to the lower limit of the effect measure confidenc e interval Medication adherence measures are developed to measure the extent of dru g utilization in pharmacy administra tive claim s databases that record medication filling and dispensing information. However, these measures have limited applications in prescribing databases, e.g. the GPRD, where dispensing information is not recorded. Unlike the US, patients in the UK are required to return to their general practitioner to get a new prescription to refill their original one (Figure 2 8 ) Indicators in the GPRD from this routine can be used as a measure of adherence and im prove exposure characterization, e.g. prescription is sue sequence number, which denotes to the number of the refill if the medication is part of a repeat prescription. In the UK, the general practitioner issues either a repeat or a one off prescription to the patient. The patient is expected to take the pres cription to the community pharmacy, where the medications are dispensed along with a repeat medications slip by the pharmacist When the patient runs off the medication (or when a refill is needed for any reason), she visits the general practitioner to get a new prescription issued based upon the repeat slip presented by the patient ( Tim Williams, GPRD, personal communication, May 22 2011 ) Furthermore, the GP is notified about prescription refills when repeat dispensing is practiced by community pharmacis ts, which usually takes place in a form similar to a
58 medication therapy management plan as a consented scheme between prescribers, pharmacists, and patients, in which t he GP prescribe a prescription for 28 day supply with a repeat order ( NPC, 2008 ) The pa tient takes the prescription to the pharmacy that is part of the scheme where the prescription is filled. When the patient needs a refill, she to her GP to reissue a repeat prescription, rather she returns to the pharmacy to get anot her refill for her medication. Upon dispensing the final batch of the medication, the patient returns to the GP to get another prescription reissued upon clinical assessment. Regardless of the way medication refills are practiced (repeat prescribing at the GP level or repeat dispensing at the pharmacist level), the GP records hence, the GPRD are well recorded where f ull prescription information for each patient is directly logged from the GP computer (Khan, Harrison & Rose, 2010) Immortal time bias Immo rtal time bias is a form of exposure misclassification that i s increasingly found in pharmacoepidemiologic studies (Suissa, 2008) ; it arises from cohort definition in which patients meet certain survival criter ion of follow up from index date in which pat ients should survive for a specific duration of time since exposure (Rothman & Greenland, 2008) (Figure 2 9 ) Failure to account for immortal person time results in biased estimates of the associati on between exposure and outcome, which is an underestimati on of the true association in the absence of this bias (If survival rate is higher in the exposed group, the HR will be reduced ), and the magnitude of bias is proportional to the duration of immortal person time (Suissa, 200 3, 200 8 ). Immortal time bias can be prevented at the design and the analysis stages of the study. At the design stage, patient follow up should start after the immortal person time period; while at the analysis stage, this period can be excluded from the analysis of the
59 denominator perso n time calculations for the risk estimates ( Suissa, 2008; Rothman & Greenland, 2008 ) In addition, Cox proportional hazards regression analyses with time dependent exposures and IPTW analyses can be used to account for immortal time bias that might arise f rom time dependent exposures (Suissa, 2008). In the present study, patients are characterized to have a m aximum of 12 months of follow up after the first prescription of the exposure of interest. This 12 month period is considered an immortal person time. T he ensuing bias will not be present in estimating the morbidity outcomes of interest; however, the bias will be induced in estimating mortality outcomes if immortal person time was not excluded from the analyses and patients follow up continued from expos ure index date, rather than from the survival criterion index date (i.e. after 12 mont hs post first prescription; Chapter 3). Diagnostic misclassification Diagnostic (disease) misclassification refers to the differential classification of the clinical cond ition among patients. Diagnostic misclassification is a form of measurement bias and should be distinguished from diagnostic bias, which is a type of selection bias in which patients are differentially diagnosed depending on exposure to specific risk facto usually the first step in pharmacoepidemiologic research design. The validity of disease diagnoses in pharmacoepidemiologic databases varies by the type and the source of the database. Some databases record diagnoses utilizing disease classification codes, e.g. international classification of diseases (ICD) and Read clinical terms; others use clinical terminologies, e.g. medical dictionary for regulatory activities (MedDRA) and systematic nomenclature of medicine (SNOMED) Incorrect coding may arise at the
60 general practice, hospital, health insurance system, or the database administrator. Patients may be misclassified as having the disease of interest, when in reality they are n ot and vice versa. In databases, the accuracy of the diagnosed condition is contingent upon the validity of the diagnosis process (at the practitioner level), and the extent of association of the classification code with the documented diagnosis (at the da tabase level), where the code serves as a surrogate measure for the true diagnosis (van Walraven & Austin, 2012). The GPRD provides the practitioners with specific guidelines on how to record diagnoses and other clinical and therapeutic events using a spec ific software system called the Vision General Practice Management Software (GPRD, 2004). Furthermore, the validity of diagnostic codes in the GPRD was established in multiple conditions, including asthma and COPD ( Hollwell, 1997; Hansell, Hollowell, Nicho ls, McNiece, & Strachan, 1999; Herrett, Thomas, Schoonen, Smeeth, & Hall 2010; Khan, Harrison & Rose, 2010 ). Moreover, the GPRD programmers were provided Read clinical terms that reflect asthma diagnoses for the present study Outcome misclassification I n retrospective database research, patient outcomes of interests are measured by a set of definitions that serve as a proxy to the true outcomes e.g. asthma morbidity is r hospit als due to asthma exacerbations, and prescribing and utilization rates of rescue inhaler medications and oral corticosteroids ( NHLBI, 2007 ) Like exposures, outcomes are prone to misclassification in pharmacoepidemiologic research. When outcome misclassification is nondifferential (Fi gures 2 8, A & 2 8, D), the association will be
61 driven towards the null ; and it will be driven away from the null if the misclassification is differential (Figures 2 8, C & 2 8, F). For example, if exposure to a specific asthma drug incr eases (or decreases) the probability of misclassifying the patient as having hospitalized for asthma exacerbation s the association between the exposure and the outcome will be biased away from the null estimate Developing algorithms to iden tify outcomes that include a mixture of clinical, referral, and prescription information can improve the validity of outcome measurement ( Khan, Harrison, & Rose, 2010; van Walraven & Austin, 2012 ). For example, if the outcome of interest is asthma exacerbations, the inv estigator can search for Read clinical terms for asthma exacerbations in addition to referral information to an emergency department and relevant medication history. Likewise, utilizing linked databases can help in the identification of relevant outcomes. For example, the mortality database that is linked to the GPRD is sensitive to the causes of death, and each record has more than one cause of death field ( ONS, 2009 ) which improve outcome ascertainment compared to similar datasets with one cause of death field Modeling Techniques to Control Bias and Confounding In observational designs, the application of correct methodologies at study design and analysis stages minimize s or prevent s bias and confounding ( Rubin, 2007 ) (Figure 2 10 ) Randomized, controlle d trials are not always feasible to conduct, ha ve limited external validity and sometimes, limited internal validity as stated earlier T herefore it is important to design and conduct observational studies that mimic randomized trials in order to achieve causal interpretation of the observed association ( Cain, Robins, Lanoy, Logan, Costagliola, & Hernn 2010 ; Hernn, 2011 ) Approaches
62 at the design stage include: restriction (excludin g patients with the confounder); matching (compare patients with identi cal confounder distribution) ; n ew user design (incident users); crossover design (exposed patients become their own control as they move in and out the exposure time period) ; active comparator design (compare exposure groups that have identical safety and effectiveness) ; and validation studies (two stage sampling and external adjustment) ( Schneeweiss, 2008 ) The last three methods are used to account for unmeasured confounding. Moreover, the following approaches are applied at the analysis stage: regression modeling; stratification (conduct separate analyses for subgroups of confounder); standardization (stratify by the confounder and apply stratum specific event rates for each exposure group to the number of patients in the corresponding stratum in the stan dard population); instrumental variables (identify a variable that is associated with the exposure but not the outcome) ; and sensitivity analyses ( examine the effect of changing variable values on the modeling results ) (Schneeweiss, 2008) R egression m odel ing is the most common approach to surmount confounding during t he analysis stage of a study The following is a review of regression modeling approaches to control bias and confounding in observational designs with longitudinal time dependent data Multip le Regression Analysis Multiple regression analysis is a statistical technique that is used to estimate the association between the expos ure and the outcome by virtue of a mathematical model that better predicts the value of the outcome as a function of th e exposure after holding all other covariates constant. The association estimates are adjusted when confounding variables are included in the covariate pool, which will minimize or eliminate the effect of
63 the differential distribution of the confounders be tween exposure comparison groups. If confounders have large influence on the association, the adjusted estimates may differ significantly from the unadjusted (crude) estimates. Examples of multiple regression analyses include linear, logistic, Poisson, and Cox proportional hazards The validity and causal interpretation of the calculated estimates are contingent upon model fitting speci fications and other assumptions, e.g. absence of unmeasured confounding ( Christenfeld, Sloan, Carroll, & Greenland 2004; R obins, Hernn, & Brumback 2000 ) Conventional Cox proport ional hazards regression In 1972, the British statistician David R. Cox developed the Cox proporti onal hazards regression (PHREG) (Cox, 1972), which today is the most commonly used regression analys is in longitudinal studies that involve survival and time to event analyses. Cox PHREG is used to estimate the hazard of an outcome occurrence conditional on confounders and other covariates including exposure Hazard is the instantaneous risk that an out come occurs at time given the patient survives t o time or later The hazard is estimated by the hazard function in E quation 2 1: ( 2 1) Where is the hazard functio n (sometimes denote d as ) which refers to the limit of the risk that an outcome will occur at the interval given patient survival to the beginning of the interval (the numerator), as appro aches zero. Since the likelihood of an outcome to occur is proportional with the length of the interval the denominator adjusts for the interval width by shrinking close to zero; therefore, the risk estimates a re instantaneously calculated at time T he outcome time for a particular
64 patient is expressed as in the numerator Figure 2 11, A shows the required data structure for the analysis using Cox PHREG models where t duration of time from the start of follow up until either outcome occurrence or last patient at the recorded time, whether the patient e xperienced the outcome (Status=1) or the patient was censored (Status=0) Even though h azards are defined in terms of probabilities, they are not probabilities, rather are rates expressed in terms of number of events per time interval, and they have values greater than 1 and less than 0, but never negative values. The reciprocal of the hazard yields the expected length of time until outcome occurrence Of course, the interpretation of the hazard and its reciprocal is contingent upon the assumption that ever ything about the patient and the ambient environment, including the hazard itself remains constant over time. However, this is not the case in reality, especially in the presence of time dependent covariates ( Allison, 2010 ) Compared to parametric survival analysis methods ( e.g. the Kaplan Meier, actuarial, and accelerated failure time methods ) Cox PHREG is a semiparametric method of creating models that hand le data with parametric and nonparametric distributions, tied outcome times and time dependent cov ariates; in addition to having efficient stratification analyses (Allison, 2010) Cox PHREG model is expressed in E quation s 2 2 and 2 3 : (2 2 ) ( 2 3) Where the hazard function for an individual at time is a product of the baseline hazard which is the hazard function for t he individual when the exposure and
65 covariate values are zero. The exponentiat ed coefficients yield the Hazard Ratios (HRs), which for the exposure variable, is interpreted as the ratio of the estimated hazard for one exposure group to the estimated hazard for other exposure group, after controlling for other c ovariates. This mo del assumes the ratio of hazards for all patients is proportional across exposure groups, and stays constant during the study follow up period (Therneau & Grambsch, 2000) Fortunately, this assumption is unnecessary when the model is exte nded to include ti me dependent covariates. Cox PHREG with time dependent covariates Time dependent covariates are explanatory variables that may change in value over the course of follow up. Repeated me asurements on the same variable for the same patient o r treatment switching are examples of time dependent variables. Incorporating time to the exposure and confounding variables in E quation 2 2 produces the Cox PHREG model with time dependent covariates in E quation 2 4 Equation 2 5 sh ows a model with time dependent exposure; E quation 2 6 shows a model with time dependent confounder; and E quation 2 7 depicts a model with one fixed, time invariant confounder, one time dependent confounder and one time varying exposure. ( 2 4) ( 2 5 ) ( 2 6 ) ( 2 7 )
66 The Cox PHREG model with time dependent covariates requires specific data structure to handle time dependent covariates. counting process ge nerates data structure similar to Figure 2 11, B where there are multiple records per Start variable s which denote to the interval (start, stop] of time in w hich patient status and time dependent and time independent variables are measured. During the interval, the value of the time dependent variable will be treated as a fixed variable just like in conventional Cox PHREG models ( Therneau & Grambsch, 2000 ) On the other hand, t he programming statements generates data structure similar to Figure 2 11, C where there is one record per patient, and the time dependent variables are continued as multiple variables per patient (Allison, 2010) B oth methods gi ve similar results when carefully and accurately programmed. Although the extended Cox PHREG models with time dependent covariates can efficiently handle time varying exposures (Figure 2 1, C) it fails to adequately control for time dependent confounding that is affected by previous exposure (Figure 2 1, B) ( Robins, Hernn & Brumback, 2000 ; Greenland, 2008 ) ; and a systematic review showed that survival analysis studies did not apply appropriate methodology to account for time dependent confounders and tim e dependent exposures ( van Walraven, Davis, Forster, & Wells 2004) Alternatively, Marg inal Structural Models technique overcomes these hurdles, and under specific assumptions, the technique allows causal interpretations for exposu re outcome association ( Robins, Hernn & Brumback, 2000 )
67 Marginal s tructural m odels In pharmacoepidemiology, drug effects are time dependent, and are affected by time dependent confounders that themselves are affected by previous drug exposure, and affect subsequent drug expos ure and subsequent outcome. In such situations, conventional statistical methods (including Cox PHREG models) produce biased estimates of exposure effect, because they fail to account for the time dependent nature of the confounders and exposures ( Suarez, Borrs & Basagaa 2011). Moreover, lack of randomization in observational designs prevents bestowing causal interpretations upon observed associations. To overcome these problem s Marginal Structural Models (MSM) technique was developed in 1997 as a new class of causal models, followed by two companion articles in 2000 that paved the way for practical applications of these models, which allow for a n unbiased, causal, population level (marginal) effect estimate of exposure on the outcome in the presence of time dependent confounding and time dependent exposure in observational data (Robins, 1998; Hernn, Brumback, & Robins, 2000; Robins, Hernn & Brumback, 2000 ). is sometime s used to reflect the extension to accoun t for censoring MSM is a weighted repeated measures technique that is based on the inverse probability of treatment weighting as an extension of the exposure propensity scores approach. Although MSM is effective in handling time dependent confounding, the technique is l ess efficient in estimating effect modification by time varying covariate than structural neste d models (SNM) (Robins, Hernn & Rotnitzky, 2007 ). MSM technique creates at each point of time (risk set) a pseudo population of counterfactuals (a hypothetical population in which all patients seem as exposed and
68 unexposed to the drug) in which time invariant and time dependent confound ers are balanced and therefore, causal association between the exposure and the outcome is the same as in the original study population ( Hernn, Brumback & Robins, 2000 2002 ). The pseudo population is created by weighting every patient in the population by the inverse of the conditional probability of being exposed to the treatment that the patient actually rece ived MSM compare two counterfactuals: outcome if the entire study population is expos ed to the treatment and outcome if the entire study population is not exposed to the treatment (Figur e 2 12 ) Thus, it gives valid causal interpretations between the exposure and the outcome. Nonetheless c ausal inference from MSM is contingent upon inherent assumptions of exchangeability, positivity, consistency, and correct modeling for weighting (expos ure model) and analysis (outcome model) ( Brumback, Hernn, Haneuse, & Robins, 2004; Mortimer, Neugebauer, van der Laan, & Tager, 2005; Cole & Hernn, 2008 ) The following rules are termed the causal effect identifiably assumptions (Cole & Hernn, 2008) : 1. Ex changeability: lack of unmeasured confounding is referred to conditional exchangeability, where all the variables explaining the exposure and the outcome are included in the analysis. Formally, conditional exchangeability defined as the independence betwee n the prospective counterfactual outcomes and the observed exposures given previous exposure and confounder history To illustrate, the risk of experiencing the outcome under exposure in the exposed patients equals the risk of exper iencing the outcome under exposure in the unexposed patients Similarly, the risk of experiencing the outcome under no exposure in the exposed patients equals the risk of experiencing the outcome under no exposure in the unexposed pa tients Exchangeability assumption can be assessed by conducting sensitivity analysis proposed by Brumback & colleagues (2004).
69 2. Positivity: experimental treatment assump tion is referred to positivity, which implies the presence of po sitive probability for patients to receive each exposure category for a set of covariates, including prior exposure and confounding history (i.e. no perfect confounding) Positivity assumption c an be assessed by Mortimer and associat es approach (2005). 3. Consistency: consistency refers to the outcome for every exposed patient equals the outcome if the patient had remained exposed when and the outcome for every unexposed patient equals the ou tcome if the patient had remained unexposed when Marginal structural modeling involves two consecutive steps : estimating stabilized inverse probability weights by propensity scoring ( exposure selection model) and conducting weighted repeated measures analysis by generalized estimating e quations ( outcome analysis model) ( Cole & Hernn, 2008 ) Binary logistic regression is used for binary exposure (Rosenbaum & Rubin, 1983 ; Hernn, Brumback, & Robins 2000 ) and m ultinomial logistic regression with the generalized logit link (GLOGIT) is used for multicategory exposure ( E quations 2 9 and 2 10 ) (Rubin, 1997 ; Allison, 2012; Chitnis, Aparasu, Chen, & Johnson 201 2 ; Desai et al 2012 ) Similarly, binary logistic regres sion or generalized linear model is used to est imate censoring probabilities (E quations 1 2 and 1 3 ). Weighted g eneralized estimating equation is used to estimate the hazard ratio for the causal association between the exposure and the outcome for each patient at every visit weigh ted by the stabilized weights (E quations 2 1 5 and 2 1 6 ). The final model includes time dependent exposure, but not time dependent confounders. Time dependent confounde rs are accounted for in the weighting process. Exposure selection model: ( 2 8 )
70 Numerator: ( 2 9 ) Denominator: ( 2 10 ) Censoring model: ( 2 1 1 ) Numerator: ( 2 1 2 ) Denominator : ( 2 1 3 ) Stabilized exposure selection & censoring weight: ( 2 1 4 ) Outcome analysis model: ( 2 15) ( 2 1 6 ) : Probability of patient to receive the observed exposure at time interval given exposure, covariates, and censoring history. : Outcome at time if outcome is observed; 0 otherwise. : Outcome at time : E xposure at time for patient for ICS; for LABA; for ICS/LABA. : Exposure history prior to time for ICS; for LABA; for ICS/LABA. : Patient censoring status at time if patient lost to follow up; if patient continued to time
71 : Patient censoring history prior to time if patient lost to follow up; if patient continued to time : Measured time independent confounders at baseline. : Mea sured tim e dependent confounders at time : Measured time dependent confounders at time Exposure Propensity Scores Technique. Rosenbaum and Rubin (1983) developed e xposure propensity sco res (EPS) technique as a mean to account for confounding by indication ( especially confounding by disease severity) in the presence of measured confounding variables. The score is defined as the likelihood of a patient being exposed to treatment given a se t of measured confounding variables; on average, exposure groups with similar scores are expected to have similar baseline information with respect to confounding variables. Thus, a quasi randomization state is achieved (Ali, 2011). The inverse of the EPS for exposure groups yields the inverse probability of treatment weighted (IPTW), which creates weights for each patient at each unit of time based on the propensity scores. The weights are interpreted as the number of copies for each observation that are r equired to form a pseudo population of counterfactuals in which no time dependent confounding exists ( Hernn, Brumback, & Robins 2000). The IPTW approach gives unbiased estimates of exposure effects in the presence of time depend ent confounding (Greenland 2008). Figure 2 13 illustrates the process of attaining quasi randomization by virtue of MSM. Suppose in a sample of 100 patients, 80 are exposed to the treatment of interest and 20 are unexposed (Figure 2 13, A). The conjecture of MSM is to recreate thi s
72 sample in a way that would equate the selection mechanism between the exposed and the unexposed. Within the exposed group, the probability of exposure given the covariance structure (propensity score PS) equals 0.8, and the inverse of that probability (inverse probability of treatment weighted IPTW ) equals 1.25. By multiplying this inverse probability as if it was a weight times the number of people in that group (n=80) a total sample size of 100 is obtained. The same is applied to the unexposed group, where the probability of not exposure equals 0.2, and the inve rse of the probability equals 5. Similarly, the total sample size of 100 will be obtained after multiplying the inverse probabi lity of exposure by the number of people in the unexposed group (n=20). In effect, the MSM has created two populations with selection mechanism of 50% chance one treated and the other untreated, i.e. probability of the outcome for the counterfactual group s when treated versus the probability of the outcome for the counterfactual group when not treated ; where is the outcome of interest, is the exposure of interest ( exposed; unexposed), and is a vector of confounding variables This quasi randomization approach creates selection probabilities that are the same for the exposed and the not exposed. The same principle extend s to multiple exposure groups (Figure 2 13, B). The EPS technique does not account for unmeasured confounding. Therefore, the exposure groups might seem balanced based on the likelihood of receiving treatment when they are not in the presence of unmeasured confounders (Glynn, Schneeweiss & Strrmer, 2006) Although the utilization of this te chnique has dramatically grown in the biomedical literature (Glynn, Schneeweiss & Strrmer, 2006 ; Kurth & Seeger, 2008 ),
73 the majority of studies used this technique yi elded results that are essentially similar to ones obtained from traditional regression models ( Shah, Laupacis, Hux, & Austin 2005 ). MSM approach controls time dependent confounding, accounts for informative censoring, and extends to multicategory exposur es (van der Laan & Robins, 2003). Conventional regression methods yield non interpretable coefficients after adjusting for t ime dependent confounder s, which predict future exposure s and outcome s and are affected by past exposure s Causal inference methods e.g. MSM are required to consistently estimate the causal effect of exposure s in such situations Instrumental Variable Analysis Technique. The application of instrumental variable (IV) analysis technique in pharmacoepidemiology is borrowed from econome trics. The IV is an observed variable that is incorporated in regression models in an attempt to prevent residual confounding due to the influence of unmeasured confounders in retrospective database analys e s ( Johnston, Gustafson, Levy, & Grootendorst 2008 ) A perfect IV should meet the following conditions lest give biased estimates (Figure 2 14 ) (Greenland, 2000) : The is associated with the exposure either directly, or both share a common cause The is not associated with the outcome. Not directly, nor through a common cause It affects the outcome only through the exposure. The is not associated with the confounding varia bles In retrospective database analysis, it is difficult to find a variable that meets these assumptions. In scenarios where exposure is time dependent, the identification of a time varying IV is nearly impossible (Hernn & Robins, 2006). Some researchers used
74 to prescribe one medication class over the other as an IV ( Brookhart, Wang, Solomon, & Schneeweiss 2006 a ); however, the proxy measure gave biased estimates (Hernn & Robins, 2006) Such bias can also be raised when the magnitude of confounding due to unmeasured confounder is strong ( Martens, Pestman, de Boer, Belitser, & Klungel 2006). Moreover, additional strong assumptions are requir ed when IV is used to estimate population average causal effect e.g. lack of additive effect modification by the IV within exposure groups ( Brookhart, Wang, Solomon, & Schneeweiss 2006 b ; Hernn & Robins, 2006 ).
75 A ) B ) C ) Figure 2 1. Illustration of confounding. A) Classical time independent confounding, B) time dependent confounding and time varying exposure and C) time varying exposure with no time dependent confounding. Z t O E t Z t+1 O E t+1 E t Z t E t 1 Z t 1 Z t+1 O E t+1 E t Z t E t 1 Z t 1
76 Figure 2 2. Illustra tion of the relationship between study sample and target population in pharmacoepidemiologic studies Figure 2 3. Illustration of informative censoring in pharmacoepidemiology Target Populati on Base Population Study Sample Study Findings Representative Representative Interpret Extrapolate Time since Exposure (Year) Patients 0 1 2 3 4 5 F E D C B A End of follow up R R O O Censored Outcome time O O Uncensored outco me time Randomly censored outcome time Informative Censoring
77 Figure 2 4 Illustrat ion of (exposure) with an adverse drug reaction (outcome) At earlier time of the study (Period 1), patients at risk of experiencing the outcome are expected to devel op the outcome and are excluded from follow up, thereby leaving a group of patients who have low risk of experiencing the outcome at later time (Periods 2 3). If the drug causes the outcome (effective/less safe), susceptible patients will differentially be excluded from the drug exposure group (Drug A, gray shaded blocks) than other comparator group (Drug B). The overtime depletion of susceptibles from one exposure group (Drug A) leads to selection of susceptibles (who were less susceptibles in prior period s) from the other group (Drug B). Overtime (Periods 3 6), the exposure group (Drug A) will appear protective against the outcome of interest (i.e. safe), when it was not in earlier periods. This leads to the crossing of survival curves at a time point due to the depletion of susceptibles and differential selection of less susceptibles overtime (HR>1 Periods 1 3; HR<1 Periods 4 6). Therefore, period specific hazard ratios are prone to selection bias due to the survivor bias phenomenon. Susceptible patients a re those who are susceptible to the outcome that is an adverse reaction to the drug of interest. Time since Exposure Outcome Risk 0 0 + High risk patients (susceptible) Low risk patients (less susceptible) Drug A group Period 1 Period 2 Period 3 Per iod 4 // Period 5 Period 6
78 ( A ) ( B ) ( C ) ( D ) ( E ) ( F ) Figure 2 5 Illustration of measurement bias types in phar macoepidemiology. A) independent nondifferential, B) independent differential exposure, C) independent differential outcome, D) dependent nondifferential, E) dependent differential exposure and F) dependent differential outcome misclassification. O E !o !e O E !o !e O E !o !e O E !o !e O E !o !e O E !o !e
79 ( A ) ( B ) ( C ) ( D ) ( E ) ( F ) Figure 2 6. Illustration of confounder misclassification in pharmacoepidemiology. A) independent nondifferential, B) independent differential, depends on exposur e, C) independent differential, depends on outcome, D) dependent nondifferential, E) exposure misclassification depends on confounder and F) outcome misclassification depends on confounder. O E !e !z Z !o O E !o !e !z Z O E !o !e !z Z O E !o !e O E !o !e !z Z O E !o !e !z Z
80 ( A ) ( B ) Figure 2 7. Illustration of exposure misclassification in randomized controlled trials. A) Open label, non placebo controlled and B) double blind, placebo controlled. In intention to treat (ITT) analyses, the interest is the effect of the in tention to randomize exposure. It is valid only in placebo controlled trials where R is not associated with O (panel B) and in the absence of loss to follow up or censoring. However, ITT has an inherent misclassification bias. When censoring or loss to fol low up occurs, pseudo ITT effect is computed within those uncensored until outcome measurement. It gives values closer to the null and are concerning in safety studies. ITT effect is effectiveness; to distinguish it from observational effectiveness the ter R O E U R O E U
81 Pharmacy Stamp Age Title, Forename, Surname & Address Patient No. Patient name Patient No. D.O.B Address Please give the Pr actice a minimum of 2 days notice prior to collecting your repeat drugs. D.O.B N.B. Ensure dose is stated Endorsements Medication 1 There are XX items on this re order form 01/0 1/2012 Medication 2 Medication 1 Medication 2 Signature of Prescriber Date For dispenser No. of Prescrns. on form NHS Ayad 2012 Prescriber Address Prescriber No. PATIENTS please read the notes overleaf The pre scriber ticks the box next to the medication to order a repeat prescription, otherwise the prescription is a one off. Figure 2 8. Sample of a prescription in the UK general practice
82 Figure 2 9. Illustration of immortal time bias in pharmacoepidemiology If a cohort study, patients are required to survive for at least one year after exposure to study drug to be included in the study. This criterion ensures that all patients have survived the first year of exposu re and those who did not survive during that period are excluded from the study. If mortality is the outcome of interest, the 1 year immortal person time should be excluded from the analysis otherwise the estimates will be underestimated. In the illustrate d cohort study, 6 patients exposed to Rx1 and 6 patients exposed to Rx2 meet the criterion of being survived for at least 12 months after exposure. If Rx1 patients were followed up to 6 months after this index date, and Rx2 patients to 9 months. Person mon th for Rx1=36, and person month for Rx2=45. Suppose during the follow up period 3 deaths are observed for Rx1 group, and 2 deaths for the Rx2 group. The correct risk ratio is (3/36)/(2/45) = 1.87 and the risk difference is 48/1000 month 1. However, if the 12*6=72 for Rx1 and 72 for Rx2 immortal person time are included in the denominator of risk estimates, the risk ratio is (3/108)/(2/117) = 1.62 and risk difference is 10/1000 month 1 i.e. underestimated. Time (Month) Exposed Patients 0 3 12 15 Survival criterion index O O R x 1 R x 2 6 9 18 21 O O O Immortal Time Period
83 Figure 2 10. Pharmacoepidemiologic approaches to account for bias and confounding Unmeasured Measured Observational Experimenta l Study Type Randomization Double Masking Placebo Control Confounding Compliance Selection bias Measurement bias Confounder Type Study Stage Design Analysis Restriction Matching New User Design Regression Stratification Standardization Study Stage Design Analysis Crossover Design Active Comparator Validation Study Instrumental Variable Sensitivity Analysis Proxy Measures
84 Patient Time Status E Z Time: duration of time from the start of follow up until outcome occurrence or last observation (i.e. censoring). Status: patient status at the recorded time, whether the patient experienced the outcome (1), or was censored (0). E, Z: exposure and confounder values at the fixed observed time. 1 2 3 4 5 6 7 8 (A) Patient Start Stop Status E Z V Start, Stop: inte rval of time (start, stop] in which the status of the patient was observed. V: time dependent variable Multiple records per patient. 1 1 1 1 1 2 2 2 (B) Patient Time Status Time1 Time2 E1 E2 V1 V2 Z One record per patient. 1 2 3 4 5 6 7 8 (C) Figure 2 11. Data structure for Cox proportional hazards model, A) with fixed covariates, B) time dependent covariates counting process method and C) time dependent covariates programming statements method.
85 Figure 2 12. Illustration of association versus causation in pharmacoepidemiology Association Causation Conditional probability: based on actual exposure received by the sample Probability of outcome among exposed. Probability of outcome among unexposed. Probability of outcome when entire population is exposed. Probability of outcome when entir e population is unexposed. Exposed Unexposed Marginal probability: based on entire population under two exposure options
86 (A) Figure 2 13. Illustration of attaining quasi randomization by virtue of marginal structural models in a hypothetical population of size N=100 with A) binary exposure and B) mu lticategory exposure situations. PS, propensity score; IPTW, inverse probability of treatment weighted. Unexposed to treatment ( ) n=20 i. ii. iii. Exposed to treatment ( ) n=80 i. ii. iii. Where in one group is the counterfactual for in the other group Probability of ex posed given = 0.5 & Probability of unexposed given = 0.5 Every patient has 50% probability of being a member of an exposure group
87 (B) Figure 2 13. Continued Exposed to treatment C ( ) n=10 i. ii. iii. Exposed to treatment A ( ) n=70 i. ii. iii. Exposed to treatment B ( ) n=20 i. ii. iii. Probability of exposure to A given = 0.5, probability of exposure to B given = 0.5 & probability of exposure to C given = 0 .5
88 Figure 2 14. Characteris tics of an instrumental variable Z O E IV W X || // \ \
89 CHAPTER 3 STUDY AIMS AND SIGNI FICANCE Study Objective The broad objective of this study is to assess asthma r elated morbidity and mo rtalit y in patients exposed to inhaled LABA bronchodilators, as monotherapy, and ICS Research Database (GPRD) ( Figure 3 1 ). Research Questions, Specific Aims, and Hypotheses This study is aimed to answer the following research questions, and to test the corresponding hypotheses: Research Question No. 1 Is there a difference in the incidence of asthma morbidity in terms o f asth ma related accident and emergency (A & E) department visits, asthma related hospitalizations, and prescribing oral corticosteroids between asthmatic patients exposed to inhaled LABA monotherapy, ICS monotherapy, and ICS/LABA combination therapy in the UK GPR D? Specific aim n o. 1 To examine the association between ICS/LABA combination therapy and asthma morbidity rates compared with inhaled LABA monotherapy, and ICS monotherapy in asthmatic patients in the GPRD. Hypothesis n o. 1 The null hypothesis states that there is no diffe rence in asthma morbidity rates measured by hazard ratios between ICS/LABA combination therapy, inhaled LABA monotherapy, and ICS monotherapy in asthmatic patients in the GPRD.
90 LABA compared with ICS ICS/LABA compared with ICS ICS/LABA compared with LABA The alternative hypothesis states that there is a difference in asthma morbidity rates between ICS/LABA combination therapy, inhaled LABA monotherapy, and ICS monotherap y in asthmatic patients in the GPRD. LABA compared with ICS ICS/LABA compared with ICS ICS/LABA compared with LABA The null and alternative hypotheses are the same for individual measures of asthma morbidity: A & E department visits for asthma, hospitalizations for asthma, and prescriptions for oral corticosteroids. Research Question No. 2 Is there a difference in the incidence of asthma deaths between asthmatic patients exposed to inhaled L ABA monotherapy, ICS monotherapy, and ICS/LABA combination therapy in the UK GPRD? Specific a im n o. 2 To examine the association between ICS/LABA combination therapy and asthma death rates compared with inhaled LABA monotherapy, and ICS monotherapy in asth matic patients in the GPRD.
91 Hypothesis n o. 2 The null hypothesis states that there is no difference in asthma death rates between ICS/LABA combination therapy, inhaled LABA monotherapy, and ICS monotherapy in asthmatic patients in the GPRD. LABA compared with ICS ICS/LABA compared with ICS ICS/LABA compared with LABA The alternative hypothesis states that there is a difference in asthma death rates between ICS/LABA combination therapy, i nhaled LABA monotherapy, and ICS monotherapy in asthmatic patients in the GPRD. LABA compared with ICS ICS/LABA compared with ICS ICS/LABA compared with LABA Research Question No. 3 In as thmatic patients who were exposed to ICS/LABA combination therapy, is there a difference in the incidence of asthma morbidity i n terms of asthma related A &E department visits, asthma related hospitalizat ions, and prescribing oral corticosteroids between st ep down therapy approaches represented by LABA discontinuation and ICS dose reduction and continual of original ICS/LABA combination therapy in the UK GPRD? Specific aim n o. 3 To examine the association between ICS/LABA combination therapy and asthma morb idity rates compared with 50% reduced ICS dose in ICS/LABA combination
92 therapy, and ICS monotherapy due to LABA discontinuation in a subgroup of asthmatic patients who were exposed to ICS/LABA combination therapy for a minimum of 3 months in the GPRD. Hypo thesis n o. 3 The null hypothesis states that there is no difference in asthma morbidity rates between ICS/LABA combination therapy (continuers) 50% ICS dose reduced ICS/LABA combination therapy (reducers), and ICS monotherapy (LABA stoppers) in asthmatic patients in the GPRD. ICS/LABA 50% reduced ICS dos e vs ICS/LABA original ICS dose ICS (LABA stoppers) vs. ICS/L ABA original ICS dose incessant ICS/LABA 50% ICS dose reduc ers vs. ICS (LAB A stoppers) The alternative hypothesis states that there is a difference in asthma morbidity rates between ICS/LABA combination therapy continuers reducers, and LABA stoppers in asthmatic patients in the GPRD. ICS/LABA reducers vs. ICS/LABA continuers LABA stoppers vs. ICS/LABA continuers ICS/LABA reducers vs. LABA stoppers The null and alternative hypotheses are the same for individual measures of asthma morbidity: A & E department visi ts for asthma, hospitalizations for asthma, and prescriptions for oral corticosteroids. Rationale and Significance T he aforementioned specific aims will be achieved by virtue of the application of analytical techniques to account for the time dependent nat ure of the var iables in real
93 life practice settings. The findings will enable us to make semi causal interpretations of the associations between LABA products and asthma related outcomes, and draw informed conclusions regarding the safety profile of inha led LABA bronchodilators in asthmatic patients. Therefore the study will serve as a complement to the forthcoming findings from the ongoing clinical trials that are commissioned by the FDA to investigate the effects of LABA discontinuation or ICS/LABA com bination therapy on asthma control. The conclusions will inform healthcare professionals, contribute to improved patient care, and further the development of asthma management guidelines. Figure 3 1. Study profile ICS, in haled corticosteroids; LABA, long acti ng beta agonists UK GPRD Asthmatic patients Patients meet selection criteria ICS/LABA > 3 months on therapy ICS LABA LABA Stopper ICS Dose Reducer Continuer
94 CHAPTER 4 METHODS Research Ethics This research study was reviewed by the Gainesville Health Science Center Institutional Review Board (IRB) at the University of Florida on July 16, 2007 under protoc ol number H 311 2007, and the Independent Scientific Advisory Committee (ISAC) of the GPRD on March 22, 2010 under protocol number 10_040R. The research study was approved in writing by the IRB on July 31, 2007 and by the IS AC on June 8, 2010. Study Type a nd Design This study is a population based retrospective cohort new user design (inception or incidence cohort) aimed at testing a formal research hypothesis in adherence to the International Society for Pharmacoepidemiology (ISPE) guidelines for good phar macoepidemiology practices (GPP) (ISPE, 2008), and according to the recommendations from the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) Good Research Practices for Comparative Effectiveness Research ( Berger, Mamdani, Atkins, & Johnson 2009; Cox, Martin, Van Staa, Garbe, Siebert, & Johnson 2009; Johnson et al 2009 ). Novel d esign approaches and advanced statistical analysis methods of causal inference will be applied to account for the limitations in observational designs di scussed in chapter 1. Data Source ce Research Database (GPRD), the Statistics (ONS) Mortality Data, and the Index of Multiple
95 Deprivation (IMD) scores are utilized for this study (GPRD, 2011 a ; Khan, Perera, Harper, & Rose 2010; NHS, 2010). The General Practice Research Database (GPRD) medical records from public primary healthcare p ractices in the UK (Wood & Coulso n, 2001). The GPRD is a not for profit research database that was established in 1987 for the purpose of pharmacoepidemiology and multidisciplinary public health research HRA (Wood & Martinez, 2004). In the UK, general pra ctitioners are the gatekeepers for primary healthcare and referrals. Around 595 participating general practices located throughout the UK, including England, Scotland, Wales, and Northern Ireland, are included in the database with incrementally collected c linical records for approximately 12 million patients from birth to death (GPRD, 2011 a ). Data collected from participating general practitioners for the GPRD include patient demographics and practice registrat ion details; medical diagnosis, treatment and i mmunization details; treatment outcomes; patient lifestyle factors; referral details to hospitals or specialties; laboratory tests; and c onsultation information (Wood & Martinez, 2004). The quality of data in the GPRD is controlled at both patient and prac tice levels, and specific level based criteria are required to be met in order to consider the data as apposite for research purposes ( Davis, Rietbrock, Rubino, Shah, Williams, & Martinez 2003). Patient acceptable registration status, demographics, and ev purpose; practice level criteria include a standard set of conditions related to practice details that are used to derive the Up to Standard (UTS) date for each registered
96 p ractice in the GPRD. The UTS date is the first date at which the collected data from the individual practice are considered eligible for recording in the GPRD (Wood & Martinez, 2004). As of April 2011, 626 UTS general practices are registered within the GP RD with longitudinal research data for a total of acceptable 11. 26 million patients (GPRD, 2011d ). The GPRD is a prominent entity in the UK that is active in defining standards and procedures for data quality improvement in primary care. The GPRD subsidize s general practices to facilitate data extraction process; and practices receive feedback concerning data quality from GPRD based on analyses of received practice data (DH, RCGP & BMA, 2011). The comprehensiveness, quality, and size of the data make the G PRD a unique tool for population based longitudinal research designs in pharmacoepidemiology, pharmacovigilance, and outcomes research In the EU the GPRD is the most extensively used database in pharmacoepidemiological research (Garcia Rodr guez & Guttha nn, 1998). In addition, the validity of the database for epidemiological research in respiratory system and research in drug safety has been established ( Jick, H., Jick, S., & Derby, 1991; Jick, H., Terris, Derby, & Jick, S., 1992; Hansell, Hollowell, Nich ols, McNiece, & Strachan 1999; Soriano, Maier, Visick, & Pride 2001; Jick S. et al 200 3 ). Besides, recording of deaths in the GPRD is complete and consistent with national statistical figures (Meier & Jick, 1997 ), and the quality of morbidity data for public health and policy research has been evaluated and established (Hollowell, 1997 ). Furthermore, because prescribers are more motivated to record information pertinent to their patients, the information in medical records database like the GPRD is rel atively more accurate and complete in comparison to other types of databases, e.g.
97 administrative claims database (Jick, S., 2010). Also, it is estimated that about 80% of asthmatic patients in the UK are seen by general practitioners ( Vermeire, Rabe, Sori ano, & Maier 2002). All these characteristics make the GPRD an efficient source for the present study topic. The Office for National Statistics (ONS) Mortality Data The GPRD is electronically linked to other National Health Service (NHS) databases in an a nonymized manner. Currently, the GPRD is linked to a subset of hospitalization records through the Hospital Episode Statistics (HES) database, and death data through the Office for National Statistics (ONS) mortality data. Some disease registries and other census data are also linked to the GPRD (GPRD, 2011b). Deaths registered before January 1, 2001 have the original underlying cause of death and all cause of death mentions coded in the 9 th version of the International Classification of Diseases (ICD 9), w hile deaths registered since January 1, 2001 are coded in the 10 th version of the ICD ( ICD 10 ) system (NHS, 2011 a ). All the mortality records in the current study are registered after January 1, 2001 (until October 20, 2009). The mortality data are for pat ients in the cohort who are affiliated with the general practices that are participated in the GPRD linkage scheme. This is a subset of the GPRD practices, all in England. Data from 224 of 472 English practices are available. Each practice consents to have their data linked via an external independent third party to external data sources, such as the ONS mortality data for research purposes (Susan Eaton, GPRD, personal communication, September 13, 2010).
98 The Index of Multiple Deprivation (IMD) Scores The IMD scores are generated at the practice level, which was provided by the GPRD as a linked dataset of the scores that are categorized into quintiles based on the distribution of the scores within specific UK countr ies ( Khan, Perera, Harper, & Rose 2010 ). The scores are con structed by Noble et al (2000) at Oxford University as a measure of socioeconomic status which reflects such status based on general practice postcodes The scores are assigned to each patient at the general practice level and used as a proxy measure for socioeconomic status. The score is calculated differently for each of the four countries of the UK (England, Scotland, Wales, and Northern Ireland). A high score (80) and a high rank (4) indicates the most deprived area. A low score (1) and a low rank (0) in dicates the least deprived area ( McLennan, Barnes, Noble, Davies, Garratt, & Dibbon 2011). Most deprived areas correspond to high socioeconomic standings, and the reverse is true for the least deprived areas. These scores were request ed from the GPRD, and were not duly supplied; at the meantime, other proxy measures are used to reflect s socioeconomic status ( patient characteristics section ). Study Population Patients with asthma who are continuously registered with UTS genera l practices within the GPRD are identified and randomly selected by GPRD research liaison according to th e following selection criteria Asthma patients were identified using specific Read Clinical Terms, version 3 that are used by the GP to record the dia gnosis and follow up for each patient. The GPRD translates Read Clinical Terms to Medical Codes in individual GPRD datasets, which are more user friendly in data management
99 process. Asthma diagnosis was defined as having a Medical Code record for asthma in the clinical, referral, or test datasets of the database before the index date of first prescription for the exposure of interest, or during the maximum follow up duration of the stud y (12 months post index date) The GPRD provides the practitioners with specific guidelines on how to record diagnoses and other clinical and therapeutic events using a specific general practice management software system called Vision (GPRD, 2004). The following selection criteria are applied to the identified asthmatic patie nts in the GPRD to construct t he exposure cohorts of interest : Inclusion Criteria Patients with records acceptable for research (acceptable patients) who are registered with the UTS general practices in the GPRD during the study period. Both male and femal e sexes All race and ethnic groups. Patients aged 13 65 years at index date. Current asthma management guidelines define adult asthmatics as aged greater than 12 years old (BTS, 2009 ); the prevalence of chronic obstructive pulmonary disease (COPD) is more pronounced in the elderly, particularly those older than 5 5 (NHLBI, 2003) with higher prevalence among patients older than 60 year s of age (GOLD, 2010); and differential diagnosis of asthma from other respiratory disorders is more accurate in individuals aged 5 45 than other age groups (Williams et al 1998) Nevertheless, including nonsmoker patients aged 40 65 years diminishes asthma misclassification as COPD (Morales, Jackson, Fielding, & Guthrie, 2012). Patients with first ever therapy event (prescri ption) record for one of the study drug classes recorded in the therapy dataset of the datab a se : inhaled LABA ICS or both (ICS/LABA) The event date associated with the first prescription will be the index date for the study. For those patients who have both drugs prescribed, the earliest of the first events will be the index date. Therapy events are recorded in the therapy dataset of the GPRD by virtue of the Product Codes. Patients with at least two prescriptions for the stu dy drugs in the first 6 month s after the index date of first ever prescription
100 P atients with at least 18 months of follow up data prior to the index date are available. P atients with at least 12 months of follow up data after the index date are available. This criterion will introduc e immortal person time of 12 months which will induce measurement bias if not accounted for in the assessment of asthma mortality outcome (specific aim no. 1) Nevertheless, the bias will not be introduced in the assessment of asthma morbidity outcome (sp ecific aims no. 2 and 3) ; h owever, this selection criterion might in voke selection bias due to informative censoring. The accounting approach for both issues is discussed in detail in the cohort definition sections that are pertinent to the specific aims o f the study Exclusion Criteria Patients wit h diagnosis Read Clinical Terms and Medical Codes as medical event records for the following respiratory conditions in the clinical, referral, or test datasets of the database and Product Codes in the therapy da taset during the baseline year and the follow up year of the study : 1. Chronic obstructive pulmonary disease, including chronic bronchitis and emphysema. 2. Diagnostic and therapeutic respiratory procedures, excluding lung function tests. 3. Lung transplantation, w ith or without heart. 4. Lung lobectomy. 5. Pneumopathies due to exposure to fumes and chemicals, including occupational lung diseases and pneumoconiosis. 6. Chronic tuberculosis, including respiratory tuberculosis and prescriptions for anti tuberculosis agents. 7. Br onchopulmonary aspergillosis, including prescriptions for antifungals for aspergillosis. 8. Pneumocystis pneumonia, including prescriptions for anti pneumocystis pneumonia agents. 9. Respiratory neoplasm including benign and malignant 10. Cystic fibrosis. 11. Intersti tial (parenchymal) lung disease. 12. Obstructive sleep apnea, including nocturnal dyspnea.
101 13. Bronchiectasis and atelectasis. 14. Pulmonary hypertension, embolism and edema. 15. Respiratory obstruction by foreign objects. 16. Congenital and structural anomalies of the respir atory system. 17. Congenital heart disease. 18. Congestive heart failure. 19. Pulmonary valvular heart disease. 20. Unspecified and other respiratory diseases including injuries to the respiratory tract Patients with prescriptions for single device combination inhaled S ABA and muscarinic receptor antagonists (MRA), including ipra tropium, oxitropium, tiotropium Inhaled MRA products are more likely prescribed to patients with COPD than those with asthma ( McIvor, Tunks & Todd 2011). Patients with prescriptions for single device combination inhaled SABA/ICS, and single device combination inhaled SABA and mast cell stabilizer (MCS). Patients who are current smokers or with a history for smoking. Smoking is associated with COPD, worsens asthma outcomes, and diminishes the re sponse to ICS (Boulet, 2009). Middle age individuals (>40 years old) who are chronic smokers are more likely to develop COPD than nonsm oker counterparts (GOLD, 2010). Patients with prescriptions for smoking cessation therapy are considered current smokers. Patients with prescriptions for non selective beta adrenoceptor blockers, including ophthalmic formulations and combination antihypertensive products ever re corded during the study period. Patients with prescriptions for allergen immunotherapy vaccination Patients with prescriptions for inhaled betamethasone for lack of dosage bioequivalence information to account for ICS strength. Patients with prescriptions for t he anti IgE, omalizumab since only one patient was prescribed omalizumab. Patients with Read Clinical Term s and corresponding Medical Code s ident ifying participation in a clinical study, including asthma research
102 Patients with Read Clinical Term s and Medical Codes identifying the use of illicit drugs. Patients with indeterminate sex. Study Durat ion and Selection of Comparison Groups Asthma management guidelines were first introduced in the UK in January 1, 1993 ( BTS 1993). The event date associated with the first prescription for the study drug after January 1, 1993 is defined as the index date for the study. Three distinct index dates are defined for the three specific aims, and are discussed below. Study outcomes will be measured during the study follow up period of 12 months post index date (follow up year) The LABA group includes first user patients to whom a prescription for a LABA inhaler was written; the ICS group includes first user patients to whom a prescription for an ICS was written; and the ICS/LABA group includes first users to whom a prescription for either a single device combina tion ICS/LABA product was prescribed, or a prescription for separate ICS and LABA inhalers that were written in a single prescription on the same date Identified exposure cohorts will be retrospectively follow ed from the relevant study index date until th e end of the follow up year, latest GPRD data recording, patient death, patient transfer out of the general practice, or development of the outcome of interest whichever date comes first. Besides, patient follow up is continued upon exposure switching to examine the time varying nature of exposure s Exposure switching is permitted in study aims 1 and 2, and is considered a censoring criterion for study aim 3. When patients switch fr om one exposure to another, they will contribute exposure person time to ea ch exposure group during the relevant exposure time window. To illustrate, a patient who was prescribed an inhaled LABA at the first visit then
103 prescribed an ICS at the second visit ; she is considered a member of the LABA gr oup during the period between the first visit and the second visit then a member of the ICS gro up during the period between the second visit and the subsequent visit and so on If she were pres cribed an ICS in addition to an inhaled LABA at the second visit, she will be considered a member of the ICS/LABA group during the period from the second visit until the succeeding one The stepping up or stepping down therapy approa ch is part of asthma management, and it is partly contributed by the underlying severity and the level of symptomatic control of the disease. Thus, patients could switch from one drug to another based upon their disease severity. However the statistical a pproach will account for time varying exposure and time dependent confounding that is affected by exposure history Exposure Measurement Study exposures are identified using the Multilex Product Dictionary, which provide GPRD P roduct Codes and British Nati onal Formulary (BNF) Codes. Product Codes reflect product name, active ingredient(s), dosage strength, and dosage formulation (GPRD, 2011c); the BNF Codes reflect the BNF chapter under which the relevant product / drug class is classified (GPRD, 2011c ). The G PRD Product Browser t ool Version 1.3.2 (Multilex coding information: August 4, 2011) is utilized to retrieve P roduct and BNF C odes for study drugs. Inhaled LABA group include s formoterol fumarate, formoterol fumarate dihydrate, a nd salmeterol xinafoate; I CS group include s beclometasone dipropionate, budesonide, ciclesonide, fluticasone propionate, and mometasone furoate; and single device combination ICS/LABA group include s budesonide/formoterol fumarate dihydrate, and fluticasone propionate/salmeterol
104 xin afoate. The ICS/LABA group is defined by the prescription of a single device combination formulation, or the addition of LABA or ICS as a separate device formulation to a respective ICS or LABA monotherapy. Exposure to drugs of interest is defined on a m onthly basis by virtue of prescription information recorded in the therapy dataset of the GPRD The patient is considered exposed to the drug of interest when (1) she receives a prescription for that drug in that specific month, or (2) the anticipated end date of the previous prescription exceeds half (>1 5 days) of that specific month. If neither of these conditions are met, the patient is not considered exposed to the drug of interest at that specific month. Contingent on the three study specific aims, th e following exposure cohort definitions are proposed in an attempt to account for selection and measurement bias es that might otherwise arise : Cohort d efinition for s pecific a im n o. 1 ( m orbidity o utcome) F or estimating the effect of exposures on asthma mor bidity outcomes, patients who were prescribed study drugs are followed for 12 months from the date of the first pre scription (index date, month 0 ) until observing study ou tcomes or reaching previously discussed censoring criteria (Figure 4 1 ). Time depend ent confounders and outcome variables are measured d uring the post index date follow up year (month 0 to month +12) Baseline, time independent confounders are measured at the index date and the 12 months preceding period baseline year (month 0 to 12). Th e exposure definition in this cohort produces an inception cohort with ICS, inhaled LABA, and ICS/LABA combination therapy initiators, i.e. incident users.
105 As previously explained, a 12 month immortal person time was introduced during cohort structure, and directly influence morbidity outcomes, it however might introduce selection bias du e to informative censoring mainly because most of asthma deaths are found to occur before hospitalizations (BTS, 2009) ( page 37 ) To accoun t for this bias and to serve as a sensitivity analysis, the follow up profile for the mortality outcome in the specific aim number 2 was applied to the morbidity outcome in the first specific aim of the study (Figure 4 2) Cohort d efinition for s pecific a i m n o. 2 ( m ortality o utcome) I mmortal person time of > 12 months after the first prescription was introduced during exposure definition and patient selection process (Figure 4 2 ) To estimate the effect of exposures on asthma mortality outcome, patient follo w up (index date, month +12) is started from 12 months after the first prescription and continued for an additional 12 months after the corresponding index date (follow up year) until observing the outcome or reaching any of the censoring criteria mentione d above Time dependent confounders and outcome variables are measured during the post index date follow up period of > 12 months follow up year (month +12 to +24). Baseline, time independent confounders are measured at the index date and the 12 months prec eding period baseline year (month 0 to +12). The exposure definition in this cohort produces a prevalence cohort with ICS, inhaled LABA, and ICS/LABA combination therapy prevalent users. The degree of influence of the immortal time bias on the point estima te is contingent on the length of the immortal person time window and the incidence rate for asthma deaths in that window ( Zhou, Rahme, Abrahamowicz, & Pilote 2005). In the
106 current study, t he immortal person time window is relatively narrow for an outcome which is multifactorial in reason, e.g. behavioral and psychosocial factors (BTS, 2009) ; if these factors are accounted for at baseline, the magnitude of the bias on asthma deaths is expected to be small in this cohort of inc ident users who survived for 12 months post first use of LABA or ICS products Nevertheless, the bias is addressed in the follow up pr ofile discussed above (Figure 4 2 ). Per earlier discussion in the methodological challenges section, including prevalent u sers could induce selection bias, which can be minimized by restricting follow up to those patients who were not exposed to the drug of interest for a period of time, then became exposed after the end of the specific period, when they will be considered dr ug initiators after being unexposed. Figure 4 3 illustrates the proposed approach that will be used as a sensitivity analysis to previous approach which will be employed in future work The effect of exposures on asthma mortality outcome is assessed by fo llowing patients for a m aximum of 12 months after the index date of (1) 12 months after the first prescription for study drugs (index date, month +12) and (2) 3 months exposure free period preceding the index date (month +9 to +12), until observing the ou tcome of interest or reaching any of the censoring criteria stated above. The exposure free period is defined as the duration of time in which patients exposed to one exposure group (e.g. ICS) are switched to another exposure group (e.g. LABA or ICS/LABA) at the beginning of the exposure free period (month +9) and continued on the new regimen for three months, then switched back to the penultimate group (ICS) at the end of the exposure free period (month +12), at which patients are deemed re initiators of I CS. The same
107 scenario is applied to the LABA and ICS/LABA groups. Similar to earlier approach, time dependent confounders and outcome variables are measured during the post index date follow up period of > 12 months (month +12 to +24) and b aseline, time in dependent confounders are measured at the index date and the 12 months preceding period (month 0 to +12). A 3 month period is used in tandem with recommendations from asthma therapeutic guidelines to monitor patients for a minimum of 3 months to evaluate t herapeutic response and disease control (NHLBI, 2007; GINA, 2009). Therefore, the immortal person 12 month period is relatively short to have medication switching among incident users, and such patients are deemed quasi incident users. Accordingly, the mag nitude of the selection bias introduced by the quasi incident users will be minuscule, and the presented method will serve as a sensitivity analysis to assess this bias. In both scenarios, treatment switching as time dependent exposure is allowed aft er ind ex date to fit the proposed analytical models. Cohort d efinition for s pecific a im n o. 3 (s ubgroup m orbidity o utcome) To examine the effect of step down th erapy approaches and continuing combination therapy on asthma morbidity a subgroup of asthmatic patie nts who meet the following criteria is followed (Figure 4 4 ) : Initiators of ICS/LABA combination therapy who have their first prescription of ICS/LABA issued at the index date proposed in specific aim no. 1 (month 0). Continued exposure to ICS/LABA as sin gle device or separate device combinations for a minimum of 3 months after the first prescription until divergence to the three exposure subgroups (month 0 to +3). The divergence point will be the study index date (month +3), at which patients are grouped to t hree exposure groups:
108 1. Reduction of the original ICS dose by half and maintenance of inhaled LABA (50%ICS/LABA dose reducers). 2. Discontinuation of inhaled LABA and maintenance of ICS monotherapy original dose (ICS LABA stoppers). 3. Maintenance of ICS/LABA combination therapy with original ICS dose (Original ICS/LABA Incessant). Have evidence of asthma control during the immortal person time period (month 0 to +3) defined by the absence of prescription records for oral corticosteroids, and absence of event r ecords for asthma related A & E department visits or asthma related hospitalizations. If this criterion reduces the sample size, these variables will be accou nted for as baseline time independent covariates during the stated period to maintain statistical ef ficiency Patients w ill be followed for 12 months from index date (month +3 to +15) until observin g the outcome of interest or reaching censoring criteria stated above in addition to switching exposure to another exposure category. Thus, e xposure switching between the three groups is not permitted as a time dependent exposure, where patients might step up therapy from LABA discontinuation to ICS add on therapy (i.e. ICS/LABA) or from ICS dose reduction to ICS dose escalation Time dependent confounders and outcome variables are measured during the post index date follow up year (month +3 to +15) and the time independent confounders are measured at the index date and the 3 months prior period (month 0 to +3). ICS d ose q uantification. The average daily dose o f ICS is calculated from prescribed product strength and usage instructions recorded in the therapy dataset of the data base. Exposure group classification according to ICS dosage is based on the estimated equipotent daily doses of ICS according to GINA rec ommendations in Table 4 1 (GINA, 2009), where the higher stre ngth class is 50% of the medium strength class, which is also 50% of the low strength class. Accordingly, patients within the high and the medium strength ICS classes are considered original ICS dose continual
109 individuals, and when the high strength class members switch to the medium strength class and the medium strength class members switch to the low strength class, those patients are considered members of the dose reducer group. Table 4 2 lis ts the commercially available dosage strengths of ICS and ICS/LABA products in the UK Outcome Measurement The outcomes of interest for the study are asthma related mortality and asthma related morbidity rates, which are measured during the follow up peri od after the specified index dates for the corresponding specific aims. Refer to cohort definition sections corresponding to mortality and morbidity outcomes for details on outcome follow up post index dates. Mortality outcome Asthma related mortality is d efined as asthma death and identified from the linked ONS Mortality Data, and from the GPRD clinical data files of the medical records. Asthma related mortality is assessed in two methods. First, for patients who are in general practices that are linked wi th the ONS Mortality Data, the International Classification of Diseases, tenth version (ICD 10) is used to classify causes of death in the ONS Mortality Data. The following ICD 10 codes for asthma related mortality are used to identify cases with causes of death that are asthma related : J 45 for asthma; J 45 .0 for predom inantly allergic asthma; J45.1 for non allergic asthma; J45.8 for mixed asthma; J45.9 for unspecified asthma; and J46 for status asthmaticus (WHO, 2007). Non neonatal causes of deaths are iden tified in the linked subset of mortality data. The ICD 10 codes in the ONS Mortality data are recorded without the decimal, e.g. J24 0
110 instead of J24.0 which are recorded in such manner for technical purposes (ONS, 2009). The second method includes all gen eral practices using the following algorithm, which involves using the GPRD Medical Browser tool Version 1.3.2 (Read coding information: November 27, 2009) to extract Read Clinical Terms and corresponding Medical Codes that identify asthma related codes i n clinical, referral, and patient data sets within +/ 21 days from the da te of death (GPRD, 2011c). The 21 day period reflects t he average lag in recording cause s of death in medical records by the general practice (Tjeerd van Staa, GPRD, personal communica tion, May, 9, 2011). Identifying cause of death in the GPRD using this method is representative to the UK national death registry where about 65% of causes of death retrieved from the GPRD records are found identical to national records (Shah & Martinez, 2004 ). A study validated asthma deaths in GPRD records versus death certificates, showed that 34 cases of the identified 77 asthma deaths were verif ied by death certificates (Lanes, Garcia Rodr g uez & Huerta, 2002). Yet, the linked ONS mortality dataset im proves the corroboration. The linked dataset method will be used a s a sensitivity analysis to the records method Morbidity outcomes Asthma r elated morbidity reflects poor asthma control, and defined by asthma related A & E department attendances, asthma rel ated hospitalizations, and prescriptions for oral corticosteroids (OCS) including short courses for the treatment of asthma exacerbations The GPRD Medical Browser tool is used to identi fy relevant Medical Codes for Read Clinical Terms from clinical, cons ultatio n, and referral GPRD data files
111 to reflect asthma A & E visits and asthma hospitalizations Prescriptions of oral corticosteroids are identified by the P roduct C odes from therapy datase ts using the GPRD Product Browser tool. In addition, an algorithm was developed to identify prescriptions for short courses of oral corticosteroids based on whether a prescription was part of a repeat schedule or a one off prescription, where the latter is considered an indicator of a short course corticosteroids regimen This algorithm is used as a sensitivity analys is for outcome definition Covariate Measurement Both GPRD Product and Medical Browsers are utilized to respectively identify Product and Medical Codes for the covariates of interest Table 4 3 serves as a co variate definition exhibit. The following domains and corresponding variables are included in the analysis models as covariates that influence the relationship between study exposure s and outcomes : Patient characteristics Patient characteristics at index d ate include the following demographic, behavioral, lifestyle, and socioeconomic factors: (13 65 years ) sex (female; male) and marital status ( unmarried; married ; and unknown status) ; duration of registration with the practice, which is defi ned as the duration in months from registration date until the study index date; exemption status for prescription payment (not exempted; exempted; and unknown status) level of capit ation supplement ( low; medium; high; not applicable; and unknown status ) ; and baseline weight status, which is determined by a mixture of measures including Read clinical terms and calculated values of the B ody M ass I ndex (BMI) in kg/m 2 categorized according to the
112 guidelines for the identification of individuals with weight problems ( underweight, BMI<18.5, normal weight, BMI=18.5 2 4.9 ; overweight, BMI= 25 29.9; obese, BMI= 30+; and unknown weight status ) (NIH, 1998) yet, weight status was classified in to obese, non obese, and unknown status in order to suppress categories with zero values in the non obese class, which include normal weight, underweight, and overweight statuses In addition, former smoker, passive smoker, and unknown smoking status) is included as covariate after excluding patients who were active smokers during the study follow up Noteworthy lifestyle information including body weight and smoking are not consistently recorded in the GPRD (Garcia Rodr guez & Gutthann, 1998) therefore added to corresponding variables Practice characteristics There are differences in population demographics, healthcare service utilization, and levels of primary care capitation payment across the four countries of the UK ( Rhys Beerstecher & Morgan 2010 ) Therefore, practice location is included as a covariate (England; Scotland; Wales; Northern Ireland; and unknown location). In addition, c onsultation length has variable effect on prescribing behavior Duration of consultation is defined as the length of time in minutes between the opening and closing of the consultation record by the GP. In general, the average duration of consultation in the UK is 9.4 minutes (95%CI, 0.47 18.3) (Deveugele, Derese, van den Brink Mu inen, Bensing, & De Maeseneer, 2002). In another study, the mean duration varied according to patient satisfaction with the GP: highest satisfaction, 9.48 minutes (95%CI, 1.79 20.75); lowest satisfaction, 9.4 minutes (95%CI, 7.92 26.72) (Cape, 2002). Based on
113 these values, the length of consultation with the patient is categorized to < 10 minutes and >10 minutes. covariate (non urgent visit; urgent visit; unknown status ). Although the visi ts could be unrelated to asthma unscheduled emergency visits might influence prescribing behavior. Asthma severity As previously discussed confounding by disease severity and channeling bias are forms of confounding by indication that are commonly presen ted in chronic disease pharmacoepidemiology and f ailure to account for baseline and time varying severity of asthma could result in casual association s between the exposure of interest and worsened asthma outcomes (Nelson, 2006). The following measures ar e used to account for asthma severity at baseline and study follow up In addition to disease symptoms, spirometry and peak expiratory flow rates (PEFR) are the best measures of asthma severity ( Ng, 2000; NHLBI, 2007; BTS, 2009 ; Kim & Mazza, 2011 ); however spirometry information and patient reported outcomes, e.g. symptoms and PEFR are scarcely and inconsistent ly reported in the GPRD (Garcia Rodr guez & Gutthann, 1998; Griffin, Lee, Caiado, Kesten, & Price 2008 ). Therefore, other proxy measures of asthma severity are utilized, which are applied in similar observational designs ( Meier & Jick, 1997; Lanes Lanza & Wentworth, 1998; W illiams et al 1998; Lanes Garcia Rodr g uez & Huerta, 2002; Anderson et al 2005; Blais, Beauchesne & Forget, 2009 ; Thomas von Ziegenweidt, Lee, & Price 2009; de Vries, Setakis, Zhang, & van Staa 2010; Guo, Tsai, Kelton, Bian, & Wigle 2011 ; Wells, Peterson, Ahmedani, Severson, Gleason Comstock, & Williams 2012 ). Nevertheless,
114 these designs failed to account for the time dependent nature of asthma severity, which are predicted by exposure history, and act as predict ors of future exposure and outcomes A lthough one study ( Wells, Peterson, Ahmedani, Severson, Gleason Comstock, & Williams 2012) utilized Cox PHREG with SABA p rescriptions as time dependent covariate the approach inadequately controlled for time dependent confounding ( Suarez, Borrs & Basagaa 2011) These measures consist of indicators of drug and healthcare utilization Drug utilization measures include pres criptions for inhaled SABA ; prescriptions for oral corticosteroids within 12 months prior to the index date (baseline year) ; number of prescriptions for inhaled SABA in the baseline year; and number of prescribed categories of asthma drugs at ind ex date. W hen combination products are prescribed in a single device formulation, the prescription is deemed as for two asthma drug classes where every ingredient serves as a separate pharmacological entity e.g. inhaled ICS/LABA Inhaled SABA and asthma drug categ ory utilization are updated every month as time dependent confounders in addition to their measurements in the baseline year SABA utilization is measured as average daily doses of inhaled SABA ( Thomas, von Ziegenweidt, Lee, & Price 2009), and frequently categorized by the average number of canisters prescribed or dispensed (Suissa, Blais & Ernst, 1994), or by the number of refills in a specific time period ( Wells, Peterson, Ahmedani, Severson, Gleason Comstock, & Williams 2012). The present study will in clude SABA utilization as a categorical measure that is updated in timely manner similar to the exposure of interest. This approach will minimize the
115 variation in calculating the average daily doses that are contingent upon the individual SABA product and estimated doses per actuations. Healthcare utilization measures that reflect asthma severity include asthma related hospitalization and A & E visits during the baseline year Hospitalization due to asthma exacerbations in the preceding year predicts the risk s for asthma exacerbations, hospitalizations, and asthma mortality in the future (NHLBI, 2007; Omachi, 2009). Prescribing >2 classes of asthma medications, >2 prescriptions for oral corticosteroids, >6 prescriptions for inhaled SABA, and a hospitalization or an A & E visit for asthma in one year prior to exposure or one year prior to a worsened outcome are recognized risk factors for increased asthma morbidity and mortality (Lanes Lanza & Wentworth, 1998; Williams et al 1998; NHLBI, 2007; BTS, 2009; Blais, Beauchesne & Forget, 2009 ; Wells, Peterson, Ahmedani, Severson, Gleason Comstock, & Williams 2012 ). Patients with controlled asthma at baseline are defined as not having more than 2 asthma drug classes or any inhaled SABA as rescue bronchodilators presc ribed at the index date, and not having any of the following during 12 months before the index date: prescriptions for oral corticosteroids, more than 6 prescriptions for inhaled SABA, or attending accident and emergency departments or hospitalizations for asthma. An indicator variable for asthma control is included as a covariate in baseline variable pool. C oncurrent asthma drug prescriptions Concurrent prescriptions for the following asthma medication cla sses are included as covariates : inhaled SABA; oral leukotriene receptor antagonists (LTRA); oral sustained release methylx anithines; mast cell stabilizers (MCS) ; and inhaled muscarinic
116 receptor antagonists ( MRA ) Oral SABA, the 5 lipoxygenase inhibitor, zileuton; and the anti IgE, omalizumab are not inclu ded as covariates. Oral SABA is not part of asthma treatment in the clinical guidelines, zileuton is not marketed in the UK, and there was only one patie nt who was prescribed omalizumab who was excluded from the sample Other concurrent prescriptions Presc riptions for t he following medication classes could act as effectiveness modifiers and might interfere with asthma outcomes; hence, are included as covariates in the analysis models: oral and parenteral antibiotics for respiratory tract infec tions (RTI); o ral parenteral and inhalational antivirals for RTI e.g. oseltamivir ; nasally administered products, including nasal corticosteroids, nasal MCS e.g. sodium cromoglicate nasal antihistamines, and nasal decongestants ; antitussives including expectorants a nd opioid based antitussives ; selective beta 1 adrenoceptor blockers, including oral and ophthalmic formulations and combination antihypertensive products ; all routes non steroidal anti inflammatory drugs (NSAIDs); oral and rectal aspirin ; oral, rectal, an d parenteral acetaminophen (paracetamol); all routes opioid analgesics, including nasal opioids and acetaminophen combination analgesics, excluding codeine based antitussives, which are included in the antitussive covariate ; and oral cholinergic agents, e. g. neostigmine bromide Although acetaminophen is not considered an NSAID, the epidemiological and clinical studies suggest an association between exposure to acetaminophen and worsened atopy, including wheezing and asthma (Beasley et al., 2011; McBride, 2 011); however, there are arguments against these findings (Chang, Leung, Tam, & Kong,
117 2011). Therefore, acetaminophen is included as a covariate to account for any potential residual confounding. Since o phthalmic cholinergic agents (miotics) have small sys temic bioavailability, they are not included as covariates. Antihistamines have no effect on lung allerg ies compared to skin allergies, where loratadine blocked skin allergy but failed to show an effect on the lungs (Town & Holgate, 1990) ; therefore, they are not considered as covariates However, n asally administered antihistamines, including combination with nasal corticosteroids are included as covariates because the route of administration might have an effect on triggering asthma exacerbation. In addit ion, placebo controlled clinical trials in asthmatic patients showed contradicting results for the effect of tumor necrosis f actor e.g. etanercept on asthma outcomes ( Morjaria, Babu, Holgate, & Polosa, 2006; Berry, Brightling, Pavord, & Wardlaw, 2007; Holgate et al., 2011 ). Large scale studies with longer follow up duration are recommended to fully unde rstand the role of anti asthma control. In the current study, there were no prescriptions for anti at the index date, and only 14 prescriptions during the study follow up. The effect of these agents on asthma remains a matter for future work. Con comitant immunizations Influenza vaccination protects against all cause mortality in COPD patients (GOLD, 2010), and it is recommended in asthmatic patients older than 6 months of age to protect against flu related complications of asthma (NHLBI, 2007). Pneumococcal polysaccharide vaccination is recommended in COPD patients 65 years and older (GOLD, 2010). However, in June 2008, the US Advisory Committee on Immunization
118 Practices (ACIP) advised healthcare providers to administer the vaccine to as thmatic adults aged 19 64 years (CDC, 2010, 2012) Immunization information at the time of prescribing study d rugs is included as a covariate, which include: prescribed influenza vaccine, including nasal and parenteral preparations; pneumococcal polysaccharide vac cine; and other immunizations regardless of vaccine ty pe. Comorbid conditions Various coexisting medication conditions are associated with asthma and could interfere with asthma control and therapy outcomes ( Sherrill, Guerra, Bobadilla, & Barbee 2003; Bou let, 2009; Bush & Zar, 2011). The following comorbid conditions are identified as covariates: atopy, respiratory tract infections (RTI), and psychosocial pathologies. Atopic conditions include allergic rhinosinusitis, allergic conjunctivitis, atopic dermat itis, psoriasis, respiratory allergies, and other allergic conditions, e.g. angioedema and food allergies RTI include otitis media, pharyngolaryngitis, influenza, acute bronchitis, pneumonia, and other unspecified RTI P sy chosocial pathologies include anx iety, depression, affective personality disorders like schizophrenia and psychosis, and other conditions such as deliberate self harm; suicidal ideation and attempts; alcohol and drug abuse; learning difficulties; employment, income, marital and legal prob lem; and social isolation. P sychological problems are further identified by prescriptions for tranquilizers, antidepressants, and antipsychotics. Psychosocial problems put asthmatic patients at risk of fatal or near fatal asthma (NHLBI, 2007). G astro es opha geal reflux disease (GERD) was not included as a covariate, since scientific evidence argues against the influence of GERD on asthma symptoms (DiMango et al., 2009), and studies showed that treating GERD with proton pump
119 inhibitors had no effect on asthma control in children and adults with coexisting asthma and GERD ( ACRC et al., 2009, 2012 ). Other factors The BTS asthma management guidelines were updated approximately eight times from 1993 to 2009 (BTS, 2011); and the Quality and Outcomes Framework (QOF) was implemented by the National Health Service (NHS) in April 1, 2004 as an annual incentive program that rewards general practitioners based on point indicators achieved by ma naging some chronic conditions including asthma, in addition to other factors, e.g. level of practice organization; patient experience at the practice; and the availability of additional services that are offered at the practice, e.g. maternity and child health services (NHS, 2011 b ) ; it is known that changes in asthma severity are in ter seasonal, and asthma deaths are found to increase in July and August among patients younger than 45 years and in December and January among older patients in the UK ( BTS, 2009 ) Therefore, the influence of seasonality on disease severity and changes in practice guidelines on disease management can be accounted by including calendar time in annual quarters as a covariate in the analyses. A n annual quarter is defined as the quarter of the calendar year in which the prescr iption of study drug was issued w hich will be measured at the index date and during the follow up year e.g. for a LABA prescription issued on February 12, 20 05 and an ICS prescription issued on April 28, 2010, the corresponding quarter s are QT1 and QT2, respectively (QT1: January March; QT2: April June; QT3: July September; & QT4: October December) The presence of asthma personal management plan (also known as asthma action plan) for well motivated patients plays a role in reducing asthma hospitalizations and
120 improving overall asthma co ntrol (Toelle & Ram, 2004) and motivating patients to maintain correct inhaler technique (Ovchinikova, Smith & Bosnic Anticevich, 2011) Patients without asthma action plans are four times at risk of hospitalization due to asthma attacks than counterpart s with asthma action plans (Asthma UK, 2010). Read clinical term signifying the presence or absence of asthma action plans at index date is identif ied to represent this covariate (Plan unavailable; plan available; & unknown) In addition, clinical terms re flecting the utilization of humidifying inhalational therapy (unavailable versus available) medications (regardless of the medication type), and the level of general compliance (satisfactory; unsatisfactory; & unknown) are identified by relevant Read clinical terms and included as covariates at the study index date. No recommendation for humidifying inhalational therapy was made at the index date, thus, the variable is not included in the covariate pool. The ty pe of inhaler device used to deliver asthma medications into the lungs affects patient adherence, inhaler technique, and asthma clinical outcomes ( Takizawa, 2009; Price et al 2011 a 2011 b ); likewise, prescribing and utilizing a spacer device in addition to the inhaler product plays a role in patient adherence and subsequent asthma outcomes (Berger, 2009). Therefore, the following variables are included in the covariate pool: Prescription for a compact spacer or holding chamber device at the index date (no t issued versus issued); prescription for a nebulizer at the index date (not issued versus issued); and device type at the index date, including pressurized metered dose inhaler (pMDI) ; breath actuated inhaler (BAI), e.g. aerosols ; and dry powder inhaler ( DPI), e.g. disks and capsules This information is derived from the
121 prescribed exposure product and extracting relevant information from the BNF (www.bnf.org) or when available W hen the ICS/LABA combination therapy is prescribed in two separate devices with two separate device types, the ICS device type is assigned to the category. Clinical judgment is used to compensate the combination device type with the anti inflammatory agent, which is the mainstay of asthma therapy. The pres cribed device types are classified according to the available formulations in the UK, and are categorized by the study dr ugs of interest: ICS ( pMDI; BAI; & DPI); LABA (pMDI & DPI); & single device ICS/LABA (pMDI & DPI). Note that BAI formulations are presc ribed for ICS product s, and in order to efficiently use the device type variable as a covariate across exposure categories, both the pMDI and BAI are merged to form a metered dose inhaler category (MDI) that will be used as a value for the device type cova riate at the index date (MDI versus DPI). Sample Size and Power Calculations Assuming similar number of patients in the three exposure groups and the samples are independently selected, the necessary sample sizes to detect the difference at two sided signi T able 4 4 The calculations di d not account for MSM procedure which permits exposure switching by default; therefor e, the estimated sample size is an overestimation of the size required f or the MSM Statistical Analysis Procedures Statistic al analyses and data presentations are performed using SAS software, Version 9.3 of the SAS System for Windows (2011 SAS Institute Inc., Cary, NC, USA ), R s oftware, Version 2.1 4 1 of the R Environment fo r Windows (2011 The R Foundation
122 for Statistical Computing, Vienna, Austria) and Microsoft Office Word software (2010 Microsoft Corporation, Redmond, WA, USA) Two sided tests with alpha=0.05 a priori level of statistical significance are used throughout the analysis procedures. Descriptive Statistics Within each study specific aim b aseline characteristics of the three exposure groups are compared using standard univariate statistical methods. For categorical variables, tests are pe rformed to compare the three groups. Alternatively, exact test is used to compare categorical variables when the expected frequencies of observations is small er than 5 ( Glantz, 2005 ) For continuous variables, one way analysis of variance (ANOVA) is employed for comparing the groups. Categorical variables are described by proportions and respective 95% confidence intervals; means and co rresponding standard deviations are reported for continuous variables. Inferential Statistics For every specific a im, three regression models are constructed and the unadjusted and adjusted estimates of association between exposure and outcomes are compared across the models. The models include conventional Cox PHREG, time dependent covariate Cox PHREG, and MSM. In ea ch model, the hazards ratio (HR) and the respective 95% confidence interval (CI) are calculated f or the three comparison groups: LABA monotherapy vs. ICS monotherapy ICS/LABA combination therapy vs. ICS monotherapy ICS/LABA combination therapy vs. LABA mon otherapy Among ICS/LABA combination therapy :
123 1. ICS dose reducers vs. ICS original dose continuers 2. ICS dose reducers vs. LABA stop p ers 3. LABA stoppers vs. ICS original dose continuers The incidence rates of study outcomes are compared across exposure groups and are calculated in terms of person time at risk, which defined as the period of time in which the observed individuals are at risk of developing the outcome of interest ( Waning & Montagne, 2001; Greenland & Rothman, 2008) The incidence rates are presented as the number of events per 1000 person month for the four outcomes of interest in the respective study research questions : asthma deaths, asthma A & E visits, asthma hospitalizations, and prescriptions for oral corticosteroids Conventional Cox PHREG model varying variables; the model in E quation 3 1 will evaluate the degree and extent of exposure association with the outcomes of interest by terminating patient fol low up upon exposure switching in additi on to other censoring criteria and all the covariates will be measured at the index date as baseline measures and in the 12 months prior (3 1 ) Time dependent covariate Cox PHREG m odel Time dependent Cox model allows for exposure switching and partially accounts for time dependent confounding and evaluate time varying nature of exposure on the association between exposure and outcomes The f ollowing two models are tested: The model in E quation 3 2 with setting exposure as time depe ndent variable, but all the other covariates as fixed, time independent variables. These covariates are measured
124 at the index date, and in the 12 months before. We will call this model time dependent exposure and fixed confounders ( 3 2 ) The variables in the vector include patient characteristics; practice characteris tics; asthma severity measures; asthma co medications; other co medications ; co immunizations; comorbidities; annual quarter in which therapy pre scriptions are issued; presence of asthma action plan ; inhaler device type; and prescription for compact spacer The second model in E quation 3 3 involves setting both exposure and covariates as time dependent factors, except for the baseline covariates th at are measured at the index date and during the 12 months prior. We will call this model time d ependent exposure and confounders ( 3 3 ) The variables in the vector include patient characteristics; practice l ocation; prescription for oral corticosteroids and number of inhaled SABA prescriptions within 12 months before the index dat e of exposure prescription; presence of asthma action plan ; and prescription for compact spacer The variables in the time depende nt vector include the duration of GP consultation with the patient at every prescribing session; prescription for inhaled SABA; number of prescribed asthma medication classes; asthma co medications ; other co medications; comorbiditie s; annual quarter in whi ch the prescription was issued ; and inhaler device type
125 The time independent exposure variable in the vector include following patients who were prescribed one exposure group until the development of the out come of interest, reaching censoring criteria, or switching to another exposure group with a new prescription for the alternative exposure. On the other hand, the time dependent vector for the exposure involves continuous patient fo llow up when exposure switching takes place, until the development of the study outcome or reaching censoring criteria. Marginal structural model The effect of the exposure on the outcomes of interest are assessed by fitting a MSM that accounts for time de pendent confounding that is affected by time dependent exposure. E quation 2 16 serves as the final outcome model that includes time varying exposure and baseline confounders weighted by the influence of time dependent confounders. Time dependent covariates Time dependent confounders representing the vector include the duration of GP consultation with the patient at every prescribing session; prescription for inhaled SABA; number of prescribed asthma medication classes; asthma co medi cations; other co medications; comorbidit ies; annual quarter in which the prescription was issued ; and inhaler device type These variables are updated on monthly basis. Time independent covariates. The fixed confounders representing the vector include patient characteristics; practice location; prescription for oral corticosteroids and number of inhaled SABA prescriptions within 12 months before the index dat e of
126 exposure prescription; presence of asthma action plan ; and prescriptio n for compact spacer Sensitivity Analyses Sensitivity analyses imply examining the effect of changing study parameters on the results ( Greenland, 1996 ). Sensitivity analyses are conducted to a ssess the uncertainty of study findings to changes in cohort de finition in terms of including patients with COPD diagnosis characterization of asthma related outcomes in terms of including selective clinical terms that reflect different asthma definitions characterization of asthma mortality outcome in the GPRD reco rds for patients not included in the ONS data, characterization of asthma morbidity outcome in terms of hospitalization and A & E visits by including prescriptions for oral corticosteroids as a severity measure instead of an outcome, and testing MSM assumpti ons of exchangeability (Brumback, Hernn, Haneuse, & Robins 2004) and positivity ( Mortimer, Neugebauer, van der Laan, & Tager 2005)
127 Table 4 1. Estimated equipotent daily doses of inhaled corticosteroids for asthmatic adults (>12 years old) according to the GINA 2009 recommendations Inhaled corticosteroid Daily dose range (mcg) Low strength Medium strength High strength Beclometasone dipropionate 200 500 501 1000 1001 2000 Budesonide 200 400 401 800 801 1600 Ciclesonide 80 160 161 320 321 1280 Fluticasone propionate 100 250 251 500 501 1000 Mometasone furoate 200 400 401 800 801 1200 GIN A, global initiative for asthma Table 4 2. Dosage strengths of inhaled corticosteroids (ICS) monotherapy and ICS/long acting beta agonist (LAB A) combina tion therapy available in the UK ICS Strength (mcg/dose) Beclometasone dipropionate 50, 100, 200, 250, 400 Budesonide 50, 100, 200, 400 Ciclesonide 80, 160 Fluticasone propionate 25, 50, 100, 125, 250, 500 Mometasone furoate 200, 400 ICS/LABA Budes onide/Formoterol fumarate dihydrate 100/6, 200/6, 400/12 Fluticasone propionate/Salmeterol xinafoate 50/25, 125/25, 250/25, 100/50, 250/50, 500/50
128 Table 4 3. Variable definitions Domain Variable Values Measurement Time Exposure Switc hing during follow up year 0: ICS 1: LABA 2: ICS/LABA Study aim no. 1 index date & follow up year Switching during follow up year 0: ICS 1: LABA 2: ICS/LABA Study aim no. 2 index date & follow up year No switching during follow up year 0: Original dose ICS/LABA 1: 50% ICS dose reducer 2: LABA stopper Study aim no. 3 index date Outcome Prescription for oral corticosteroid s 0: Not issued 1: Issued Study follow up year Asthma related visit for accident & emergency dept. 1: Ev ent occurred Asthma related inpatient hospital attendance 1: Event occurred Asthma related death 1: Event occurred Asthma severity Prescription for oral corticosteroids 0: Not issued 1: Issued Study bas eline year Asthma related visit for accident & emergency dept. 1: Event occurred Asthma related inpatient hospital attendance 1: Event occurred Number of prescriptions for inhaled SABA >0 prescriptions issued Prescription for inhaled SABA 0: Not issued 1: Issued Study index date & follow up year Number of asthma drug classes prescribed >0 classes Concurrent immunization Prescription for influenza vaccine 0: Not issued 1: Issued Study index da te & follow up year Prescription for pneumoco ccal polysaccharide vaccine 0: Not issued 1: Issued Prescription for other vaccines 0: Not issued 1: Issued
129 Table 4 3 C ontinued Domain Variable Values Measurement Time Concurrent asthma drugs Prescri ption for oral leukotriene receptor blocker 0: Not issued 1: Issued Study index date & follow up year Prescription for oral methyl xanthines 0: Not issued 1: Issued Prescription for inhaled mast cell stabilizer 0: Not issued 1: Issued Prescription for inhaled muscarinic receptor blocker 0: Not issued 1: Issued Other co medications Prescription for antibiotics for respiratory infections 0: Not issued 1: Issued Study index date & follow up year Prescription for antivirals for respiratory infe ctions 0: Not issued 1: Issued Prescription for nasal corticosteroids 0: Not issued 1: Issued Prescription for nasal mast cell stabilizers 0: Not issued 1: Issued Prescription for nasal antihistamines 0: Not issued 1: Issued Prescription for na sal decongestants 0: Not issued 1: Issued Prescription for antitussives & expectorants 0: Not issued 1: Issued Prescr iption for oral & ophthalmic selective beta 1 blockers 0: Not issued 1: Issued Prescription for NSAIDs 0: Not issued 1: Issued Prescription for aspirin 0: Not issued 1: Issued Prescription for opioid analgesics 0: Not issued 1: Issued Prescription for oral cholinergics 0: Not issued 1: Issued Prescription for acetaminophen 0: Not issued 1: Issued Prescription for tumor necrosis factor 0: Not issued 1: Issued
130 Table 4 3 C ontinued Domain Variable Values Measurement Time Concomitant clinical conditions Atopy 0: Not recorded 1: Recorded Study index date Respiratory infections 0: Not recorded 1: Recorded Psychosocial pat hologies 0: Not recorded 1: Recorded Patient & practice characteristics 13 65 years Study index date 0: Female 1: Male 0: Unmarried 1: Married 2: Unknown 0 : Non obe se 1 : Obese 2 : Unknown 0: Nonsmoker 1: Former smoker 2: Passive smoker 3: Unknown Patient exempted from prescription payment 0: Not exempted 1: Exempted 2: Unknown Level of capitation supplement the patient has 0: Low 1: M edium 2: High 3: Not applicable 4: Unknown registration with the practice >0 months consultation with patient 0: 10 minutes & shorter 1: Longer than 10 minutes Urgency the prac tice 0: Non urgent visit 1: Urgent visit 2: Unknown General practice location in the UK 0: England 1: Scotland 2: Wales 3: Northern Ireland 4: Unknown
131 Table 4 3 C ontinued Domain Variable Values Measurement Time Other factors Annual quarter in which prescription for exposure of interest was issued 0: 1 st January March 1: 2 nd April June 2: 3 rd July September 3: 4 th October December Study index date & follow up year Presence of asthma action plan 0: Not available 1: Available 2: Unknown Study ind ex date Humidifying inhalational therapy 0: Not available 1: Available asthma medications 0: Satisfactory 1: Unsatisfactory 2: Unknown compliance 0: Poor 1: Good 2: Unknown Prescription for sp acer 0: Not issued 1: Issued Prescription for nebulizer 0: Not issued 1: Issued Inhaler device type for exposure of interest ICS 0 : pMDI 1 : BAI 2 : DPI 3: Unknown Study index date LABA 0: pMDI 1: DPI 2: Unknown ICS/LABA single device 0: pMDI 1: DPI 2: Unknown Exposure of interest 0: MDI 1: DPI 2: Unknown
132 Table 4 4. Estimated sample size of study population Outcome Exposure group (incidence, %) Sample size per group LABA ICS/LABA Asthma death 0.067 0.012 20,489 Asthma hospitali zation 1.516 0.394 1,179 ICS ICS/LABA Asthma death 0.016 0.012 -Asthma hospitalization 1.000 0.394 2,958 LABA ICS Asthma death 0.067 0.016 25,035 Asthma hospitalization 1.516 1.000 7,323 ICS, inhaled corticosteroids LABA, l ong acting beta agonists
133 Figure 4 1. Study follow up profile for morbidity outcome (Study Aim No. 1) Months 0 12 Index Date Maximum Follow up +12 LABA Initiators ICS Initiators ICS/LABA Initiators +24 First Prescription Baseline Year Follow up Year Study Duration
134 Figure 4 2. Study follow up profile for mortality outcome (Study Aim N o. 2) Months 0 12 Index Date Maximum Follow up +12 LABA Prevalent Users ICS Prevalent Users I CS/LABA Prevalent Users +24 First Prescription Baseline Year Follow up Year Study Duration Immortal Person Time Period
135 Figure 4 3. Study follow up profile for mortality outcome among prevalent users illustrating the design to adjust for selection bias Months 12 Index Date Maximum Follow up +12 LABA Reinitiators ICS Reinitiators ICS/LABA Reinitiators +24 First Prescription Baseline Year Follow up Year Study Duration Immortal Person Time Period ICS Initiators LABA Initiators ICS/LABA Initiators +9 Exposure Free (Switching) Period
136 Figure 4 4. Study follow up profile for subgroup analysis for morbidity outcome (Study Aim No. 3) Months 0 12 Index Date Maximum Follow up +15 ICS/LABA Original Dose Continuers 50%ICS/LABA Dose Reducers ICS/L ABA Initiators +24 First Prescription Baseline Year Follow up Year Study Duration +3 ICS Monotherapy: LABA Stoppers Immortal Person Time Controlled Asthma Period
137 CHAPTER 5 RESULTS Descriptive Statistics Characteristics of study population are described in the following sections of exclusion c ohort original cohort study sample, and step down therapy sub group study sample. Exclusion C ohort Medical records for patients with asthma ar e retrieved from the GPRD with 308,839 initiators of the study drugs ( inhaled LABA, ICS, or combined ICS/LABA inhalers ) who meet the inclusion criteria described above Patient disposition is shown in figure 5 1 where exclusion numbers in the extension rectangles are mutually not exclusive because patients might have more than one exclusion criterion. The majority of patients are excluded from the study cohort (257,73 6 ). About 1 0% of the patients received study drugs on or before January 1, 1993. 22.6% of the patients were aged either younger than 13 (12,669) or older than 65 (57,162). Patients with COPD comprised less than fourth of the retrieved records (70,394). Respiratory di agnostic and therapeutic procedures except lung function tests (5,424) lung transplantation with or without heart (40) and total or partial removal of lung lobes (830) involved in 2% of the patients. Lung diseases due to occupational exposure to chemical s and fumes, e.g. w ere reported in 2,482 patients. Patients with infectious lung diseases accounted for 4.7% of the population including prescriptions for relevant antibacterial and antifungal agents ever prescribed ( 5,241 asthmatics with tuberculosis, 9, 393 asthmatics with aspe rgillosis, and 9 patients with pneumocystis pneumonia). Three thousand 250 patients had benign or malignant
138 neoplasm of the respiratory system, including tumors affecting the larynx, trachea, bronchi, bronchioles, lungs, and pleural cavity. Cystic fibrosis was found in 323 patients, parenchymal lung disease, e.g. pneumoconiosis was found in 537 patients, and obstructive sleep apnea was reported in 4,069 patients. Respiratory insufficiency due to obstruction by foreign objects was rec orded in 251 asthmatic patients. Among asthmatics, 5,427 had destruction and widening of the bronchi (bronchiectasis) and partial or complete lung collapse (atelectasis). Congenital and structural anomalies of the respiratory system, including anomalies of the ribs, diaphragm, and intercostal muscles are found in 1,439 patients. Asthmatics with c ardiovascular and respiratory circulatory disorders accounted for 5.3% of the patients: pulmonary hypertension, pulmonary embolism, or pulmonary edema (5,349); cong estive heart failure (10,441); congenital heart disease (535); and valvular heart disease affecting the pulmonary valve, e.g. pulmonary atresia (108). Approximately 9% of the patients had unspecified respiratory disorders (26,595), including injuries to th e respiratory system, e.g. rib fracture, and conditions other than the ailments specified above. Eight thousand 250 patients were reported to participate in clinical studies, including asthma research. Nearly half of the patients were active cigarette smok ers (152,677 ), and only 5,430 patients were reported to use recreational drugs and illicit substances, including glue sniffing and marijuana smoking. Patients with prescription based exclusion criteria comprised approximately 16% of the population. About 7 % of asthmatics were prescribed non selective beta adrenoceptor blockers, in volving ophthalmic products for glaucoma (2,674) and oral products (19,954), including combination antihypertensive
139 formulations (e.g. diuretics combined or calcium channel blocker combined) Among the population, 8.5% asthmatics were prescribed inhaled SABA single device combination products, including 23,431 patients with inhaled MRA/SABA; 2,172 patients with inhaled ICS/SABA; and 698 patients with MCS/SABA. Patients with prescrip tions for inhaled betamethasone (145), allergen immunotherapy vaccination (27), and omalizumab (1) were excluded. In addition, four asthmati c patients with indeterminate sex were excluded from the study. Original C ohort Limiting the GPRD datasets based on the exclusion criteria described above produced a study cohort of 51,10 3 patients with asthma who have initiated the study drugs between January 4, 1993 and August 2 0 2010, and who have survived for at least one year after initiation of study drug (Figure 5 1) T he majority of the cohort were prescribed ICS (46,92 8 ) followed by ICS/LABA combination products (3,461), and inhaled LABA products (714). Among the combination group, the majority of patients were prescribed single device combination formulations (2,692) compared to separate devices (769). Beclomethasone accounted for the majority of ICS monotherapies (n=42,328; 90.2%), followed by budesonide (n=3,097; 6.6%), fluticasone (n=1,361; 2.9%), ciclesonide (n=94; 0.2%), and mometasone (n=47; 0.1%). Salme terol was prescribed more than formoterol as inhaled LABA (n=674; 94.4% versus n=4; 5.6%). Most of the ICS/LABA combination therapy prescriptions were for fluticasone/salmeterol (n=2,208; 63.8%) compared to budesonide/formoterol (n=1,253; 36.3%). The study population was followed for an average of 1 year after exposure. During the follow up year, patients received a total of 176,723 ICS prescriptions, 7,468 inhaled
140 LABA prescriptions, and 33,371 prescriptions for ICS/LABA combination inhalers. There was a c onsistent distribution of individual products within exposure class, e.g. about 92% of ICS prescriptions were for beclomethasone, 94% of LABA prescriptions were for salmeterol, and 64% of ICS/LABA prescriptions were for fluticasone/salmeterol. However, the distribution of switching products was different across exposure classes. During the follow up year, the majority of ICS initiators continued on ICS monotherapy, and only 15 asthmatics switched to LABA monotherapy as salmeterol (0.03%) compared with 10,37 1 (22.1%) asthmatics switched to ICS/LABA combination therapy (about 87% as fluticasone/salmeterol). Among LABA monotherapy initiators, 478 (67%) substituted LABA with ICS monotherapy (mostly as beclometasone), 223 (31.2%) patient added ICS (ICS/LABA combi nation therapy mostly as fluticasone/salmeterol), and only 13 (1.8%) patients continued LABA monotherapy (majority as salmeterol) 2,045 (59.1%) of ICS/LABA combination therapy initiators switched to ICS monotherapy, mainly as beclomethasone and budesonide About 41% of combination therapy initiators continued on this regimen during the follow up year. There was no stepping down from combination to LABA monotherapy. Baseline characteristics of exposure groups are shown in table 5 1, which are measured at st ud y index date and baseline year. The mean age for the cohort was 39 years, and patients in the LABA monotherapy and the ICS/LABA combination therapy group s were relatively older than the patients in the ICS monotherapy group ( 40.4 years, 40.0 years, and 3 7.2 years, respectively ). The majority of the patients in all exposure groups were females with rather similar distribution across the three exposure groups. Among patients with recorded pertinent information 60% were married at the
141 time of prescribing st udy drugs, with about 65%, 62%, and 60% of the LABA, ICS/LABA, and ICS groups were married Most of the asthmatics were obese (BMI > 30) ; however, only one patient at the LABA exposure group had a known weight status, which was obese. About 84% of the ICS e xposure group and 78% of the ICS/LABA exposure group were obese at baseline. The majority of the cohort (89%) was non active smoker; about 10% of the patients were reported as former smokers, and only 1% of the patients were classified as passive smokers a t the time when exposure drugs are prescribed. The distribution of former smoking status among exposure groups was the highest in the LABA monotherapy group (13.1%), followed by the ICS/LABA combination therapy group (12%), and th e ICS monotherapy group (9 .6%). On the other hand, passive smoking status was slightly different among exposure groups (ICS, 1%; LABA, 0.7%; and ICS/LABA, 0.6%). The vast majority of asthmatics were exempted from payment for their prescribed medications with low levels of capitatio n supplements. With minor differences, patients who were prescribed ICS/LABA combination therapy were exempted from payment more than patients who were prescribed monotherapy alternatives (ICS/LABA, 83%; ICS, 80%; LABA, 71%). H owever, the distribution of c apitation supplement was the highest among ICS monotherapy group compared with LABA monotherapy and ICS/LABA combination therapy groups. Yet, most of the patients in each exposure group had low capitation supplements for their prescription payment. On aver age, the duration of patient registration with the practice was ten years, and was relatively longer in the combination group compared with the monotherapy groups. About 63% of the visits to the practice s during which study drugs were
142 prescribed w ere ten m inutes or shorter. Approximately 72%, 64%, and 9% of the patients who were respectively prescribed LABA monotherapy, ICS monotherapy, and ICS/LABA combination therapy had < 10 minute consultation with the general practitioner at the time the corresponding s tudy drugs were prescribed. The preponderance of these visits was classified as not urgent for all asthmatics across all exposure groups. About 79% of the practices were in England, 10% in Wales, 9% in Scotland, and 2% in Northern Ireland. Figure 5 2 depi cts the distribution of initiators of ICS and LABA monotherapies, and ICS/LABA combination therapy. Among asthmatics who were prescribed ICS monotherapy, about 8 0 %, 9 %, 9%, and 2% received their prescriptions in England, Wales, Scotland, and Northern Irela nd, res pectively. In the LABA monotherapy group, about 71% 14%, 12%, and 3% of asthmatics correspondingly received their prescriptions in England, Wales, Scotland, and Northern Ireland. Within the ICS/LABA combination therapy group, about 75%, 13%, 7%, an d 5% of the prescriptions were issued in England, Wales, Scotland, and Northern Ireland. Among the cohort, most of the prescriptions in England were for ICS monotherapy. In Scotland and Wales, most prescriptions were issued for LABA monotherapy. Conversely ICS/LABA combination therapy contributed to the majority of prescriptions for the study exposure of interest in Northern Ireland. The location of practices visited by 125 (0.2%) patients wa s not recorded in the database corresponding to 120 patients in the ICS monotherapy group and 5 patients in the ICS/LABA combination group Baseline asthma was un controll ed in the majority of patients (60%); however, that was contributed by the ICS monotherapy group, which had 61% of patients with uncontrolled asthma Conversely, the disease in 48% of the ICS/LABA combination
143 therapy and 40% of the LABA monotherapy groups was classified as controlled at the time exposures were prescribed. O nly 6% of the whole cohort was prescribed oral corticosteroids and 0.3% attended accident and emergency departments for asthma during the 12 months before receiving prescriptions for the exposure of interest. Only 1 patient was hospitalized for asthma during the 12 months period before receiving a prescription for ICS monotherapy. On a verage, the cohort received 2.5 prescriptions for inhaled SABA as rescue bronchodilators during the 12 months pre index date period, which corresponds to 91% of asthmatics received < 6 SABA prescriptions. About 53% of patients were prescribed inhaled SABA c oncomitantly with the exposure of interest. The mean number of asthma mediation classes prescribed at the index date was 1.7 for the patient; t he vast majority of them had one additional asthma medication class prescribed with the exposure of interest at t he index date Merely 3% of asthmatics had more than 2 asthma medication classes prescribed on the same day in which the exposure of interest was prescribed. Leukotriene receptor antagonists, e.g. montelukast accounted for the majority of asthma medication s that were concurrently prescribed with LABA monotherapy and ICS/LABA combination therapy; however, oral methylxanthines, e.g. theophylline accounted for the majority of asthma medications prescribed with ICS monotherapy. Methylxanthines ranked second to LTRA as co medications prescribed with LABA monotherapy and ICS/LABA combination therapy, followed by inhaled muscarinic receptor antagonists, e.g. ipratropium and inhaled mast cell stabilizers, e.g. nedocromil Among asthmatics who were prescribed ICS mon otherapy, LTRA ranked second to
144 methylxanthines, followed by inhaled MRA and inhaled MCS as asthma medications that were concurrently prescribed with ICS monotherapy. Among other medications prescribed on the same day the patients rec eived their study drug s, oral and parenteral antibiotics for respiratory tract infections accounted for about 58% the medications prescribed in the cohort, corresponding to about 59%, 48%, and 50% of the medications concomitantly prescribed with ICS monotherapy, ICS/LABA combin ation therapy, and LABA monotherapy, respectively. N asal corticosteroids, e.g. t riamcinolone were the second mostly prescribed co medications at the index date with distributions of 25% of the medications in the ICS/LABA combination group respectively comp ared with 21% and 22% of the medications in the LABA and ICS monotherapy groups Opioid analgesics, e.g. m orphine accounted for 11% of medications prescribed with ICS/LABA combination inhalers, 10% of medications prescribed with LABA monotherapy inhalers, and 7% of medications prescribed with ICS monotherapy inhalers. About 4% of medications (other than asthma drugs) prescribed together with study drugs were NSAIDs, e.g. ibuprofen. Aspirin and acetaminophen accounted for half of that figure each. Approxima tely 11% of the medications that were prescribed in tandem with ICS/LABA combination therapy were for NSAIDs compared with 10% and 7% of medications prescribed with ICS and LABA monotherapies. Patients received LABA monotherapy had 6.5% of co medications a s orally or rectally administered aspirin, while aspirin correspondingly accounted for 4% and 2% of co medications in ICS/LABA combination therapy and ICS monotherapy groups Antivirals for respiratory tract infections, e.g. oseltamivir ; nasally administer ed antihistamines, e.g. azelastine and decongestants, e.g. oxymetazoline ; ophthalmic selective beta 1 blockers, e.g.
145 betaxolol; and oral cholinergics, e.g. neostigmine bromide were only prescribed in patients received ICS monotherapy prescriptions. Nasally administered MCS, e.g. cromolyn sodium were only prescribed in monotherapy groups with larger distributions in the LABA group. Expectorants and opioid based antitussive formulations accounted for 2% of concurrently prescribed medications (other than asth ma drugs) for patients in the monotherapy groups, compared with 1.2% of the medications in the ICS/LABA combination group. Orally administered selective beta 1 blockers, e.g. atenolol and atenolol/chlorthalidone accounted for 2% of concomitantly prescribed medications with ICS/LABA combination therapy and LABA monotherapy groups, and 1.5% of co medications in the ICS monotherapy group. About 78% of the prescriptions for immunizations that were issued at the index date were for influenza vaccine s. Pneumococc al polysaccharide vaccine accounted for about 4% of concomitant vaccines. Other immunizations accounted for the rest of concurrent vaccines that were prescribed on the same date the exposures of interest were prescribed. Patients received LABA monotherapy (n=2) were only prescribed influenza vaccines at baseline. About 78% of the concomitantly prescribed vaccines that were prescribed to asthmatics in the ICS monotherapy and the ICS/LABA combination therapy groups were for influ enza. Only 6 patients in the I CS monotherapy group and 3 patients in the ICS/LABA combination group were prescribed pneumococcal polysaccharide vaccine. Other immunizations were tallied to 19% and 13% of all vaccines prescribed to the ICS monotherapy and ICS/LABA combination therapy gr oups, respectively. The trend of prescribing study drugs in the cohort was similar across the first two and the fourth annual quarters with most prescriptions in the fourth
146 quarter ; conversely, the third quarter had the least number of prescriptions. This trend was similar across patients prescribed controller medications with ICS monotherapy and ICS/LABA combination therapy; however, inhaled LABA bronchodilator monotherapy was least and most prescribed during the first and fourth quarters of the year, res pectively (Figure 5 3 ) Atopic conditions in asthmatics accounted for the majority of the selected comorbid conditions (49%), followed by respiratory tract infections (32%) and psychosocial pathologies (19%). Allergic rhinitis and sinusitis, other and uncl assified respiratory tract infections, and use of antidepressants were respectively contributed to most of atopies, respiratory inf ections, and psychosocial problems The distribution of atopic conditions across exposure groups was similar in ICS monothera py (3.6%) and ICS/LABA combination therapy groups (3.5%), but was more in the LABA monotherapy group (4.1%). However, the distribution of respiratory tract infections was more prominent in the ICS monotherapy group (2.4%) compared with LABA monotherapy and ICS/L ABA combination therapy groups (2.1% & 2.2%, respectively) ; and psychosocial pathologies were mostly distributed in ICS/LABA combination group (2.4%) compared with the monotherapy groups (ICS, 1.3%; LABA, 2%) Information about the availability of pe rsonalized asthma action plan was only available for about 1% of asthmatics initiated ICS monotherapy and about 2% of patients initiated ICS/LABA combination therapy All these patients reported havin g asthma action plans during the study period Only one patient received ICS monotherapy reported no asthma action plan was available during the study period.
147 medications (regardless of the medication class or route of administration ) was only available for a small number of asthmatics in the ICS monotherapy and the ICS/LABA combination groups. The distribution between both groups was similar with the majority of patients having satisfactory compared with unsatisfactory compliance lev els. Regardless of medication type, good general medication compliance was recorded in about 0.1% of the patients in each of the three exposure groups (one patient in the LABA monotherapy group) and only 7 patients in the ICS monotherapy group reported po or compliance level. The vast majority of the prescribed exposures were in inhaled aerosol forms compared with inhaled dry powders. Aerosols were most prescribed for ICS monotherapy (91%) than LABA monotherapy (74%) or ICS/LABA combination therapy (53%). C onversely, dry powder inhalers were most prescribed for ICS/LABA combination therapy (47%) than monotherapy (ICS, 9%; LABA, 26%) Device type was not determined for about 0.1% of initiators of ICS monotherapy. Most of asthmatics did not receive a spacer or holding chamber at the time when the exposure was prescribed. A bout 10% of ICS monotherapy initiators had spacer devices prescribed at baseline, compared to 6.6% of ICS/LABA combination therapy and 3.2% of LABA monotherapy initiators. Although most of the prescribed to ICS monotherapy initiators, and only one patient in the ICS/LABA combination therapy group was prescribed a nebulizer at baseline. Additionally, p atients who initiated LABA monothera py was not recommended a nebulizer for their inhaled bronchodilator therapy. The differences in baseline characteristics among exposure groups were statistically significant in age; marital status; smoking status; level of capitation
148 supplement; consultati on duration; urgency of visit; location of practice in the UK; asthma disease control at baseline; asthma severity indicators at index date and baseline year (except for asthma related hospitalization) ; concurrent prescriptions for asthma medications; conc urrent prescriptions for nasal corticosteroids, nasal mast cell stabilizers, oral and rectal aspirins, a nd all routes opioid analgesics; concomitant prescriptions for influenza and pneumococcal polysaccharide vaccines; whether a spacer was prescribed; alle rgic conjunctivitis among comorbidity with atopic conditions; pneumonia among comorbidity with respiratory tract infections; psychosocial conditions, including prescriptions for tranquilizers and antidepressant s; availability of asthma action plan; and the type of inhaler device for the prescribed exposure S tep D own T herapy C ohort Among initiators of ICS/LABA combination products, the effect of step down therapy approaches are reported according to the daily dosage strength of the ICS in the initiated comb ination product, whether single device or separate devices The majority of ICS/LABA combination initiators were prescribed m edium dose ICS ( 2,581 ) between January 4, 1993 and August 11, 2010 compared to high dose ICS ( 645 ) that were prescribed between Jan uary 5, 1993 and July 12 2010 Among the cohort, 59 initiators of high dose ICS and 176 initiators of m edium dose ICS had uncontrolled asthma during the three months before initiation date (Figure 5 1) Uncontrolled asthma defined as having prescriptions for oral corticosteroids (n= 56 and 170 respectively) asthma related attendance to accident and emergency departments (n= 3 and 6 respectively) or hospitalization due to asthma ( no event occurred for either group ) Those patients were excluded from the cohort. The sample was followed for an average
149 of 1 year after exposure. During the follow up year, patients received a total of 2,203 prescriptions for high ICS dose ICS/LABA combination therapy inhalers, 5,999 prescriptions for m edium ICS dose ICS/LABA c ombination therapy inhalers, 6,591 prescriptions for low ICS dose ICS/LABA combination therapy inhalers, 43,410 prescriptions for m edium dose ICS monotherapy inhalers, and 4,175 prescriptions for low dose ICS monotherapy inhalers. The distribution of basel ine characteristics for patients included in the analyses is described below: Or iginal high dose ICS initiators Table 5 2 shows baseline characteristics of patients according to the type of step down therapy approach among 645 initiators of high dose ICS c ombination products. Most of these patients were ICS dose reducers to m edium dose ICS ( 337 ), followed by LABA stoppers with original high dose ICS ( 172 ), and continuers of original high dose ICS strength ( 136 ). The mean age of the cohort was 42.1 years. LA BA stoppers were relatively older than original and reduced ICS dose combination groups (43.5 versus 41.5 and 41.4, respectively). The majority of patients were females with relatively more distribution among the LABA stoppers compared with the other two g roups. Among asthmatics with recorded social and behavioral information, about 65% were married, 86% obese, 87% nonsmokers, 12% former smokers, 85% exempted from prescription payment, and 55% with low capitation supplement. On average patients were regist ered with the practices for 9.4 years with shortest duration among LABA stoppers and longest duration among ICS dose re d u c ers Most of the visits at which the exposure was prescribed were for ten minutes or less, and the majority was not urgent. Most of t he shorter visits were among LABA stoppers followed by original dose and reduced
150 dose groups. Visits longer than 10 minutes were mostly for dose reducer patients. Original ICS dose combination group contributed to the majority of urgent visits. About 67% of LABA stoppers, 66% of ICS dose reducers, and 61% of original ICS dose patients were married. Only one patient in the original ICS dose group was classified as non obese. Most of nonsmoker patients were LABA stoppers; conversely, most of the former smoke rs were ICS dose reducers, and most of the original ICS dose patients were passive smokers. Payment for prescriptions was mostly exempted with low capitation levels among LABA stoppers compared to ICS/LABA combination groups. About 78% of the practices wer e in England, 12% in Wales, 6% in Scotland, and 4% in Northern Ireland. Figure 5 4 depicts the distribution of initiators of original high dose ICS and reduced m edium dose ICS ICS/LABA combination therapies and LABA stopper ( high dose ICS mono therapy ) A mong asthmatics who were prescribed high dose ICS combination therapy, about 7 7 %, 14 %, 5 %, and 4 % received their prescriptions in England, Wales, Scotland, and Northern Ireland, respectively. In the reduced m edium dose ICS combination therapy group, about 7 7 %, 1 2 %, 6 %, and 5 % of asthmatics correspondingly received their prescriptions in England, Wales, Scotland, and Northern Ireland. Within the LABA stopper group, about 80 %, 9 %, 8 %, and 3 % of the prescriptions were issued in England, Wales, Scotland, and No rthern Ireland. Among the cohort, most of the prescriptions in England and Scotland were for high dose ICS monotherapy (LABA Stoppers). In Wales most prescriptions were issued for high dose ICS combination therapy Conversely, medium dose ICS combination therapy contributed to the majority of exposure prescriptions in Northern Ireland. Practice
151 location for one patient in the reduced ICS dose combination group was not recorded in the database. Approximately 44% of asthmatics received prescriptions for SABA rescue inhalers, corresponding to 52% LABA stoppers, 43% m edium dose ICS combination therapy patients, and 39% high dose ICS combination therapy patients. The mean number of asthma medication classes prescribed at index date for the cohort was 2.2 with t he least number among LABA stoppers The vast majority of asthmatics received one additional asthma medication class to their exposure of interest. Nearly 1 in 3 patients had more than two asthma medication classes prescribed at the index date. The vast ma jority of prescriptions issued to LABA stoppers had < 2 asthma medication classes (99%), compared with 58% of original high dose ICS and 56% of reduced m edium dose ICS combination groups. Most of the prescriptions with >2 asthma medication classes were issu ed for reduced ICS dose combination group, followed by the original high dose ICS combination group. Only one LABA stopper had >2 asthma medication classes prescribed on the same date LABA was withdrawn and high dose ICS was continued The majority of asth ma medications prescribed at baseline was LTRA (45.4%), followed by oral methylxanthines (27.3%), inhaled MRA (25.8%), and inhaled MCS (1.5%). Anti inflammatory LTRA medications and inhaled MRA bronchodilators were mostly prescribed to high dose ICS combin ation therapy patients, and least prescribed to high dose ICS monoth erapy patients (LABA stoppers). Equally, oral xanthines were more prescribed with high dose ICS combination therapy compared with m edium dose ICS combination therapy and high dose ICS mono therapy.
152 Among the cohort, about 38%, 22%, and 11% of prescriptions issued for concomitant medications at baseline ( other than asthma drugs ) were respectively for RTI antibiotics, nasal corticosteroids, and opioid analgesics. Antibiotics and nasal corticos teroids were mostly prescribed to high dose ICS groups (combination and monotherapy) and least prescribed to m edium dose ICS combination therapy (dose reducers). Conversely, opioid analgesics were mostly prescribed to dose reducers than original dose patie nts or LABA stoppers. About 5%, 3%, and 2% of the concurrent medications prescribed to the cohort were for NSAIDs, aspirin, and acetaminophen, respectively. Most of the NSAIDs prescriptions wer e issued to high dose ICS groups compared to m edium dose ICS gr oup. On the other hand, aspirin and acetaminophen were mostly prescribed to m edium dose ICS combination therapy group, followed by LABA stoppers and high dose ICS combination therapy group. Approximately 1.7%, 1.2% of concomitant prescriptions were issued for orally administered selective beta blockers and antitussives, respectively. Oral selectiv e beta blockers were least prescribed to LABA stoppers compared with other two groups ; antitussives were least prescribed to dose reducers Only one patient in the LABA stopper group received a prescription for nasally administered MCS. N o patient in this step down therapy cohort received prescriptions for RTI antivirals, nasal antihistamines, nasal decongestants, ophthalmic selective beta blockers, or oral choliner gics. Influenza vaccines accounted for the majority of the prescriptions of concurrent immunizations at the time exposures of interest were prescribed, which was mainly contributed by patients in the ICS dose reducer combination therapy group (n= 4 ). Only one pre scription for influenza vaccine w as prescribed to high dose ICS combination
153 therapy patients, one prescription for pneumococcal polysaccharide vaccine and one prescription for other vaccines were issued to dose reducer group, and one prescription fo r other vaccines was issued to the LABA stopper group. About 27% of the prescriptions were issued to the cohort in the fourth quarter of the year; the distribution of prescriptions was relatively equivalent across the other three quarters. Most of the high dose ICS combination therapy prescriptions were issued during the second quarter, whereas the step down approaches (dose reducers and LABA stoppers) were mostly prescribed during the fourth quarter ( Figure 5 6, A ) Approximately 45% of the identified como rbidities in the cohort were classified as atopic conditions, and 55% were evenly distributed between respiratory tract infections and psychosocial pathologies. Atopies accounted for the majority of comorbid conditions in original ICS dose combination grou p; conversely, psychosocial pathologies and respiratory infections accounted for the majority of comorbidities in the step down therapy approaches (dose reducers and LABA stoppers, respectively). Most of atopic conditions in these asthmatics were allergic rhinitis and sinusitis. Unclassified respiratory tract infections accounted for the majority of RTI in the cohort. Utilization of antidepressants contributed to most of psychosocial problems identified in the cohort. Personalized asthma action plan was ava ilable in 2 ori ginal high ICS dose users, 6 medium ICS dose reducers, and 1 LABA stoppers. Only one patient in the dose reducer group had unsatisfactory compliance level with any prescribed asthma medications, compared to four patients in the same group ha d satisfactory compliance with asthma medications and one patient had good compliance with any medication prescribed regardless of the class or indication. Only one patient in the original ICS dose group had
154 satisfactory asthma drug compliance and two pati ents in the same group had good compliance with any medication prescribed. No compliance related information was recorded in the LABA stopp er group. Nearly two third prescribed exposures were in form of aerosols (pMDI and BAI) and one third was in form of dry powder inhalers. Aerosols were mainly prescribed to the original high ICS dose ICS/LABA combination group, followed by the LABA stopper and ICS dose reducer groups. On the other hand, dry powder inhalers were mainly prescribed to the ICS dose reducer I CS/LABA combination group, followed by the LABA stopper and original high ICS dose ICS/LABA combination groups. At baseline, only 8% of asthmatics had a spacer or a holding chamber device prescribed, corresponding to about 10% of original ICS dose users, 8 % of LABA stoppers, and 7% of ICS dose reducers. One patient in the ICS dose reducer group received a prescription for nebulizer. Baseline differences across the groups were statistically significant in the following characteristics: Age ; smoking status ; c apitation supplement level ; consultation duration ; prescription for SABA rescue inhaler ; number of asthma medication classes prescribed ; LTRA and methylxanthines concomitant asthma medications prescribed; respiratory tract infections as concurrent clinical conditions ; and inhaler device type for the exposure of interest. Origin al m edium dose ICS initiators Table 5 3 describes baseline characteristics of patients by the type of step down therapy approach among 2,581 initiators of m edium dose ICS combination products. The majority of these patients were LABA stoppers with original m edium dose ICS ( 1,742 ), followed by ICS dose reducers to low dose ICS ( 519 ), and continuers of original
155 medium dose ICS strength ( 320 ) The mean age of the cohort was 4 0 3 years. LA BA stoppers and original ICS dose combination group were rel atively older than reduced ICS dose co mbination group (4 1 3 and 41. 4 versus 38 2 respectively). The majority of patients were females with relatively more distribution among the low ICS dose redu cers compared with the other two groups. Among asthmatics with recorded social and behavioral information, about 6 7% were married, 92 % obese, 8 8 % nonsmokers, 1 1 % former smokers, 77 % exempted from prescription payment, and 5 6 % with low capitation supplement On average, patients were registered with the practices for 9. 6 years, with longest duration among LABA stoppers and shortest duration among ICS dose reducers. Similar to the previous step down cohort, m ost of the visits at which the exposure was prescri bed were for ten minutes or less, and the majority was not urgent. Most of the shorter visits were among LABA stoppers, followed by reduced dose and original dose groups. Visits longer than 10 minutes were mostly for original medium ICS dose ICS/LABA combi nation therapy patients. Combination therapy groups contributed to the majority of urgent visits. About 6 9 % of LABA st oppers and 6 6 % of original ICS dos e patients, and 58% of ICS dose reducers were married. Only one patient in the LABA stopper group was c lassified as non obese. Most of nonsmoker and passive smoker patients were LABA stoppers; conversely, most of the former smokers were in the original m edium dose ICS group Payment for prescriptions was mostly exem pted among original m edium dose ICS combin ation group compared to step down therapy groups. However, most of the low capitation supplements were in the step down therapy groups, especially among LABA stoppers.
156 About 7 6 % of the practices were in England, 1 1 % in Wales, 9 % in Scotland, and 4% in Nort hern Ireland. Figure 5 5 illustrates the distribution of initiators of original m edium dose ICS and reduced low dose ICS ICS/LABA combination therapies, and LABA stopper ( m edium dose ICS monotherapy). Among asthmatics who were prescribed m edium dose ICS c ombination therapy, about 7 7 %, 1 2 %, 6 %, and 5 % received their prescriptions in England, Wales, Scotland, and Northern Ireland, respectively. In the reduced low dose ICS combination therapy group, about 77%, 12%, 7 %, and 4 % of asthmatics correspondingly rec eived their prescriptions in England, Wales, Scotland, and Northern Ireland. Within the LABA stopper group, 76 %, 10 %, about 10 %, and 4 % of the prescriptions were issued in England, Wales, Scotland, and Northern Ireland. Among the cohort, most of the prescr iptions in Northern Ireland and Wales were for medium dose ICS and low dose ICS combination therapy groups, compared to LABA stoppers with least distribution in both countries The distribution of exposure prescriptions issued in England was virtually equa l across exposure groups. Conversely, most of the prescriptions issued in Scotland were for medium dose ICS monotherapy (LABA stoppers) Practice location for 5 patient s was not recorded in the database corresponding to one patient in the original ICS dos e group, and 1 and 4 patients in the step down therapy approach groups (ICS dose reducers and LABA stoppers, respectively) Approximately 4 9 % of asthmatics received prescriptions for SABA rescue inhalers, corresponding to 5 3 % LABA stoppers, 39 % low dose IC S combination therapy patients, and 43 % m edium dose ICS combination therapy patients. The mean number of asthma medication classes prescribed at index date for the cohort was 2. 1 with the least
157 number among LABA stoppers. The vast majority of asthmatics r eceived one additional asthma medication class to their exposure of interest. About 14% of patients had more than two asthma medication classes prescribed at the index date. The vast majority of prescriptions issued to LABA stoppers had < 2 asthma medicatio n classes (99%), compared with 60 % of reduced low dose ICS and 56% of original m edium dose ICS combination groups. Most of the prescriptions with >2 asthma medication classes were issued for reduced ICS dose combination group, followed by the original m edi um dose ICS combination group. Only 1% of LABA stoppers had >2 asthma medication classes prescribed on the same date LABA was withdrawn and m edium dose ICS was continued. In contrast to the previous step down therapy cohort, t he majority of asthma medicati ons prescribed at baseline was oral methylxanthines ( 33.6 %), followed by oral LTRA ( 30.5 %), inhaled MRA ( 29 %), and inhaled MCS ( 7 %). Anti inflammatory LTRA medications and inhaled MRA bronchodilators were mostly prescribed to m edium dose ICS combination th erapy patients, and least prescribed to m edium dose ICS monotherapy patients (LABA stoppers). Equally, oral xanthines were more prescribed with ICS/LABA combination therap ies than ICS monotherapy. Among the cohort, about 54 %, 2 3 %, and 1 0 % of prescriptions issued for concomitant medications at baseline (other than asthma drugs) were respectively for RTI antibiotics, nasal corticosteroids, and opioid analgesics. Antibiotics were mostly prescribed to LABA stoppers Conversely, nasal corticosteroids were least prescribed to LABA stoppers. O pioid analgesics were mostly prescribed to original m edium dose ICS ICS/LABA combination group compared About 4 %, 3%, and 2% of the concurrent medications prescribed to the cohort were for NSAIDs, aspirin, and acetaminophen,
158 respectively. Most of the NSAIDs prescriptions were issued to ICS dose reducers compared to original m edium dose ICS group s (combination and monotherapies) On the other hand, aspirin and acetaminophen were mostly prescribed to original dose ICS combinatio n therapy group, and least prescribed to LABA stopper group Approximately 2% of concomitant prescriptions were issued for orally administered selective beta blockers and additional 2% for antitussive s Oral selective beta blockers were mostly prescribed t o original m edium dose ICS combination group than other two groups. Equally, antitussives were mostly prescribed to LABA stoppers. Like the previous step down therapy cohort, o nly one patient in the LABA stopper group received a prescription for nasally ad ministered MCS. Yet, three patients in this group received prescriptions for nasally administered antihistamines. No patient in this step down therapy cohort received pr escriptions for RTI antivirals nasal decongestants, ophthalmic selective beta blockers or oral cholinergics. Influenza vaccines accounted for the majority of the prescriptions of concurrent immunizations at baseline which was mainly contributed by patients in the ICS dose reducer combination therapy group I nfluenza vaccine and other type s of vaccines w ere least prescribed to LABA stoppers compare to combination therapy users. Only two prescription s for pneumococcal polysaccharide vaccine were issued to the cohort, one in the original medium dose group and another in the LABA stopper group About 2 6 % of the prescriptions were issued to the cohort in the fourth quarter of the year; unlike the other step down therapy cohort, the distribution of prescriptions was not similar across the other three quarters. About 26%, 25%, and 23% of prescript ions were correspondingly issued in the first, second, and third quarters. Most of the medium
159 dose ICS prescriptions (original combination therapy and LABA stoppers monotherapy) were issued during the fourth quarter, whereas the low dose ICS combination pr escriptions (dose reducers) were mostly prescribed during the first quarter (Figure 5 6, B ). Approximately 25 % of the identified comorbidities in the cohort were classified as atopic conditions and 32 % and 22% were respiratory tract infections and psychos ocial pathologies respectively Atopies accounted for the majority of comorbid conditions in all the three groups ; however, psychosocial pathologies ranked the second mostly encountered comorbidities in the m edium ICS dose groups (combination and monother apies), whereas respiratory infections accounted for the second most comorbidities in the low reduced dose ICS combination group Similar to other step down therapy cohort, m ost of atopic conditions in these asthmatics were allergic rhinitis and sinusit is, and u nclassified respiratory tract infections accounted for the majority of RTI in the cohort. Likewise, u tilization of antidepressants contributed to most of the psychosocial problems identified in the cohort. Personalized asthma action plan was availabl e in 6 original medium ICS dose users, 14 low ICS dose reducers, and 10 LABA stoppers. Only one patient in the original medium dose ICS group and one patient in the reduced low dose ICS group had unsatisfactory compliance level with any prescribed asthma m edications, compared to four and 2 patients in the same group s had satisfactory compliance with asthma medications. Additionally, three LABA stoppers had satisfactory compliance with asthma medications, and only one patient in the same group had poor gener al compliance level One original ICS dose user, 2 ICS dose reducers, and 4 LABA
160 stoppers had good overall compliance with medications regardless of medication type or clinical indication. Approximately 73% of prescribed exposures were aerosol formulations and 37% w ere dry power disk inhalers. Unlike other step down therapy cohort, a erosols were mainly prescribed to LABA stoppers (low dose ICS monotherapy) followed by the combi nation groups (original medium dose ICS and reduced low dose ICS). D ry powder in haler s were mainly prescribed to low dose ICS reducers; followed by original medium ICS dose groups (original combination and LABA stoppers) O nly 9.5 % of asthmatics had a spacer or a holding chamber device prescribed at baseline corresponding to about 1 1 % of LABA stoppers 7 % of original ICS dose users and 6 % of ICS dose reducers. One original ICS dose user and one LABA stopper received a prescription for nebulizer. Baseline differences across the groups were statistically significant in the following ch aracteristics: Age; sex; marital status; smoking status; capitation supplement level; consultation duration; practice location in the UK; prescription for SABA rescue inhaler; number of asthma medication classes prescribed; LTRA, antibiotics for RTI, aceta minophen, and opioids as concomitant medications; concurrently prescribed influenza vaccines; psychosocial pathologies as concurrent clinical conditions; presence of asthma action plan; asthma medication compliance; inhaler device type for the exposure of interest; and whether a spacer was prescribed. Inferential Statistics The results of testing study hypotheses are described in the following sections of asthma related morbidity outcomes for original cohort and step down therapy sub groups, and asthma relat ed and all cause mortality outcomes for original cohort
161 Asthma related M orbidity Asthma related morbidity rates are identified by prescription rates for oral corticosteroids and asthma related accident and emergency department attendance rates. No asthma related hospitalization was recorded in the sample during the 1 year study follow up, and no asthma related A & E department visits for initiators of high dose ICS combination therapy step down cohort; thus, these outcomes were not included in outcome assess ment for the corresponding original cohort and the relevant step down therapy cohort. Table 5 4 shows the incidence rates of prescriptions for oral corticosteroids and A&E department visits for asthma exacerbations during study follow up after exposure ini tiation. Table 5 5 gives the incidence rates of prescribing short courses OCS for asthma exacerbations. Among the original cohort, initiators of LABA monotherapy had higher incidence rates for asthma related morbidity than ICS monotherapy and ICS/LABA comb ination therapy initiators Among step down therapy cohorts, continuers of high dos e ICS and continuers of medium dose ICS combination therapies had higher incidence rates for asthma related morbidity than dose reducers or LABA stoppers. Prescriptions for oral corticostero ids The study population received a total of 7 108 prescriptions for oral corticosteroids during the 1 year follow up 91.4% of which were for short courses to treat asthma exacerbations. The distribution of prescriptions for OCS and short courses of OCS in the original cohort was 4,597 and 4,197, respectively for ICS monotherapy initiators; 326 and 310, respectively for LABA monotherapy initiators; 2,185 and 1,988, respectively for ICS/LABA combination therapy initiators Among h igh dose I CS step
162 down subgroup the number of prescriptions for OCS and short courses of OCS prescribed during the 1 year follow up was 286 and 265 for original dose continuers ; and 465 and 310 for dose reducers. LABA stoppers received 171 OCS prescriptions that al l classified as short courses On the other hand, the distribution of prescriptions for OCS and short courses of OCS among m edium dose subgroup was as following: original dose continuers (862 and 690, respectively); dose reducers (856 and 568, respectively ); and LABA stoppers (1,317 and 1,219, respectively ) (Tables 5 4 and 5 5) Main cohort. During the first year after exposure to study drugs, the incidence rate of prescribing relatively long courses of OCS was higher in LABA monotherapy initiators (19.9 pe r 100 person yrs 95%CI, 13.8 28.6 ) than ICS/LABA combination therapy initiators ( 17.7 per 100 person yrs, 95%CI 14.9 20.9 ) or ICS monotherapy initiators (10.6 per 100 person yrs, 95%CI 10.0 11.2) Likewise, initiators of LABA monotherapy had higher incide nce rates of prescribing short courses of OCS for asthma exacerbations (18.5 per 100 person yrs, 95%CI 12.7 27.0) compared to ICS/LABA combination therapy initiators (14.8 per 100 person yrs, 95%CI 12.4 17.8) or ICS monotherapy initiators (9.8 per 100 pers on yrs, 95%CI 9.3 10.4). Table 5 8 lists the distribution of mean time to event for outcomes of interest stratified by exposure type. On average, initiators of LABA monotherapy had the shortest mean time to receive prescriptions of long or short courses OC S for asthma exacerbations (11.3 months) compared with initiators of ICS monotherapy (11.7 months) and ICS/LABA combination therapy ( 11.4 months ). Figure 5 7 shows a statistically significant difference in the probability of receiving long and short course s of OCS among initiators of ICS, LABA or ICS/LABA. Also, Bonferroni correction for
163 multiple comparisons showed statistically highly significant differences between all the three comparison groups; yet, comparing survival probability between LABA monothera py and ICS/LABA combination therapy yielded statistically not significant results (Bonferroni p value 0.1247). High dose ICS step down subgroup. Continuers of ICS/LABA combination therapy with high dose ICS had higher incidence rates for prescribing long a nd short courses of OCS (respectively: 28.9 per 100 person yrs, 95%CI, 19.2 43.5 and 27.7 per 100 person yrs, 95%CI, 18.2 42.0) compared with initiators of reduced medium dose ICS combination therapy ( 21.9 per 100 person yrs, 95%CI, 16.0 29.9 and 17.9 per 100 person yrs, 95%CI, 12.7 25.4 ) and LABA stoppers w ho are continuing high dose ICS monotherapy ( respectively: 19.3 per 100 person yrs, 95%CI, 11.8 31.5 and 15.1 per 100 person yrs, 95%CI, 12.9 17.5 ). The trend of incidence rates was similar for original dose continuers between long and short OCS co urses. However, the figure was reversed between step down therapy approaches, where high dose ICS monotherapy users had the lowest incidence rate of prescribing long courses; in contrast, medium dose ICS combina tion therapy users had the lowest incidence rate of prescribing short courses. Asthmatics who withdrew LABA and continued high dose ICS as monotherapy had the shortest average time to receive prescriptions of long or short courses OCS for asthma exacerbati ons ( 3.8 months) compared with asthmatics who continued LABA while reducing ICS dose to medium as a combination therapy ( 8 months) and asthmatics who continued ICS/LABA combination therapy with original high dose ICS ( 5.4 months) However, no statistically significant difference was observed between
164 exposure groups in terms of prescribing long (Long rank test p value 0.2118) or short courses of OCS (Long rank test p value 0.1421) and Bonferroni adjustment failed to show statistically significant results fo r the three comparisons (Table and Figure 5 8) Medium dose ICS step down subgroup. Continuers of ICS/LABA combination therapy with medium dose ICS had higher incidence rates for prescribing long and short courses of OCS (respectively: 2 1 .9 per 100 person yrs, 95%CI, 16.0 29.9 and 17.9 per 100 person yrs, 95%CI, 12.7 25.4 ) compared with initiators of reduced low dose ICS combination therapy (12.0 per 100 person yrs, 95%CI, 8.9 16.1 and 1 0 .9 per 100 person yrs, 95%CI, 8.1 14.9 ) and LABA stoppers who are cont inuing medium dose ICS monotherapy (respectively: 15.5 per 100 person yrs, 95%CI, 13.4 17.9 and 15.1 per 100 person yrs, 95%CI, 12.9 17.5 ). Unlike the previous step down cohort, t he trend of incidence rates was similar across exposure groups between long a nd short OCS courses regardless of the step down therapy approach Unlike patients started ICS/LABA combination therapy with high dose ICS, asthmatics who started combination therapy with medium dose ICS had the shortest time to receive prescriptions for O CS (8 months) compared with counterparts who had any of the two step down approaches (11.6 months). However, in tandem with patients commenced high dose ICS, there was no statistically significant differences between exposure groups in terms of OCS prescri bing (Table and Figure 5.8). Asthma related A & E visits During one year of patient follow up, a total of 100 A&E department visits for asthma exacerbations were recorded for the cohort, corresponding to 68 events for ICS monotherapy initiators, 16 events fo r LABA monotherapy initiators, and 16 events for
165 ICS/LABA combination therapy initiators Among medium dose step down cohort, low ICS dose reducers encountered 19 asthma related A&E visits compared to 7 visits for low dose ICS monothera py (LABA stoppers). No asthma related A&E visits were reported in the original medium dose continuers in medium dose step down cohort or among asthmatics in high dose step down cohort (Table 5 4) Main cohort. Asthmatics who started LABA monotherapy had higher annual incidenc e rate s for A&E department visits for asthma exacerbations (0.5 per 100 person yrs, 95%CI, 0.08 3.9) than counterparts who started ICS monotherapy (0.1 per 100 person yrs, 95%CI, 0.08 0.2) or ICS/LABA combination therapy (0.3 per 100 person yrs, 95%CI, 0.0 8 1.3). Compared to ICS based therapy approaches, patients who started LABA monotherapy had the shortest time to be admitted to A&E department for asthma exacerbations (about 1 week versus 11 12 months). However, there was no statistically significant diff erence between therapy approaches in terms of asthma related A&E department visits ( l og rank test p value 0.2185) and Bonferroni test did not show any significant difference for all three comparisons (Table and Figure 5 8) Medium dose ICS step down subgr oup. No asthma related visits to A&E departments were reported for initiators of high dose ICS step down therapy cohort. Among initiators of medium dose ICS step down therapy, no similar events were reported for continuers of medium dose ICS combination th erapy. Among the step down therapy approaches, patients who reduced ICS dose to low dose while continuing LABA had higher incidence rates for asthma related A&E visits (0.3 per 100 person yrs, 95%CI, 0.05 2.4) than patients who withdrew LABA and continued medium dose ICS
166 monotherapy (0.2 per 100 person yrs, 95%CI, 0.04 0.7). Patients who continued LABA as ICS/LABA combination therapy while reducing ICS dose to low had, on average, one day after exposure to be admitted to A&E department for asthma exacerbati ons compared with patients who withdrew LABA and continued medium dose ICS monotherapy (6.4 months). Similar to original cohort, log rank and Bonferroni tests showed no statistically significant difference between exposure groups (Table and Figure 5 8) As thma related M ortality Incidence rates for asthma related mortality are calculated among initiators of study drugs in general practices that are linked to the ONS mortality database (n=117), and among initiators of study drugs in general practices that are part of the ONS mortality database linkage scheme i.e. algorithm based (n=50,986) All the linked practices were in England, while the unlinked practices were spread across all four UK countries with the majority in England (79%), followed by Wales (10%) Scotland (9%), and Northern Ireland (2%). Table 5 6 shows the incidence rates of mortality outcomes during study follow up after exposure initiation (from the beginning of second year after first prescription of study exposure Figure 4 2) Among the algo rithm derived asthma death practices, initiators of LABA monotherapy had higher incidence rates for asthma deaths (0.6 per 100 person yrs, 95%CI, 0.2 1.5) compared with ICS/LABA combination therapy (0.5 per 100 person yrs, 95%CI, 0.3 0.7) and ICS monothera py initiators (0.1 per 100 person yrs, 95%CI, 0.1 0.2) Among the ONS mortality linked practices, only 5 cases of asthma related deaths were identified, and all were initiators of ICS monotherapy with uncontrolled asthma at baseline and mean age of 31 year s. The
167 majority of th o se asthmatics were nonsmoker males initiated their ICS in the form of aerosols during the first quarter of the year (Table 5 7 ). The ONS derived incidence rate of asthma related mortality among initiators of ICS monotherapy was 5.0 pe r 100 person yrs (95%CI, 2.1 12.0). Furthermore, among study sample initiators of ICS/LABA combination therapy had higher incidence rates for all cause deaths (1.1 per 100 person yrs, 95%CI, 0.8 1.5) followed by LABA monotherapy (0.7 per 100 person yrs, 95%CI, 0.3 1.7) and ICS monotherapy initiators (0.5 per 100 person yrs, 95%CI, 0.4 0.6) Although the average time to asthma mortality and all cause mortality was identical across exposure groups (12 months after the end of first year following issuing fir st prescription for ICS, LABA, or ICS/LABA), there was statistically significant difference between exposure groups and between all the three comparisons (log rank test p value <0.0001; Bonferroni p value <0.001 for all comparisons ) Comparison of Models T ables 5 9 to 5 14 gives haza rd ratios of asthma related morbidity, asthma related mortality, and all cause mortality stratified by regression models and exposure group comparisons. Conventional Cox PHREG models Unadjusted Cox regression models showed that the hazard of receiving prescriptions for OCS for LABA initiators is 2.92 times the hazard for ICS monotherapy initiators; the hazard for ICS/LABA combination therapy initiators is 9.87 times the hazard for ICS monotherapy initiators, and 1.32 times the h azard for LABA monotherapy initiators. The hazard trend was relatively similar for short courses OCS. However, adjusting the model for time independent covariates,
168 including variables measured during the baseline year and index date accounted for the varia tion i n the unadjusted hazard ratios. Initiators of LABA monotherapy had 78% more likelihood of receiving prescriptions for OCS than initiators of ICS monotherapy ( HR, 1.78; 95% CI, 1.17 2.54), initiators of ICS/LABA combination therapy had 54% higher like lihood of receiving OCS than initiators of ICS monotherapy ( HR, 1.54; 95%CI, 1.30 1.86); however, they were 5% less likely to receive OCS prescriptions than initiators of LABA monotherapy ( HR, 0.95; 95%CI, 0.63 1.41). Similarly, prescribing short courses O CS was 71% more likely to initiators of LABA than ICS monotherapies ( HR, 1.71; 95%CI, 1.16 2.49), 40% more likely to initiators of combination therapy than ICS monotherapy ( HR, 1.4; 95%CI, 1.15 1.70), and 15% less likely to initiators of combination therap y than LABA monotherapy ( HR, 0.85; 95%CI, 0.56 1.30). Comparisons including ICS monotherapy were statistically significant, compared to ICS/LABA versus LABA monotherapy and the adjusted models fitted better than the unadjusted models (unadjusted model Aka ike information criterion AIC, 30568.065; baseline adjusted model AIC, 28127.583) (Table 5 9) Unadjusted models showed LABA initiators had 4.04 times the hazard of asthma related A&E department visits for ICS initiators; ICS/LABA initiators had 2.25 times the hazard for ICS initiators; and ICS/LABA initiators were about 1% more likely to be admitted to A&E departments for asthma exacerbations than LABA initiators. On the other hand, adjusting for baseline time independent covariates yielded statistically n ot significant estimates of increased risks for asthma related A&E department attendance by initiators of LABA monotherapy compared to initiators of ICS monotherapy (HR, 3.96; 95%CI, 0.4 30), and decreased risks by initiators of ICS/LABA combination therap y than
169 initiators of LABA monotherapy (HR, 0.78; 95%CI, 0.2 6.04). The hazard of A&E department visits between ICS/LABA combination therapy and ICS monotherapy initiators was not different (HR, 0.99; 95%CI, 0.21 7.14). The baseline adjusted model did not f it better than the unadjusted model for predicting A&E department visits (unadjusted model AIC, 349.218; baseline adjusted model AIC, 381.544) (Table 5 12). Among asthmatics in practices that are unlinked to ONS mortality database the unadjusted estimates showed LABA monotherapy and ICS/LABA combination therapy initiators had higher risks for asthma related deaths than ICS monotherapy initiators, and combination therapy initiators had higher asthma mortality risks than LABA monotherapy initiators. Adjustin g for baseline characteristics showed the hazard of asthma deaths for LABA initiators was 3.2 times the hazard for ICS initiators ( HR, 3.2; 95%CI, 1.65 8.92), and the hazard for ICS/LABA initiators was 5.1 times the hazard for ICS initiators ( HR, 5.1; 95%C I, 1.01 5.02) and 1.62 times the hazard for LABA initiators ( HR, 1.62; 95%CI, 0.1 22.1). Compared to the unadjusted model, the baseline adjusted was better fitted (unadjusted model AIC, 1782.478; baseline adjusted model AIC, 1697.062). The comparisons betw een LABA and ICS monotherapies were statistically significant, while the comparisons between combination therapy and ICS monotherapy were borderline significant, compared with the not significant comparisons between combination therapy and LABA monotherapy (Table 5 13). With regard to all cause mortality outcome, the unadjusted model showed that LABA based therapies had higher risks for all cause deaths than ICS monotherapy, and ICS/LABA combination therapy had higher risks than LABA monotherapy regimens. L ikewise, the direction was similar across the estimates after adjusting for baseline time independent covariates. The
170 hazard of all deaths for LABA initiators was 32% higher than the hazard of all deaths for ICS initiators ( HR, 1.32; 95%CI, 0.49 5.95); the hazard of all deaths for ICS/LABA initiators was 77% higher than the hazard for ICS initiators ( HR, 1.77; 95%CI, 0.71 4.0); and the hazard for ICS/LABA initiators was 39% higher than the hazard for LABA initiators ( HR, 1.39; 95%CI, 0.44 4.34). The adjuste d model was better fitted than the unadjusted one (unadjusted model AIC, 5504.828; baseline adjusted AIC, 5191.972), but the estimates lacked statistical significance (Table 5 14). Among asthmatics who initiated high dose ICS in ICS/LABA combination therap y, the unadjusted models showed relatively no difference in the hazard of receiving long or short courses of OCS prescriptions for ICS dose reducers compared with the hazard for continuers of original high dose ICS combination therapies, or the hazard for LABA stoppers compared with dose reducers. However, the models showed that LABA stoppers (high dose ICS monotherapy) were 37 39% less likely to receive long or short courses OCS than continuers of original high dose ICS/LABA combination therapy. After adju sting for baseline covariates, initiators of medium dose ICS/LABA combination therapy (dose reducers) were 21% less likely to receive OCS prescriptions than continuers of original high dose ICS/LABA combination therapy ( HR, 0.79; 95%CI, 0.68 1.07). Prescri ptions of OCS were 64% and 59% less likely to be issued for initiators of high dose ICS m onotherapy (LABA stoppers) than continuers of original high dose ICS/LAAB combination therapy ( HR, 0.36; 95%CI, 0.08 1.17) or initiators of medium dose ICS/LABA combin ation therapy ( HR, 0.41; 95%CI, 0.12 1.07), respectively. Likewise, relative rates and corresponding confidence intervals were similar across comparisons involving short courses of OCS. Compared to the unadjusted model, the
171 baseline adjusted model fits wel l (Unadjusted model AIC, 1080,772; baseline adjusted model AIC, 974.684); however, none of the comparisons were statistically significant (Table 5 10) Among initiators of medium dose ICS/LABA combination therapy, unadjusted models showed relatively no dif ference in the hazards for prescribing long or short courses of OCS for dose reducers than continuers of original ICS dose combination therapies; yet, continuers of original medium dose ICS monotherapy (LABA stoppers) were 8 10% more likely to receive pres criptions for short or long courses OCS than medium dose ICS/LABA combination therapy initiators (continuers), and 2.24 2.26 times more likely to receive short or long OCS prescriptions than low dose ICS/LABA combination therapy initiators (dose reducers). Adjusting for baseline covariates yielded no difference in the hazard of prescribing lon g courses OCS between dose reducers and continuers of original ICS dose, but showed that low dose ICS reducers were 17% less likely to receive short courses OCS than c ontinuers of original medium dose ICS combination therapies. Similarly, initiators of medium dose ICS monotherapy (LABA stoppers) were 2% more likely to receive long courses OCS than continuers of medium dose ICS combination therapy ( HR, 1.02; 95%CI, 0.32 3.23), but 39% less likely to receive short courses OCS than continuers of medium dose ICS combination therapy ( HR, 0.61; 95%CI, 0.39 0.94). Prescribing long courses OCS for LABA stoppers was 52% more likely than reducers of ICS dose combination therapy ( H R, 1.52; 95%CI, 1.14 7.0), and prescribing short courses OCS for LABA stoppers was 36% more likely than dose reducers ( HR, 1.36; 95%CI, 0.97 6.37). Statistical significance was observed between LABA stoppers and dose reducers in terms of prescribing long c ourses OCS,
172 and between LABA stoppers and continuers of original dose in terms of prescribing short courses OCS. Relative to the unadjusted model, the baseline adjusted model fitted better (unadjusted model AIC, 4321.999; baseline adjusted model AIC, 4066. 097) (Table 5 11) Crude estimates showed initiators of medium dose ICS monotherapy (LABA stoppers) had 2.53 times the hazard of asthma related attendance to A&E departments for initiators of low dose ICS as ICS/LABA combination therapy. Adjusting for base line covariates showed LABA stoppers were 21% more likely to be admitted to A&E departments for asthma exacerbations than dose reducers ( HR, 1.21; 95%CI, 0.07 13.2). The prediction model did not fit well in comparison to the unadjusted model (unadjusted mo del AIC, 51.318; baseline adjusted model AIC, 128.0) (Table 5 12). Time dependent Cox PHREG models Adjusting the model for time dependent covariates in addition to baseline variables showed initiators of LABA monotherapy were 34% more likely to receive pr escriptions for long courses OCS than initiators of ICS monotherapy ( HR, 1.34%; 95%CI, 1.27 2.02); and initiators of ICS/LABA combination therapy are 17% more likely to receive long courses of OCS than initiators of ICS monotherapy ( HR, 1.17; 95%CI, 1.04 2 .3), and 30% less likely to receive OCS than initiators of LABA monotherapy ( HR, 0.7; 95%CI, 0.4 1.64) Also, LABA initiators were 47% more likely to receive short courses OCS for asthma exacerbations than ICS initiators ( HR, 1.47; 95%CI, 1.22 2.44 ); there was no difference in prescribing short course OCS between combination therapy initiators and ICS monotherapy initiators (HR, 1.0; 95%CI, 0.66 2.1); however, ICS/LABA initiators were 9% less likely to receive short courses OCS for asthma exacerbations than LABA monotherapy initiators ( HR,
173 0.91; 95%CI, 0.81 1.35) Statistical significance was observed in comparisons between LABA and ICS initiators. Compared to baseline adjusted model, time dependent covariate adjusted model had AIC values of 72818.935 for OC S and 70778.622 for short course OCS (Table 5 9) In addition, LABA initiators were 4% more likely to be admitted to A&E departments for asthma exacerbations compared to ICS initiators ( HR, 1.04; 95%CI, 0.32 15). On the other hand, initiators of combinatio n therapy are less likely to visit A&E departments for asthma exacerbations compared to ICS monotherapy initiators ( HR, 0.72; 95%CI, 0.08 5.01 ) and LABA monotherapy initiators ( HR, 0.56; 95%CI, 0.11 4.12 ). Statistical significance was not attained in the r egression model (AIC, 7631.89) (Table 5 12). The model for predicting asthma related deaths among general practices that are not part of the ONS mortality database linkage scheme showed statistically significant differences between initiators of monotherap ies by LABA and ICS (HR, 2.67; 95%CI, 1.44 4.93; model AIC, 2408.12). Initiators of combination therapy were more likely to experience asthma deaths than initiators of ICS monotherapy (HR, 4.2; 95%CI, 0.32 54.4) or LABA monotherapy (HR, 1.58; 95%CI, 0.11 2 1.8) (Table 5 13). Conversely, there was no statistically significant difference for all cause deaths between comparison groups (model AIC, 8575.759), where LABA monotherapy initiators had 26% increase in the risk of death from any cause compared with ICS monotherapy initiators ( HR, 1.26; 95%CI, 0.83 1.92); and combination therapy initiators were less likely to die from any cause than ICS monotherapy initiators (HR, 0.61; 95%CI, 0.21 1.74) or LABA monotherapy initiators (HR, 0.48; 95%CI, 0.16 1.46).
174 Among i nitiators of high dose ICS/LABA combination therapy, s tatistical significance was observed between all comparison groups for OCS prescribing (model AIC, 2334.382 ) (Table 5 10). The hazard of receiving prescriptions for OCS for asthmatics on medium dose ICS combination therapy (dose reducers) is 78 % of the hazard for asthmatics continued high dose ICS combination therapy (original dose continuers) ( HR, 0.78; 95%CI, 0. 6 0.95 ); the hazard of prescribing OCS to asthmatics who withdrew LABA and continued high do se ICS monotherapy is 34 % of the hazard for original dose continuers ( HR, 0.34; 95%CI, 0.0 7 0.55 ); and the hazard o f issuing OCS prescriptions to LABA stoppers is 33 % the hazard for dose reducers ( HR, 0.33; 95%CI, 0. 07 0.52 ). The patterns for hazard ratios and corresponding confidence intervals for prescribing short courses OCS were relatively similar to the patterns of prescribing long courses OCS with model AIC value of 1973.53 Among asthmatics who initiated medium dose ICS/LABA combination therapy, stat istical significance was not observed in comparing OCS prescribing rates between asthmatics reduced ICS dose to low dose ICS/LABA combination therapy and continuers of original medium dose ICS/LABA combination therapy (HR, 0.85; 95%CI, 0.67 1.07). Initiato rs of medium dose ICS monotherapy (LABA stoppers) were 79% less likely to receive prescriptions for OCS than continuers of medium dose ICS/LABA combination therapy ( HR, 0.21; 95%CI, 0.09 0.52); and LABA stoppers were 28% more likely to receive OCS prescrip tions than dose reducers ( HR, 1.28; 95%CI, 1.09 5.08). Model AIC equals to 8708.432 (Table 5 11). Short courses OCS for asthma exacerbations were less likely prescribed to low do se ICS/LABA than medium dose ICS/LABA combination therapies (HR, 0.75; 95%CI, 0.59 0.97). Consistent with
175 prescribing long courses OCS, LABA stoppers were less likely to receive short courses OCS than original dose continuers (HR, 0.19; 95%CI, 0.1 0.38), but more likely to receive short courses OCS than dose reducers (HR, 1.24; 95%C I, 1.07 5.01). Model AIC value equals to 7661.84 (Table 5 11). Moreover, asthmatics who discontinued LABA and continued medium dose ICS as monotherapy had 3% more likelihood to be admitted to A&E departments for asthma exacerbations than asthmatics who red uced the dose of ICS to low dose while maintaining LABA as an alternative step down therapy approach (HR, 1.03; 95%CI, 0.04 7.02; model AIC, 432.701) (Table 5 12). Marginal structural models In the original cohort, LABA initiators were 14 % more likely to receive OCS than ICS initiators ( HR, 1. 14 ; 95%CI, 1.03 1.22 Conversely, combination therapy initiators were less likely to receive OCS than ICS initiators (HR, 0.91; 95%CI, 0.41 1. 0 ) or LABA initiators (HR, 0. 23 95%CI, 0. 1 0.34 ). Compared to previous mod el, MSM model was better fitted (Cox AIC, 72818.935; MSM AIC, 62275.601) and statistical significance was borderline between combination therapy and ICS monotherapy comparisons Likewise, initiators of LABA monotherapy had 10 % higher risk of receiving pres criptions for short courses OCS than initiators of ICS monotherapy (HR, 1. 10; 95%CI, 1.07 1.18 ). Prescribing short courses OC S for asthma exacerbations was 62 % less likely in ICS/LABA combination therapy initiators than ICS monotherapy initiators (HR, 0. 38 ; 95%CI, 0. 12 0.66 ) and 50% less likely in combination therapy initiators than LABA monotherapy initiators (HR, 0.5; 95%CI, 0.14 0.78). Compared to previous Cox model, MSM model for predicting short courses OCS was better fitted (Cox AIC, 70778.622; MSM AI C, 59429.489). Comparisons between combination therapy and LABA monotherapy groups were statistically significant (Table
176 5 9). Additionally, the hazard of visiting A&E departments for asthma exacerbations for LABA monotherapy initiators is 1% of the hazard for ICS monotherapy initiators (HR, 1.01; 95%CI, 0.05 8.02). Combination therapy initiators were less likely to have asthma related visits to A&E departments compared to ICS monotherapy initiators (HR, 0.41; 95%CI, 0.03 3.14) or LABA monotherapy initiator s (HR, 0.31; 95%CI, 0.05 2.06). Noticeably, none of the A&E visit estimates were statistically significant although the model was better fitted compared with the previous Cox model (Cox AIC, 7631.89; MSM AIC, 6899.302) (Table 5 12). In tandem with time de pendent Cox model, MSM showed statistical significance differences in asthma deaths between LABA monotherapy initiators and ICS monotherapy initiators (HR, 1.25; 95%CI, 1.11 3.01; MSM AIC, 2089.231). Combination therapy initiators were more likely to exper ience asthma deaths than ICS monotherapy counterparts (HR, 2.12; 95%CI, 0.13 35.9) or LABA monotherapy initiators (HR, 1.2; 95%CI, 0.04 15.3) (Table 5 13). Nevertheless, the model did not detect statistically significant differences between exposure groups in terms of all cause mortality ( MSM AIC, 7911.483 ) (Table 5 14), where LABA initiators had 15% more likelihood of death regardless the cause compared to ICS initiators (HR, 1.15; 95%CI, 0.63 1.78). On the other hand, ICS/LABA combination therapy initiato rs had less likelihood of dying from any reason than ICS monotherapy initiators (HR, 0.4; 95%CI, 0.15 1.52) or LABA monotherapy initiators (HR, 0.31; 95%CI, 0.1 1.31). Model selection. Unadjusted models are not useful to consider given the nature of observ ational data, and models adjusted for baseline covariates do not adequately control for variations contributed by time dependent covariates and exposures.
177 Estimates from marginal structural models and Cox models with time dependent covariates are compared and contrasted in terms of AIC values for model fi t evaluations, and graphically to determine if time dependent confounding was present. For all models, MSM had lower AIC values than time dependent Cox mod els, indicating better fit. With regard to models i n step down therapy approaches, AIC values were higher in time dependent Cox models than baseline adjusted models, an expected finding since AIC penalizes models with larger number of covariates. Figure s 5 13 and 5 14 depicts hazard ratios and correspondin g confidence intervals of OCS, short course of OCS, A&E visits, asthma deaths, and all cause deaths classified by exposure group comparisons and regression models. In general, confidence intervals for estimates derived by MSM were narrower than correspondi ng confidence intervals derived by time dependent Cox models. Interestingly, MSM confidence intervals overlapped with outcome specific intervals derived by Co x models for mortality outcomes and A&E department visits, but did not overlap for prescribing OCS outcomes, suggesting the presence of time dependent confounding with regard to exposure effect on prescribing OCS, and absence of such confounding on deaths and A&E department visits. Sensitivity Analyses Prescriptions for oral corticosteroids were includ ed as a covariate instead of an outcome to account for asthma severity, and HR estimates were not significantly different in terms of asthma related A&E department visit s Time dependent Cox model: LABA vs. ICS (HR, 1.05; 95%CI 0.29 16.1); ICS/LABA vs. ICS (HR, 0.71; 95%CI, 0.1 4.91); and ICS/LABA vs. LABA (HR, 0.61; 95%CI, 0.06 6.89) MSM: LABA vs. ICS (HR, 1.0; 95%CI 0.03 6.71); ICS/LABA vs. ICS (HR, 0.39; 95%CI, 0.04 3.0); and ICS/LABA
178 vs. LABA (HR, 0.28; 95%CI, 0.0 6 1.98 ) However, there was a noticeabl e difference in terms of asthma related mortality, where MSM estimates had narrower confidence intervals and weaker HR estimates and the statistical significance between LABA initiators and ICS initiators was disappeared. However, Cox model estimates did n ot change significantly, suggesting prescribing OCS can be a time dependent confounder when asthma mortality is the outcome of interest. MSM: LABA vs. ICS (HR, 1.04; 95%CI, 0.95 2.0); ICS/LABA vs. ICS (HR, 1.07; 95%CI, 0.08 3.01); and ICS/LABA vs. LABA (HR 0.95; 95%CI, 0.02 5.61). Cox model: LABA vs. ICS (HR, 2.59; 95%CI 1.31 4.01); ICS/LABA vs. ICS (HR, 3.99; 95%CI, 0.28 50); and ICS/LABA vs. LABA (HR, 1.41; 95%CI, 0.08 19.1). M oreover, the addition of patients with coexisting COPD did not impart signific ant changes to morbidity or mortality estimates, where the direction and the magnitude of HR were similar to estimates obtained in original cohort with exclude d COPD patients; however, s ome estimates gained statistical significance in MSM models: prescribi ng OCS for ICS/LABA vs. ICS (HR, 0.82; 95%CI, 0.32 0.92); ICS/LABA vs. LABA (HR, 0.21; 95%CI, 0.1 0.33); LABA vs. ICS (HR, 1.11; 95%CI, 0.98 1.19). Asthma A&E visits: LABA vs. ICS (HR, 0.91; 95%CI 0.1 2.38); ICS/LABA vs. ICS (HR, 0.32; 95%CI, 0.11 0.57); I CS/LABA vs. LABA (HR, 0.28; 95%CI, 0.17 1.08). These findings suggest LABA products have beneficial effects on COPD outcomes Asthma deaths: LABA vs. ICS (HR, 1.19; 95%CI, 1.05 2.05); ICS/LABA vs. ICS (HR, 1.97; 95%CI, 0.1 21.4); ICS/LABA vs. LABA (HR, 1.1 ; 95%CI, 0.1 9.14). Figure 5 15 illustrates the distribution of stabilized weights over study follow up year by every month of exposure measurement. On average, the weights have a mean
179 of 1.02 (min, 0.36; max, 2.66) compared to unstabilized weights (mean, 2.33; min, 1.13; max, 3.32) indicating model satisfaction with positivity assumption. I n addition, there was no tangible differences in MSM derived morbidity outcomes when patient follow up changed to 12 months after first prescriptions of study drugs (i.e study index date for morbidity outcome followed index date for mortality outcome Figure 4 2). Prescribing OCS: LABA vs. ICS (HR, 1.08; 95%CI, 0.99 1.18); ICS/LABA vs. ICS (HR, 0.87; 95%CI, 0.37 0.98); and ICS/LABA vs. LABA (HR, 0.18; 95%CI, 0.07 0.21). A sthma related A&E department visits: LABA vs. ICS (HR, 0.97; 95%CI, 0.03 6.7); ICS/LABA vs. ICS (HR, 0.37; 95%CI, 0.03 3.0); and ICS/LABA vs. LABA (HR, 0.28; 95%CI, 0.05 1.9).
180 Table 5 1. Patient characteristics for original study cohort stratified by exp osure type (January 4, 1993 August 20, 2010) Characteristic Exposure group, n=51,103 p value ICS 46,92 8 (91.8) LABA 714 (1.4) ICS/LABA 3,461 (6.8) A ge (years) 37.2 (15.3) 40.4 (15.2) 40.0 (14.7) <0.0 01 Sex (Female) 26,641 ( 56.8 ) 404 (56.6) 1,989 (57. 5) 0.7197 Marital status Unmarried Married Unknown 4,537 (9.7) 6,843 (14.6) 35,548 (75.7) 70 (9.8) 127 (17.8) 517 (72.4) 342 (9.9) 556 (16.1) 2,563 (74.0) 0.0191 Weight status Non obese Obese Unknown 12 (0.03) 63 (0.1) 46,853 (99.8) 0 1 (0.1) 713 (9 9.9) 2 (0.06) 7 (0.2) 3,452 (99.7) 0. 6454 Smoking status Nonsmoker Former smoker Passive smoker Unknown 40,798 (87.0) 4,385 (9.3) 473 (1.0) 1,272 (2.7) 599 (83.9) 91 (12.7) 5 (0.7) 19 (2.7) 2,957 (85.4) 406 (11.7) 20 (0.6) 78 (2.3) <0.001 Prescriptio n payment Not exempted Exempted Unknown 133 (0.3) 521 (1.1) 46,274 (98.6) 2 (0.3) 5 (0.7) 707 (99.0) 11 (0.3) 54 (1.6) 3,396 (98.1) 0.1265 Capitation supplement level Low Medium High Not applicable Unknown 334 (0.7) 222 (0.5) 80 (0.2) 21,396 (45.6) 24 ,896 (53) 4 (0.6) 2 (0.3) 1 (0.1) 239 (33.5) 468 (65.5) 21 (0.6) 14 (0.4) 5 (0.1) 1,264 (36.5) 2,157 (62.4) <0.001 Registration duration (months) n=46,0 39 1 19.8 (12 3 7 ) n=691 11 7.0 (133.0 ) n=3,436 1 20 6 (131.0) 0.7837 Consultation duration < 10 minutes >10 minutes Unknown 29,819 (63.5) 17,075 (36.4) 34 (0.1) 515 (72.1) 196 (27.5) 3 (0.4 2,028 (58.6) 1,426 (41.2) 7 (0.2) <0.001 Urgency of visit to practice Not urgent visit Urgent visit Unknown 46,493 (99.1) 401 (0.8) 34 (0.1) 710 (99.5) 1 (0.1) 3 (0 .4) 3,423 (98.9) 31 (0.9) 7 (0.2) <0.001 General practice location England Scotland Wales Northern Ireland Unknown 37,393 (79.7) 4,082 (8.7) 4,393 (9. 3 ) 940 (2.0) 120 (0.3) 507 (71.0) 84 (11.8) 99 (13.8) 24 (3.4) 0 2,581 (74.6) 256 (7.4) 440 (12.7) 17 9 (5.2) 5 (0.1) <0.001
181 Table 5 1. Continued Characteristic Exposure group, n=51,103 p value ICS 46,92 8 (91.8) LABA 714 (1.4) ICS/LABA 3,461 (6.8) Asthma severity 12 months prior to index date Prescription for OCS A&E department visit Hospitalizatio n No. of SABA prescriptions < 6 > 6 2,673 (5.7) 126 (0.3) 1 (<0.1) 2.2 (7.7) 42,717 (91) 4,211 (9.0) 46 (0.1) 2 (0.3) 0 3.2 (12.0) 637 (89.2) 77 (10.8) 257 (0.5) 20 (0.6) 0 2.2 (9.8) 3,179 (91.9) 282 (8.1) <0.001 0.0048 0.9565 0.0042 0.0575 Asthma s everity at index date Prescription for SABA No. of asthma drug classes < 2 >2 25,500 (54.3) 1.5 (0.5) 46,698 (99.5) 230 (0.5) 219 (30.7) 1.3 (0.5) 703 (98.5) 11 (1.5) 1,339 (38.7) 2.4 (0.5) 2,083 (60.2) 1,378 (39.8) <0.001 <0.001 <0.001 Asthma controll ed at baseline 18,293 (39.0) 429 (60.1) 1,787 (51.6) <0.001 Concurrent asthma medications LTRA Oral methylxanthines Inhaled MCS Inhaled MRA 86 (0.2) 99 (0.2) 64 (0.1) 94 (0.2) 14 (2.0) 6 (0.8) 4 (0.6) 5 (0.7) 53 (1.5) 24 (0.7) 2 (0.06) 24 (0.7) <0.001 <0.001 0.0042 <0.001 Other concurrent medications Antibiotics for RTI Antivirals for RTI Nasal CS Nasal MCS Nasal antihistamines Nasal decongestants Antitussives Selective beta 1 blockers Oral Ophthalmic NSAIDs Aspirin Acetaminophen Opioid analgesics Ora l cholinergics 4,509 (9.6) 6 (0.01) 1,686 (3.6) 15 (0.03) 17 (0.04) 17 (0.04) 161 (0.3) 116 (0.2) 115 (0.2) 1 (<0.1) 288 (0.6) 145 (0.3) 191 (0.4) 546 (1.2) 1 (<0.1) 54 (7.6) 0 23 (3.2) 2 (0.3) 0 0 2 (0.3) 2 (0.3) 2 (0.3) 0 5 (0.7) 7 (1.0) 2 (0.3) 11 (1. 5) 0 306 (8.8) 0 164 (4.7) 0 0 0 8 (0.2) 13 (0.4) 13 (0.4) 0 32 (0.9) 26 (0.7) 21 (0.6) 73 (2.1) 0 0.0658 0.7657 0.002 <0.001 0.4693 0.4693 0.5285 0.3508 0.3355 0.9565 0.0830 <0.001 0.1810 <0.001 0.9565 Concurrent immunizations Influenza vaccine Pneumoc occal PS vaccine Other vaccines 154 (0.3) 6 (0.01) 37 (0.1) 2 (0.3) 0 0 24 (0.7) 3 (0.1) 4 (0.1) 0.0021 0.0063 0.5701 Annual quarter at index date 1 st (January March) 2 nd (April June) 3 rd (July September) 4 th (October December) 11,635 (24.8) 12,230 ( 26.0) 10,767 (23.0) 12,296 (26.2) 160 (22.4) 176 (24.7) 182 (25.5) 196 (27.4) 902 (26.1) 876 (25.3) 779 (22.5) 904 (26.1) 0.2568
182 Table 5 1. Continued Characteristic Exposure group, n=51,103 p value ICS 46,92 8 (91.8) LABA 714 (1.4) ICS/LABA 3,461 (6. 8) Comorbidities Atopic conditions Allergic rhinosinusitis Allergic conjunctivitis Atopic dermatitis Psoriasis Respiratory allergies Other allergies Respiratory tract infections Otitis media Pharyngolaryngitis Influenza Bronchitis Pneumonia Other infecti ons Psychosocial pathologies Anxiety APD Depression Other conditions Tranquilizer use Antipsychotic use Antidepressant use 1,697 (3.6) 968 (2.1) 25 (0.05) 282 (0.6) 26 (0.06) 7 (0.01) 448 (1.0) 1,139 (2.4) 31 (0.07) 148 (0.3) 53 (0.1) 3 (0.01) 4 (0.01) 61 0 (1.3) 618 (1.3) 15 (0.03) 1 (0.01) 129 (0.3) 21 (0.04) 65 (0.1) 32 (0.07) 484 (1.0) 29 (4.1) 13 (1.8) 2 (0.3) 4 (0.5) 1 (0.1) 0 10 (1.4) 15 (2.1) 0 3 (0.4) 1 (0.1) 0 0 8 (1.1) 14 (2.0) 0 0 1 (0.1) 1 (0.1) 3 (0.4) 0 10 (1.4) 120 (3.5) 66 (2.0) 2 (0.06) 18 (0.5) 2 (0.06) 0 35 (1.0) 76 (2.2) 0 11 (0.3) 1 (0.03) 0 2 (0.06) 44 (1.3) 83 (2.4) 2 (0.06) 0 14 (0.4) 4 (0.1) 11 (0.3) 5 (0.1) 65 (1.9) 0.7315 0.7492 0.0411 0.8307 0.6410 0.7324 0.4612 0.5998 0.2516 0.8849 0.3348 0.8751 0.0342 0.9076 <0.001 0.6417 0. 8751 0.2945 0.1158 0.0067 0.2105 <0.001 Asthma action plan Available Not available Unknown 564 (1.2) 1 (<0.1) 46,363 (98.8) 0 0 714 (100) 74 (2.1) 0 2,287 (97.9) <0.001 Asthma medication compliance Satisfactory Unsatisfactory Unknown 101 (0.2) 27 (0. 1) 46,800 (99.7) 0 0 714 (100) 8 (0.2) 2 (0.1) 3,451 (99.7) 0.7359 General compliance level Good Poor Unknown 40 (0.1) 7 (0.01) 46,881 (99.9) 1 (0.1) 0 713 (99.9) 5 (0.1) 0 3,456 (99.9) 0.7208 Inhaler device type for exposure pMDI BAI DPI Unknown Ae rosol Powder Unknown 37,230 (79.3) 5,433 (11.6) 4,158 (8.9) 107 (0.2) 42,696 (91.0) 4,169 (8.9) 63 (0.1) 527 (73.8) n/a 187 ( 26.2 ) 0 527 (73.8) 187 (26.2) 0 1,088 (31.4) n/a 1,517 (43.8) 856 (24.8) 1,823 (52.7) 1,638 (47.3) 0 <0.001 <0.001 Spacer w as prescribed 4,944 (10.5) 23 (3.2) 230 (6.6) <0.001 Nebulizer was prescribed 36 (0.08) 0 1 (0.03) 0.4622
183 ICS, inhaled corticosteroids; LABA, long acting beta agonists OCS, oral corticosteroids; A&E, accident and emergency SABA, inha led short acting beta agonists LTRA, le ukotriene receptor antagonists MCS, mast cell stabilizers MRA, m uscarinic receptor antagonists RTI, respiratory tract infections NSAIDs, non steroidal anti inflammatory d rugs PS, polysaccharide APD, affective personality disorders pMDI, p r essurized metered dose inhaler BAI, breath actuated inhaler DPI, dry powder inhaler Chi squared test used for categorical characteristics, and analysis of variance (ANOVA) test is used for continuous characteristics Numbers and percentages are reported for categorical factors, and means and corresponding standard deviations are reported for continuous factors
184 Table 5 2. Patient characteristics for step down therapy cohort stratified by approach type among original hi gh ICS/LA BA dose initiators (January 5, 1993 July 12 2010) Characteristic Exposure group + n= 645 p value Original d ose 136 (21. 1 ) Dose r educer 337 (52. 2 ) LABA stopper 172 (26. 7 ) Age (years) 4 1 5 (13. 6 ) 41. 4 (14.0) 43.5 (13. 3 ) 0.02 1 Sex (Female) 76 ( 5 5 9 ) 190 (56.4 ) 98 (56. 9 ) 0.966 Marital status Unmarried Married Unknown 14 (10. 3 ) 22 (16. 2 ) 100 (73. 5 ) 29 (8. 6 ) 58 (17. 2 ) 250 (74. 2 ) 12 (7. 0 ) 24 (14. 0 ) 136 (7 9 0 ) 0. 1704 Weight status Non obese Obese Unknown 1 (0. 7 ) 1 (0. 7 ) 134 (9 8 6 ) 0 4 ( 1.2 ) 333 (9 8 8 ) 0 1 (0. 6 ) 171 (99. 4 ) 0. 3736 Smoking status Nonsmoker Former smoker Passive smoker Unknown 115 (8 4 5 ) 17 (1 2 5 ) 2 (1. 5 ) 2 (1. 5 ) 287 (85. 1 ) 45 (13. 3 ) 2 (0. 6 ) 3 ( 1.0 ) 154 (89. 5 ) 13 (7. 6 ) 2 ( 1.2 ) 3 (1. 7 ) 0.0 246 Prescription payment Not exempted Exempted Unknown 1 (0. 7 ) 3 (2. 2 ) 132 (97. 1 ) 3 (0. 9 ) 8 ( 2. 4 ) 326 (97. 7 ) 3 ( 1.7 ) 5 (2. 9 ) 164 ( 95.3 ) 0.7 881 Capitation supplement level Low Medium High Not applicable Unknown 2 ( 1. 5 ) 0 1 (0. 7 ) 42 (30. 9 ) 91 (66.9 ) 5 ( 1.5 ) 1 (0. 3 ) 2 (0. 6 ) 125 ( 37. 1 ) 204 (62. 5 ) 3 ( 1.7 ) 0 1 (0. 7 ) 105 (61. 0 ) 63 ( 36.6 ) <0.001 Registration duration (months) n= 121 11 2 5 (14 4 5 ) n= 307 1 16.1 (134. 4 ) n= 153 1 09 0 (13 1 5 ) 0. 6382 Consultation duration < 10 minutes >10 minutes Unknown 90 ( 66.2 ) 46 ( 33 8 ) 0 189 (56. 1 ) 148 (43. 9 ) 0 140 (81. 4 ) 31 (18. 0 ) 1 (0. 6 ) <0.001 Urgency of visit to practice Not urgent visit Urgent visit Unknown 132 (97. 0 ) 4 (3.0 ) 0 334 (99. 1 ) 3 ( 0.9 ) 0 171 (99. 4 ) 1 (0. 6 ) 0 0.06 1 General practice location England Scotland Wales Northern Ireland Unknown 105 (7 7 2 ) 7 (5. 1 ) 19 ( 14.0 ) 5 (3. 6 ) 0 258 (76. 6 ) 20 ( 6.0 ) 41 (12. 2 ) 17 (5. 0 ) 1 (0. 2 ) 138 (80. 2 ) 14 (8. 1 ) 15 ( 8 7 ) 5 ( 3.0 ) 0 0. 1186
185 Table 5 2. Continued Characteristic Exposure group + n= 645 p value Original dose 136 (21. 1 ) Dose reducer 337 (52. 2 ) LABA stopper 1 72 (26. 7 ) Asthma severity at index date Prescription for SABA No. of asthma drug classes < 2 >2 53 (3 9 0 ) 2. 5 (0.6) 79 (58. 0 ) 57 (4 2 0 ) 145 ( 43.0 ) 2.5 (0.6) 189 (56. 0 ) 148 (4 4 0 ) 89 (51. 7 ) 1.5 (0.5) 171 (9 9 4 ) 1 ( 0 6 ) <0.001 <0.001 <0.001 Concurrent asthma medications LTRA Oral methylxanthines Inhaled MCS Inhaled MRA 6 (4. 4 ) 3 (2. 2 ) 1 (0. 7 ) 2 (1. 4 ) 5 (1.5 ) 2 (0.6) 0 3 (0. 9 ) 1 (0. 6 ) 1 (0. 6 ) 0 2 ( 1.1 ) <0.001 0.0485 0.1547 0. 1237 Other concurrent medications Antibiotics for RTI Antivirals for RTI Na sal CS Nasal MCS Nasal antihistamines Nasal decongestants Antitussives Selective beta 1 blockers Oral Ophthalmic NSAIDs Aspirin Acetaminophen Opioid analgesics Oral cholinergics 14 (10. 3 ) 0 8 (6.0 ) 0 0 0 1 (0. 7 ) 2 ( 1.4 ) 2 ( 1.4 ) 0 2 (1. 4 ) 2 ( 1.4 ) 1 (0. 7 ) 4 ( 3.0 ) 0 26 (7. 7 ) 0 15 (4.4 ) 0 0 0 1 (0.3) 4 ( 1.2 ) 4 ( 1.2 ) 0 3 (0. 9 ) 3 ( 0. 9 ) 2 (0. 6 ) 10 (3.0) 0 18 (10.4 ) 0 9 (5. 2 ) 1 (0. 6 ) 0 0 1 (0 .6 ) 1 (0. 6 ) 1 (0. 6 ) 0 3 (1. 7 ) 1 (0.6) 1 (0. 6 ) 3 ( 1. 7 ) 0 0. 1476 n/a 0.44 48 0.2 505 n/a n/a 0.9 321 0.7 361 0.7 361 n/a 0. 1675 0. 7878 0.50 83 0. 2192 n/a Concurrent immunizations Influenza vaccine Pneumococcal PS vaccine Other vaccines 1 (0. 7 ) 0 0 3 (0. 9 ) 1 (0. 3 ) 1 (0. 3 ) 0 0 1 (0. 6 ) 0.11 55 0.633 8 0.6 57 Annual quarter at index date 1 st (January March) 2 nd (April June) 3 rd (July September) 4 th (October December) 3 3 (24. 3 ) 37 (27. 2 ) 30 (22. 0 ) 36 (2 6 5 ) 82 (24. 3 ) 84 (25. 0 ) 81 (2 4 0 ) 90 (26. 7 ) 40 (23. 2 ) 42 (24. 4 ) 42 (24. 4 ) 48 (2 8.0 ) 0. 8884
186 Table 5 2. Continued Characteristic Exposure group + n= 645 p value Original dose 136 (21. 1 ) Dose reducer 337 (52. 2 ) LABA stopper 172 (26. 7 ) Comorbidities Atopic conditions Allergic rhinosinusitis Allergic conjunctivitis Atopic dermatitis Psoriasis Respiratory allergies Other allergies Respiratory tract infections Otitis media Pharyngola ryngitis Influenza Bronchitis Pneumonia Other infections Psychosocial pathologies Anxiety APD Depression Other conditions Tranquilizer use Antipsychotic use Antidepressant use 6 (4. 4 ) 3 (2. 2 ) 0 2 (1. 5 ) 0 0 1 ( 0.7 ) 6 ( 4.4 ) 0 2 ( 1.4 ) 0 0 1 (0. 7 ) 3 (2. 2 ) 6 ( 4.4 ) 0 0 1 (0. 7 ) 0 2 ( 1.4 ) 0 3 ( 2.2 ) 13 (3. 8 ) 7 ( 2.0 ) 0 2 (0.6) 0 0 4 (1. 2 ) 5 (1. 5 ) 0 1 (0. 3 ) 0 0 0 4 ( 1.2 ) 15 ( 4.4 ) 0 0 3 (0. 9 ) 1 (0. 3 ) 1 (0. 3 ) 1 (0. 3 ) 9 ( 2.6 ) 5 ( 3.0 ) 3 (1. 7 ) 0 1 (0. 6 ) 0 0 2 ( 1.1 ) 6 ( 3. 5 ) 1 (0. 6 ) 0 0 0 0 5 ( 3.0 ) 8 ( 4.6 ) 1 (0. 6 ) 0 2 ( 1. 1 ) 1 (0. 6 ) 0 1 (0. 6 ) 3 (1. 7 ) 0. 1258 0. 3672 n/a 0. 2768 n/a n/a 0.4 383 0.0 175 0.2 505 0.1 386 n/a n/a 0.1 547 0.0 898 0. 5838 0.2 505 n/a 0.7 389 0.66 85 0. 2660 0. 6685 0. 3227 Asthma action plan Available Not available Un known 2 (1. 5 ) 0 134 (98. 5 ) 6 (1.8) 0 3 31 ( 98.2) 1 (0.6) 0 171 (99.4) 0. 1991 Asthma medication compliance Satisfactory Unsatisfactory Unknown 1 (0. 7 ) 0 135 (99. 3 ) 4 ( 1.2 ) 1 (0. 3 ) 332 (9 8 .5) 0 0 172 (100) 0.56 06 General compliance level Good Poor Unknown 2 ( 1.4 ) 0 134 (9 8.6 ) 1 (0. 3 ) 0 336 (9 9 7 ) 0 0 172 (100) 0.1 386 Inhaler device type for exposure Aerosol ( pMDI & BAI) Powder (DPI) Unknown 115 (8 4 .6) 21 ( 15.4 ) 0 209 ( 62. 0 ) 128 (38.0 ) 0 114 (66. 3 ) 58 (33. 7 ) 0 <0.001 Spacer was prescribed 14 (10. 3 ) 24 ( 7. 1 ) 14 (8. 1 ) 0. 1401 Nebulizer was prescribed 0 1 (0.1) 0 0.633 8 ICS, inhaled corticosteroids L ABA, long acting beta agonists OCS, oral corticosteroids A&E, accident and emergency
187 SABA, inha led short acting beta agonists LTRA, le ukotriene receptor antagonists MCS, mast cell stabilizers MR A, m uscarinic receptor antagonists RTI respiratory tract infections NSAIDs, non steroidal anti inflammatory d rugs PS, polysaccharide APD, affective p ersonality disorders pMDI, pr essurized metered dose inhaler BAI, breath actuated inhaler DPI, dry powder i nhaler Chi squared test used for categorical characteristics, and analysis of variance (ANOVA) test is used for contin uous characteristics Numbers and percentages are reported for categorical factors, and means and corresponding standard deviations are reported for continuous factors + Original dose users are patients with ICS/LABA combination therapy with high dose ICS, dose reducers are patients who step down to ICS/LABA combination therapy with medium dose ICS, and LABA stopper s are high dose ICS monotherapy users
188 Table 5 3. Patient characteristics for step down therapy cohort stratified by approach type among original medi um ICS/LABA dose initiators (January 4, 1993 August 11, 2010) Characteristic Exposure group + n= 2,581 p value Original dose 320 (12. 4 ) Dose reducer 519 (20. 1 ) LABA stopper 1,742 (67. 5 ) Age (years) 41. 4 (14.0) 38. 2 (15.0) 41. 3 (14.0) <0.001 Sex ( Female ) 181 (56. 5 ) 301 ( 58.0 ) 944 (54. 2 ) 0.02 64 Marital status Unmarried Married Unknown 28 (8. 7 ) 56 (17. 5 ) 236 (7 3 8 ) 55 (10. 6 ) 78 (15. 0 ) 386 (74.4) 136 (7. 8 ) 307 (17. 6 ) 1,299 (74. 6 ) 0.00 34 Weight status Non obese Obese Unknown 0 4 ( 1.3 ) 316 (9 8 7 ) 0 1 (0. 2 ) 518 (99. 8 ) 1 (0.0 5 ) 6 (0. 3 ) 1,735 (99. 6 ) 0. 1629 Smoking status Nonsmoker Former smoker Passive s moker Unknown 272 (85.0 ) 43 (13. 4 ) 2 (0. 6 ) 3 (1. 0 ) 447 (86. 1 ) 53 (10.2) 3 (0. 6 ) 16 (3. 1 ) 1,512 (86. 8 ) 188 (10. 8 ) 14 (0.8) 28 (1.6) 0.000 6 Prescription payment Not exempted Exempted Unknown 1 (0.3) 1 0 ( 3.1 ) 309 ( 96. 6 ) 3 (0. 6 ) 6 (1. 1 ) 510 (98. 3 ) 2 (0. 1 ) 24 (1. 4 ) 1,716 (98. 5 ) 0. 5353 Capitation supplement level Low Medium High Not applicable Unknown 2 (0. 6 ) 1 (0. 3 ) 1 (0. 3 ) 118 (3 6. 8 ) 198 (6 2 0 ) 3 (0. 6 ) 2 (0.4) 1 (0. 2 ) 177 (3 4 .1 ) 336 (6 4 7 ) 12 (0.7) 7 (0. 4 ) 2 (0.1 ) 927 (53. 2 ) 794 (45. 5 ) <0.001 Regist ration duration (months) n= 317 1 16 2 (134. 4 ) n= 503 1 08 2 (1 17 1 ) n= 1,697 12 1 6 (13 6 2 ) 0.0 0 21 Consultation duration < 10 minutes >10 minutes Unknown 179 (5 6.0 ) 140 (4 3 8 ) 1 (0.2) 303 (58. 4 ) 215 (41. 4 ) 1 (0.2) 1,261 (72.4) 479 (27.5) 2 (0.1) <0.001 Urge ncy of visit to practice Not urgent visit Urgent visit Unknown 317 (99.1) 2 (0. 6 ) 1 (0. 3) 514 (99. 0 ) 4 (0. 8 ) 1 (0. 2 ) 1,730 (99. 3 ) 10 (0. 6 ) 2 (0.1) 0. 8892 General practice location England Scotland Wales Northern Ireland Unknown 245 (76. 5 ) 19 ( 6.0 ) 39 (12. 2 ) 16 (5. 0 ) 1 (0. 3 ) 398 (7 6 7 ) 36 (7. 0 ) 63 (12. 1 ) 21 (4.0 ) 1 (0.2) 1,324 (76.0 ) 171 (9. 8 ) 174 (10. 0 ) 70 (4.0) 3 (0.2) 0.0003
189 Table 5 3. Continued Characteristic Exposure group + n= 2,581 p value Original dose 320 (12. 4 ) Dose reducer 519 (20. 1 ) LA BA stopper 1,742 (67. 5 ) Asthma severity at index date Prescription for SABA No. of asthma drug classes < 2 >2 137 (42. 8 ) 2.5 (0.5 ) 179 (56. 0 ) 141 (4 4.0 ) 201 (38. 7 ) 2.4 (0.5) 310 ( 59 7 ) 209 ( 40. 3 ) 923 (5 3 0 ) 1.5 (0.5) 1,724 (99.0) 18 (1.0) <0.001 <0.00 1 <0.001 Concurrent asthma medications LTRA Oral methylxanthines Inhaled MCS Inhaled MRA 5 (1.5) 2 (0.6) 0 3 (1.0 ) 6 (1. 1 ) 3 (0. 6 ) 1 (0. 2 ) 2 (0. 4 ) 3 (0.2) 9 (0.5) 2 (0.1) 7 (0.4) <0.001 0.9 353 0.3 509 0. 4335 Other concurrent medications Antibiotics fo r RTI Antivirals for RTI Nasal CS Nasal MCS Nasal antihistamines Nasal decongestants Antitussives Selective beta 1 blockers Oral Ophthalmic NSAIDs Aspirin Acetaminophen Opioid analgesics Oral cholinergics 24 (7. 5 ) 0 14 (4. 4 ) 0 0 0 3 ( 1.0 ) 4 ( 1.2 ) 4 ( 1.2 ) 0 2 (0. 6 ) 4 ( 1.2 ) 3 ( 1.0 ) 10 (3. 1 ) 0 40 (7. 7 ) 0 24 (4. 6 ) 0 0 0 3 (0. 6 ) 5 (1.0 ) 5 ( 1.0 ) 0 4 (0. 8 ) 3 (0. 6 ) 3 (0.6 ) 8 (1. 5 ) 0 174 (10.0 ) 0 64 (3. 7 ) 1 (0.0 6 ) 3 (0. 2 ) 0 10 (0. 6 ) 1 0 (0. 6 ) 1 0 (0. 6 ) 0 10 (0.6) 7 (0.4) 5 (0.3) 24 (1.4) 0 0.00 87 n/a 0. 1617 0.78 57 0.4 85 n/a 0. 577 0.8 571 0.8 571 n/a 0. 6331 0.0 88 0.0 098 0.00 3 n/a Concurrent immunizations Influenza vaccine Pneumococcal PS vaccine Other vaccines 7 ( 2.2 ) 1 (0. 3 ) 1 (0. 3 ) 1 5 ( 3.0 ) 0 2 (0. 4 ) 1 3 (0. 7 ) 1 (0.1 ) 3 (0.2 ) 0. 0 00 4 0. 2495 0.6 445 Annual quarter at index date 1 st (January March) 2 nd (April June) 3 rd (July September) 4 th (October December) 78 (24. 3 ) 79 (2 4 7 ) 77 (2 4. 1 ) 86 (26.9 ) 147 (2 8 3 ) 121 (2 3 3 ) 114 (2 2 0 ) 137 (26. 4 ) 437 (25.1) 439 (25.2) 409 (23. 5 ) 457 (26. 2 ) 0. 1734
190 Table 5 3. Continu ed Characteristic Exposure group + n= 2,581 p value Original dose 320 (12. 4 ) Dose reducer 519 (20. 1 ) LABA stopper 1,742 (67. 5 ) Comorbidities Atopic conditions Allergic rhinosinusitis Allergic conjunctivitis Atopic dermatitis Psoriasis Respiratory alle rgies Other allergies Respiratory tract infections Otitis media Pharyngolaryngitis Influenza Bronchitis Pneumonia Other infections Psychosocial pathologies Anxiety APD Depression Other conditions Tranquilizer use Antipsychotic use Antidepressant use 12 (3 8 ) 6 ( 2.0 ) 0 2 (0.6) 0 0 4 (1. 2 ) 4 (1. 2 ) 0 1 (0. 3 ) 0 0 0 3 ( 1.0 ) 8 (2. 5 ) 0 0 2 (0. 6 ) 1 (0. 3 ) 1 (0. 3 ) 1 (0. 3 ) 3 ( 1.0 ) 17 (3. 3 ) 9 (1. 7 ) 1 (0. 2 ) 3 (0. 6 ) 1 (0. 2 ) 0 3 ( 0.6 ) 11 (2. 1 ) 0 2 (0. 4 ) 0 0 1 (0. 2 ) 8 (1. 5 ) 13 (2. 5 ) 1 (0. 2 ) 0 2 (0.4) 1 (0. 2 ) 2 (0. 4 ) 1 (0 2 ) 6 ( 1.1 ) 52 (3.0) 28 (1.6) 1 (0.0 6 ) 9 (0. 5 ) 1 (0.0 8 ) 1 (0.0 8 ) 12 (0. 7 ) 42 (2.4) 0 3 (0.2) 1 (0.0 6 ) 1 (0.0 6 ) 1 (0.0 6 ) 36 ( 2.0 ) 19 (1.1) 0 0 5 (0. 3 ) 1 (0.0 6 ) 3 (0.2) 1 (0.0 6 ) 9 (0.5 ) 0. 3428 0. 5968 0.29 41 0. 8352 0.6 919 0.78 57 0.2 131 0.0 923 n/a 0. 5477 0.4 85 0.78 57 0.54 24 0. 5309 <0.001 0.019 n/a 0. 3171 0.0 57 1 0.5 271 0. 2963 <0.001 Asthma action plan Available Not available Unknown 6 (1.8) 0 314 (98.2) 14 (2. 7 ) 0 505 (97. 3 ) 10 (0.6) 0 1732 (99.4) <0.001 Asthma medication compliance Satisfactory Unsatisfa ctory Unknown 4 ( 1.3 ) 1 (0. 3 ) 315 (98.4 ) 2 (0. 4 ) 1 (0. 2 ) 516 (99. 4 ) 3 (0.2 ) 0 1,739 (99. 8 ) 0.0 088 General compliance level Good Poor Unknown 1 (0. 3 ) 0 319 (99. 7 ) 2 (0. 4 ) 0 517 (99. 6 ) 4 (0. 2 ) 1 (0.0 5 ) 1,737 (99. 7 ) 0. 9302 Inhaler device type for expo sure Aerosol ( pMDI & BAI) Powder (DPI) Unknown 198 (6 1.9 ) 122 (3 8 1 ) 0 252 (48.5 ) 267 (5 1 .5 ) 0 1,432 (82. 2 ) 308 (17. 7 ) 2 (0. 1 ) <0.001 Spacer was prescribed 22 ( 7.0 ) 32 (6. 2 ) 192 (11. 0 ) <0.001 Nebulizer was prescribed 1 (0. 3 ) 0 1 (0.0 6 ) 0.2 495 ICS, in haled corticosteroids L ABA, long acting beta agonists OCS oral corticosteroids A&E, accident and emergency
191 SABA, inha led short acting beta agonists LTRA, le ukotriene receptor antagonists MCS, mast cell stabilizers MRA, m uscarinic receptor antagonists RTI respiratory tract infections NSAIDs, non steroi dal anti inflammatory drugs PS, polysaccharide APD, a ffective personality disorders pMDI, pr essurized metered dose inhaler BAI, breath actuated inhaler DPI, dry powder inhaler Chi squared test act test are used for categorical characteristics, and analysis of variance (ANOVA) test is used for continuous characteristics Numbers and percentages are reported for categorical factors, and means and corresponding standard deviations are reported for c ontinuous factors + Original dose users are patients with ICS/LABA combination therapy with medium dose ICS, dose reducers are patients who step down to ICS/LABA combination therapy with low dose ICS, and LABA stoppers are me dium dose ICS monotherapy users
192 Table 5 4. Incidence rates of morbidity outcomes among initiators of study drugs Exposure Outcome Oral corticosteroid prescription Asthma A&E department visit No. of cases Rate (95%CI) + No. of cases Rate (95%CI) + Original cohort I CS monotherapy 1,321 10.6 ( 10.0 11.2 ) 15 0.1 ( 0.08 0.2 ) LABA monotherapy 29 19.9 ( 13.8 28.6 ) 1 0.5 ( 0.08 3.9 ) ICS/LABA combination therapy 138 17.7 ( 14.9 20.9 ) 2 0.3 ( 0.08 1.3 ) Step down therapy cohort High dose ICS combination therapy 23 28.9 ( 19.2 43.5 ) 0 n/a Medium dose ICS combination therapy 39 21.9 ( 16.0 29.9 ) 0 n/a High dose ICS monotherapy 16 19.3 ( 11.8 31.5 ) 0 n/a Medium dose ICS combination therapy 39 21.9 ( 16.0 29.9 ) 0 n/a Low dose ICS combination therapy 45 12.0 ( 8. 9 16.1 ) 1 0.3 ( 0.05 2.4 ) Medium dose ICS monotherapy 178 15.5 ( 13.4 17.9 ) 2 0.2 ( 0.04 0.7 ) + c ases per 100 person years A&E, accident and emergency ICS, inhaled corticosteroids LABA, long acting beta ago nists n/a, not applicable
193 Table 5 5. Incid ence rates of prescriptions for short courses of oral corticosteroids for asthma exacerbations among initiators of study drugs Exposure Short course oral corticosteroids No. of cases Rate (95%CI) + Original cohort ICS monotherapy 1,226 9.8 (9.3 10.4) LABA monotherapy 27 18.5 (12.7 27.0) ICS/LABA combination therapy 116 14.8 (12.4 17.8) Step down therapy cohort High dose ICS combination therapy 22 27.7 (18.2 42.0) Medium dose ICS combination therapy 32 17.9 (12.7 25.4) High dose I CS monotherapy 16 19.3 (11.8 31.5) Medium dose ICS combination therapy 32 17.9 (12.7 25.4) Low dose ICS combination therapy 41 10.9 (8.1 14.9) Medium dose ICS monotherapy 173 15.1 (12.9 17.5) + c ases per 100 person years A&E, accident and emergency ICS, inhaled corticosteroids LABA, long acting beta agonists
194 Table 5 6 Incidence rates of mortality outcomes among initiators of study drugs Exposure Outcome Asthma death All cause death No. of cases Rate (95%CI) + No. of cases Rate (95%CI) + O NS linked practices ICS monotherapy 5 5.0 ( 2.1 12.0 ) LABA monotherapy 0 n/a ICS/LABA combination therapy 0 n/a ONS unlinked practices ++ ICS monotherapy 63 0.1 ( 0.1 0.2 ) LABA monotherapy 4 0.6 ( 0.2 1.5 ) ICS/LABA co mbination therapy 16 0.5 ( 0.3 0.7 ) Original cohort ICS monotherapy 215 0.5 ( 0.4 0. 6 ) LABA monotherapy 5 0.7 ( 0.3 1.7 ) ICS/LABA combination therapy 36 1.1 ( 0.8 1.5 ) + c ases per 100 person years ++ algorithm derived outcome ICS, inhaled corticosteroids LABA, long acting beta agonists ONS, office for na tional statistics n/a, not applicable
195 Table 5 7 Characteristics of asthmatics died of asthma that are identified in the ONS Mortality linked database Characteristic No. (%) Initiat ors of inhaled corticosteroids monotherapy 5 (100) >2 asthma medication classes prescribed at index date 5 (100) >6 inhaled SABA prescriptions issued in preceding year 5 (100) Asthma status uncontrolled at index date 5(100) Age (year), Mean (SD) 13 35 44 50 31 (17.3) 2 (40) 1 (20) 1 (20) 1 (20) Sex Female Male 1 (20) 4 (80) Marital status Unmarried Married Unknown 1 (20) 1 (20) 3 (60) Smoking status Nonsmoker Unknown 3 (60) 2 (40) Concomitant medications at index date Inhaled short acting beta ag onists Antibiotics for respiratory infections Aspirin 5 (100) 1 (20) 1 (20) Consultation duration (minute) < 10 >10 4 (80) 1 (20) Inhaler device type for inhaled corticosteroids Aerosols Dry powder 3 (60) 2 (40) Capitation supplement level Not applica ble Unknown 4 (80) 1 (20) Annual quarter at index date 1 st (January March) 2 nd (April June) 3 rd (July September) 4 th (October December) 2 (40) 1 (20) 1 (20) 1 (20) SABA, short acting beta agonist SD, standard deviation
196 Table 5 8. Distributi on of average time to event among exposure groups Exposure T ime to event in days, Mean (SD) Outcome OCS OCS short course A&E visit Asthma death All cause death ICS 351.2 ( 0.4 ) 352.1 ( 0.4 ) 335.0 ( 0.05 ) 365 (0) 364 ( 0.01 ) LABA 340.2 (5.1) 341.3 ( 5.0) 7.0 (n/a) 365 (0) 365 (0) ICS/LABA 343.5 (2.0) 347.4 (1.8) 365.0 (0.1) 365 (0) 365 (0) High dose ICS/LABA 163.0 ( 3.0 ) 164.0 ( 2.7 ) Medium dose ICS/LABA 240.4 ( 2.6 ) 243.7 ( 2.3 ) High dose ICS + 113.6 ( 1.2 ) 113.6 ( 1.2 ) Medium do se ICS/LABA 240.4 ( 2.6 ) 243.7 ( 2.3 ) n/a Low dose ICS/LABA 349.2 ( 2.6 ) 351.0 ( 2.4 ) 1 .0 ( n/a ) Medium dose ICS + 347.4 ( 1.4 ) 348.0 ( 1.4 ) 191.0 () + LABA stopper A&E, accident and emergency ICS, inhaled corticosteroids LABA, long acting beta agonists O CS, oral corticosteroids n/a, not applicable SD, standard deviation
197 Table 5 9. Hazard ratios of oral corticosteroid prescriptions among original cohort Model Outcome Exposure Comparison HR 95%CI AIC Unadjusted OCS LABA vs. ICS 2.92 2.54 3 .36 30568.065 ICS/LABA vs. ICS 9.87 7.05 13.8 ICS/LABA vs. LABA 1.32 1.14 1.52 Short course OCS LABA vs. ICS 2.71 2.40 3.02 28127.583 ICS/LABA vs. ICS 8.83 6.19 12.0 ICS/LABA vs. LABA 1.29 1.08 1.43 Adjusted conve ntional Cox PHREG + OCS LABA vs. ICS 1.78 1.17 2.54 29763.723 ICS/LABA vs. ICS 1.54 1.30 1.86 ICS/LABA vs. LABA 0.95 0.63 1.41 Short course OCS LABA vs. ICS 1.71 1.16 2.49 27505.543 ICS/LABA vs. ICS 1.40 1.15 1.70 ICS/LABA vs. LABA 0.85 0.56 1.30 Adjusted time dependent Cox PHREG ++ OCS LABA vs. ICS 1. 34 1. 27 2. 02 72818.935 ICS/LABA vs. ICS 1.17 1.04 2. 30 ICS/LABA vs. LABA 0.70 0. 4 0 1.64 Short course OCS LABA vs. ICS 1. 47 1. 22 2.44 70778.62 2 ICS/LABA vs. ICS 1.00 0. 66 2.10 ICS/LABA vs. LABA 0. 91 0. 81 1.35 Marginal structural model +++ OCS LABA vs. ICS 1. 14 1.03 1. 22 62275.601 ICS/LABA vs. ICS 0.91 0.41 1. 00 ICS/LABA vs. LABA 0. 23 0. 09 0. 34 Short cour se OCS LABA vs. ICS 1. 10 1. 07 1. 18 59429.489 ICS/LABA vs. ICS 0. 38 0.12 0 66 ICS/LABA vs. LABA 0.50 0.14 0.78 + adjusted for baseline year and index date covariates ++ adjusted for baseline year, index date and time dependent covariates +++ adjusted for all covariates, including time dependent confounders
198 AIC, Akaike information criterion CI, confidence interval ICS, inhaled corticosteroids HR, hazard ratio LABA, long acting beta agonists OCS, oral corticosteroids
199 Table 5 10. Hazard ratios of oral corticosteroid prescription s among step down cohort with high dose ICS/LABA initiators Model Outcome Exposure Comparison HR 95%CI AIC Unadjusted OCS Reducer vs. Original 1.00 0.72 1.68 1082.443 Stopper vs. Original 0.63 0.39 1.67 Stopper vs. Reducer 1.03 0.57 1.89 Short course OCS Reducer vs. Original 0.98 0.70 1.68 974.846 Stopper vs. Original 0.61 0.39 1.66 Stopper vs. Reducer 1.01 0.55 1.84 Adjusted conventional Cox PHREG + OCS Reducer vs. Ori ginal 0.79 0.68 1.07 1080.772 Stopper vs. Original 0.36 0.08 1.17 Stopper vs. Reducer 0.41 0.12 1.07 Short course OCS Reducer vs. Original 0.78 0.68 1.04 974.684 Stopper vs. Original 0.35 0.07 1.16 Stopper vs. Reducer 0. 39 0.08 1.03 Adjusted time dependent Cox PHREG ++ OCS Reducer vs. Original 0.78 0.60 0.95 2334.382 Stopper vs. Original 0.34 0.07 0.55 Stopper vs. Reducer 0.33 0.07 0.52 Short course OCS Reducer vs. Original 0.71 0.60 0. 90 1973.530 Stopper vs. Original 0.32 0.06 0.50 Stopper vs. Reducer 0.35 0.06 0.51 + adjusted for baseline year and index date covariates ++ adjusted for baseline year, index date and time dependent covariates AIC, Akaike information criterio n CI, confidence interval HR, hazard ratio OCS, oral corticosteroids
200 Table 5 1 1 Hazard ratios of oral corticosteroid prescriptions among step down cohort with medium dose ICS/LABA initiators Model Outcome Exposure Comparison HR 95%CI AIC Unadjus ted OCS Reducer vs. Original 1.01 0.44 4.11 4378.684 Stopper vs. Original 1.10 0.76 1.35 Stopper vs. Reducer 2.26 2.02 8.85 Short course OCS Reducer vs. Original 0.94 0.41 3.07 4117.586 Stopper vs. Original 1.08 0.75 1.14 Stopper vs. Reducer 2.24 1.90 2.71 Adjusted conventional Cox PHREG + OCS Reducer vs. Original 0.99 0.78 2.01 4321.999 Stopper vs. Original 1.02 0.32 3.23 Stopper vs. Reducer 1.52 1.14 7.00 Short course OCS Reducer vs. Or iginal 0.83 0.26 2.65 4066.097 Stopper vs. Original 0.61 0.39 0.94 Stopper vs. Reducer 1.36 0.97 6.37 Adjusted time dependent Cox PHREG ++ OCS Reducer vs. Original 0.85 0.67 1.07 8708.432 Stopper vs. Original 0.21 0.09 0.52 Stopper vs. Reducer 1.28 1.09 5.08 Short course OCS Reducer vs. Original 0.75 0.59 0.97 7661.840 Stopper vs. Original 0.19 0.10 0.38 Stopper vs. Reducer 1.24 1.07 5.01 + adjusted for baseline year and index date covariates ++ adjusted for baseline year, index date and time dependent covariates AIC, Akaike information criterion CI, confidence interval HR, hazard ratio OCS, oral corticosteroids
201 Table 5 12. Hazard ratios of asthma related accident and emergency department visits among original cohort and step down cohort with medium dose ICS/LABA initiators Model Outcome Exposure Comparison HR 95%CI AIC Unadjusted Asthma A&E visit LABA vs. ICS 4.04 0.52 35.7 349.218 ICS/LABA vs. ICS 2.25 0.51 9.87 ICS/LABA vs. L ABA 1.01 0.33 7.27 Stopper vs. Reducer 2.53 0.21 22.0 51.318 Adjusted conventional Cox PHREG + Asthma A&E visit LABA vs. ICS 3.96 0.40 30.0 381.544 ICS/LABA vs. ICS 0.99 0.21 7.14 ICS/LABA vs. LABA 0.78 0.20 6.04 Stopper vs. Reducer 1.21 0.07 13.2 128.000 Adjusted time dependent Cox PHREG ++ Asthma A&E visit LABA vs. ICS 1.04 0.32 15.0 7631.890 ICS/LABA vs. ICS 0.72 0.08 5.01 ICS/LABA vs. LABA 0.56 0.11 4.12 Stopper vs. Reducer 1.03 0.04 7.02 432.7 01 Marginal structural model +++ Asthma A&E visit LABA vs. ICS 1.01 0.05 8.02 6899.302 ICS/LABA vs. ICS 0.41 0.03 3.14 ICS/LABA vs. LABA 0.32 0.05 2.06 + adjusted for baseline year and index date covariates ++ adjusted for baseline ye ar, index date and time dependent covariates +++ adjusted for all covariates, including time dependent confounders A&E, accident and emergency AIC, Akaike information criterion CI, confidence interval ICS, inhaled corticosteroids HR, hazard ratio LABA, lon g acting beta agonists
202 Table 5 13. Hazard ratios of asthma deaths among original cohort in practices unlinked to ONS mortality database Model Outcome Exposure Comparison HR 95%CI AIC Unadjusted Asthma death LABA vs. ICS 4.18 1.52 11.5 1782.478 ICS/LABA vs. ICS 5 .47 2.01 6.01 ICS/LABA vs. LABA 1 .83 0.28 4 2.49 Adjusted conventional Cox PHREG + Asthma death LABA vs. ICS 3.20 1. 6 5 8.92 1697.062 ICS/LABA vs. ICS 5 .10 1.01 5.02 ICS/LABA vs. LABA 1 .62 0.09 2 2.1 A djusted time dependent Cox PHREG ++ Asthma death LABA vs. ICS 2. 67 1.44 4.93 2408.120 ICS/LABA vs. ICS 4.20 0.32 54.4 ICS/LABA vs. LABA 1.58 0.11 21.8 Marginal structural model +++ Asthma death LABA vs. ICS 1. 25 1.11 3 01 2 0 89.231 ICS/LABA vs. ICS 2 12 0. 1 3 35 9 ICS/LABA vs. LABA 1 20 0.04 15 3 + adjusted for baseline year and index date covariates ++ adjusted for baseline year, index date and time dependent covariates +++ adjusted for all covariates, including time depend ent confounders AIC, Akaike information criterion CI, confidence interval ICS, inhaled corticosteroids HR, hazard ratio LABA, long acting beta agonists
203 Table 5 14. Hazard ratios of all cause deaths among original cohort Model Outcome Exposure Comparison HR 95%CI AIC Unadjusted All cause death LABA vs. ICS 1.54 0.63 6.01 5504.828 ICS/LABA vs. ICS 2.23 1.57 4.77 ICS/LABA vs. LABA 1.45 0.57 4.98 Adjusted conventional Cox PHREG + All cause death LABA vs. ICS 1. 32 0.49 5.95 5191. 972 ICS/LABA vs. ICS 1.77 0.71 4.00 ICS/LABA vs. LABA 1.39 0.44 4.34 Adjusted time dependent Cox PHREG ++ All cause death LABA vs. ICS 1. 26 0. 83 1 .9 2 8575.759 ICS/LABA vs. ICS 0 61 0.2 1 1.74 ICS/LABA vs. LABA 0.48 0.16 1.46 Marginal structural model +++ All cause death LABA vs. ICS 1.15 0. 63 1.78 7911 .483 ICS/LABA vs. ICS 0. 40 0. 15 1.52 ICS/LABA vs. LABA 0.31 0.0 9 1.31 + adjusted for baseline year and index date covariates ++ adjusted for baseline year, index date and time dependent covariates +++ adjusted for all covariates, including time dependent confounders AIC, Akaike information criterion CI, confidence interval ICS, inhaled corticosteroids HR, hazard ratio LABA, long acting beta agonists
204 First Rx for study drugs on or before January 1, 1993 (n = 29,911 ) Patients aged <13 (n = 12,669) and > 65 years (n = 57,162 ) Chronic obstruct ive pulmonary disease (n = 70,394 ) R espiratory procedures (n = 5,424 ) L ung transplantation (n = 40 ) Lung lobectomy (n = 830 ) Occup ational pneumopathies (n = 2,482 ) Tuberculosis, including Rx fo r anti tuberculosis (n = 5,241 ) Aspergillosis, including Rx f or anti aspergillosis (n = 9,393 ) Pneumocystis pneumonia, including Rx for relevant agents (n = 9 ) Respiratory neoplasm (n = 3,250 ) C ystic fibrosis (n = 323 ) Par enchymal lung disease (n = 53 7 ) Obstructive sleep apnea (n = 4,069 ) \ \ Patients with asthma Initiators of study drugs (n = 308,839) UK GPRD Fig ure 5 1. Cohort sample disposition ICS, inhaled corticosteroids; LABA, long acting beta agonists; Rx, Prescription. Exclusion n umbers are mutually not exclusive, where p atients might have more than one criterion.
205 Respiratory obstruct ion by foreign objects (n = 251 ) \ \ Bronchiec tasis and atelectasis (n = 5,427 ) Congenita l and structural anomalies of respiratory system (n = 1,439 ) Pulmonary h ypertension, embolism, and edema (n = 5,349 ) Con gestive heart failure (n = 10,441 ) Con genital heart disease (n = 535 ) Pulmonary v alvular heart disease (n = 108 ) Unspe cified and other disorders of resp iratory system (n = 26,595 ) Participation in clinical study, including asthma research (n = 8,250 ) Active smokers, including Rx fo r smoking cessation (n = 152,677 ) Illicit drug use (n = 5,430 ) Rx for oral nonselective beta blockers (n = 19,954 ) Rx for ophthalmic non selective beta blockers (n = 2,674 ) Rx for ICS/SABA sing le device combination (n = 2,172 ) Rx for MCS/SABA s ingle device combination (n = 698 ) Rx for MRA/SABA single device combination (n = 23,431 ) Rx for inhaled betamethasone ( n = 145 ) \ \ Figure 5 1. Continued
206 Figure 5 1 Continued Rx for allergen immunotherapy vaccination (n = 27 ) \ \ Rx for omalizumab (n = 1 ) Pa tients with indeterminate sex (n = 4 ) Asthmatic patients included in study (n = 51,10 3 ) ICS initiators (n = 46 92 8) (91.8% ) LABA initiators (n = 714) (1 .4% ) Single device combination (n = 2 692 ) (77.8%) Separate devices (n = 769 ) (22.2%) ICS/LABA initiators (n = 3 461 ) (6.8% ) Medium dose ICS initiators (n = 2,581 ) (80%) Step down therapy cohort (n = 3 226 ) Uncontrolled asthma (n = 235) Original hi gh dose ICS (n = 136) (21.1% ) High dose ICS initiators (n = 645 ) (20%) Medium dose ICS reducers (n = 337) (52.2% ) High dose ICS LABA stoppers (n = 172) (26.7% ) Original medium dose ICS (n = 320) (12.4% ) Low dose ICS reducers (n = 519) (20.1% ) Medium dose ICS LABA stoppers (n = 1,742) (67.5% )
207 Figure 5 2. Distribution o f exposure initiators across UK countries. ICS, inhaled corticosteroids; LABA, long acting beta agonists.
208 Figure 5 3 Prescri bing trend of study exposures ICS, inhaled corticosteroids; LABA, long acting beta agonists.
209 Figure 5 4. Distribution of prescriptions in the step down therapy cohort with original high ICS/LABA dose initiators. ICS, inhaled corticosteroids; LABA, long acting beta agonists
210 Figure 5 5. Distribution of prescriptions in the step down thera py cohort with original medium ICS/LABA dose initiators. ICS, inhaled corticosteroids; LABA, long acting beta agonists
211 A) B) Figure 5 6 Prescribi ng trend of study exposures in step do wn therapy cohort with A) original high ICS/LABA dose initiators and B) original medium ICS/LABA dose initiators. ICS, inhaled corticosteroids; LABA, long acting beta agonists.
212 A) B) Figure 5 7. Product limit surv ival estimates of prescribing A) long courses and B) short courses oral corticosteroids among original cohort of inhaled corticosteroids (ICS), long acting beta agonists (LABA), and ICS/LABA initiators
213 A) B) Figure 5 8. Product limit survival estimates of p rescribing A) long courses and B) short courses oral corticosteroids among high inhaled corticosteroids (ICS)/long acting beta agonists (LABA) dose initiators
214 A) B) Figure 5 9. Product limit survival estimates of prescribing A) long courses and B) short courses oral corticosteroids among medium inhaled corticosteroids (ICS)/long acting beta agonists (LABA) dose initiators
215 Figure 5 10. Product limit survival estimates of attending accident and emergency departments for asthma exacerbations among original cohort of inhaled corticosteroids (ICS), long acting beta agonists (LABA), and ICS/LABA initiators
216 Figure 5 1 1 Product limit survival estimates of asthma related deaths among original cohort of inhaled corticosteroids (ICS), long acting beta agonists (LABA), and ICS/LABA initiators in practices unlinked to the Office of National Statistics mortality database
217 Figure 5 1 2 Product limit survival estimates of all cause deaths among original cohort of inhaled corticosteroids (ICS), long acting beta agonists (LABA), and ICS/LABA initiators
218 Comparison Groups 0 0.5 1.0 1.5 2.0 2.5 Hazard Ratio (95%CI) ICS/LABA vs. LABA ICS/LABA vs. ICS LABA vs. ICS Outcome OCS Short A&E Course Visit OCS Short A&E Course Visit OCS Short A&E Course Visit Time dependent Cox Model Marginal structural model Figure 5 13. Hazard ratios of asthma related morbidity outcomes stratified by comparison groups and regression models
219 Comparison Groups 0 0.5 1.0 1.5 2.0 2.5 Hazard Ratio (95%CI) ICS/LABA vs. LABA ICS/LABA vs. ICS LABA vs. ICS Outcome Asthma All Cause Death Death Time dependent Cox Model Marginal structural model Asthma All Cause Death Death Asthma All Cause Death Death Figure 5 14. Hazard ratios of asthma related and all cause mortali ty outcomes stratified by comparison groups and regression models
220 Figure 5 1 5 Distribution of s tabilized weights estimated by marginal structural models across study follow up year Box and Whiskers: Min, 1st Quartile, Median, 3rd Quartile, Max; Shaded Square: Mean Mean = 1.02 Range = 0.36 2.66 0 0.5 1.0 1.5 2.5 3.0 1 2 3 4 5 6 7 8 9 10 11 12 Follow up Time (Month) Stabilized Weight
221 Figure 5 1 6 D istribution of un stabilized weights estimated by marginal structural models across study follow up year Box and Whiskers: Min, 1st Quartile, Median, 3rd Quartile, Max; Shaded Square: Mean Mean = 2.33 Range = 1.13 3.32 1.0 1.5 2.0 2.5 3. 0 3.5 1 2 3 4 5 6 7 8 9 10 11 12 Follow up Time (Month) Uns tabilized Weight
222 CHAPTER 6 DISCUSSION AND CONCLUSIONS Discussion Medical records of asthmatics in the GPRD were utilized to conduct a population based cohort study to as sess asthma related morbidity and mortality after exposure to inhaled LABA bronchodilators as monot herapy and ICS based regimens From 1993 to 2010, a total of 51,103 asthmatics were followed for a maximum of 12 months after initiating ICS monotherapy, LAB A monotherapy, or ICS/LABA combination therapy. Incidence rates of asthma deaths were separately calculated for England general practices that are linked to ONS mortality database, and practices that are not part of the O NS mortality linkage scheme Asthma related morbidity was measured by incidence rates of prescribing oral corticosteroids and asthma related attendance to accident and emergency departments Likewise, asthma morbidity outcomes were evaluated across step down therapy approaches among a subgr oup of asthmatics who initiated ICS/LABA combination therapy. Marginal structural models were applied to evaluate asthma outcomes in the cohort, and Cox proportional hazards regression with time dependent covariates was applied to step down therapy subgrou p to evaluate asthma related morbidity in terms of prescribing oral corticosteroids for asthma exacerbations. Socioeconomic and behavioral information was not completely recorded for the majority of patients, except for smoking status, which was ascertaine d for most of the patients because the nature of the disease and smoking being a known risk factor for it. Un known pertinent information was covariate. Most of asthmatics with uncontrolled asthma at baseline were prescribed ICS
223 monotherapy (two third of ICS initiators had uncontrolled asthma), which was consistent with step therapy approach recommended by disease management guidelines ( NHLBI, 2007; BTS, 2009; & GINA, 2009) Conversely, the majority of LABA ini tiators had controlled disease at baseline, and the ICS/LABA combination regime was rather equally prescribed to asthmatics with controlled ( 52% ) and uncontrolled disease (48%), but mostly to patients with controlled disease This might s uggest inhaled LAB A based regimes, particularly monotherapy, were reserved for asthmatics with controlled disease (Table 5 1). However, anti inflammatory LTRA were mostly prescribed to initiators of LABA based therapies than ICS monotherapy, suggesting controller medication s like LTRA are necessary as add on therapies to bronchodilator formulations with LABA, particularly LABA monotherapy In contrast, xanthine bronchodilators were mainly prescribed to ICS monotherapy initiators Furthermore, most of LABA stoppers compared t o other patients in step down therapy cohort received inhaled SABA as rescue bronchodilators (Tables 5 2 and 5 3), which is consistent with recommendations from disease management guideline s of not using LABA inhalers as rescue bronchodilators. Similarly, LTRA anti inflammatory medications were mostly co prescribed to ICS/LABA combination therapy users than LABA stoppers (ICS monotherapy users), indicating more controlled disease on ICS monotherapy, and stepping off LABA when disease is controlled, which is consistent with practice guidelines. Prescribing study drugs was the lowest in asthmatics during the third quarter of the year, which is the summer season in UK climate. During summer there is a reduction in asthma exacerbations, and thus the need to pres cribe bronchodilators and controller
224 medications. In contrast, prescribing trend was higher in other quarters of the year that are associated with higher triggers and allergens, e.g. pollens in spring and cold in winter. This trend was similar across ICS b ased therapies (ICS monotherapy and ICS/LABA combination therapy); however, LABA monotherapy was least prescribed during first quarter, when winter is ending and spring is beginning. During these times, there might be a lower need for bronchodilator therap y compared to anti inflammatory therapy; therefore, LABA bronchodilators were mostly prescribed during autumn and the beginning of winter (third and fourth quarters, respectively) when fluctuations in temperatures, cold weather and chances of contracting v iral respiratory infections are high. These factors increases exacerbations and the need for bronchodilator maintenance could be necessary. Among step down therapy approaches, stopping LABA and continuing ICS monotherapy was mostly prescribing during the fourth quarter of the year, regardless of ICS dose. However, prescribing trends for the step down therapy approach with ICS dose reduction while maintaining LABA as ICS/LABA combination therapy was different across ICS dos age Combination therapies contain ing medium ICS dose were mainly prescribed during the fourth quarter, while combination therapies containing low ICS dose were mainly prescribed during the first qua rter. These quarters encompass winter season in the UK, which is associated with increased triggers for asthma exacerbations. In particular, the fourth quarter covers the beginning of winter and higher doses of ICS might be necessary to control exacerbation risks than the end of winter and beginning of spring which spans the middle and late part s of first quarter. Although a randomized clinical trial showed that ICS do not have an effect on the intensity or duration of
225 wheezing episodes in asthmatics (Doull et al., 1997), prescribing higher doses of ICS during winter in the current study might re flect a prescribing behavior in general practice in the UK. Additionally, the study showed that asthmatics who stopped LABA and continued ICS as high dose monotherapy encountered asthma exacerbations faster than asthmatics who continued LABA as add on to I CS (i.e., combination therapy). Particularly, patients who continued LABA with medium dose ICS had more months free of exacerbations (T able 5 8, Figures 5 8 and 5 9). However, the trend was not the same in patients with medium dose ICS at baseline (medium dose ICS step down therapy cohort), suggesting patients with worse asthma might benefit more from adding LABA but contemporarily reducing ICS dose to medium from original high dose. Asthma related mortality Among practices linked to the ONS mortality datab ase, t he current st udy found an incidence rate of 5 per 100 person yrs for asthma deaths among asthmatics initiated ICS monotherapy; the figure reduced to 0.1 per 100 person yr for the same group of patients when asthma death was identified among practices unlinked to the mortality database. The incidence of asthma deaths among LABA initiators was 6 times the incidence among ICS initiators (0.6 per person yrs). There was no power to detect differences between exposure groups in terms of asthma deaths among linked practices; yet, among unlinked practices, asthmatics who initiated LABA monotherapy and survived for one year had 25% increased risk of dying from asthma compared to counterparts who initiated ICS monotherapy and survived for one year after exposure (HR, 1.25; 95%CI, 1.11 3.01). There was no statistically different finding between
226 asthmatics received combination therapy and monotherapies. Equally, the study revealed no statistically significant results in terms of all cause mortality between exposure groups. These findings should be interpreted with caution, since there was no asthma deaths identified in the LABA group among practices linked to ONS mortality database, and the outcome in unlinked practices was derived by an untested algorithm that migh t introduced outcome classification bias, where identified asthma deaths could be due to any other reason but was misclassified as such because of the presence of asthma related Read clinical terms within 21 days of death date, e. g., severe asthma, asthma attack, or endotracheal intubation. Such clinical terms could be recorded routinely at every visit In concordance with previous studies using the GPRD ( de Vries, Setakis, Zhang, & van Staa, 2010 ), we conclude there was no statistical power to detect asth ma deaths among linked practices, or all cause deaths among all practices. However, among unlinked practices, we carefully conclude that inhaled LABA monotherapy is increases asthma deaths compared to ICS monotherapy. There has been conflicting discussions about the role of inhaled LABA monotherapy in increasing asthma deaths. Post hoc analyses of the SMART clinical trial showed the increased asthma deaths in salmeterol users compared to placebo users was mainly attributed by lack of concomitant ICS use at baseline (RR=4.37; 95%CI=1.24 15.3) ; however, information about ICS use was measured by prescribing rate of ICS rather than actual ICS used by patients (Nelson et al., 2006). Likewise, one meta analysis concluded that about 80% of asthma related deaths in the US was attributed to LABA products, regardless of ICS use ( Hagan, 2006; Salpeter, S, Buckley, Ormiston, & Salpeter, E., 2006 ). It is important to remember that SMART study
227 contributed to the majority of patients in this meta analysis. Subsequent meta a nalyses did not find enough power to detect any difference in asthma mortality rates between LABA products and other asthma medications, including ICS ( Jaeschke et al., 2008; Nelson et al., 2010 ). Contrarily, inhaled LABA monotherapy was found to increase asthma deaths compared to ICS/LABA monotherapy or placebo in one meta analys i s (RR, 3.83 ; 95%CI, 1.21 12.1) (Rodrigo, Moral, Marcos & Castro Rodriguez 2009). Another systematic review of placebo controlled trials (Weatherall et al., 2010) which was 86% w eighted by data from the SMART and SNS trials found an increased risk of asthma deaths in patients who used salmeterol but were not prescribed ICS (OR, 7.3; 95%CI, 1.8 29.4), but a decreased risk of asthma deaths in counterparts who used salmeterol and wer e prescribed ICS (OR, 2.1; 95%CI, 0.6 7.9). Nevertheless, a systematic review suggested no difference in asthma deaths between LABA monotherapy or ICS/LABA combination therapy (OR, 1.05; 95%CI, 0.32 3.47) (Cates Las serson, & Jaeschke 2009). In sum, it is recommended to retest the hypothesis among practices that are linked to ONS mortality database bu t with broader coverage in ; as well as, including incident users who are free from immortal person time after exposure Asthma rela ted morbidity The current study showed that inhaled LABA monotherapy is associated with 10 14% increase in prescribing short OCS for asthma exacerbations compared with ICS monotherapy, but did not produce significant difference in terms of asthma related A &E department visits. However, prescribing ICS concomitantly with LABA as a single device inhaler or separate devices is associated with 9 62% decrease in prescribing
228 OCS for asthma exacerbations compared with ICS monotherapy, and 50 77% decrease in prescr ibing OCS compared with LABA monotherapy. Equally, there was no difference in asthma related visits to A&E departments between exposure groups. The findings suggest prescribing inhaled LABA bronchodilators as an add on therapy to ICS to reduce asthma exace rbations. These findings are consistent with the results of clinical trials and observational studies. A meta analysis (Bateman et al., 2008) comparing ICS/salmeterol combination therapy with ICS monotherapy showed a significant reduction in asthma exacerb ations requiring oral corticosteroids (Risk Difference, 0.02; 95%CI, 0.04 to 0.01). Another meta analysis ( Rodrigo, Moral, Marcos, & Castro Rodriguez, 2009 ) showed ICS/LABA combination therapy is associated with less asthma exacerbations requiring syste mic steroids ( OR, 0.73; 95%CI, 0.67 0.79 ). Amon g step down therapy approaches, there was no difference in asthma related visits to A&E departments between LABA stoppers (medium ICS monotherapy) and ICS dose reducers (low ICS/LABA combination therapy) The study showed that a ny step down therapy approach is better than continuing original dose regimen, but within step down therapy approaches there are differences: among high dose ICS, stopping LABA while continuing high dose ICS is associated with lower exac erbation rates than reducing ICS dose to medium and continuing LABA (OCS: HR, 0.33; 95%CI, 0.07 0.52 and short courses OCS: HR, 0.35; 95%CI, 0.06 0.51). However, discontinuing LABA while maintaining medium dose ICS is associated with higher exacerbation ra tes than reducing ICS dose to low and continuing LABA (OCS: HR, 1.28; 95%CI, 1.09 5.08 and short courses OCS: HR, 1.24; 95%CI, 1.07 5.01). These findings are consistent with a similar study in the GPRD ( Thomas, von Ziegenweidt, Lee, & Price, 2009 ) compared
229 stepping up therapy approaches in initiators of ICS monotherapy showed that continuing ICS monotherapy while increasing the dose is associated with 25% less likelihood for prescriptions of OCS for asthma exacerbations compared with adding LABA as a combin ation therapy (OR, 0.75; 95%CI, 0.71 0.78), and the difference was more prominent for prescribing short courses of OCS (OR, 0.5; 95%CI, 0.46 0.55). The relationship between inadequate doses of ICS and increased asthma morbidity rates is well documented, ev en before the introduction of LABA products (Suissa, S. & Ernst, P, 2001), and the findings of the current study recommend continuing LABA as an add on therapy to ICS while maintaining adequate controller strength to achieve better asthma outcomes. Limitat ions The present study has many limitations and the findings must be interpreted in light of them. Given the observational nature of the study design, lack of randomization precludes equal distribution of known and unknown risk factors among exposure grou ps. Although attempts are made to account for all potential and actual confounders, residual confounding due to unmeasured factors is highly likely in observational datasets. Therefore, the estimated average causal effects of LABA products on asthma morbid ity and mortality outcomes should be interpreted with caution. Furthermore, external validity of the findings is limited to the UK population, which could affect extrapolations of findings to asthmatics in other countries, e.g. US. Similarly regarding age and race factors, where findings cannot be related to children younger than 13 or elderlies >65 years old, or African Americans. Another limitation in the design is including patients who survived for a minimum of 12 months after first receiving
230 pr escripti ons for study drugs. Although the potential bias attributed by this criterion was accounted for by the analysis stage, patients are no longer considered incident users. In addition, inconsistency and incompleteness of records pertinent to potential confoun ding variables are the main drawback of retrospective database analysis. Yet, in attempts to account for the scarcity of information within variables, categories with unknown information are included to satisfy model convergence in statistical analyses. It should be noted that the data in the GPRD are prescribing rather than dispensing adherence. Low adherence with inhaled pharmaceutical dosage forms, particularly ICS is a wid ely recognized problem. It is reported among new users of chronic medications, ICS users had the highest treatment discontinuation rate within one year of initiation (Beekveldt Postma et al., 2004) w ith asthma medications and any other medication were identified and included as covariates; yet, the distribution these terms were relatively scarce. Likewise, lack of information on over the counter (OTC) products casts more limitations, especially when t he OTC products influence asthma medication choices or asthma outcomes, e.g. NSAIDs or aspirin. Information about socioeconomic characteristics is included as supplement leve ls. Such information might not be adequate surrogates for socioeconomic characteristics of patients. In contrast, indices of multiple deprivations at the general practi ce level. Practice specific scores are requested from the GPRD but were not generated duly, and the scores can be included in future work.
231 Information about prescriptions issued to patients in venues other than general practices are not recorded, and there fore, time dependent exposure might not be fully categorized and exposure misclassification might happen when patients received a prescription for their subsequent exposure from an outpatient clinic or a hospital. Also, confounding misclassification could happen because lung function tests were not used as a severity measure, although the alternative measures are deemed sufficient given the nature of the database. Similarly, outcome misclassification is highly anticipated with regard to asthma related death s among practices that are unlinked to the ONS mortality database, where a patient who died due to an etiology other than asthma, might be erroneously classified as died due to asthma when a Read clinical term denoting to asthma was found within 21 days of death date. Therefore, relying on findings from the linked practices is more informative and valid; unfortunately there were no asthma deaths in LABA based groups and too few cases in ICS monotherapy group to establish the effect of LABA products on asthm a mortality. Conclusions In tandem with recommendations from regulatory stakeholders and asthma management guidelines, this study showed that inhaled long acting beta agonist bronchodilators should be used with inhaled corticosteroids as either single devi ce or separate devices combination therapy. Such approach is associated with lower rates of asthma exacerbations defined by receiving prescrip tions for oral corticosteroids or attending acci dent and emergency departments. Inhaled LABA should not be used as monotherapy and when used in combination with ICS LABA should be added to low dose ICS ins tead of medium or high doses. When an increase in ICS dose is
232 necessary it is recommended to discontinue LABA and continue ICS monotherapy in medium or high doses In conclusion, combination therapy with ICS/LABA has better asthma control than either ICS or LABA alone, and LABA monotherapy is associated with increased risk of death from asth ma attacks than ICS monotherapy. There was no sufficient statistical power to establish the effect of ICS/LABA combination therapy on asthma deaths. LABA stoppers are associated with worsened asthma than ICS/LABA dose reducers when ICS monotherapy is in medium strength; however, when ICS monotherapy is in high strength, withdrawi ng LABA is associated with better asthma control than continuing LABA as reduced ICS/LABA regimen. Future Work The current study utilized marginal structural models to account for time dependent confounding in an attempt to quantify the average causal effe ct of LABA bronchodilators on asthma morbidity and mortality. The methodology was compared with Cox proportional hazards model with time dependent covariates, which essentially yield similar estimates and overlapping confidence intervals when time dependen t confounding is not present (Figures 5 13 and 5 14). The technique however was not tested for adequate confounding control (i.e. lack of residual confounding), and this can be extended for future assumption testing. Furthermore, the technique is inefficie nt to test for exposure effect modification, and it is recommended to compare the results with structural nested models to test for effect modification by including patients with coexisting COPD and asthma, or prescribing for inhaled muscarinic receptor an tagonists as a proxy indicator for COPD. Also, the research can be repeated by
233 further accounting for socioeconomic characteristics by including indices of multiple deprivations after acquiring the scores from the GPRD. This can further adjust for unmeasur ed confounding by socioeconomic factors. In addition, the analyses for mortality outcomes can be repeated in a cohort without immortal person time after initiating study drugs to test the proposed design level approach to account for depletion of susceptib les by prevalent users (Figure 4 3). Given the low power to detect differences in mortality outcomes between exposure groups, testing related hypotheses can be strengthened by linking the cohort with a larger number of practices that are linked to the nati onal mortality database across different countries within the UK.
234 LIST OF REFERENCES Albano P M & Ramos J D. (2011). Association of house dust mite specific IgE with asthma control, medications and household pets. Asia Pacific Allergy 1 (3), 145 1 51 Ali A K. (2011). Analytical Approaches to Achieve Quasi Randomization in Retrospective Database Analysis. ISPOR CONNECTIONS 17 (2), 10 11. Allison P D. (2010). Survival Analysis Using SAS: A Practical Guide (2nd ed.) C ary, NC: SAS Institute Inc Al lison, P. D. (2012). Logistic Regression Using SAS: Theory and Application (2nd ed.). Cary, NC: SAS Institute Inc. American Lung Association Asthma Clinical Research Centers (ACRC), Mastronarde, J. G., Anthonisen, N. R., Castro, M., Holbrook, J. T., Leone F. T., Teague, W. G., & Wise, R. A. (2009). Efficacy of Esomeprazole for Treatment of Poorly Controlled Asthma. New England Journal of Medicine, 360 (15), 1487 1499. American Lung Association Asthma Clinical Research Centers (ACRC), Holbrook, J. T., Wise, R. A., Gold, B. D., Blake, K., Brown, E. D., Castro, M., Dozor, A. J., Lima, J. J., Mastronarde, J. G., Sockrider, M. M., & Teague, W. G. Lansoprazole for Children with Poorly Controlled Asthma: A Randomized Controlled Trial. Journal of the American Medic al Association, 307 (4), 373 381. Anderson H R Ayres J G Sturdy P M Bland J M Butland B K Peckitt C Taylor J C & Victor C R. (2005). Bronchodilator Treatment and Deaths from Asthma: Case Control Study. British Medical Journal, 3 30 (7483), 117. Arrighi H M & Hertz Picciotto I. (1994). The evolving concept of the healthy worker survivor effect. Epidemiology 5 (2), 189 1 96. Asthma UK. (2010) Media Resources f or Journalists: Key Facts and Statistics Retrieved May 11, 2011, from http://www.asthma.org.uk/news_media/media_resources/for_journalists_key.html Barnes, P. J. (2000). Mechanisms in COPD: differences from asthma. Chest, 117 (2 Suppl), 10S 14S. Barnes P J. (2006). Against the Dutch hypothesis: asthma and chronic obstructiv e pulmonary disease are distinct diseases. American Journal of Respiratory and Critical Care Medicine, 174 (3), 240 24 3; discussion 243 24 4. Barnes P J. (2007). Scientific rationale for using a single inhaler for asthma control. European Respiratory Journ al, 29, 587 595.
235 Bateman E D Boushey H A Bousquet J Busse W W Clark T J Pauwels R A Pedersen S E & GOAL Investigators Group. (2004). Can Guideline Defined Asthma Control be Achieved? The Gaining Optimal Asthma ControL Study. Amer ican Journal of Respiratory and Critical Care Medicine, 170 (8), 836 8 44. Bateman, E., Nelsom, H., Bousquet, J., Kral, K., Sutton, L., Ortega, H., & Yancey, S. (2008). Meta Analysis: Effects of Adding Salmeterol to Inhaled Corticosteroids on Serious Asthma Related Events. Annals of Internal Medicine, 149 (1), 33 42. Beasley, R. W., Clayton, T. O., Crane, J., Lai, C. K., Montefort, S. R., Mutius, E., Stewart, A. W., & ISAAC Phase Three Study Group. (2011). Acetaminophen Use and Risk of Asthma, Rhinoconjunctivi tis, and Eczema in Adolescents: International Study of Asthma and Allergies in Childhood Phase Three. American Journal of Respiratory and Critical Care Medicine, 183 (2), 171 178. Beekveldt Postma, N. S., Gerrits, C. M., Lammers, J. W., Raaijmakers, J. A., & Herings, R. M. (2004). Persistence with Inhaled Corticosteroid Therapy in Daily Practice. Respiratory Medicine, 98 (8), 752 759. Bgaud B. (2000). Dictionary of Pharmacoepidemiology. West Sussex UK : John Wiley & Sons, Ltd. Berger M L Mamdani M At kins D & Johnson M L. (2009). Good Research Practices for Comparative Effectiveness Research: Defining, Reporting and Interpreting Nonrandomized Studies of Treatment Effects Using Secondary Data Sources: The ISPOR Good Research Practices for Retrospec tive Database Analysis Task Force Report Part I. Value in Health, 12 (8), 1044 10 52. Berry, M., Brightling, C., Pavord, I, & Wardlaw, A. (2006). TNF alpha in Asthma. Current Opinion in Pharmacology, 7 (3), 279 282. Bisgaard H & Szefler S. (2006). Long Ac ting Beta2 Agonists and Paed iatric Asthma. Lancet, 367 (9507), 286 28 8. Blais L Beauchesne M F & Forget A. (2009). Acute Care Among Asthma Patients Using Budesonide/Formoterol or Fluticasone Propionate/Salmeterol. Respiratory Medicine, 103 (2), 237 2 43. Bleecker, E. R. (2004). Similarities and differences in asthma and COPD. The Dutch hypothesis. Chest, 126 (2 Suppl), 93S 95S; discussion 159S 161S. Bleecker, E. R., Nelson, H. S., Kraft, M., Corren, J., Meyers, D. A., Yancey, S. W., Anderson, W. H., Emm ett, A. H., & Ortega, H. G. (2010). Beta2 Receptor Polymorphisms in Patients Receiving Salmeterol with or without Fluticasone Propionate. American Journal of Respiratory and Critical Care Medicine, 181 (7), 676 687.
236 Bleecker, E. R., Yancey, S. W., Baitinger L. A., Edwards, L. D., Klotsman, M., Anderson, W. H., & Dorinsky, P. M. (2006). Salmeterol response is not affected by beta2 adrenergic receptor genotype in subjects with persistent asthma. Journal of Allergy and Clinical Immunology, 118 (4):809 816. Boul et L. P. (2009). Influence of Comorbid Conditions on Asthma. European Respiratory Journal, 33, 897 906. Brenner H & Blettner M. (1997). Controlling for continuous confounders in epidemiologic r esearch. Epidemiology, 8 (4), 429 4 34. British Thoracic Soc iety (BTS). (2006). The Burden of Lung Disease: A Statistics Report from the British Thoracic Society (2nd ed.). London, UK: BTS. Retrieved December 27, 2011, from http://www.brit thoracic.org.uk/Portals/0/Library/BTS%20Publications/burdeon_of_lung_disease 2 007.pdf British Thoracic Society (BTS). (2011). Past Asthma Guidelines. Retrieved May 8, 2011, from http://www.brit thoracic.org.uk/Clinical Information/Asthma/Asthma Guidelines/Past Asthma Guidelines.aspx British Thoracic Society (BTS), British Paediatri c Association, Research Unit of the Campaign, Royal College of General Practitioners, General Practitioners in Asthma Group, British Association of Accident and Emergency Medicine, & British Paediatric Respiratory Group. (1993). Guidelines on the Management of Asthma. Thorax, 48 (2 Suppl), S1 24. British Thoracic Society (BTS) & Scottish Intercolle giate Guidelines Network (SIGN) (2009). British Guideline on the Management of Asthma. A National Clinical Guideline. Retrieved March 13, 2011, from http:// www.brit thoracic. org.uk Brookhart M A Wang P S Solomon D H & Schneeweiss S. (2006 a ). Evaluating short term drug effects using a physician specific prescribing preference as an instrumental vari able. Epidemiology, 17 (3), 268 2 75. Brookhart M A Wang P S Solomon D H & Schneeweiss S. (2006 b ). Instrumental variable analysis of secondary pharmacoepidemiologi c data. Epidemiology, 17 (4), 373 37 4. Brumback B A Hernn M A Haneuse S J & Robins J M. (2004). Sensitivity analyses for unmeasured confounding assuming a marginal structural model for repeated measures. Statistics in Medicine, 23 (5), 749 7 67. Buist A S. (2003). Similarities and differences between as thma and chronic obstructive pulmonary disease: treatment and early outcomes. European Respiratory Journal Suppl ement, 39, 30s 35s.
237 Bush A & Zar H J. (2011). WHO Universal Definition of Severe Asthma. Current Opinion in Allergy and Clinical Immunology 11 (2):115 1 21. Cain L E Robins J M Lanoy E Logan R Costagliola D & Hernn M A. (2010). When to start treatment? A systematic approach to the comparison of dynamic regimes using observational data. International Journal of Biostatistics 6 (2), Article 18. Cape, J. (2002). Consultation Length, Patient Estimated Consultation Length, and Satisfaction with the Consultation. British Journal of General Practice, 52 (485), 1004 1006. Castle W Fuller R Hall J & Palmer J. (1993). Sereven t Nationwide Surveillance Study: Comparison of Salmeterol with Salbutamol in Asthmatic Patients who Require Regular Bron chodilator Treatment. British Medical Journal, 306, 1034 10 37. Cates, C. J., Lasserson, T. J., & Jaeschke, R. (2009). Regular Treatment with Salmeterol and Inhaled Steroids for Chronic Asthma: Serious Adverse Events. Cochrane Database of Systematic Reviews, 3 CD00 6922 Cazzola M Calzetta L & Matera M G. (2011). (2) adrenoceptor agonists: current and future direction. British Jour nal of Pharmacology, 163 (1), 4 17. Centers for Disease Control and Prevention (CDC). (2010). Updated recommendations for prevention of invasive pneumococcal disease among adults using the 23 valent pneumococcal polysaccharide vaccine (PPSV23). Morbidity an d Mortality Weekly Report, 59 (34), 1102 110 6. Retrieved February 16, 2012, from http://www.cdc.gov/mmwr/pdf/wk/mm5934.pdf Centers for Diseas e Control and Prevention (CDC) (2011 a ). Asthma in the US. CDC Vital Signs, 2011 5. Retrieved May 11, 2011, from htt p://www.cdc.gov/VitalSigns/pdf/2011 05 vitalsign s.pdf Centers for Diseas e Control and Prevention (CDC) (2011 b ). Public Health Strategic Framework for COPD Prevention. Atlanta, GA: CDC. Retrieved December 27, 2011, from http://www.cdc.gov/copd/pdfs/Framewo rk_for_COPD_Prevention. pdf Centers for Disease Co ntrol and Prevention (CDC). (2012). Vaccines and Preventable Diseases: Pneumococcal Vaccination. Retrieved February 12, 2012, from http://www.cdc.gov/vaccines/vpd vac/ pneumo/default.htm Chanez P Vignola A M T Enander I Li D Jeffery P K & Bousquet J. (1997). Corticosteroid reversibility in COPD is related to features of asthma. American Journal of Respiratory and Critical Care Medicine, 155 (5), 1529 15 34.
238 Chang, K. C., Leung, C. C., Tam, C. M., & Kong, F. Y. (2011). Acetaminophen and Asthma: Spurious Association? American Journal of Respiratory and Critical Care Medicine, 183 (11), 1570 1571. Chitnis A S Aparasu R R Chen H & Johnson M L. (201 2 ). Effect of Certain An giotensin Converting Enzyme Inhibitors on Mortality in Heart Failure: A Multiple Propensity Analysis. Research in Social and Administrative Pharmacy 8 (2), 145 156. Christenfeld N J Sloan R P Carroll D & Greenland S. (2004). Risk factors, confo unding, and the illusion of statistical control. Psychosomatic Medicine, 66 (6), 868 8 75. Cole S R & Hernn M A. (2004). Adjusted survival curves with inverse probability weights. Computer Methods and Programs in Biomedicine, 75 (1), 45 4 9. Cole S R & Hernn M A. (2008). Constructing inverse probability weights for marginal structural models. American Journal of Epidemiology, 168 (6), 656 6 64. Colice G L. (2011). Emerging Therapeutic Options for Asthma. American Journal of Managed Care, 17, S82 S8 9. Cox D R. (1972). Regression models and life tables. Journal of the Royal Statistical Society. Series B, Statistical Methodology, 34 (2), 187 202; discussion 202 220. Cox E Martin B C Van Staa T Garbe E Siebert U & Johnson M L. (2009). Good Research Practices for Comparative Effectiveness Research: Approaches to Mitigate Bias and Confounding in the Design of Nonrandomized Studies of Treatment Effects Using Secondary Data Sources: The International Society for Pharmacoeconomics and Outcom es Research Good Research Practices for Retrospective Database Analysis Task Force Report Part II. Value in Health, 12 (8), 1053 10 61. Cserhti E. (2004). The History of Bronchial Asthma from the Ancient Times Till the Middle Ages. Acta Physiologica Hungar ica, 91 (3 4), 243 261. Danaei G Tavakkoli M & Hernn M A. (2012). Bias in observational studies of prevalent users: Lessons for comparative effectiveness research from a meta analysis of statins. American Journal of Epidemiology, 175 (4), 250 262. D avis S Rietbrock S Rubino A Shah A D Williams T & Martinez C. (2003). Auditing the quality of data in the general practice research database [abstract]. Pharmacoepidemiol ogy and Drug Saf ety, 12, S61. de Souza C J & Burkey B F. (2001) Beta 3 adrenoceptor agonists as anti diabetic and anti obesity drugs in humans. Current Pharmaceutical Design, 7 (14), 1433 14 49.
239 de Vries F Setakis E Zhang B & van Staa T P. (2010). Long Acting 2 Agonists in Adult Asthma and the Pattern of Ris k of Death and Severe Asthma Outcomes: A Study Using the GPRD. European Respiratory Journal, 36 494 502. Decramer M & Selroos O. (2005). Asthma and COPD: differences and similarities. With special reference to the usefulness of budesonide/formoterol i n a single inhaler (Symbicort) in both diseases. International Journal of Clinical Practice, 59 (4), 385 398. Desai, R. J., Ashton, C. M., Deswal, A., Morgan, R. O., Mehta, H. B., Chen, H., Aparasu, R. R., & Johnson, M. L. (201 2 ). Comparative effectiveness of individual angiotensin receptor blockers on risk of mortality in patients with chronic heart failure. Pharmacoepidemiology and Drug Safety, 21 (3), 233 240. Deveugele M Derese A van den Brink Muinen A Bensing J & De Maeseneer J. (2002). Con sultation length in general practice: cross sectional s tudy in six European countries. British Medical Journal, 325 (7362), 472. DiMango, E., Holbrook, J. T., Simpson, E., Reibman, J. Richter, J., Narula, S., Prusakowski, N., Mastronarde, J. G., Wise, R. A. & American Lung Association Asthma Clinical Research Centers. (2009). Effects of Asymptomatic Proximal and Distal Gastroesophageal Reflux on Asthma Severity. American Journal of Respiratory and Critical Care Medicine, 180 (9), 809 816. Doull, I. J. M., La mpe, F. C., Smith, S., Schreiber, J., Freezer, N. J., & Holgate, S. T. (1997). Effect of Inhaled Corticosteroids on Episodes of Wheezing Associated with Viral Infection in School Age Children: Randomised Double Blind Placebo Controlled Trial. British Medic al Journal, 315 858 862. Electronic Medicines Compendium (EMC). (2011). Information about UK Licensed Medicines Retrieved February 4, 2011, from http:// www.medic ines.org.uk/EMC European Medicines Agency (EMA). (2010). European Medicines Agency 2011 Prior ities for Drug Safety Research. Patient Health Protection. Retrieved June 20, 2011, from http://www.ema.europa.eu/docs/en_GB/document_library/Other/2010/07/WC5000 94267 .pdf. Fitzpatrick S Joks R & Silverberg J I. (2012). Obesity is association with increased asthma severity and exacerbations, and increased serum immunoglobulin E in inner city adults. Clinical and Experimental Allergy, 42 (5), 747 759. Food and Drug Administration (FDA). (2010a) Postmarketing Drug Safety Information for Patients and Pr oviders: Advair Diskus, Advair HPA, Brovana, Foradil, Perforomist, Serevent Diskus, and Symbicort Information (Long Acting Beta Agonists). Retrieved February 4, 2011, from http://www.fda.gov/drugs/drugsafety/postmarketdrugsafetyinformationforpatientsan dpro viders/ucm108111.htm
240 Food and Drug Administration (FDA). (2010b). Postmarketing Drug Safety Information for Patients and Providers: FDA Drug Safety Communication: New Safety Requirements for Long Acting Inhaled Asthma Medications Called Long Acting Beta Ag onists (LABAs). Retrieved February 4, 2011, from http://www.fda.gov/drugs/drugsafety/postmarketdrugsafetyinformationforpatientsan dproviders/ucm200776.htm Food and Drug Administration (FDA). (2011 a ). FDA Approved Drug Products (Drugs @ FDA) Retrieved Febru ary 4, 2011, from http://www.accessdata.fda.gov/scripts/cder/drugsatfda/index.cf m Food and Drug Administration (FDA). (2011 b ). Safety Information : Long Acting Beta Agonists (LABAs): New Safe Use Requirements. Retrieved June 11, 2011, from http://www.fda.go v/Safety/MedWatch/SafetyInformation/SafetyAlertsforHumanMed icalProducts/ucm201003 .htm Garcia Rodrguez L A & Gutthann S P. (1998). Use of the UK General Practice Research Database for Pharmacoepidemiology. British Journal of Clinical Pharmacology, 45 419 425. Garde D. (2012). Boehringer gets positive PhII for COPD inhalant. Fierce Drug Delivery Retrieved May 23, 2012, from http://www.fiercedrugdelivery.com/story/boehringer gets positive phii copd inhalant/20 12 05 23 General Practice Research Databa se (GPRD). ( 2011 a ) Database Homepage. Retrieved March 23, 2011, from http://www.gprd.com/h ome/ General Practice Research Database (GPRD). (2011b). External Record Linkage Retrieved May 4, 2011, from http://www.gprd.com/products/lin ks.asp General Practice Research Database (GPRD). (2011c). GPRD On Line Data (G.O.L.D), Hierarchical Dictionaries Retrieved May 8, 2011, from http://www.gprd.com/services/onli ne.asp General Practice Research Database (GPRD). (2011d). The GPRD Practice & Patient Populations. Ret rieved May 4, 2011, from http://www.gprd.com/_docs/GPRD%20Practice%20Patient%20Populations%20Ap r2011.pdf General Practice Research Database (GPRD). (2004). Recording Guidelines for Vision Users. London, UK: GPRD. Gibson P G Powell H & Ducharme F M (2007). Differential effects of maintenance long acting beta agonist and inhaled corticosteroid on asthma exacerbations. Journal of Allergy and Clinical Immunology, 119 (2) 334 3 50.
241 Glantz, S. A. (2005). Primer of Biostatistics (6th ed.). New York, NY: M cGraw Hill Companies, Inc. Glassroth J. (2006). The role of long acting beta agonists in the management of asthma: analysis, meta analysis, and more analysis. Annals of Internal Medicine, 144 (12), 936 93 7. Global Initiative for Asthma (GINA). (2009). Glob al Strategy for Asthma Management and Prevention. Retrieved March 13, 2011, from http://www.ginasthma.org Global Initiative for Chronic Obstructive Lung Disease (GOLD). (2010). Global Strategy for the Diagnosis, Management, and Prevention of Chronic Obstru ctive Pulmonary Disease. Retrieved March 26, 2011, from http:// w ww.goldcopd.org Glymour M M & Greenland S. (2008). Causal Diagrams. In K. J. Rothman, S. Greenland & T. L. Lash (Eds.), Modern Epidemiology (3rd ed ) (pp. 183 209). Philadelphia, PA: Lip pincott Williams & Wilkins. Glynn R J Schneeweiss S & Strrmer T. (2006). Indications for propensity scores and review of their use in pharmacoepidemiology. Basic & Clinical Pharmacology & Toxicology, 98 (3), 253 25 9. Greenland, S. (1996). Basic met hods for sensitivity analysis of biases. International Journal of Epidemiology, 25 (6), 1107 1116. Greenland, S. (2000). An introduction to instrumental variables for epidemiologists. International Journal of Epidemiology, 29 (4), 722 729. Greenland, S. (200 8). Introduction to Regression Modeling. In K. J. Rothman, S. Greenland, & T. L. Lash (Eds.), Modern Epidemiology (3rd ed.) (pp. 418 455). Philadelphia, PA: Lippincott Williams & Wilkins. Greenland, S., & Rothman, K. J. (2008). Measures of Occurrence. In K J. Rothman, S. Greenland, & T. L. Lash (Eds.), Modern Epidemiology (3rd ed.) (pp. 32 50). Philadelphia, PA: Lippincott Williams & Wilkins. Greenland S Rothman K J & Lash T L. (2008). Measures of Effect and Measures of Association. In K. J. Rothm an, S. Greenland, & T. L. Lash (Eds.), Modern Epidemiology (3rd ed.) (pp. 51 70). Philadelphia, PA: Lippincott Williams & Wilkins. Greenstone I R Chroinin M N Masse V Danish A Magdalinos H Zhang X & Ducharme F M. (2005). Combination o f Inhaled Long Acting Beta2 Agonists and Inhaled Steroids Versus Higher Dose of Inhaled Steroids in Children and Adults with Persistent Asthma. Cochrane Database of Systematic Reviews, 4, CD005533.
242 Griffin J Lee S Caiado M Kesten S & Price D. (2008). Comparison of tiotropium bromide and combined ipratropium/salbutamol for the treatment of COPD: a UK General Practice Research Database 12 month follow up study. Primary Care Respiratory Journal, 17 (2), 104 1 10. Grimes D A & Schulz K F. (2002 ). Bias and causal associations in observ ational research. Lancet, 359 (9302), 248 2 52. Guerra S. (2005). Overlap of asthma and chronic obstructive pulmonary disease. Current Opinion in Pulmonary Medicine, 11 (1), 7 13. Guilbert T & Krawiec M. (2003). N atural history of asthma. Pediatric Clinics of North America, 50 (3), 523 5 38. Guo J J Tsai K Kelton C M Bian B & Wigle P R. (2011). Risk of Serious Asthma Exacerbations Associated with Long Acting Beta Agonists among Patients with Asthma: A Retrospective Cohort Study. Annals of Allergy, Asthma & Immunology, 106 (3), 214 2 22. Gupta, R., Sheikh, A., Strachan, D. P., & Anderson, H. R. (2004). Burden of Allergic Disease in the UK: Secondary Analyses of National Databases. Clinical and Experimenta l Allergy, 34, 520 526. Gupta A Sjoukes A Richards D Banya W Hawrylowicz C Bush A & Saglani S. (2011). Relationship between Serum Vitamin D, Disease Severity, and Airway Remodeling in Children with Asthma. American Journal of Respiratory and Critical Care Medicine, 184 (12), 1342 134 9. Hagan, P. (2006). Four in Five Asthma Deaths may be due to Long Acting Agonists. British Medical Journal, 332, 1467. Hansell A Hollowell J Nichols T McNiece R & Strachan D. (1999). Use of the General Practice Research Database (GPRD) for Respiratory Epidemiology: a Comparison with the 4th Morbidity Survey in General Practice (MSGP4). Thorax, 54 (5), 413 41 9. Hernn, M. A. (2010). The hazards of hazard ratios. Epidemiology, 21 (1), 13 15. Hernn, M. A. (2011). With great data comes great responsibility: publishing comparative effectiveness research in epidemiology. Epidemiology, 22 (3), 290 291. Hernn M A Brumback B & Robins J M. (2000). Marginal Structural Models to Estimate the Causal Ef fect of Zidovudine on the Survival of HIV Positive Men. Epidemiol ogy, 11 (5), 561 570.
243 Hernn M A Brumback B A & Robins J M. (2002). Estimating the Causal Effect of Zidovudine on CD4 Count with a Marginal Structural Model for Repeated Measures. S tatistics in Medicine, 21 689 1709. Hernn M A & Cole S R. (2009). C ausal diagrams and measurement bias. American Journal of Epidemiology, 170 (8), 959 9 62; discussion 963 96 4. Hernn M A Hernndez Daz S & Robins J M. (2004). A structural a pproach to selection bias. Epidemiolog y, 15 (5), 615 6 25. Hernn M A & Robins J M. (2006). Instruments for causal inference: an Epidemiology 17 (4), 360 3 72. Herrett E Thomas S L Schoonen W M Smeeth L & Hall A J (2010). Validation and validity of diagnoses in the General Practice Research Database: a systematic review. British Journal of Clinical Pharmacology, 69 (1), 4 14. Hill A. (2012). tr ack to file its respiratory drug Relovair in Europe and the US despite mixed results in late stage trials. InPharm News. Retrieved May 20, 2012, from http://www.inpharm.com/news/170759/underw helming results gsk s relovair Holgate S T Price D & Valov irta E. (2006). Asthma Out of Control? A Structured Review of Recent Patient Surveys. BMC Pulmonary Medicine, 6 (Suppl 1):S2. Holgate, S. T., Noonan, M., Chanez, P., Busse, W., Dupont, L., Pavord, I., Hakulinen, A., Paolozzi, L., Wajdula, J., Zang, C., Nel son, H., & Raible, D. (2011). Efficacy and Safety of Etanercept in Moderate to Severe Asthma: A Randomised, Controlled Trial. European Respiratory Journal, 37 (6), 1352 1359. Hollowell J. (1997). The General Practice Research Database: quality of morbidity data. Population Trends, 87, 36 40. Hopp R J & Townley R G. (2006). The Origins and Characteristics of Asthma. In M. E. Gershwin & T. E. Albertson ( Eds.), Bronchial Asthma: A Guide for Practical Understanding and Treatment ( 5th e d. ) (pp. 3 28). Tot owa, NJ: Humana Press Inc. Hudson M & Suissa S. (2010). Avoiding common pitfalls in the analysis of observational studies of new treatments for rheumatoid arthritis. Arthritis Care & Research, 62 (6), 805 8 10. Ichinose M. (2009). Differences of inflamm atory mechanisms in asthma and COPD. Allergology International, 58 (3), 307 3 13. International Society for Pharmacoepidemiology (ISPE). (2008). Guidelines for Good Pharmacoepidemiology Practices (GPP). Pharmacoepidemiol ogy and Drug Saf ety, 17 (2), 200 20 8.
244 J aeschke R P M Mejza F Nair P Lesniak W Brozek J Thabane L Cheng J Schunemann H J Sears M R & Guyatt G. (2008). The Safety of Long Acting Beta Agonists among Patients with Asthma Using Inhaled Corticosteroids. Syste matic Review and Meta Analysis American Journal of Respiratory and Critical Care Medicine, 178, 1009 10 16. Jepsen P Johnson S P Gillman M W & Srensen H T. (2004). Interpretation of observa tional studies. Heart, 90 (8), 956 9 60. Jick H Jick S S & Derby L E. (1991). Validation of Information Recorded on General Practitioner Based Computerised Data Resource in the United Kingdom. British Medical Journal, 302, 766 76 8. Jick H Terris B Z Derby L E & Jick S S. (1992). Further V alidation of Information Recorded on a General Practitioner Based Computerized Data Resource in the United Kingdom. Pharmacoepidemiol ogy and Drug Safety, 1, 347 34 9. Jick S. (2010). Ph armacoepidemiology. In J. Olsen R. Saracci & D. Trichopoulos (Eds.), Teaching Epidemiology: A Guide for Teachers in Epidemiology, Public Health and Clinical Medicine ( 3rd ed. ) (pp. 143). New York, NY: Oxford University Press Inc. Jick S S Kaye J A Vasilakis Scaramozza C ., Garcia Rodrguez L A Ruigmez A Meie r C R Schlienger R G Black C & Jick H. (2003). Validity of the General Practice Research Data base. Pharmacotherapy, 23 (5), 686 68 9. Johnson M. (1998). The beta adrenoceptor. American Journal of Respiratory and Critical Care Medicine, 158 (5 Pt 3), S146 S1 53. Johnson M L Crown W Martin B C Dormuth C R & Siebert U. (2009). Good Research Practices for Comparative Effectiveness Research: Analytic Methods to Improve Causal Inference from Nonrandomized Studied of Treatment Effects Usin g Secondary Data Sources: The ISPOR Good Research Practices for Retrospective Database Analysis Task Force Report Part III. Value in Health 12 (8), 1062 10 73. Johnston K M Gustafson P Levy A R & Grootendorst P. (2008). Use of instrumental varia bles in the analysis of generalized linear models in the presence of unmeasured confounding with applications to epidemiological research. Statistics in Medicine, 27 (9), 1539 15 56. Jonas D E Wines R DelMonte M Amick H Wilkins T Einerson B Schuler C L Wynia B A & Shilliday B B. (2011). Drug class review: Controller medications for asthma. Drug Class Reviews. Retrieved February 13, 2012, from http://www.ncbi.nlm.nih.gov/books/NBK56695/pdf/TOC. pdf
245 Kattan M W & Cowen M E. (Eds .). (2009). Causal Inference in Medical Decision Maki ng. In Encyclopedia of Medical Decision Making (Vol. 1 ) (pp. 117 121). Thousands Oaks, CA : SAGE Publications, Inc. Kemp J Armstrong L Wan Y Alagappan V K Ohlssen D & Pascoe S. (2011). Sa fety of formoterol in adults and children with asthma: a meta analysis. Annals of Allergy, Asthma & Immunology, 107 (1), 71 7 8. Khan N F Harrison S E & Rose P W. (2010). Validity of diagnostic coding within the General Practice Research Database: a systematic review. British Journal of General Practice, 60 (572), e128 e1 36. Khan N F Perera R Harper S & Rose P W. (2010). Adaptation and validation of the Charlson index for Read/OXMIS coded databases. BMC Family Practice, 11, 1. Kim H & Mazza J. (2011). Asthma. Allergy, Asthma, and Clinical Immunology, 7 ( Suppl 1 ), S2. Kurth T & Seeger J D. (2008). Propensity score analys es in pharmacoepidemiology. In A. G. Hartzema, H. H. Tilson, & K. A. Chan (Eds.), Pharmacoepidemiology and Therape utic Risk Management (pp. 301 324) Cincinnati, OH : Harvey Whitney Books Company. Lanes S F ., Garcia Rodrguez L A & Huerta C. (2002). Respiratory Medications and Risk of Asthma Death. Thorax, 57 (8), 683 68 6. Lanes S F Lanza L L & Wentworth C E 3rd. (1998). Risk of Emergency Care, Hospitalization, and ICU Stays for Acute Asthma Among Recipients of Salmeterol. American Journal of Respiratory and Critical Care Medicine, 158 (3), 857 8 61. Lange P Parner J Vestbo J Schnohr P & Jense n G. (1998). A 15 year follow up study of ventilator function in adults with asthma. New England Journal of Medicine, 339 (17), 1194 1 200. Lasserson T J Ferrara G & Casali L. (2011). Combination fluticasone and salmeterol versus fixed dose combinat ion budesonide and formoterol for chronic asthma in adults and children. Cochrane Database of Systematic Reviews, 12, CD004106. Laupacis A & Mamdani M. (2004). Observational studies of treatment effectiveness: some cautions. Annals of Internal Medicine 140 (11), 923 92 4. Levenson M for the Foo d and Drug Administration Joint Meeting of the Pulmonary Allergy Drug Advisory Committee, Drug Safety and Risk Management Advisory Committee, and Pediatric Advisory Committee on December 10 11, 2008. Retrieved Fe bruary 4, 2011, from http://www.fda.gov/ohrms/dockets/ac/08/briefing/2008 4398b1 01 FDA.pd f
246 Liu T Valdez R Yoon P W Crocker D Moonsinghe R & Khoury M J. (2009). The association between family history of asthma and the prevalence of asthma among US adults: National Health and Nutrition Examination Survey, 1990 2004. Genetics in Medicine, 11 (5), 323 32 8. Loerbroks A Apfelbacher C J Amelang M & Strmer T. (2008). Obesity and adult asthma: potential effect modification by gender, bu t not by hay fever. Annals of Epidemiology, 18 (4), 283 28 9. Long A A. (2011). The Burden of Asthma and Improving Patient Outcomes. American Journal of Managed Care, 17, S75 S81. Lopez A D & Murray C C. (1998). The global burden of disease, 1990 202 0. Nature Medicine, 4 (11), 1241 124 3. Martens E P Pestman W R de Boer A Belitser S V & Klungel O H. (2006). Instrumental variables: application and lim itations. Epidemiology. 17 (3), 260 26 7. Masoli M Weatherall M Holt S & Beasley R. (2005). Moderate Dose Inhaled Corticosteroids Plus Salmeterol Versus Higher Doses of Inhaled Corticosteroids in Symptom atic Asthma. Thorax, 60 (9), 730 73 4. McBride, J. T. (2011). The Association of Acetaminophen and Asthma Prevalence and Severity. Pedi atrics, 128 (6), 1181 1185. McIvor R A Pizzichini E Turner M O Hussack P Hargreave F E & Sears M R. (1998). Potential Masking Effects of Salmeterol on Airway Inflammation in Asthma. American Journal of Respiratory and Critical Care Medic ine, 158 (3), 924 9 30. McIvor R A Tunks M Todd D C. COP D. Clinical Evidence, 6201, 1502. McLennan D Barnes H Noble M Davies J Garratt E & Dibbon C. (2011). The English Indices of Deprivation 2010. Technical Report. London, UK: Depar tment for Communit ies and Local Government. Retrieved June 9, 2011, from http://www.communities.gov.uk/publications/corporate/statistics/indice s2010 Meier C R & Jick H. (1997). Drug Use and Pulmonary Death Rates in Increasingly Symptomatic Asthma Patie nts in the UK. Thorax, 52 (7), 612 61 7. Miller E K Hernandez J Z Wimmenauer V Shepherd B E Hijano D Libster R Serra M E Bhat N Batalle J P Mohamed Y Reynaldi A Rodriguez A Otello M Pisapia N Bugna J Bellabar ba M Kraft D Coviello S Ferolla F M Chen A London S J Siberry G K Williams J V & Polack F P. ( 201 2 ). A Mechanistic Role for Type III Interferon 1 in Asthma Exacerbations Mediated by Human Rhinoviruses. American Journal of Res piratory and Critical Care Medicine, 185 (5), 508 516.
2 47 Morales, D. R., Jackson, C., Fielding, S., & Guthrie, B. (2012). Long Agonist Prescribing in People with Asthma in Primary Care. Thorax, 67 (10), [published online ahead of print September 1, 20 12]. Retrieved September 6, 2012 from http://thorax.bmj.com/content/early/2012/08/31/thoraxjnl 2012 202071.full.pdf Morjaria, J. B., Babu, K. S., Holgate, S. T., & Polosa, R. (2006). Tumour Necrosis Factor Drug Discover y Today: Therapeutic Strategies, 3 (3), 309 316. Mortimer K M Neugebauer R van der Laan M & Tager I B. An application of model fitting procedures for marginal structural models. American Journal of Epidemiology, 162 (4), 382 38 8. Murray C J L opez A D. (1997). Alternative projections of mortality and disability by cause 1990 2020: Global Burden of Dis ease Study. Lancet, 349 (9064), 1498 1 504. Nadel J A. (1985). Role of Infl ammation in Asthma. Chest, 87, 171S. Nair P. (2011). Early Intervent ions with Inhaled Corticosteroids in Asthma: Benefits and Risks. Current Opinion in Pulmonary Medicine, 17 (1), 12 1 5. National Health Service (NHS). (2010). Data Dictionary: Mortality. Retrieved March 23, 2011 from http://www.hesonline.nhs.uk/Ease/servlet /ContentServer?siteID=1937&categoryI D =1299 National Health Service (NHS). (2011 a ). Hospital E pisode Statistics (HES) Online: Linked ONS HES Morta lity Data. Retrieved May 5, 2011, from http://www.hesonline.nhs.uk/Ease/servlet/ContentServer?siteID=1937&categ oryI D =1299 National Health Service (NHS). (2011 b ). Introduction to Quality and Outcomes Framewo rk (QOF). Retrieved May 8, 2011, from http://www.ic.nhs.uk/statistics and data collections/supporting information/audits and performance/the quality and outcomes framework/qof information/introduction to qof National Heart, Lung, and Blood Institute (NHLBI). (2003). Chronic Obstructive Pulmonary Disease Data Fact Sheet. National Institute of Health, publication number: 03 5229. National Heart Lung and Blood Insti tute (NHLBI). (2007). Expert Panel Report 3 (EPR3): Guidelines for the Diagnosis and Management of Asthma Retrieved June 5, 2011, from http://www.nhlbi.nih.gov/guidelines/asthma/asth gdln.pdf National Heart Lung and Blood Institute (NHLBI). (2009). Morbid ity & Mortality : 2009 Chart Book on Cardiovascular, Lung, and Blood Diseases. Retrieved December 27, 2011, from http://www.nhlbi.nih.gov/resources/docs/2009 _ChartBook.pdf
248 National Institutes of Health (NIH). (1998). Clinical Guidelines on The Identificatio n, Evaluation, and Treatment of Overweight and Obesity in Adults. The Evidence Report 98, 4083. Retrieved June 10, 2011, from http://www.nhlbi.nih.gov/guidelines/obesity/ob_gdlns .pdf National Prescribing Center (NPC). (2008). Dispensing with Repeats: A pr actical guide to repeat dispensing. (2nd ed.). London, UK: National Health Services. Retrieved February 6, 2012, from http://www.npc.nhs.uk/repeat_medication/repeat_dispensing/ resources/dwr_for_w eb.pdf Nelson, H. S. (2006). Is there a Problem with Inhaled Long Acting Beta Adrenergic Agonists? Journal of Allergy and Clinical Immunology, 117, 3 16. Nelson H Bonuccelli C Radner F Ottosson A Carroll K J Andersson T L G & LaForce C. (2010). Safety of Formoterol in Patients with Asthma: Comb ined Analysis of Data from Double Blind, Randomized Controlled Trials. Journal of Allergy and Clinical Immunology, 125, 390 39 6. Nelson H S Weiss S T Bleecker E R Yancey S W Dorinsky P M & SMART Study Group. (2006). The Salmeterol Multi center Asthma Research Trial. A Comparison of Usual Pharmacotherapy for Asthma or Usual Pharmacothera py Plus Salmeterol. Chest, 129, 15 26. Ng T P. (2000). Validity of symptom and clinical measures of asthma severity for primary outpatient assessment of adult asthma. British Journal of General Practice, 50 (450), 7 12. Noble M Smith G A N Penhale B Wright G Dibben C Owen T & Lloyd M. (2000). Measuring multiple deprivation at the small area level: the indices of deprivation 2000. London, UK: Department of the Environment, Transportation, an d the Regions. Novartis Briefing Document. (2011). Indacaterol (QAB149) in Chronic Obstructive Pulmonary Disease (NDA 22 383) Pulmonary Allergy Drugs Advisory C ommittee Meeting: March 8, 2011. Retrieve d May 20, 2012, from http://www.fda.gov/downloads/AdvisoryCommittees/CommitteesMeetingMaterials/ Drugs/Pulmonary AllergyDrugsAdvisoryCommittee/UCM24563 9.pdf Office for National Statistics (ONS). (2009). Mortality Statist ics: Deaths Registered in 2008: Revie w of the National Statistician on Deaths in England and Wales, 2008. South Wales, UK: Office for National Statistics. Retrieved May 5, 2011, from http://www.statistics.gov.uk/downloads/theme_health/DR2008/DR_ 08.pdf Omachi T A. (2009). Poor Outcomes and A sthma Hospitalizations: How Important is Asthma Severi ty and How Do We Measure it? Allergologia et Immunopathologia, 37 (5), 223 224.
249 T B. (2007). Inhaled beta agonists. Respiratory Care, 52 (7), 820 8 32. Orie N Sluiter H DeVries K Tramm eling G & Wiktop J. (1961). The host factor in bronchitis. In N. Orie & H. Sluiter (E ds. ), Bronchitis (pp.43 59) Assen, the Netherlands: Royal Van Gorcum Ovchinikova L Smith L & Bosnic Anticevich S. (2011). Inhaler technique maintenance: gain Journal of Asthma, 48 (6), 616 6 24. Palmer C N Lipworth B J Lee S Ismail T Macgregor D F & Mukhopadhyay S. (2006). Arginine 16 Beta2 Adrenoceptor Genotype Predisposes to Exacerbations i n Young Asthmatics Taking Regular Salmeterol. Thorax 61 (11) 940 94 4. Paul G Brehm J Alcorn J F Holguin F Aujla S & Celed n J C. (2012). Vitamin D and Asthma. American Journal of Respiratory and Critical Care Medicine, 185 (2), 124 132. P rice D Haughney J Sims E Ali M von Ziegenweidt J Hillyer E V Lee A J Chisholm A & Barnes N. (2011 a ). Effectiveness of Inhaler Types for Real World Asthma Management: Retrospective Observational Study Using the GPRD. Journal of As thma and Allergy, 4, 37 47. Price D Roche N Virchow J C Burden A Ali M Chisholm A Lee A J Hillyer E V von Ziegenweidt J. (2011 b ). Device Type and Real World Effectiveness of Asthma Combination Therapy: An Observational Study. Re spiratory Medicine, 105 (10), 1457 1466. Psaty B M & Siscovick D S. (2010). Minimizing bias due to confounding by indication in comparative effectiveness research: the importance of restriction. Journal of the American Medical Association, 304 (8), 897 89 8. Rang H P Dale M M Ritter J M & Flower R J. (Eds.). (2007). The Respiratory System. In ( 6th e d. ) (pp. 356 367). Philadelphia, PA: Churchill Livingstone & Elsevier Ray, W. A. (2003). Evaluating Medication Eff ects Outside of Clinical Trials: New User Designs. American Journal of Epidemiology, 158 (9), 915 920. Rhys G Beerstecher H J & Morgan C L. (2010). Primary Care Capitation Payments in the UK. An Observational Study. BMC Health Services Research, 10 156. Robins, J. M. (1998). Marginal Structural Models. In 1997 Proceedings of the Section on Bayesian Statistical Science (pp. 1 10). Alexandria, VA: American Statistical Association.
250 Robins J M Hernn M A & Brumback B. (2000). Marginal Structur al Models and Causal Inference in E pidemiology. Epidemiology, 11 550 560. Robins J M Hernn M A & Rotnitzky A. (2007). Effect modification by time varying covariates. A merican Journal of Epidemiology, 166 (9), 994 1002. Rodrigo, G. J., Moral, M. P ., Marcos, L. G., & Castro Rodriguez, J. A. (2009). Safety of Regular Use of Long Acting Beta Agonists as Monotherapy or Added to Inhaled Corticosteroids in Asthma. A Systematic Review. Pulmonary Pharmacology and Therapeutics, 22 (1), 9 19. Rosenbaum P & Rubin D. (1983). The Central Role of the Propensity Score in Observational Studies for C ausal Effects. Biometrika, 70, 41 55. Rothman, K. J., & Greenland, S. (2008). Cohort Studies. In K. J. Rothman, S. Greenland, & T. L. Lash (Eds.). Modern Epidemiology (3rd ed.) (pp. 100 110). Philadelphia, PA: Lippincott Williams & Wilkins. Rothman K J Greenland S & Lash T L. (Eds.). (2008). Validity in Epidemiologic Studie s. In Modern Epidemiology ( 3rd e d. ) (pp. 128 147). Philadelphia, PA: Lip pincott Williams & Wilkins. Rubin D B. (1997). Estimating Causal Effects from Large Data Sets Using Propensity Scores. Annals of Internal Medicine, 127, 757 763. Rubin D B. (2007). The design versus the analysis of observational studies for causal effects: parallels wi th the design of randomized trials. Statistics in Medicine, 26 (1), 20 36. Ryu J H & Scanlon P D. (2001). Obstructive lung diseases: COPD, asthma, and many imitators. Mayo Clinic Proceedings, 76 (11), 1144 11 53. Salpeter S R Buckley N S Ormisto n T M & Salpeter E E. (2006). Meta analysis: effect of long acting beta agonists on severe asthma exacerbations and asthma related deaths. Annals of Internal Medicine, 144 (12), 904 9 12. Salpeter S R Wall A J & Buckley N S. (2010). Long Acti ng Beta Agonists with and without Inhaled Corticosteroids and Catastrophic Asthma Events. American Journal of Medicine, 123 (4) 322 32 8. Schneeweiss S. (2008). Confounding. In A. G. Hartzema H. H. Tilson & K. A. Chan (Eds.) Pharmacoepidemiology and The rapeutic Risk Management (pp. 273 299). Cincinnati, OH: Harvey Wh itney Books Company. Sears M R. (2011) Safe Use of Long Acting Agonists: What Have We Learnt? Expert Opinion on Drug Safety, 10 (5), 767 778.
251 Shah A D & Martinez C. (2004). A Comparison of the Cause of Death Recorded in the General Practice Research Database with National Mortality Statistics in England and Wale s [abstract] Pharmacoepid emiology and Drug Saf ety, 13, S2 S3. Shah B R Laupacis A Hux J E & Austin P C. (2005). Propensity score methods gave similar results to traditional regression modeling in observational studies: a systematic review. Jo urnal of Clinical Epidemiology, 58 (6), 550 55 9. Shaw D Haldar P & Pavord I. (Eds.). (2007). Treatment of Stable Asthma. In S. L. Johnson. Asthma: An Atlas of Investigation and Management (pp. 85 101). Oxford, UK: Clinical Publ ishing. Sherr i ll D G uerra S Bobadilla A & Barbee R. (2003). The role of concomitant respiratory diseases on the rate of decline in FEV1 among adult asthmatics. European Respiratory Journal, 21 (1), 95 100. Shuster, J. J. (2010). Empirical vs natural weighting in random effects meta analysis. Statistics in Medicine, 29 (12), 1259 1265. Shuster J J Jones L S & Salmon D A. (2007). Fixed vs random effects meta analysis in rare event studies: the rosiglitazone link with myocardial infarction and cardiac death. Stat is tics in Medicine, 26 (24), 4375 43 85. Silva G E Sherrill D L Guerra S & Barbee R A. (2004). Asthma as a risk factor for COPD in a longitudinal study. Chest, 126 (1), 59 65. Simons F E Gerstner T V & Cheang M S. (1997). Tolerance to the Bronchoprotective Effect of Salmeterol in Adolescents with Exercise Induced Asthma Using Concurrent Inhaled Glucocorticoid Treatment. Pediatrics, 99 (5), 655 65 9. Soriano J B Maier W C Visick G & Pride N B. (2001). Validation of General Practi tioner Diagnosed COPD in the UK General Practice Research Database. Eur opean Journal of Epidemiology, 17, 1075 1080. Stram D O Huberman M & Wu A H. (2002). Is residual confounding a reasonable explanation for the apparent protective effects of bet a carotene found in epidemiologic studies of lung cancer in smokers? American Journal of Epidemiology, 155 (7), 622 62 8. Suarez D Borrs R & Basagaa X. (2011). Differences Between Marginal Structural Models and Conventional Models in Their Exposure Effect Estimates. A Systemati c Review. Epidemiology, 22 (4), 586 588. Suissa, S. (2003). Effectiveness of inhaled corticosteroids in chronic obstructive pulmonary disease: immortal time bias in observational studies. American Journal of Respiratory and Crit ical Care Medicine, 168 (1), 49 53.
252 Suissa, S. (2008). Immortal time bias in pharmaco epidemiology. American Journal of Epidemiology, 167 (4), 492 499. Suissa S Blais L & Ernst P. (1994). Patterns of increasing beta agonist use and the risk of fatal o r near fatal asthma. European Respiratory Journal, 7 (9), 1602 160 9. Suissa, S., & Ernst, P. Inhaled Corticosteroids: Impact on Asthma Morbidity and Mortality. Journal of Allergy and Clinical Immunology, 107 (6), 937 944. Suissa S Ernst P & Kezouh A. (2002). Regular use of inhaled corticosteroids and the long term prevention of hospitalisati on for asthma. Thorax, 57 (10), 880 88 4. Szklo M & Nieto F J. (Eds.). (2004). Understanding l ack of validity: Bias. In Epidemiology, B eyond the B asics (pp. 125 176). Sudbury, MA: Jone s and Bartlett Publishers, Inc. Takizawa H. (2009). Recent Development of Drug Delivery Systems for the Treatment of Asthma and Related Disorders. Recent Patents on Inflammation & Allergy Drug Discovery, 3 (3), 232 23 9. Tan K S G rove A McLean A Gnosspelius Y Hall I & Lipworth B. (1997). Systemic Corticosteroid Rapidly Reverses Bronchodilator Subsensitivity Induced by Formoterol in Asthmatic Patients. American Journal of Respiratory and Critical Care Medicine, 156, 28 35. Therneau T M & Grambsch P M. (Eds.). (2000). Testi ng Proportional Hazards. In Modeling Survival Data: Extending the Cox Model (pp. 127 153). New York, NY: Springer Sc ience and Business Media, LLC. Thomas M von Ziegenweidt J Lee A J & Pri ce D. (2009). High dose inhaled corticosteroids versus add on long acting beta agonists in asthma: an observational study. Journal of Allergy and Clinical Immunology, 123 (1), 116 121. Thompson W D. (1991). Effect modification and the limits of biologica l inference from epidemiologic data. Journal of Clinical Immunology, 44 (3), 221 2 32. Toelle B G Ram F S. (2004). Written Individualised Management Plans for Asthma in Children and Adults. Cochrane Database of Systematic Reviews, 2, CD002171. Toh S, & Hernn M A. (2008). Causal inference from longitudinal studies with baseline randomization. International Journal of Biostatistics, 4 (1), Article 22. Town, G. I., & Holgate, S. T. Comparison of the Effect of Loratadine on the Airway and Skin Responses to Histamine, Methacholine, and Allergen in Subjects with Asthma. Journal of Allergy and Clinical Immunology, 86 (6 Pt 1), 886 893.
253 United Kingdom Department of Health (DH), Royal College of General Practitioners (RCGP), & British Medical Association (BMA). (2011). The Good Practice Guidelines for GP Electronic Patient Records, Version 4 (2011). Leeds, UK: Department of Health. Retrieved June 9, 2011, from http://www.dh.gov.uk/prod_consum_dh/groups/dh_digitalassets/documents/digital asset/dh_12535 0.pdf United States Department of Health and Human Services (DHHS). (2011). Research Prog ram Announcement No. PA 11 245: Obesity and Asthma: Awareness and Management (R01). Retrieved December 23, 2011, from http://grants.nih.gov/grants/guide/pa files/PA 11 245.h tml va n der Laan M J & Robins J M. (2003). Unified M ethods for C ensored L ongitudinal D ata and C ausality. New York, NY: Springer Verlag. van Huisstede A & Braunstahl G J. (2010). Obesity and asthma: co morbidity or causal relationship? Monaldi Archives for Chest Disease, 73 (3), 116 1 23. van Walraven C & Austin P. (2012). Administrative database research has unique characteristics that can risk biased results. Journal of Clinical Epidemiology, 65 (2), 126 1 31. van Walraven C Davis D Forster A J & Wells G A. (2004). Time dependent bias was common in survival analyses published in leading clinical journals. Journal of Clinical Epidemiology, 57 (7), 672 6 82. van Wonderen K E van der Mark L B Mohrs J Geskus R B van der Wal W M van Aalderen W M Bindels P J & ter Riet G. (2009). Prediction and treatment of asthma in preschool children at risk: study design and baseline data of a prospective cohort study in general practice (ARCADE). BMC Pulmonary Medicine, 9, 13. Vermeir e P A Rabe K F Soriano J B & Maier W C. (2002). Asthma Control and Differences in Management Practices Across Seven European Countries. Respiratory Medicine, 96 (3), 142 14 9. Virchow J C Crompton G K Dal Negro R Pedersen S Magnan A Seidenberg J & Barnes P J. (2008). Importance of inhaler devices in the management of airway disease. Respiratory Medicine, 102 (1), 10 1 9. Visness C M London S J Daniels J L K aufman J S Yeatts K B Siega Riz A M Calatroni A & Zeldin D C. (2010). Association of childhood obesity with atopic and nonatopic asthma: results from the National Health and Nutrition Examination Survey 1999 2006. Journal of Asthma, 47 (7), 822 82 9. Waning B & Montagne M. (Eds.). (2001). Pharm acoepidemiology: Principles and Practice New York, NY: The McGraw Hill Companies Inc.
254 Weatherall, M., Wijesinghe, M., Perrin, K., Harwood, M., & Beasley, R. (2010). Meta Analysis of the Risk of Mortality with Salmeterol and the Effect of Concomitant Inha led Corticosteroid Therapy. Thorax, 65 (1), 39 43. Wechsler M E Kunselman S J Chinchilli V M Bleecker E Boushey H A, Calhoun W J Ameredes B T et al. (2009). Effect of Beta2 Adrenergic Receptor Polymorphism on Response to Long Acting Beta2 Agonist in Asthma (LARGE Trial): A Genotype Stratified, Randomised, Placebo Controlled, Cross over Trial. Lancet, 374 (9703), 1754 17 64. Wells K E Peterson E L Ahmedani B K Severson R K Gleason Comstock J & Williams L K. (2012). T he relationship between combination inhaled corticosteroid and long acting agonist use and severe asthma exacerbations in a diverse population. Journal of Allergy and Clinical Immunology, 129, 1274 127 9. Williams C Crossland L Finnerty J Crane J Holgate S Pearce N & Beasley R. (1998). Case Control Study of Sa lmeterol and Near Fatal Atta cks of Asthma. Thorax, 53 (1), 7 13. Wood L & Coulson R. (2001). Revitalizing the General Practice Research Database: plans, challenges, and opportunities. Pharmacoepidemiol ogy and Drug Saf ety, 10 (5):379 3 83. Wood L & Mart inez C. (2004). The General Practice Research Database. Role in Pharmacovigilance. Drug Saf ety, 27 (12), 871 8 81. World Health Organization (WHO). (2007). International Statistical Classification of Diseases and Related Health Problems. 10th Revision. Retr ieved May 5, 2011, from http://apps.who.int/classifications/apps/icd/icd10online/ World Health Organization (WHO). (2011). Chronic Respiratory Diseases: Asthma Retrieved May 11, 2011, from http://www.who.int/respirato ry/asthma/en/ Ye X Gutierrez B Za rotsky V Nelson M & Blanchette C M. (2009). Appropriate Use of Inhaled Corticosteroids and Long Acting Beta(2) Adrenergic Agonist Combination Therapy Among Asthma Patients in a US Commercially Insured Population. Current Medical Research and Opinio n, 25 (9), 2251 225 8. Zhang Y Moffatt M F & Cookson W O. (2012). Genetic and genomic approaches to asthma: new insights for the origins. Current Opinion in Pulmonary Medicine, 18 (1), 6 13. Zhou Z Rahme E Abrahamowicz M. & Pilote L. (2005). Survival Bias Associated with Time to Treatment Initiation in Drug Effectiveness Evaluation: A Comparison of Methods. American Journal of Epidemiology, 162 (10), 1016 10 23.
255 BIOGRAPHICAL SKETCH Dr. Ayad Ali earned his bachelor of p harmaceutical s ciences de gree from University of Mosul, Iraq in 2002. He was awarded the prestigious Fulbright Scholarship from the United States Department of State, Bureau of Educational and Cultural Affairs to further his educational and professional prospects by receiving his m aster of s ci ence in p harmacy degree from the University of Florida in 2007 He was admitted to the doctoral program at the University of Florida in 2008 and earned his d octor of p hilosophy in p harmaceutical s ciences in 2012 with concentration in p harmaceu tical o utcomes and p olicy and specialization in p harmacoepidemiology & p harmacovigilance Dr. Ali has been the recipient of numerous awards and honors from professional and educational organ izations for his leadership, service, and productivity in research Also, he has multiple publications and presented his research at local, national, and international venues He received many research and travel grants and has professional experience in community, hospital, academic, and industry pharmacy sectors. Dr. A li is passionate about pharmacy and public health, and longs to contribute to the global improvement in pharmaceutical systems and drug safety, especially in developing and transitional countries including his home country, Iraq