<%BANNER%>

Risk of Suicide and Suicide Attempt Associated with Atomoxetine Compared to Central Nervous System Stimulant Treatment

Permanent Link: http://ufdc.ufl.edu/UFE0044893/00001

Material Information

Title: Risk of Suicide and Suicide Attempt Associated with Atomoxetine Compared to Central Nervous System Stimulant Treatment
Physical Description: 1 online resource (187 p.)
Language: english
Creator: Linden, Stephan
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2013

Subjects

Subjects / Keywords: adhd -- adolescents -- atomoxetine -- children -- medicaid -- stimulant -- suicidality -- suicide -- youths
Pharmaceutical Outcomes and Policy -- Dissertations, Academic -- UF
Genre: Pharmaceutical Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: BACKGROUND: As diagnosis and treatment of ADHD have increased over the last decade, new pharmacotherapies and safety concerns have emerged. We aimed to describe longitudinal trends, determinants of ADHD treatment and estimated the risk of suicidal events in youths treated with atomoxetine compared to central nervous system stimulants. METHODS: We used healthcare claims data from youths eligible for 26 state Medicaid programs from 1999-2006 linked to the National Death Index. To analyze utilization, subjects age 5-18 entered the cohort at the first mental health diagnosis commonly treated with atomoxetine or stimulants. We analyzed utilization trends per calendar month, adjusted for seasonality with segmented and Joinpoint® regression. We assessed sociodemographic and clinical determinants of treatment choice with logistic regression at first atomoxetine and stimulant dispensing. Finally, we used Cox proportional hazard models to estimate the hazard of suicidal events. RESULTS: The utilization analysis included 1,013,556 youths with 343,912 (33.9%) ever exposed to atomoxetine or stimulants. Atomoxetine utilization steeply increased after market introduction, followed by utilization decline originating late in 2003 with only a significant level-off in September 2005 (boxed warning). Strongest determinants of atomoxetine initiation were tic disorder (odds ratios ORs for new and subsequent users=2.7 to 3.3, p=0.01) and substance abuse (ORs=1.5-1.4,p=0.01). Strongest determinants of stimulant use were non-Caucasian race/ethnicity (ORs=0.5-0.8, p=0.01) and calendar year (OR=0.2-0.3, p=0.01, 2006 vs. 2003). Increasing age increased odds for new and decreased odds for subsequent atomoxetine use(6-8 vs. 15-19 age, ORs=1.7 and 0.6). Compared to stimulant use, the adjusted suicide event hazard ratio (HR) for atomoxetine was 0.94 (95%CI 0.5-1.7) in the initial user cohort and 0.6 (95%CI 0.3-1.3) for subsequent use. CONCLUSION: Decline of atomoxetine utilization significantly preceded the boxed warning, but corresponded with publicized safety concerns. Sociodemographic and clinical characteristics were associated with ADHD treatment choice. Initial and subsequent treatment of youths with atomoxetine compared to stimulants was not significantly associated with an increased risk of suicide events in current practice. Current practice, precautions and warnings appear sufficient and no further regulatory actions concerning suicidal events are necessary. The small incidence resulted in wide confidence intervals and did not allow stratified analysis of high-risk groups.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Stephan Linden.
Thesis: Thesis (Ph.D.)--University of Florida, 2013.
Local: Adviser: Winterstein, Almut Gertrud.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2013
System ID: UFE0044893:00001

Permanent Link: http://ufdc.ufl.edu/UFE0044893/00001

Material Information

Title: Risk of Suicide and Suicide Attempt Associated with Atomoxetine Compared to Central Nervous System Stimulant Treatment
Physical Description: 1 online resource (187 p.)
Language: english
Creator: Linden, Stephan
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2013

Subjects

Subjects / Keywords: adhd -- adolescents -- atomoxetine -- children -- medicaid -- stimulant -- suicidality -- suicide -- youths
Pharmaceutical Outcomes and Policy -- Dissertations, Academic -- UF
Genre: Pharmaceutical Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: BACKGROUND: As diagnosis and treatment of ADHD have increased over the last decade, new pharmacotherapies and safety concerns have emerged. We aimed to describe longitudinal trends, determinants of ADHD treatment and estimated the risk of suicidal events in youths treated with atomoxetine compared to central nervous system stimulants. METHODS: We used healthcare claims data from youths eligible for 26 state Medicaid programs from 1999-2006 linked to the National Death Index. To analyze utilization, subjects age 5-18 entered the cohort at the first mental health diagnosis commonly treated with atomoxetine or stimulants. We analyzed utilization trends per calendar month, adjusted for seasonality with segmented and Joinpoint® regression. We assessed sociodemographic and clinical determinants of treatment choice with logistic regression at first atomoxetine and stimulant dispensing. Finally, we used Cox proportional hazard models to estimate the hazard of suicidal events. RESULTS: The utilization analysis included 1,013,556 youths with 343,912 (33.9%) ever exposed to atomoxetine or stimulants. Atomoxetine utilization steeply increased after market introduction, followed by utilization decline originating late in 2003 with only a significant level-off in September 2005 (boxed warning). Strongest determinants of atomoxetine initiation were tic disorder (odds ratios ORs for new and subsequent users=2.7 to 3.3, p=0.01) and substance abuse (ORs=1.5-1.4,p=0.01). Strongest determinants of stimulant use were non-Caucasian race/ethnicity (ORs=0.5-0.8, p=0.01) and calendar year (OR=0.2-0.3, p=0.01, 2006 vs. 2003). Increasing age increased odds for new and decreased odds for subsequent atomoxetine use(6-8 vs. 15-19 age, ORs=1.7 and 0.6). Compared to stimulant use, the adjusted suicide event hazard ratio (HR) for atomoxetine was 0.94 (95%CI 0.5-1.7) in the initial user cohort and 0.6 (95%CI 0.3-1.3) for subsequent use. CONCLUSION: Decline of atomoxetine utilization significantly preceded the boxed warning, but corresponded with publicized safety concerns. Sociodemographic and clinical characteristics were associated with ADHD treatment choice. Initial and subsequent treatment of youths with atomoxetine compared to stimulants was not significantly associated with an increased risk of suicide events in current practice. Current practice, precautions and warnings appear sufficient and no further regulatory actions concerning suicidal events are necessary. The small incidence resulted in wide confidence intervals and did not allow stratified analysis of high-risk groups.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Stephan Linden.
Thesis: Thesis (Ph.D.)--University of Florida, 2013.
Local: Adviser: Winterstein, Almut Gertrud.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-05-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2013
System ID: UFE0044893:00001


This item has the following downloads:


Full Text

PAGE 1

1 RISK OF SUICIDE AND SUICIDE ATTE MPT ASSOCIATED WITH ATOMOXETINE COMPARED TO CENTRAL NER VOUS SYSTEM STIMULANT TREATMENT By STEPHAN LINDEN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTI AL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 3

PAGE 2

2 201 3 Stephan Linden

PAGE 3

3 To my parents, my family, mi carinyo and friends w ithout you I would not be who and where I am today, t hank you

PAGE 4

4 ACKNOWLEDGMENTS I express my infinite gratitude to my Doctor of Philosophy supervisory committee: Almut G. Winterstein, Richard Segal, Tobias Gerhard, Regina Bussing and Jonathan Shuster. The supervisory committee guide d me through this dissertation with their expertise, advice and inspiration. I especially thank my advisor, Almut G. Winterstein, for all her guidance, patience, support, trust, never ending challenges and encouragement given to me throughout my time at th e PhD program at the University of Florida. She is an outstanding mentor and I will forever be indebted appreciation through his experience, support patience and advice with data programming and management I also would like to thank, Nicole Corwine, the support with Institutional review board ( IRB ) applications and data use r agreements For the provisio n of data, I thank the Centers for Medicare and Medicaid Services (CMS) and the National Death Index (NDI) at the National Center for Health Statistics. The Centers for Medicaid and Medicare Services granted free dat a reuse for doctoral research. This stud y was partially funded by a grant from the Agency for Healthcare Research an d Quality (AHRQ, R01 HS0185606). My graduate studies were supported by the Mary Kay Owens Healthcare Innovation and the DuBow Family f ellowship s Additionally, I would like to than k the faculty, graduate students and staff of the Department of Pharmaceutical Outcomes and Policy. Last but not least I would like to thank my family friends and my girlfriend Carla for their emotional support, guidance, encouragement warmth and (mostly ) the right amount of distraction.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 9 LIST OF FIGURES ................................ ................................ ................................ ........ 12 LIST OF EQUATIONS ................................ ................................ ................................ ... 14 LIST OF ABBREVIATIONS ................................ ................................ ........................... 15 ABSTRACT ................................ ................................ ................................ ................... 18 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 20 Background ................................ ................................ ................................ ............. 20 Need for Study ................................ ................................ ................................ 22 Purpose of Study ................................ ................................ .............................. 23 Study Significance ................................ ................................ ............................ 23 Research Questions and Hypotheses ................................ ................................ ..... 24 Part I: Change(s) in Atomoxetine and CNS Stimulant Utilization ...................... 24 Part II: Determinants of Treatment Choice ................................ ....................... 25 Part III: R isk of Suicide and Suicide Attempt Associated with Atomoxetine Compared to CNS Stimulant Pharmacotherapy ................................ ............ 26 2 LITERATURE REVIEW ................................ ................................ .......................... 28 Assessment and Treatment of ADHD ................................ ................................ ..... 28 Attention Deficit/Hyperactivity Disorder (ADHD) ................................ ............... 28 Assessment of ADHD ................................ ................................ ....................... 28 Severity of ADHD ................................ ................................ ............................. 29 Treatment of ADHD ................................ ................................ .......................... 29 Central Nervous System (CNS) stimulants ................................ ................ 31 Atomoxetine ................................ ................................ ............................... 32 Propensity of ADHD treatment and treatment choice ................................ 35 Treatment persist ence ................................ ................................ ............... 36 Effects of Publicized Safety Concerns ................................ ................................ .... 36 Suicide, Suicide Attempt and Suicidal Ideation ................................ ....................... 37 Risk Factors of Suicidal Events ................................ ................................ ........ 38 Sociodemographics ................................ ................................ ................... 38 Antidepressants & depression ................................ ................................ ... 40 Anticonvulsants & epilepsy ................................ ................................ ........ 42

PAGE 6

6 Validity of Study Endpoints ................................ ................................ ............... 43 Suicide ................................ ................................ ................................ ....... 43 Suicide attempt ................................ ................................ .......................... 44 3 METHODOLOGY ................................ ................................ ................................ ... 47 Study Design ................................ ................................ ................................ .......... 47 Part I: Change(s) in Atomoxetine and CNS Stimulants Utilization .................... 48 Part II: Determinants of Treatment Choice ................................ ....................... 48 Part III: Risk of Suicide and Suicide Attempt Associated with Atomoxetine Compared to CNS Stimulant Treatment. ................................ ....................... 49 Data Sources ................................ ................................ ................................ .......... 49 Medicaid Analytic eXtract (MAX) ................................ ................................ ...... 50 MAX Data Validation ................................ ................................ ........................ 52 National Drug Code (NDC) & Multum Lexicon Database ............................... 52 Death Master File (DMF) and National Death Index (NDI) ............................... 53 Death Records and Cause of Death ................................ ................................ 54 Covariates ................................ ................................ ................................ ............... 54 Sociodemographic Characteristics ................................ ................................ ... 55 Co morbidities ................................ ................................ ................................ .. 56 Concomitant Medication Use ................................ ................................ ............ 57 Suicide Attempt & Suicidal Ideation ................................ ................................ .. 57 Source Population ................................ ................................ ................................ .. 58 Part I: Change(s) in Atomoxetine and CNS Stimulant Utilization ............................ 59 Study Population ................................ ................................ .............................. 59 Inclusion criteria ................................ ................................ ......................... 59 Exclusion criteria ................................ ................................ ........................ 60 Censoring criteria ................................ ................................ ....................... 60 Drug Exposure ................................ ................................ ................................ 60 Data Analysis ................................ ................................ ................................ ... 61 Seasonality adjustment ................................ ................................ .............. 61 Segmented regression ................................ ................................ ............... 62 Joinpoint regression ................................ ................................ ................. 63 Part II: Determinants of Treatment Choice ................................ .............................. 64 Study Popul ation ................................ ................................ .............................. 64 Inclusion criteria ................................ ................................ ......................... 64 Initial and subsequent treatment choice ................................ ..................... 64 Subsequent CNS Stimulant Initiation ................................ ............................... 65 Florida Provider Specialty ................................ ................................ ................. 66 Data Analysis ................................ ................................ ................................ ... 66 Part III: Risk of Suicide and Suicide Attempt Associated with Atomoxetine Compared to CNS Stimulant Treatment ................................ .............................. 68 Study Population ................................ ................................ .............................. 68 Inclusion criteria ................................ ................................ ......................... 69 Exclusion criteria ................................ ................................ ........................ 69 Censoring criteria ................................ ................................ ....................... 70 Initial and subsequent treatment ................................ ................................ 70

PAGE 7

7 Time dependent drug exposure ................................ ................................ ....... 71 Sensitivity Analysis of T ime dependence of Co morbidities ............................. 72 Study Endpoints ................................ ................................ ............................... 72 Data Analysis ................................ ................................ ................................ ... 73 Propensity score ................................ ................................ ........................ 73 Cox proportional hazard regression ................................ ........................... 75 General Tables and Figures ................................ ................................ ................... 77 Part II: Tables and Figures ................................ ................................ ...................... 79 Part III: Tables and Figures ................................ ................................ ..................... 80 4 RESULTS ................................ ................................ ................................ ............... 88 Part I: Change(s) in Atomoxetine and CNS Stimulant Utilization ............................ 88 Sensitivity Analysis of Maximum Follow up ................................ ...................... 88 Baseline Covariates ................................ ................................ ......................... 88 Sociodemographics ................................ ................................ ................... 88 Treatment initiators and index diag nosis ................................ .................... 89 Co morbidities ................................ ................................ ............................ 89 Treatment Initiation and Prevalent Use ................................ ............................ 90 Treatment Initiations Rates ................................ ................................ ............... 90 Segmented Regression ................................ ................................ .................... 91 Joinpoint Regression ................................ ................................ ...................... 91 Part II: Determinants of Treatment Choice ................................ .............................. 92 Initial Treatment Choice ................................ ................................ .................... 92 Sociodemographics ................................ ................................ ................... 93 Mental disorders & other co morbidities ................................ ..................... 93 Drug exposure ................................ ................................ ........................... 94 Logistic regression analysis of determinants of initial treatment choice ..... 94 Subsequent Treatment Choice ................................ ................................ ......... 95 Sociodemographics ................................ ................................ ................... 95 Mental disorders & other co morbidities ................................ ..................... 95 Drug exposure ................................ ................................ ........................... 96 Logistic regression analysis of determinants of subsequent treatment choice ................................ ................................ ................................ ..... 96 Comparison of Initial and Subsequent Treatment Cohorts ............................... 97 Part III: Risk of Suicide and Suicide Attempts ................................ ......................... 98 Initial Treatment ................................ ................................ ................................ 98 Drug exposure ................................ ................................ ........................... 99 Study endpoints ................................ ................................ ......................... 99 Risk of suicide and suicide attempt a ssociated with atomoxetine compared to CNS stimulant treatment ................................ .................... 99 Subsequent Treatment ................................ ................................ ................... 100 Drug exposure ................................ ................................ ......................... 101 Study endpoints ................................ ................................ ....................... 102 Risk of suicide and suicide attempt a ssociated with subsequent atomoxetine initiation compared to CNS stimulant treatment ............... 102

PAGE 8

8 Part I: Tables and Figures ................................ ................................ ..................... 104 Part II: Tables and Figures ................................ ................................ .................... 116 Part III: Tables and Figures ................................ ................................ ................... 130 5 DISCUSSION ................................ ................................ ................................ ....... 145 Part I ................................ ................................ ................................ ..................... 145 Part II ................................ ................................ ................................ .................... 146 Part III ................................ ................................ ................................ ................... 148 Covariates ................................ ................................ ................................ ...... 149 Exposure ................................ ................................ ................................ ........ 150 Study Endpoints ................................ ................................ ............................. 151 Strength and Limitations ................................ ................................ ....................... 152 Conclusion ................................ ................................ ................................ ............ 154 APPENDIX A MAX DATA ................................ ................................ ................................ ........... 156 B 26 LARGEST MEDICAID FEE FOR SERVICE STATES ................................ ..... 157 C NATIONAL DRUG CODES ................................ ................................ ................... 162 D OPERATIONAL DEFINITIONS ................................ ................................ ............. 169 LIST OF REFERENCES ................................ ................................ ............................. 174 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 187

PAGE 9

9 LIST OF TABLES Table page 1 1 Recent regulatory actions on increased suicidality associated with psychotropic medications ................................ ................................ ................... 27 2 1 Atomoxetine boxed warning ................................ ................................ ............... 46 2 2 State specific report on completeness of inpatient E codes ............................... 46 2 3 State specific report on completeness of Emergency Dep t artment E codes ...... 46 3 1 Patients newly initiating treatment under five years of age ................................ 77 3 2 Part II: Inclusion and exclusion criteria with associated cohort size .................... 79 4 1 Part I: Sensitivity analysis of 12, 24 and 36 month of maximum follow up ....... 104 4 2 Part I: Inclusion and exclusion criteria with associated cohort size ................... 104 4 3 Part I: Baseline sociodemographi c chara cteristics of the study cohort, 03 06 .. 105 4 4 Part I: Baseline treatment initiation and index diagnosis ................................ .. 106 4 5 Part I: B aseline co morbidities ................................ ................................ .......... 107 4 6 Part II: Baseline sociodemographic and clinical characteristics, initial treatment cohort ................................ ................................ ............................... 116 4 7 Part II: Baseline drug exposure, initial treatment cohort ................................ ... 119 4 8 Part II: Determinants of atomoxetine versus CNS stimulant initiation ............... 120 4 9 Part II: Baseline characteristics of subsequent atomoxetine initiators and matched CNS stimulant users ................................ ................................ .......... 123 4 10 Part II: Baseline drug exposure of subsequent atomoxetine initia tors and matched CNS stimulant users ................................ ................................ .......... 126 4 11 Part II: Determinants of subsequent atomoxetine versus matched CNS stimulant use ................................ ................................ ................................ .... 127 4 12 Part III: Distribution of exclusion and inclusion criteria, initial treatment cohort 130 4 13 Part III: Baseline sociodemographic and clinical characteristics, initial treatment cohort ................................ ................................ ............................... 131 4 14 Part III: Baseline concomitant drug exposure, initial treatment cohort .............. 134

PAGE 10

10 4 15 Part III: Distribution of mono and dua l therapy, initial treatment cohort ............ 134 4 16 Part III: Drug exposure, suicide & suicide attempts after 3, 6, 12, 24 month and overall, initial treatment cohort ................................ ................................ ... 135 4 17 Part III: Hazard ratios (HR) for suicide and suicide attempt, initial treatment cohort ................................ ................................ ................................ ............... 136 4 18 Part III: Distribution of exclusion and inclusion crite ria, subsequent treatment cohort ................................ ................................ ................................ ............... 137 4 19 Part III: Baseline sociodemographic and clinical characteristics, subsequent treatment cohort ................................ ................................ ............................... 138 4 20 Part III: Baseline drug exposure, subsequent treatment cohort ........................ 141 4 21 Part III: Distribution of mono and dual therapy, subsequent treatment cohort .. 142 4 22 Part III: Drug exposure, suicide & suicide attempts after 3, 6, 12, 24 months & overall, subsequent treatment ................................ ................................ ....... 143 4 23 Part III: Hazard ratios (HR ) for suicide and suicide attempt, subsequent treatment cohort ................................ ................................ ............................... 144 B 1 State abbreviations ................................ ................................ ........................... 157 B 2 26 largest fee for service st ate enrollees in 1999 & 2000 ................................ 15 8 B 3 26 largest fee for service state enrollees in 2001 & 2002 ................................ 159 B 4 26 largest fee for servic e state enrollees in 2003 & 2004 ................................ 160 B 5 26 largest fee for service state enrollees in 2005 & 2006 ................................ 161 C 1 Atomoxetine active in gredient ................................ ................................ ........... 162 C 2 CNS stimulant active ingredients ................................ ................................ ...... 162 C 3 Antidepressant active ingredients ................................ ................................ ..... 163 C 4 Anticonvulsant active ingredients ................................ ................................ ..... 164 C 5 Antipsychotic active ingredients ................................ ................................ ........ 165 C 6 A nxiolyti c (including sedatives and hypnotics) active ingredients ..................... 167 C 7 Alpha agonist active ingredients ................................ ................................ ....... 168 C 8 Lithium active ingredi ents ................................ ................................ ................. 168

PAGE 11

11 D 1 I ndex date ................................ ................................ ................................ ......... 169 D 2 Mental health diagnoses associated with ADHD pharmacotherapy ................. 169 D 3 Exclusion criteria ................................ ................................ .............................. 170 D 4 Censoring (exclusion during baseline) ................................ .............................. 170 D 5 Study drug exposur e ................................ ................................ ........................ 171 D 6 Study endpoints ................................ ................................ ................................ 171 D 7 Potential confounders and other covariates ................................ ..................... 172

PAGE 12

12 LIST OF FIGURE S Figure page 3 1 29 U.S. states with the largest fee for services populations. 101 Note: List of state abbreviations Appendix B ................................ ................................ .......... 51 3 2 Overview study design ................................ ................................ ....................... 77 3 3 Overview cohort selection ................................ ................................ .................. 78 3 4 Part III: Distribution of cohort eligibility, dr ug exposure, substance use, tic and oppositional defiant disorder by treatment group over time; initial treatment cohort ................................ ................................ ................................ ................. 80 3 5 Part III: Distribution of mental disorder diagnoses commonly tre ated with atomoxetine or CNS stimulants, by treatment group over time; initial treatment cohort ................................ ................................ ................................ 81 3 6 Part III: Distribution of other disorder diagnoses, by treatment group over time; initial treatment ................................ ................................ .......................... 82 3 7 Part III: Distribution of cohort eligibility, drug exposure, substance use, tic and oppositional defiant disorder by treatment group over time; subsequent treatment ................................ ................................ ................................ ............ 83 3 8 Part III: Distribution of mental disorder diagnoses commonly treated with atomoxetine or CNS stimulants, by treatment group over time; subsequent treatment cohort ................................ ................................ ................................ 84 3 9 Part III: Distribution of other disorder diagnoses, by treatment group over time; subsequent treatment cohort ................................ ................................ ..... 85 3 10 Exposure propensity score distribu tion of initial user cohort ............................... 86 3 11 Exposure propensity score of subsequent user cohort ................................ ....... 87 4 1 Eligible and at risk patients of treatment initiation per calendar month and treatment group, 2003 2006 ................................ ................................ ............. 108 4 2 Atomoxetine treatment initiation and prevalent utilization per calendar month, 2003 2006 ................................ ................................ ................................ ........ 109 4 3 CNS stimulant treatment initiation and prevalent utilization per calendar month, 2003 2006 ................................ ................................ ............................ 110 4 4 Atomoxetine treatment initiations/1,000 patie nt months, 2003 2006, seasonality adjusted ................................ ................................ ......................... 111

PAGE 13

13 4 5 CNS stimulant treatment initiations/1,000 patient months, 2003 2006, seasonality adjusted ................................ ................................ ......................... 112 4 6 Segmented linear regression analysis of atomoxetine treatment initiations/1,000 patient months, 2003 2006 ................................ ..................... 113 4 7 Joinpoint analysis of atomoxetine treatment initiations /1,000 patient months, 2003 2006 ................................ ................................ ................................ ........ 114 4 8 Joinpoint analysis of CNS stimulant treatment initiations/1,000 patient months, 2003 2006 ................................ ................................ ........................... 115

PAGE 14

14 LIST OF EQUATIONS Equat ion page 3 1 Atomoxetine Initiations per 1,000 Patient Months at Risk ................................ ... 61 3 2 CNS Stimulant Ini tiations per 1,000 Patient Months at Risk ............................... 61 3 3 Seasonal ARIMA ................................ ................................ ................................ 62 3 4 Segmented Regression ................................ ................................ ...................... 62 3 5 Joinpoint Regression ................................ ................................ ........................ 63 3 6 Logistic Regression ................................ ................................ ............................ 67 3 7 Propensity Score ................................ ................................ ................................ 74 3 8 Cox Proportional Hazard Regression ................................ ................................ 76

PAGE 15

15 LIST OF ABBREVIATION S AACAP American Academy of Child and Adolescent Psychiatry ADHD Attention Deficit/Hyperactivity Disorder ADHDRS ADHD Rating Scale IV Parent Version AHCA Florida Agency for Health Care Administration AHRQ Agency for Healthcare Research and Quality ANOVA Analysis of Variance ARIMA Autoregressive integrated moving average CAARS Conners' Adult ADHD Rating Scales CD Con duct Disorder CI Confidence Interval CMS Centers for Medicare & Medicaid Services CNS Central Nervous System DMF Death Master File DSM IV Diagnostic and Statistical Manual 4th edition ED Emergency Department EMEA European Medicines Agency EUNETHYDIS Europe an Network for Hyperkinetic Disorders FDA Food and Drug Administration FFS Fee For Service FPL Federal Poverty Level GPRD General Practice Research Database HMO Health Maintenance Organization HCUP Healthca re Cost and Utilization Project ICD 10 The Interna tional Statist ical Classification of Diseases 10 th Revision

PAGE 16

16 ICD 9 CM International Classification of Diseases, 9 th Revision, clinical modification IP MAX Inpatient records file IQ I ntelligence quotient IRB Institutional Review Board ITT Intention to treat L T MAX Long term care records file MAO I Monoamine Oxidase Inhibitors MAS Mixed Amphetamine Salts MAX Medicaid Analytic eXtract MCHIP Medicaid Expansion SCHIP Program MDD Major Depressive Disorder M ED PAR Medicare Provider Analysis and Review MEPS Medical Ex penditure Panel Survey MPH Methylphenidate MTA Multimodal Treatment Study of Children With ADHD NAMCS National Ambulatory Medical Care Survey NCHS National Center for Health Statistics NDC National Drug Code NDI National Death Index NPV Negative Predictiv e Value NIMH National Institute of Mental Health NRI (Selective) Norepinephrine Reuptake Inhibitors NSCH National Survey of Children's Health ODD Oppositional Defiant Disorder OT MAX Other Services records file

PAGE 17

17 OTC Over the C ounter PPV Positive Predictive Value PS MAX Personal summary records file RCT Randomized Controlled clinical Trials R ES DAC Research Data Assistance Center RX MAX Drug records file SARIMA Seasonal Autoregressive Integrated Moving Average SCHIP State Child Health Insurance Program S EDD State Emergency Department Databases SID State Inpatient Databases SNRI Serotonin Norepinephrine Reuptake Inhibitor SSA Social Security Agency SSI Supplemental Security Income SSN Social Security Number SSRI Selective Serotonin Reuptake Inhibitors SD Standard Deviation S DT Standardized difference test TANF Temporary Assistance for Needy Families U S United States

PAGE 18

18 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requir ements for the Degree of Doctor of Philosophy RISK OF SUICIDE AND SUICIDE ATTEMPT ASSOCIATED WITH ATOMOXETINE COMPARED TO CENTRAL NERVOUS SYSTEM STIMULANT TREATMENT By Stephan Linden May 201 3 Chair: Almut G. Winterstein Major: Pharmaceutical Sciences As d iagnos i s and treatment of ADHD ha ve increased over the last decade, new pharmacotherapies and safety concerns have emerged. We aimed to describe longitudinal trends determinants of ADHD treatment and estimated the risk of suicidal events in youths t reated with atomoxetine compared to central nervous system stimulant s We used healthcare claims data from youths eligible for 26 state Medicaid programs from 1999 2006 linked to the National Death Index To analyze utilization, subjects age 5 18 entered t he cohort at the first mental health diagnosis commonly treated with atomoxetine or stimulants We analyzed utilization trends per calendar month adjusted for seasonality with segmented and Joinpoint regression We assessed sociodemograph ic and clinical determinants of treatment choice with logistic regression at first atomoxetine and stimulant dispensing Finally w e used Cox proportional hazard models to estimate the hazard of suicidal events. The utilization analysis included 1,013,556 youths with 343, 912 (33.9%) ever exposed to atomoxetine or stimulants A tomoxetine utilization steeply increase d after

PAGE 19

19 market introduction, followed by utilization decline originating late in 2003 with only a significant level off in September 2005 (boxed warning). Strong est determinants of atomoxetine initiation were tic disorder ( odds ratio s [OR] s for new and subsequent users = 2.7 to 3.3 p<0.01) and substance abuse (OR s =1.5 1.4, p<0.01 ) Strongest determinants of stimulant use were non Caucasian race/ethnicity ( ORs=0.5 0 8, p<0.01 ) and calendar year (OR= 0. 2 0. 3 p<0.01, 200 6 vs. 200 3). Increasing age increased odds for new and decreased odds for subsequent atomoxetine use (6 8 vs. 15 19 age OR s =1 7 and 0.6). Compared to stimulant use, t he adjusted suicide event hazard ra tio (HR) for atomoxetine was 0.94 (95% CI 0.5 1.7 ) in the initial user cohort and 0. 6 (95%CI 0. 3 1. 3 ) for subsequent use. D ecline of atomoxetine utilization significantly preceded the boxed warning, but corresponded with publicized safety concerns. Sociodem ographic and clinical characteristics were associated with ADHD treatment choice Initial and subsequent treatment of youths with atomoxetine compared to stimulants was not significantly associated with an increased risk of suicide events in current practi ce. Current practice, precautions and warnings appear sufficient and no further regulatory actions concerning suicidal events are necessary. The small incidence resulted in wide confidence intervals and did not allow stratified analysis of high risk groups

PAGE 20

20 CHAPTER 1 INTRODUCTION Background Attention deficit/hyperactivity d isorder (ADHD) is the most common mental health disorder in children and adolescents. In 2007, the National Survey of Children's Health (NSCH) reported that 5.4 million children between 5 to 17 years old were diagnosed with ADHD, almost 1 in 10 in the general population. 1 3 In 2 005, the United States ( U.S. ) Food and Drug Administration (FDA) and the European Medicines Agency (EMA) directed the man ufacturer of atomoxetine (Strattera ) to include a boxed warning and additional warning statements regarding an increased risk of suicidal ideation in childr en and adolescents treated for ADHD I n addition, pharmacists were to provide patients and their ca regivers with a medication guide with each prescription or refill. 4 7 The FDA based the box ed warning on a meta analysis of 14 short term (6 18 weeks) placebo controlled trials. 8 The 14 trials involved 2 208 patients (1 357 on atomoxetine and 851 on placebo) pooled from the manufacturer's clinical trial database. The analysis showed a statistically significant higher risk of suicidal ideation in children assigned to atomoxetin e when c ompared to placebo The average risk of suicidal ideation was about four per thousand patients treated with atomoxetine (5 in 1 357) paralleled by no patients treated with placebo (0 in 851). Of all 2 208 patients, only one patient attempted suici de while being treated w ith atomoxetine. Concomitantly, concerns arose among health professionals about an increased risk of suicidality in patients using antidepressants which also resulted in boxed warnings indicating an increased risk of suicidality i n children and adolescents ( Table 1 1 ). 9 The addition of the boxed warning for antidepressants was also based on the

PAGE 21

21 result s of a meta analysis which concluded nearly twice the rate of suicidality in antidepressant users when compared to patients assigned to placebo treatment 10 Although atomoxetine is not approved by the FDA for d ep ression, it is structurally closely related to the antidepressant fluoxetine and was originally developed as an antidepressant. 11 Thus, the observed increase in suicidal ideation and a potential drug class effect support the concern that atomoxetine might increase the risk of suicidality in children and adolescents. Each year suicide attempts of children and adolescents result in about 4,400 deaths, making suicide the third leading cause of death in children and a dolescents, a nd 149,000 emergency department visits in the U.S. 12 Certain psychotropic medications are thought to reduce suicidal ideation in youth, but several have also show n to increase this risk. 10,13 Despite the importance of this safety c oncern current literature lac ks comprehensive and population based information whether th e increased suicidal ideation risk translates to general practice and to more severe clinical outcomes such as suicide attempt and manifest suicide 14 The core objective of this study was to evaluate whet her atomoxetine is associated with an increased risk of suicide attempt and suicide in patients newly treated with atomoxetine when compared to use of Central Nervous System (CNS) stimulants in a population based cohort study. To support the methodological design of the safety study, and to provide information on current practice pattern to clinicians, we supplemented the safety core of this dissertation with information on secular utilization trends and determinants of atomoxetine and st imulant treatment i nitiation

PAGE 22

22 Need for Study While the diagnosis and treatment of ADHD has increased over the last decade, new pharmacotherapies have been approved and new safety concerns have emerged. In 2002, atomoxetine was newly approved as non stimulant and non controll ed substance alternative to the standard ADHD treatment stimulants. A pproximately 2.7 million children and adolescents diagnosed with ADHD receive d pharmacotherapy in the U.S. and an estimated 15% of treated children and adolescents r eceived atomoxetine i n 2003 3,15,16 Shortly thereafter, atomoxetine was implicated with a potential risk of increased suicidal ideation in children and adolescents. C urrent literature provides evidence of an increased risk of suicidal i deation associated with atomoxetine treatment in clinical trial populations. However, no research has conclusively investigated whether the risk translates to general practice and to more severe clinical outcomes such as suicide attempts and manifest suici des. 8 14 Because of the generally low incidence of suicide attempt and suicide, no previous research has been adequately powered to detect a clinically significant risk difference while adequately adjusting for confounding factors Furthermore, secular utilization trends and treatment pattern resulting from increased ADHD treatment, emerging safety concerns and regulatory actions are unknown. This study was aimed at evaluating utilization trends, treatment determinants, the suicidality risk and if applic able, to identify high risk populations The results of this analysis are essential to clinicians in assessing risk and benefit of ADHD treatment options and to tailor regimen to individual patient needs

PAGE 23

23 Purpose of Study The first part of this study analy zed secular trends in utilization of atomoxetine versus stimulants with particular focus on changes in atomoxetine treatment initiation associated with the implementation of the boxed warning in September 2005. The second part assessed patient demographics mental co morbidities and concomitant drug use that alter the propensity of treatment initiation of atomoxetine versus stimulants. Moreover, t he first two parts not only inform clinical practice, but also deliver important insights about inclusion criter ia, potential confounders and high risk groups for the design of the safety study in part three The core safety analysis investigated whether the increased risk of suicidal ideation reported in randomized controlled clinical trials (RCT s ) translates into current practice and extends into an increased risk of suicide attempt and manifest suicide. We anticipate that our results will enable clinicians to make better informed decisions in prescribing atomoxetine and stimulants. Study Significance In 2008 the F DA began mandating drug manufacturers to provide data on suicidality in clinical trials during the new drug approval process. 17 However, the benefit of this information is limited by the nature of clinical trial s with common exclusion of high risk patients, limited sample size and newly approved medications In 2007, six of the 20 most prescribed drugs were labeled with warnings for increased suicidality, emphasizing the importance of formal studies on suicide atte mpt and suicide 18 Therefore, large population based observational research is essential to investigate a possibly elevated risk of suicidal events associated with atomoxetine in general practice. 19

PAGE 24

24 We used a population based sample representing a large proportion of youths in the United States, with overrepresentation of minorities, youths in Foster c are and youths with complex mental dis orders and thus, increased risk for suicidality. With about half of U.S. youths covered by Medicaid we expect broad generalizability of our results Given the massive sampling frame, rigorous methodological approaches and comprehensive access to healthcare utilization data we expect to adequately assess the risk for suicidal events associated with atomoxetine and establish the evidence needed for clinical risk benefit decisions. Therefore, t his study will contribute e vidence to the current drug safety controversy and provide the necessary evidence for adequate risk benefit decisions with special consideration of vulnerable pediatric populations, such as those with complex psychiat ric needs as well as minorities. Furthe rmore no previous research has examined atomoxetine utilization compared to stimulants and the effects of publicized safety concerns on prescribing patterns and utilization. Such analysis will aid clinicians in assessing their own treatment approaches in the context of their peers. It will further allow general estimates of the impact of safety concerns and regulatory action on ADHD treatment pattern and safety Research Questions and Hypotheses The research questions in this dissertation are categorized i n three parts. Where formal statistical hypothesis testing applies, statistical significance is assumed at an alpha level of 0.05, if not otherwise specified. H 0 refers to a null hypothesis and H A to an alternative hypothesis. Part I: Change(s) in Atomoxet ine and CNS Stimulant Utilization Research question 1 : What is the monthly prevalent utilization of atomoxetine and CNS stimulant s from 2003 to 2006 in pediatric Medicaid patients in 26 U.S. states?

PAGE 25

25 Research question 2 : What is the monthly treatment initia tion of atomoxetine and CNS stimulant s in 2003 to 2006 in pediatric Medicaid patients in 26 U.S. states? Research question 3 : Is the implementation of a boxed warning associated with a significant change in atomoxetine initiation after September 2005? Hyp othesis 3: H A : There is a significant decline in atomoxetine initiation after implementation of a boxed warning in September 2005. H 0 : There is no significant decline in atomoxetine initiation after implementation of a boxed warning in September 2005. Rese arch question 4 : Are there changes in the trend of atomoxetine initiation from 200 3 to 2006? Hypothesis 4 : H A : There are k max change point(s) in the trend of atomoxetine initiation. H 0 : There are k min change point(s) in the trend of atomoxetine initiation Research question 5 : Are there changes in the trends of CNS stimulant initiation from 200 3 to 2006? H ypothesis 5 : H A : There are k max change point(s) in the trend of stimulant initiation. H 0 : There are k min change point(s) in the trend of stimulant initi ation. Note that for hypotheses 4 and 5 sequential Joinpoint testing will be used. Testing starts from k min =0 and k max =4. If H 0 is rejected, the testing moves on to k min =1 and k max =4; if H A is rejected the testing moves on to k min =0 and k max =3; test ing continues until the model with the lowest number of change point is determined. If k max Joinpoint model is selected, testing proceeds to k max vs k max +1. Part II: Determinants of Treatment Choice Research question 1 : Are sociodemographic and clinical characteristics six months prior to treatment initiation (baseline) similar for initial atomoxetine and CNS stimulant users from 200 3 to 2006? Hypothesis 1 : H A : Baseline characteristics of initial atomoxetine users differ significantly from initial CNS st imulant users. H 0 : There is no significant difference of baseline characteristics of initial atomoxetine users and initial CNS stimulant users. Research question 2 : Are sociodemographic and clinical baseline characteristics of patients who switch from init ial atomoxetine to CNS stimulant pharmacotherapy similar to those who stay on the original therapy? Hypothesis 2 : H A : Baseline characteristics of patients that switch from initial atomoxetine to CNS stimulant pharmacotherapy are significantly different to those who stay on the original therapy. H 0 : Baseline characteristics of patients that switch from

PAGE 26

26 initial atomoxetine to CNS stimulant pharmacotherapy are not significantly different to those who stay on the original therapy. Research question 3 : Are soci odemographic and clinical baseline characteristics of patients who switch from initial CNS stimulant to atomoxetine pharmacotherapy similar to those who stay on original therapy? Hypothesis 3 : H A : Baseline characteristics of patients who switch from initi al CNS stimulant to atomoxetine pharmacotherapy are significantly different to those who stay on original therapy. H 0 : Baseline characteristics of patients who switch from initial CNS stimulant to atomoxetine pharmacotherapy are not significantly different to those who stay on original therapy. Part III: Risk of Suicide and Suicide Attempt Associated with Atomoxetine C ompared to CNS Stimulant P harmacotherapy We assessed the risk separately for 3 6, 12, 24 months after treatment initiation and until the end of follow up. Research Question 1 : Does treatment initiation of atomoxetine increase the risk of suicide and suicide attempt when compared to treatment with CNS stimulants Hypothesis 1 : H A : Treatment initiation of atomoxetine is associated with a signif icantly increased risk of suicide and suicide attempt compared to treatment with CNS stimulants. H 0 : Treatment initiation of atomoxetine is not associated with a signific antly increased risk of suicide and suicide attempt compared to treatment with CNS sti mulants. Research Question 2 : Does subsequent treatment initiation of atomoxetine increase the risk of suicide and suicide attempt when compared to treatment with CNS stimulants Hypothesis 2 : H A : Subsequent treatment initiation of atomoxetine is associat ed with a significantly increased risk of suicide and suicide attempt compared to treatment with CNS stimulants. H 0 : Subsequent treatment initiation of atomoxetine is not associated with a significantly increased risk of suicide and suicide attempt compare d to treatment with CNS stimulants.

PAGE 27

27 Table 1 1 Recent regulatory actions on increased suicidality associated with psychotropic m edications Year Month Event 2003 October Dear Health Care Profe ssional letters on reports of suic idality in pediatric patients t reated with antidepressant m edications for Major Depressive Disorder (MDD) 200 4 October FDA issues public health advisory on increased risk of suicidal thinking and behavior in children and adolescents treated with antid epressant medications and orders the manufacturers to revise the labeling to include a boxed warning and additional warning statement as well as to develop Medication Guides. 2005 September FDA issues public health advisory on increased risk of suic idal thinking and behavior in children and adolescents treated with atomoxetine and orders the manufacturer to revise the labeling to include a boxed warning and additional warning statement as well as to develop a Medication Guide. 200 8 December FD A requires manufacturer of antiepileptic drugs to include w arnings and information to p atients about the increased risk of suicidal thoughts and behavior as well as to develop a Medication Guide

PAGE 28

28 CHAPTER 2 LITERATURE REVIEW Assessment and Treatment of A DHD Attention Deficit/ Hyperactivity Disorder (ADHD) In 2007, the National Survey of Children's Health (NSCH) reported 5.4 million children between 5 to 17 years old ever diagnosed with ADHD, about 9.5% or almost 1 in 10 in the general population. 2 In only four years, this prevalence ha d i ncreased by 22%. 2 As a consequenc e of this constant rise over the past decade, expenditures in the United Sates for ADHD medications have increased by 594% from 1994 to 2003. Whereas the sales volume rose by 80%, pharmacotherapy prices increased by 285%. 20 On a larger scope, the worldwide prevalence of ADHD i s estimated to be around 5% for children 21 and around 3% among adults. 22 Patients with ADHD display persistent symptoms of impulsiveness and inattention, with or without hyperactivity. 23,24 Most patients are diagnosed during childhood and up to 50% of patients continue to display symptoms into early adulthood. 25 ADHD is linked to severe social, educational, occupational, and interpersonal difficulties as well as a higher risk for accidents and psychiatric co morbidities. 26 Approximately 2.7 million children and adolescents diagnosed with ADHD receive pharmacotherapy in the U.S. S timulants are the principal and most common pharmacotherapy, but an estimated 15% of children and adolescents received atomoxetine in 2003. 2 27 Assessment of ADHD The American Academy of Child and Adolescent Psychiatry (AACAP) practice parameters define an ADHD diagnosis as:

PAGE 29

29 A diagnosis of ADHD according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM IV), 24 implies the presence of hyperactive impulsive or inattentive symptoms that cause impairment and that were present before the age of seven. The symptoms must be persistent, must be more severe than is typically observed in individuals at a comparable level of development, must cause clinical ly significant impairment, e.g. in social, academic, or occupational functioning, and must be present in two or more settings, e.g., school (or work) and at home. The symptoms must not be better accounted for by another mental disorder. 28 Severity of ADHD ADHD severity is defined in three subtypes according to the DSM IV; the predominantly inattentive subtype (ADD without hyperactivity), the hyperactive and impulsive subtype with only few inattentive symptoms and the combined inattentive and hyperactive/impulsive subtype. 28 The combined subtype is most common (61%), followed by the inattentive type (30%) and the hyperactive/impulsive type (9%). Previous research indicates that children and adolescents with the combined subtype have a higher propensity for other psychiatric disorders (especially conduct disorder, ODD, and bipolar disorder). 29 ICD9 CM codes distinguish only between ADHD wit hout hyperactivity (314.00) and with hyperactivity (314.01). Treatment of ADHD Traditionally, ADHD has been treated with pharmacotherapy and/or behavioral/psychological therapy. In 1997, the American Academy of Child and Adolescent Psychiatry (AACA P) p ract ice p arameter stated: T he careful clinician balances the risks of medication, the risks of the untreated disorder, and the expected benefits of medication relative to other treatments. 30 This implies a difference between users and non users of pharmacotherapy according to the severity of ADHD and other co morbidities resulting in a potential difference in baseline risk of suicide and suicide attempt and thus confounding in the

PAGE 30

30 assessment of suicidal risk However, the se practice parameters have changed over time, mainly because of results of the Multimodal Treatment Study of Children with ADHD (MTA) initiated by the National Institute of Mental Health. MTA concluded superior ity of pharmacotherapy to behavioral treatment and routine community care, leading to the current AACAP p ractice p arameter established in 2007. 28 31 The focus of ADHD treatment is now on pharmacothe rapy and the guidelines state: B ehavior therapy may be re commended as an initial treatment if the of ADHD is uncertain, parents reject medication treatment, or there is marked disagreement about the diagnosis between parents or between paren ts and teachers. 28 The shifted focus on pharmacotherapy is confirmed by the increase of the proportion of ADHD patients receiving pharmacotherapy from 56% in 2003 to 66% in 2007. 2 Despite the shift in treatment focus, differences between users and non users of ADHD pharmacotherapy persist. This, in turn might confound the ev alu ation of the risk of suicide and suicide attempt associat ed with atomoxetine if compared to non treatment. Therefore, we have chosen stimulant users as our reference group for the atomoxetine safety evaluation. According to the AACAP p ractice p arameters, d ifferences between stimulant and atomoxetine users might surface as well. S timulants are considered the first line treatment for ADHD, particularly if no co morbidities are present. However, atomoxetine may be considered as the first medication for ADHD in individuals with an active substance abuse problem, co morbid anxiety or tic disorder Atomoxetine is a second line alternative if a patient experiences severe side effects to stimulants or more general intolerances such as hypersensitivity, incompatibili ty or non response. Typical side

PAGE 31

31 effects indicating a treatment switch to atomoxetine are mood liability, tics sleep disturbances and eating problems. About 50% of non responders to methylphenidate will respond to atomoxetine therapy and approximately 75% of responders to methylphenidate will also respond to atomoxetine. 32 The p ractice p arameters conclude: I t is the sole choice of the family and the clinician as t o which agent should individualized. Nothing in these guidelines should be construed by third party payers as justification for requiring a patient to be a treatment failure (or exper ience side effects) to one agent before allowing the trial of another. 28 Central Nervous System (CNS) s timulants CNS stimulants include methylphenidate, dexmethylphenidate and the mixed amphetamine salts: amphetamine, me thamphetamine and dexamphetamine. S timulants have been used to treat ADHD si nce the 1960s and remain the principal and most prevalent ADHD therapy 3,15,16,33 The efficacy of stimulants in treating ADHD has been established in multiple randomized controlled trials. In addition, a meta analys is concluded stimulant teachers and parents. Although academic performance was not improved, rates of delinquency and substance abuse did not change significantly either. 34 Of the different drugs summarized as stimulants, no significant differences in terms of efficacy or side effects have been documented. 35,36 In February 2005, Health Canada suspended mixed amph etamine salts for six months from the Canadian market associated with reports of cardiac sudden death. 37 39 Thereafter, t he cardiovascular safety of stimulants was controversially discussed, but did not result in a boxed warning. Instead the Food and Drug Administration ( FDA ) introduced the requirement for distribution of a medication guide to reflect concerns

PAGE 32

32 about adverse cardiovascular effects in 2007. 40 43 Although stimul ants are considered generally safe due to results of clinical trials; no large population based studies have assessed associated suicidality. 34 Atomoxetine The FDA approved atomoxetine in December 2002 for use in ch ildren older than five years, ad olescents, and adults to treat ADHD 6 The efficacy of atomoxetine in pediatric patients (age 6 to 18) in the treatment of ADHD was established in four randomized, d ouble blind, placebo controlled, short term ( 6 10 weeks) studies (N=759) with intention to treat (ITT) analyses in youths who met DSM IV criteria for ADHD 6 However, clinical trials suggest smaller effect sizes of atomoxetine compared to stimulant treatment. 44 A tomoxetine is a non stimulant, non schedule II controlled substance alternative to stimulants with a different mechanism of action to treat ADHD 7 32 C ategorized as a selective n orepinephrine reuptake inhibitor (S(NRI)) 11 atomoxetine is structurally very similar to the selective serot onin reuptake inhibitors (SSRI) antidepressant fluoxet ine. 45 The precise mechanism of a tomoxetin e in the treatment of ADHD remains unknown. inhibition of the pre synaptic norepinephrine transporter, as determined in ex vivo uptake and neurotransmitter depletion studi es analogous to SNRI antidepressants 6 7 SNRI and SSRI are currently some of the most prescribed antidepressants. In a meta analysis of all available randomized controlled clinical trials (RCTs), SSRIs and other antidepressants showed nearly twice the rate of suicidality relative to placebo. 10 This risk resulted in a boxed warning for SSRI antidepressants and might correspond to simila r problems with atomoxetine. 46

PAGE 33

33 In 2005, the FDA mandated a boxed warning for atomoxetine ( Table 2 1 ) .The initial safety analysis by the FDA was based on 14 randomized clinical trials from the 208 patients. 1 357 patients were treated with atomoxetine and 851 patients were assigned to placebo. All recorded events of suicidal ideation occurred in children younger than 12 years and within 32 days of atomoxetine treatment initiation Although the age range of study subjects was 6 to 17.9 years, the mean age was 10.5 years (SD 2.4 ) indicating a population skewed up ra nged from 6 to 18 weeks but were skewed towards shorter follow up periods. Therefore, this meta analysis could neither sufficiently analyze the risk in non clinical trial population, older adolescents nor the risk after more than three month of treatment. In the pooled analysis, the frequency of suicidal ideation was 0.37% (5/1 357) in the atomoxetine group, compared to 0% (0/851) in the placebo group. Only one suicide attempt was reported in the atomoxetine treated group and no manifest suicide was reporte d in any clinical trial. The suicide attempt was conducted ingesting multiple atomoxetine capsules. The study calculated a Mantel Haenszel incidence difference of 0.46 (95% confidence interval [CI] 0.09 0.83; p = 0.016) and Mantel Haenszel risk ratio of 2. 92 (95% CI 0.63 13.57; p = 0.172). Frequencies of suicide related events in pediatric patients with ADHD did not differ between methylphenidate and atomoxetine with a Mantel Haenszel incidence difference of 0.12 (95% CI 0.62 to 0.38; p = 0.649). The numb er needed to harm in pediatric patients for an additional suicide related event was estimated at 227. In contrast, the number needed to treat to achieve remission of

PAGE 34

34 ADHD symptoms was five. 8 In addition to clinical trial evidence several case reports document suicide ideation in patients initiating atomoxetine treatment 47,48 An expert group of the European Network for Hyperkinetic Disorders (EUNETHYDIS) reviewed the literature to summarize management of adverse effects of ADHD pharmacotherapy. The group pointed out more common self reported s uicide related events in population based studies without any ADHD diagnosis or treatment than in clinical trial population s analyzed in meta analys e s. 49 For instance, in an unrelated study the observed rate of self harm among 11 15 year olds with no mental disorder was 1.2%, 9.4% with anxiety disorders, 18.8% with depression, 12.6% with conduct disorder and 8.5% among children with hyperkinetic disorder. 50 In contrast, the meta analysis which led to the boxed warning for atomoxetine reports a proportion of only 0.4%. 8 Furthermore, the pooled clinical trials were only conducted for a maximum o f 6 18 weeks in only 2 208 patients, resulting in limited power to investigate completed suicides or suicide attempts. Thus, the risk of manifest suicide and suicide attempts in a pediatric population, and especially in those with concomitant mental dis ord er and/or concomitant psychotropic drug utilization, remains unknown. 8 A retrospective cohort study of the General Practice Research Database (GPRD) analyzed patients from 1993 to 2006 treated with stimulants or ato moxetine. From the c ohort of 18, 637 patient years (N=5 351), seven patients died. Although s uicide was not the primary outcome of the study, the standardized mortality ratio (SMR) of suicides for patients aged 11 14 years was 161.9 (95%CI 19.6 584. 9 ) and f or patients aged 15 21 years 1.84 (95% CI 0.05, 10. 3 ). The SMR is calculated to quantify a change of mortality in a study versus a reference population However, the SMR was based only on two

PAGE 35

35 confirmed and one possible suicide, as well as calculated with es timates of expected deaths among the general population as comparison group. Most importantly, the study did not address confounding adequately. Thus, the study cannot exclude confounding as the source of an increased risk of suicide among users of methylp henidate, dexamphetamine or atomoxetine by for example depression or other factors 51 Propensity of ADHD t reatment and t reatment c hoice Following the above ci ted AACAP guidelines, ADHD treatment choice is influenced by clinical characteristic s, but also socio demographi c and provider characteristics. 52 Previous research has indicat ed that male gender, school age being Caucasian rural dwelling, and Foster care are associated with a significantly higher propensity o f stimulant treatment. The propensity of stimulant initiation decrease s with co morbid mental disorders especially more complex mental disorders such as b ipolar disorder, schizophrenia or autism. In contrast, ADHD with hyperactivity increase s the propensi ty of stimulant initiation. 53 S timulant utilization among youths has been lowest in uninsured youths and W estern parts, and highest in S outhern parts of the U.S. 52 54 Amid Medicaid beneficiaries, patients in Foster care had a higher propensity to initiate stimulant treatment compared to patients in Temporary Assistance for Needy Families (TANF) or State Children' s Health Insurance Program ( SCHIP ) programs. 53 Finally, when adjusted for ADHD severity, youths diagnosed by psychiatrists were 42% less likely to receive ADHD pharmacotherapy than those diagnosed by primary care ph ysicians. 53 At the same time, the more severely impaired and complex group of ADHD patients were more likely to consult psychiatrists than primary care physicians. 55 However, no comprehensive information is a vailable for determinants of atomoxetine treatment choice.

PAGE 36

36 Treatment p ersistence Previous research describes early discontinuation of ADHD treatment as common. 56,57 For example 54% of youths treated with stimulants filled only one prescription, 19% filled only two prescriptions, 16% filled three or four prescriptions, and only 11% filled five or more prescriptions within one year. 56 However, the persistence of stimulant treat ment varies greatly. Whereas only one half of the patients continue treatment after the first year, 17% continue pharmacotherapy for five years or more. 15 Similar to treatment initiation, treatment continuation has been associated with socio demographic and clinical characteristics. Younger age, being Caucasian Foster care and disability status predict greater tre atment persistence. 58 61 Also, a high starting dose, use of extended release formulations and other psychotropic co ncomitant medications increased stimulant use persistence. 60 ,62,63 In contrast, less severe ADHD, absence of hyperactivity, ADHD family history and mental co morbidities such as bipolar disorder, tic disorder and oppositional defiant disorder were associated with lower treatment persistence. 58,60,64 Effects of Publicized Safety Concerns Previous analyses of drug safety concerns and subsequent safety alerts revealed quick responses of physicians and patients to the publicized safety concerns. For example, one study analyzed t rends in antidepressant utilization from 2002 to 2005 to capture the effect of different publicized safety concerns. The authors showed that the paroxetine safety advisories and a boxed warning resulted in a significant reduction of antidepressant utilizat ion in the targeted populations. 65 In 2003, the FDA distributed a public health advisory for an increased suicidal risk in pediatric patients treated with

PAGE 37

37 selective serotonin reuptake inhibitors (SSRI s). In the subs equent 24 month utilization of SSRIs among depressed youths and adults declined by 58%. 66,67 Other prominent examples of publicized safety concerns are the results of the Women's Health Initiative (WHI) and rosigli tazone studies. The WHI reported an increased risk of cardiac adverse events in July 2002. As a result, hormone replacement therapy (HRT) utilization dropped in the first half of 2003 to 66% compared to the first half of 2002. 68 For rosiglitazone, a meta analysis reported in May 2007 an increased cardiovascular risk associated with rosiglitazone, which resulted in a boxed warning for heart failure. As result utilization of rosiglitazone decreased to 59% in December 2007 when compared to February of the same year. 69 Changes in utilization of atomoxetine before and after the boxed warning concerning an increased risk of suicidality in youths have not been described nor analyzed so far. Suicide, Suicide Attempt and Suicidal Ideation The terms suicidal attempt and suicidal ideation were defined by an independent panel of suicide experts assembled by Columbia University on behalf of the FDA. 70 The suicidality classification project defined suicide ideation as passive thoughts about wanting to be dead or active thoughts about killing one self which are not accompanied by preparatory behavior. Suicidal thoughts are considered as a synonym for s uicidal ideation and are included in the d efinition of suicidal ideation. A suicide attempt was defined as self injurious behavior associated with some intent to die. The intent to die does not have to be explicit. For example, if a patient denies intent t o die, but thought that the self injury could be lethal, then intent can be inferred. The presence of intent to die differentiates suicidal acts from self injury.

PAGE 38

38 Suicide, suicide attempt and suicidal ideation are serious public health problems in children and adolescents. Suicide was the third leading cause of death in children and adolescents in 2007 resulting in 4 400 deaths each year. For each suicide, estimates of non fatal suicide attempts vary from 11 25 resulting in 149,000 youth treated in Emergen cy Departments for self inflicted injuries in the U.S A survey in schools across the U.S. revealed 15% of students seriously considered suicide, 11% created a plan, and 7% tried to commit suicide in the twelve months prior to the survey. 12 71 National estimates for sui cide of 1 to 20 years old youth are approximately 2.69 per 100,000 person years. 72 However, among youth with ADHD the number of suicides has been reported to be between 32 and 37 per 100,000 person years. 73 Risk Factors of Suicidal Events Sociodemographic s Suicid ality affects all y outh, but some groups are at higher risk than others. Males are more likely to commit suicide than females with 84% of all suicides occurr ing in males. Females, however, are more likely to report attempting suicide than males. Cultural variations in suicid e rates also exist, with Native American/Alaskan and Hispanic youth having the highest rates of suicide related fatalities. A nationwide survey of youth in grades 9 12 in public and private schools in the U.S. found Hispanic youth were more likely to repor t attempting suicide than their B lack and W hite, non Hispanic peers. 12 Age strongly influences the risk of suicides, with children ages 10 to 14 showing a risk of 0.9 suicides per 100,000 person years compared to 6.9 suicides per 100,000 person years in adolescents 15 to 19 years old. In contrast to the general

PAGE 39

39 population, children and adolescents are much more likely to use firearms, suffocation, and poisoning than other metho ds of suicide. 72 The National Institute of Mental Health (NIMH) determines depression, impulsiv ity and other mental disorders (especially mo od and personality disorders), prior suicide attempt and a substance abuse disorder (often in combination with other mental disorders) as the core risk factors for the vast majority of completed or attempted suicides (more than 90%). 74,75 Others report the following additional risk factors: conduct disorder, antisocial behavior, schizophrenia, 74 anxiety ,sleep disorders, mania and bipolar di sorder, attention deficit/hyperactivity disorder, psychotic disorder, delirium, sexual deviations, disturbance of emotions specific to childhood or adolescence as well as additional indicators of psychiatric severity such as number of psychiatric hospitali zations, number of psychiatric visits with a depression diagnosis and number of psychiatric drug classes (antipsychotics, benzodiazepines, and other psychotropics). 76 We consider ed analyze d and if applicable adjust ed for all these risk factors in our study design and analyses. Furthermore, the NIMH and other researchers identify these additional risk factors: family history of mental disorder or substance abuse, family history of suicide, family violence, including physical or sexual abuse, stressful life events, hopelessness, and firearms in the home. 74,75 Th e current study is incapable to adjust for or even measur e any of these additional risk factors due to the nature of administrative health claims database s However, we are not aware why any of these risk factors should be unevenly distributed in the comparison groups.

PAGE 40

40 Antidepressants & d epression Depression is more prevalent in children and adolescents with ADHD than in the same age group without ADHD. The combination of ADHD and depression in youths results in a more s evere course of psychopathology, a higher risk of long term impairment and suicide than in yo uths with either disorder alone. 16 77 The use of antidepressants in children and adolescents has rapidly increased in the last 10 to 20 years. From 1987 to 1996 alone, prevalence increased from 0.30 to 1.06 per 100 children aged 6 to 1 4 years and 0.5 to 2.1 per 100 adolescents aged 15 to 18 years. The highest use of antidepressants was observed among Caucasian, male youths aged 6 to 18 years, residing in the southern U.S. and publicly insured. The rate of psychotropic medication use am ong youth without insurance was less than one half that of children with public or private insurance. 78 Initial concerns of increased suicidality rel ated to antidepressants were raised by a review of paroxetine suicidality summary data. As a result, the FDA requested similar data for eight other antidepressants with pediatric research data. After a preliminary, review the FDA issued a Public Health Adv isory to Health Care Professionals concerning reports of suicidality in pediatric patients being treated with antidepressant medications for Major Depressive Disorder (MDD) in October 2003 ( Table 1 1 ). The Public Health Advisory u rged health care professionals to use these nine drugs with caution and called for a Psychopharmacologic Drugs Advisory Committee and the Pediatric Subcommittee meeting in February 2004. 79 This meeting culminated in the FDA issuing a strong warning and mandating labeling changes for ten, mostly newer antidepressants. The labeling changes included warnings to monitor all patients (adult and pediatric) being treated wi th these drugs for suicidality. In October

PAGE 41

41 2004, the FDA communicated a second public health advisory announcing the requirement of a boxed warning for all antidepressants regarding the risk of suicidality in pediatric patients. Finally, in February 2005, the FDA provided specific language for the warning and required a patient medication guide. 67 When the FDA recognized the possible increased risk of suicidality associated with antidepressants, a decision was made to combine the currently available information of clinical trials in a meta analysis. However, the combining of information from dif ferent clinical trials proofed to be not without difficulties. Although all studies were placebo controlled RCTs, they varied significantly, especially, in study protocol, inclusion and exclusion criteria and intervention drug. Definitions of suicidality a cross studies were heterogeneous and studies lacked detailed information and validated measures to a ssess and evaluate suicidality. As a consequence, the FDA instructed the University of Columbia to conduct a suicidality classification project to standardi ze the measures of suicidality, to review and re classify suicidality outcomes of the existing clinical trials, and finally complete a subsequent meta analysis under the supervision of Dr. Hammad. 10,70 The 24 clinic al trials of this meta analysis included over 4 500 patients and no suicides. But the researchers identified, with the newly validated measurement of suicidality, 89 occurrences of suicidal behavior or ideation and 120 possible occurrences of suicidal beha vior or ideation. Based on this meta analysis, the FDA treated with antidepressants to be 4%, a rate which was twice the placebo risk of 2%. 9

PAGE 42

42 For selective serotonin reuptake inhibitors (SSRI), th e risk ratio was 1.66 (95% CI, 1.02 2.68), and for all antidepressant drugs, the risk ratio was 1.95 (95% CI, 1.28 2.98). 10 Follo wing this meta analysis, in 2005 the FDA began a comprehensive review of 295 individual antidepressant trials which included over 77,000 adult patients with MDD and other psychiatric disorders. 80 The FDA mandated manufacturers to add a boxed warning and expanded warning statements in their labeling that alerts health care providers to an increased risk of suicidality in children and adolescents currently treated with all types of antidepressants. Moreover, the FDA mandated distribution of Patient Medication Guides (Med Guides), which advise patients of the risks and precautions which should be considered during treatment with antidepressant drugs. 80 However, the suicidality risk associated with antidepressants has been discussed controversial. 81 Anticonvulsants & e pilepsy In December 2008, the FDA issued a public health advisory on increased suicidal ideation in patients treated with anticonvulsant drugs. The FDA reviewed clinical t rials for eleven anticonvulsants in 66 randomized controlled trials (N=43,892) and discovered almost twice the risk (OR 1.8, 95% CI 1.24 to 2.66 ) of suicidal ideation compared to patients who did not take one of these drugs. The reported number needed to h arm was 530. In the clinical studies reviewed by the FDA (mean trial length was 12 weeks), this risk was reported as early as one week after starting treatment and continued for the entire length of treatment. Overall, four manifest suicides were observed, all in the anticonvulsant treated groups. As a result of this review, manufacturers were mandated to develop a Medication Guide to provide patients with information about the risk of suicidal thoughts and behaviors. 82

PAGE 43

43 Numerous pharmacoepidemiologic studies have been published since the implementation of the boxed warn ing. Although reasonable evidence exists for an increased risk of suicidal ideation following epilepsy diagnosis and anticonvulsant drug treatment, the underlying causes of the risk are not fully understood. Possible alternative causes are reappearance of prior suicidality or prior mental illness. However, researchers seem to agree on the higher prevalence of co morbid mood disorders in epilepsy and higher prevalence of suicidal ideation in patients with epilepsy than in the general population. 13 Validity of Study Endpoints Suicide During the study period, 1999 2006, ICD 10 codes were the format for codi ng causes of death on death certificates in the U.S. 83 ICD 10 codes for intentional self harm as cause of death on death certificates have not been validated (ICD 10 X60 X84). However, the predecessor ICD 9 and the current revision codes used in this analysis have been frequently used in previous research (E codes 950.x 958.x). 76,84,85 In 1989, death certi ficates cause of death coding with ICD 9 E codes were validated in Army Veterans of the Vietnam era. A physician panel determined ICD 9 E codes for intentional self harm or injury related death certificates have a specificity and sensitivity of more than 9 0%. 86 Gjertsona et al analyzed the effects of switching from ICD 9 to ICD 10 codes on intentional injury mortality statistics in Italy and Norway. ICD 10 codes did not change the overall tr ends for suicides in either country. 87 Completed suicides are rare events emphasizing the need to have sensitive measures that comprehensively capture completed suicides from death records. Ray et

PAGE 44

44 based on personal communication with the Tennessee Health Department, Bureau of Vital Statistics in 2006 (ICD 10 codes Y10 88 Patorno et al because mortality due to injuries or accidents account for a proportion of suicides, reaching 87% as being suicidal a mong accidental deaths suspected. 84 Patorno et al identified violent deaths with ICD 10 codes, S00 T78, V01 V99, W00 X59, and Y10 Y34 based on a small study from Finland which may or may not be applicable to American coding practices and our M edicaid population 89 Suicide a ttempt In the ICD 9CM coding language, E codes are used to provide supplemental information about the cause and intent of an injury. Iribarren et al validated the use of ICD9 CM E codes (E950.x E958.x ) to ascertain suicide at tempts resulting in emergency department (ED) visits or hospitalizations from electronic administrative health database against medical chart review. The study showed a sensitivity of 95%, a specificity of 87% and a positive predictive value of 86% to iden tify suicide attempts compared to medical charts review. 19 The authors used a cohort of 150,000 patients enrolled in a large HMO (Kaiser Permanente). Thus, patients and coding practices might be different to our Medicaid database, although unlikely. 19 Previous research validate d the accuracy of E codes for injury related hospitalizations with medical rec ord review. The accuracy of ICD 9 CM E codes was about 85% in broader groups of E codes 90 T he validity of E codes for self inflicted harm for combined hospitalizations and ED visits has been reported with a PPV of 86% 90 Providing E codes for reimbursement is mandatory in approximately half of the U.S. states. The Agency for Healthcare Research and Quality (AHRQ) maintains the

PAGE 45

45 largest collection of longitudinal hospital care and ED data in the U.S. The Healthcare Cost and Utilization Project (HCUP) is sponsored by AHRQ to analyze the data HCUP reports the proportion of injury reports with complete E codes to typically exceed 90%. 91 Code Evaluation included data of 33 states, of the analyzed 26 U.S. States in our study, 18 are included in the report. Of those 18 states, 1 2 have regulations or mandates on reporting E codes and 6 specifically enforce their regulat ions or mandates. The 1 8 states have an average of 87% (Minimum 53%, Maximum 99%) comple teness of Inpatient E codes ( Table 2 2 ). The report of ED discharges includes data of 9 states, of the analyzed 26 U.S. states in our study, 4 are included in the report. Of those 4 states, 3 have regulations or mandates on reporting E codes and 3 specifically enforce their regulations or mandates. The 4 states have an average of 93% (Minimum 72%, Maximum 98%) completeness of ED E codes ( Table 2 3 ). Mini Sentinel is a pilot project by the FDA to development of an active surveillance system to monitor FDA regulated medical products. In terms of suicide attempts, the Mini Sentinel study concluded a high PPV of suicide atte mpts in validated databases. However, the number of validated databases is limited and the generalizability to other clinical populations and databases is unknown. 92 In conclusion, validation of suicide attempts from electronic administrative health data in particular from Medicaid, seems to be limited. However, existing validation concludes adequate validity of manifest suicides and suicide attempts requiring hospitalization and ED visits.

PAGE 46

46 Table 2 1 Atomoxetine box ed warni ng Atomoxetine increases the risk of suicidal thinking in children and adolescents with ADHD. Anyone considering the use of Strattera must balance the increased risk of suicidal thinking with the clinical need for the drug. Patients who are started on the rapy should be observed closely for clinical worsening, suicidal thinking or behaviors, or unusual changes in behavior. Families and caregivers should be advised to closely observe the patient and to communicate changes or concerning behaviors with the pre scriber. Table 2 2 State spec ific report on completeness of i npatient E codes States Inpatient Injury Discharges with E codes [%] Mandates/Regulations E nforce Mandates/Regulations Florida 80.1 Yes No Georgia 94.2 Yes No I llinois 56.5 N/A N/A Iowa 80.7 No Kansas 59.8 No Massachusetts 98.6 Yes No Minnesota 85 No Missouri 98.3 Yes Yes New York 98.8 Yes Yes North Carolina 95.5 No Pennsylvania 96.3 Yes Yes South Carolina 97.1 Yes Yes Tennessee 94.9 Yes Yes Te xas 71.8 Yes No Vermont 98.8 Yes No Virginia 78.1 Yes No West Virginia 52.6 No Wisconsin 98.5 Yes Yes Table 2 3 State specific report on completeness of Emergency Department (ED) E codes States Injury Discharges E cod es [%] States with Mandates/Regulations Enforce Mandates/Regulations Minnesota 71.9 No Missouri 98.2 Yes Yes South Carolina 97.6 Yes Yes Tennessee 97.3 Yes Yes

PAGE 47

47 CHAPTER 3 METHODOLOGY Following the sequence of the research questions, presentation of the study methodology was divided into three parts: (1) Trends in atomoxetine and stimulant utilization, (2) Determinants of initial and subsequent treatment choice and (3) Risk of suicide and suicide attempt associated with atomoxetine compared to CNS sti mulant treatment. Unless otherwise specified, t he level of significance for all statistical tests was set at 0.05 two sided. SAS 9.2 (SAS Institute, Cary, NC) 93 was used for data management and data analyses. Graphs were created with Microsoft Excel 2010 (Microsoft Corp., Redmond, WA) 94 This study was approved by the University of Florida Institutional Review and Privacy Bo ards (#269 2012), National Death Index (# 2012 0044 ) and the Centers for Medicare and Medicaid Services Privacy Board (DUA 23778 ), including a waiver for the Health Insurance Portability and Accountability Act authorization. Data was obtained under data use rs agreements with the Centers for Medicare and Medicaid and the National Death Index and are not available for shari ng. Study Design A ll analyses were conducted with the same source popul ation of Medicaid beneficiaries. Specific inclusion exclusion and c ensoring criteria for the cohorts assembled for each study part are detailed later in this chapter. Following is a brief description of the study design for each of the three study parts ( Figure 3 2 )

PAGE 48

48 Part I: Change(s) in Atomoxet ine and CNS Stimulants Utilization We used a time series design to analyze trends of treatment initiation and prevalent use of atomoxetine and stimulant s in automated healthcare claims data from youths eligible for 26 state Medicaid programs from 200 3 to 2 006. Youths age 5 18 entered the cohort at the first diagnosis of a mental health di sorder commonly treated with atomoxetine or stimulants such as attention deficit/hyperactivity disorder (ADHD), adjustment reaction, disturbance of conduct or other, mixed or unspecified emotional disturb ance. The study period extended from shortly after approval of atomoxetine in January 2003 to December 2006. The study period was segmented into 48 months. We adjust ed for seasonality with s easonal autoregressive integrated moving average (SARIMA). 95 We used segmented regression to determine abrupt changes in the slope of atomoxetine initiation rates at the implementation of the boxed warning in September 2005 In a ddition, we empirical ly analyzed significant changes in atomoxetine and stimulant initiation rate trends with Joinpoint regression models 96 Part II: Determinants of Treatment Choice In a retrospective cohort design, we used mult ivariate logistic regression models to identify determinants of initial and subsequent treatment choice in patients treated with atomoxetine compared to stimulants. We fitted two separate logistic regression models comparing treatment initiation of atomoxe tine ( initial user cohort) and subsequent atomoxetine treatment (subsequent user cohort) to stimulant treatment. Patient s subsequently initiating atomoxetine treatment were matched to patients with current exposure to stimulants by month s since stimulant t reatment initiation Analyzed

PAGE 49

49 determinants of treatment choice were sociodemograph ic and clinical characteristics extracted from a six month baseline period preceding the initiation or match dates Part III: Risk of Suicide and Suicide Attem pt Associated w ith Atomoxetine C ompared to CNS Stimulant Treatment In a retrospective cohort design, we used exposure propensity score adjusted Cox proportional hazard models t o e valuate the risk of suicide and suicide attempt in pediatric patients initiating treatment with atomoxetine compared to stimulants The analys e s used a new users design to allow detect ion of risk increases shortly after treatment initiation and avoid bias introduced by depletion of susceptibles 97 We fitted two separate Cox proportional hazard regression models comparing treatment initiation of atomoxetine versus stimulants ( initial user cohort) as well as patie nts starting subsequent atomoxetine treatment after initial stimulant treatment (subsequent user cohort) Patient s subsequently initiating atomoxetine treatment were matched to patients with current exposure to stimulants by month s since treatment initiati on. Data Sources The study cohort s w ere assembled from Medicaid Analytic eXtract (MAX) data, consisting of automated administrative healthcare claims, obtained from the 26 U.S. states with the largest pediatric populations eligible for Medicaid fee for ser vice benefits during the years 1999 to 2006. The Medicaid Extract (MAX) data was acquired from the Centers for Medicare and Medi caid Services. MAX data provide monthly detail on Medicaid eligibility, demographic information, claims detail for all in and o utpatient services as well as medications reimbursed by Medicaid. This dataset has been linked to the Social Security Agency (SSA) Death Master File (DMF) and in a second stage, to the National Death Index to validate deaths and ascertain causes of death.

PAGE 50

50 Medicaid Analytic eXtract (MAX) The Medicaid Analytic eXtract (MAX) data is a set of person level data files on Medicaid eligibility, service utilization and payments. 98 MAX data comprise monthly Medicaid eligibility information along with demographic information and all administrative h ealth claims including ambulatory, acute, and long term health care encounters, as well as pharmacy claims and other auxiliary services ( Appendix A ). The 29 considered U.S. states were: Texas, Georgia, Florida, North Carolina, Lou isiana, Illinois, Ohio, South Carolina, Mississippi, Arkansas, Indiana, Massachusetts, New York, Tennessee, Alabama, Missouri, Pennsylvania, Virginia, West Virginia, Iowa, Kansas, Idaho, Nebraska, Wisconsin, Maryland, New Hampshire, Vermont, New Jersey, an d Minnesota ( Figure 3 1 ). MAX data was chosen because of its large publicly insured pediatric population, comprehensive coverage of health care services for youths and overrepresentation of especially vulnerable pediatric populati ons, such as those with complex psychiatric needs as well as minorities. In 2007, 39% or 28.8 million of all 74 million children in the U.S. were enrolled in Medicaid. 99 Of those 28.8 million, 18.5 million or 64% were enrolled in the 29 U .S. states considered in our study. 100 Between 1999 and 2006, a mean of 41% of Medicaid beneficiaries were enrolled in fee for service in the 2 9 states ( Appendix B ).

PAGE 51

51 Figure 3 1 2 9 U.S. states wi th the largest fee for services populations. 101 Note: List of state abbreviations Appendix B

PAGE 52

52 MAX Data Validation The validity of Medicaid claims data with respect to ADHD diagnosis accuracy severity or accuracy of atomoxetine and stimulant treatment has no t been analyzed so far. However, Medicaid pharmacy claims data have been validated to define psychotropic drug exposure with a positive predictive value and negative predictive value of over 85%. 102 A previous study analyzed the same source dataset t o validate comple teness of MAX data. 103 Medicaid fee for service enrollment several diagnoses and prescription drug utilization rates stratified by state and calendar year were analyzed This analysis revealed some incomplete data and resulted in exclusion of some calendar years for two states and exclusion of three states. The previous study detected significantly lowe r proportion of overall prescription drug utilization and/or mental health diagnoses in Tennessee from 1999 2002 Pennsylvania in 1999 Maryland, Alabama and New Jersey compared to similar proportions among other study years and states. 104 Consequently, we excluded t he afore mentioned years and states from our anal ysis, reducing the states analyzed from 29 to 2 6. However, the excluded years and states constitute d only a small proportion of our sample size. National Dr ug Code (NDC) & Multum Lexicon D atabase National Drug Codes (NDCs), provided with each dispensed pr escription in pharmacy claims, were translated into active ingredients and American Hospital Formulary Services (AHFS) drug therapeutic classes with the drug classification system provided by Cerner Multum, Inc., Denver Colorado, 2008 ( Appendix C ). 105 Although we were unable to match and subsequently translate 38,536 NDCs, or 37.7%, of all unique NDCs in the MAX data, these NDCs only accounted for 1. 7 %, of all

PAGE 53

53 pharmacy claims. Therefore, we conclude d a h igh and satisfactory translation of NDCs with the Multum Lexicon database S timulant exposure was defined as a prescription dispensed for methylphenidate, dexmethylphenidate and the mixed amphetamine salts: amphetamine or dexamphetamine. S timulant exposur e includ ed all available doses, immediate and sustained release forms. Atomoxetine exposure was defined based on pharmacy claims for atomoxetine, in cluding all available doses. Dosing differences were not considered in any analysis, because body weight, an important factor to measure dosage in children, is not available in MAX data. Death Master File (DMF) and National Death Index (NDI) We identified all death s and the cause of death in our study cohort with combined information from MAX data, U.S. Social S ecurity Agency (SSA) Death Master File (DMF) and National Death Index (NDI). The DMF is a database with date of death of SSA beneficiaries and non beneficiaries. The information provided is derived from SSA payment records and information accumulated from relatives of deceased individuals, funeral directors, financial institutions, postal authorities as well as other government agencies. 106 SSA DMF is more complete in age groups with a larger proportion of SSA beneficiaries, with 93 to 96 percent of all death included in the DMF data in persons 65 years old and older when compared with deaths reported in official U.S. vital statistics. 107 109 In contrast, deaths of persons aged 0 to 24 are less reliably captured with approximately 43 percent of actual deaths in this age group report ed in the DMF. 107 All deaths extracted from MAX and SSA DMF information were linked to the National Dea th Index (NDI). The NDI is a database maintained by the National Center for Health Statistics (NCHS) and accumulates death records collected by each

PAGE 54

54 individual U.S. state vital statistics office. 110 The NDI database includes the date of death, the death certificate number, and the cause of death. Cause of death with in the NDI database is defined by the International Statistical Classification of Diseases, Tenth Revision (ICD 10). 111 The NDI is considered the most accurate death database and the only U.S. wide source to provide additional information on cause of death. 109,112 114 Previous research validated suicide death certificates against medical charts and autopsy reports concluding a sensitivity of 90% and a specificity of 100%. 86 Because use of the NDI is charged per individual record submitted for analysis and per year searched ($350 basic fee + 21ct per record and year searched), we utilized NDI only to verify deaths and to obtain cause of death information rec onciled fro m MAX and SSA DMF data sources. Thus, the positive predictive value (PPV) of deaths in our study is expected to be high Death Records and Cause of Death W e identified 773 death s w ith the combined information from MAX and DMF data in our study c ohort NDI information validated 685 (88.6% ) with complete death information, including cause of death. Conse quently we excluded 88 patients (11. 4 %) with incomplete death information This proportion of validation is consistent with a previous study from o ur research group, which us ed the same Medicaid source population to analyze cardiac safety of stimulants in children and adolescents. 103 Here, NDI was able to match 89.2% of the identified death s Covariates Covariates were measured during the six month period preceding the index date, included age and calendar year at index date, gender, race/ethnicity, state, reasons for

PAGE 55

55 Medicaid eligibility (e.g., Foster care ) any concomitant drug exposure, prior hospitalizations and presence of other psychiatric disorders Sociodemographic Characteristics Assessed sociodemographic characteristics were: age, follow up time, gender, race/ethnicity and reason for Medicaid eligibility. State was defined as the state of residence at index date. Reason for Medicaid eligibility was categorized into Foster care eligibility for Supplemental Security Income (SSI) or Temporary Assistance for Needy Families (TANF) ( Appendix D ) 72 Children enrolled in Foster care have a high er propensity of physical and mental health problems, which are associated with higher suicidality. This may result from possible maltreatment, separation from their homes and families, and the continuing disruptions to their daily lives. Previous research id entified up to 80 percent of children and adolescents enrolled in Foster care to have some kind of emotional or behavioral disorder, developmental delay, or other problem requiring mental health services. 115,116 Al l children enrolled in Foster care are automatically eligible for Medicaid. 117 Supplemental Security Income (SSI) is a Federal income supplement program to help aged, blind, and disabled adults and children with little or no income to provide cash to meet basic needs for food, clothing, and shelter Eligibility is determined by the Social Security Administration, with an income limit of approximately 75 percent o f the Federal Poverty Level (FPL) and an individual asset limit of $2,000. Eligibility for SSI comprises eligibility for Medicaid. 118 Temporary Assistance for Needy Families (TANF) provides cash assistance to low income families with dependent children to enable care for children in their own homes or in the homes of relatives. To receive TANF, families must include at least one

PAGE 56

56 child between 6 to 18 years old attending school and parents or caretakers must participate in school conferences. Countable assets may not exceed $2,000 and the value of vehicles needed for work may not ex ceed $8,500. The quantity of temporary cash assistance is calculated individually per family, conditional on state of residence, family size and family circumstances. For example in Florida, eligibility for TANF is generally set at 200% of the FPL. In 2006 200% of the FPL was about $32,200 annually or $2,683 per month for a family of three. 119 Co morbidities Because DSM IV criteria for ADHD severity are not available in ICD9 CM coding language we used proxies to estimate the severity of ADHD. W e considered the last two digits of the primary diagnosis of ADHD if available, to identify presence ( ICD9 CM 314.01) or absence (314.00 or 314) of hyper activity with ADHD at baseline Excluded were other forms of h yperkinetic disorder of child hood (314.1/.2/.8/.9), because of very low incidence ( Appendix D ). Additional measured p sychiatric co morbidities include d substance use disorder (291 xx 292 xx, 303 xx 305 xx), anxiety (300.0, 300.2, 300.3, 308. xx 309.21, 3 09.81 ), bipolar disorder (296.xx 301.13), schizophrenia (295. xx ), depression (296.2 296.3, 300.4, 311.xx), mental ret ardation (318.xx), o ppositional defiant disorder (313.81) or psychosis (297 .xx 298 .xx ) and other (not mentioned before within 290.xx 319.xx). We also described overall number of distinct psychiatric disorder categories present during baseline as an indicator of mental health severity (one, two, three, four or more). A complete list of covariates and their definitions is provided in Appendix D

PAGE 57

57 Previous research has shown superior validity of diagnoses were based on at least two outpatient claims or one inpatient claim. 76,120 However, we alte red our algorithm to a less restrictive inclusion of diagnoses based on only one outpatient claim or one inpatient claim. We decided on this less restrictive algorithm to increase the sensitivity of detecting any indication of mental illness es However, we conducted a sensitivity analysis requiring at least two outpatient claims or one inpatient claim Co ncomitant Medication U se Concomitant medication use was determined from pharmacy claims and defined as presen ce of any pharmacy claim of a respective drug class during baseline including : antidepressants, anticonvulsants, antipsychotics, anxiolytics (including sedatives and hypnotics), alpha agonists, lithium and opioid analgesics ( Appendix D ). We also incorporated a measure of the total number of psychotropic drug classes a patient was exposed to during baseline (none, one, two, three, four or more) as an indicator of mental disease severity Suicide Attempt & Suicidal Ideation As a result of suicidal safety concerns surrounding at omoxetine use, we constructed a variable capturing any previous suicide attempts and claims with diagnosis for suicidal ideation during the six month baseline period We identified suicide attempts with ICD9 CM codes for external cause of injury codes (E c odes) of deliberate self harm (E950.x E959.x) in any diagnostic field 121 Instances of s uicidal ideation were defi ned as ICD9 CM codes V62.84 and 300.9 at any diagnosis field

PAGE 58

58 Source Population In order to accommodate the cohort set up for each study part, we describe the general cohort inclusion and exclusion criteria first and the additional criteria pertinent to each specific part in the following sections ( Figure 3 3 ) In short, we included all subjects at least 5 years and not more than 18 years of age at the index date In each study part we included only subjects with a first diagnosis of a qualifying mental disorder or a qualifying ADHD tr eatment prescription (index date) preceded by at least six months continuous Medicaid eligibility to facilitate a new user design. T his look back period prior to the index date allowed complete ascertainment of prior medical history and exposure to any oth er prescription drugs. Atomoxetine and stimulants are not approved under the age of six years and the risk of suicidality under six years of age is negligable 7,12 However, we discovered a significant amount of newly treate d patients (87,778, 13.1% ) initiating treatment with atomoxetine or stimulants between five and six years of age. In contrast, only 44,597 patients (6.6%) initiated treatment between 0 and 5 years of age ( Table 3 1 ). The included 26 states had different upper age el igibility criteria for Medicaid. Therefore, we selected the 19 th birthday as upper age restriction reflecting the lowest common denominator of Medicaid eligibility limi t s across the included states We considered only the first eligibility period where patients met all inclusion criteri a for any analysis. Hence, patients were followed until the end of the ir Medicaid fee for service eligibility period, their 19 th birthday or death, whatever occurred first, unless additional censoring criteria were specified per study part. Diagnoses were defined by any inpatient or outpatient claim as International Classification of Diseases, Ninth Revision, clinical modification (ICD9 CM) codes. 121 A

PAGE 59

59 significant proport ion of patients initiating atomoxetine or stimulants did not have an ADHD diagnosis at or before the index date. Therefore, we empirically identified m ental health disorders commonly treated with atomoxetine and stimulant s to minimize confounding by indica tion (e.g. stimulants for obesity) Therefore, the mental health diagnosis related inclusion criteria were extended to the following diagnoses: adjustment reaction (309.xx), disturbance of conduct, not elsewhere classified (312.xx except 312.3x), other or mixed emotional disturbances of childhood or adolescence (313.8x) and unspecified emotional disturbance of childhood or adolescence (313.9) ( Appendix D ). 103 These expanded diagnosis related inclusion criteria allowed assignment of a mental health diagnosis that might results in atomoxetine or stimulant treatment fo r about 80% of all treatment initiators and more than 50% of patients initiating treatment prior to an ADHD diagnosis. 103 Part I: Change(s) in Atomoxetine and CNS Stimulant Utilization Study Population Inclusion c riteria S ubjects entered the cohort at the first diagnosis of mental diseases commonly treated with atomoxetine or stimulants (in dex date) ( Figure 3 2 ) If a prescription for atomoxetine or stimulants preceded the index diagnosis by less than one year, the prescription fill date became the index date. In either case, the index date had to be preceded by at least six months continuous Medicaid eligibility ( Appendix D ). We permitted cohort entry with mental diagnoses following the treatment initiation date because some health care practitioners may prescribe a medication empirically a s part of the diagnostic work up. 122 Hence, the diagnosis might be preceded by the prescription. However, for the prescription and diagnosis to be associated, we allowed a

PAGE 60

60 prescription to precede a diagnosis by a ma ximum of one year We determined one year as the maximum for a prescription preceding a diagnosis because this is a feasible time frame to consider pharmacotherapy as a part of the diagnostic work up and included the majority of subjects with prescriptions p receding the initial diagnosis. Exclusion criteria Subjects were excluded if they initiated atomoxetine and stimulants on the same day or follow up time after the index date did not exceed one day. Censoring c riteria In addition to the previously describ ed censoring criteria we censored subjects at the beginning of the 37 th follow up month The selection of a maximum follow up period reflects an attempt to balance optimal sample size with the duration of a qualifying mental health diagnosis. As patients without a recurring mental health diagnosis are included, extensive follow up might bias initiation rates when the disorder acuity decreases On the other hand, b ecause atomoxetine is often second line treatment, this selection is especially important to n ot incorrectly distort the amount of atomoxetine treatment initiations included in our analysis by selecting a too short follow up time cut off The mean months between index date and stimulant init iation w ere 3.8 (SD 8.4) and for a tomoxetine 20.5 (SD15.2 ) months. Consequently, we decided to confirm the maximum follow up period with a sensitivity analysis of 12, 24 and 36 month of follow up. Drug Exposure We defined treatment initiation as a new prescription filled for one drug treatment category without a prescription of the same treatment cat egory in the previous six month Prevalent drug use was defined accordingly, as a prescription filled with the

PAGE 61

61 presence of a dispensed prescription of the same drug category within the p receding six month We describe per calendar month the prevalent utilization, stratified by atomoxetine and stimulants (Research question 1 & 2). We standardized treatment initiations to a monthly treatment initiation rate to prevent skewed results by an altering denominator of patients at risk of initiation per calendar month. T o standardize atomoxetine and stimulant treatment initiation, we calculated for each drug category and calendar month the number of patients initiating treatment per 1,000 patients a t risk for initiating treatmen t (Equation (3 1 ) & (3 2 ) ) Atomoxetine initiation rate = (monthly) Atomoxetine treatment initiations per calendar month 1,000 p atient month at risk for atomoxetine initiation per calendar mo nth: 0 to 36 mont h after the cohort entry date and no atomoxetine exp osure in the previous six month (3 1 ) CNS stimulant initiation rate = (monthly) CNS stimulant treatment initiations per calendar month 1,00 0 p atient month at risk for stimulant initiation per calendar month: 0 to 36 month after the cohort entry date and no stimulant exposure in the previous six month (3 2 ) Data Analysis We utilized a time series design to ana lyze monthly changes of treatment initiation and describe prevalent use of atomoxetine and stimulant Seasonality a djustment To address pronounced seasonality trends in utilization due to school breaks we adjusted treatment initiation rates using seasonal autoregressive integrated moving averages (SARIMA ) This technique utilizes the association in the sequentially lagged

PAGE 62

62 relationships that usually exist in data collected periodically. 95 A seasonal ARIMA model is described in short below (Equation (3 3 )). S ARIMA (p,d,q) x (P,D,Q) s (3 3 ) Where: P = number of seasonal autoregressive (SAR) terms D = number of seasonal differences Q = number of seas onal moving average (SMA) terms Segmented r egression The resulting seasonality adjusted treatment initiation rates were analyzed with segmented linear regression to investigate presence of an abrupt change in the intercept and slope of atomoxetine tre atment initiation rate s after the addition of a boxed warning in September 2005 (E quation (3 4 )) (R esearch question 3). Segmented regression model: Yt = 1 + 1 T + 2 D + 2 P + e t (3 4 ) Whe re: Y t = Atomoxetine initiation per person at risk, per month at time t 1 = Intercept at t month0 1 = Regression coefficient of atomoxetine initia tion before September 2005 T = Time month 1 48 (Jan. 2003 Dec. 2006) D = Dummy variable for pre/ post boxe d warning time (0 until Sep 2005, 1 afterwards) 2 = Intercept at t month34 P = Time since boxed warning (0 until Sept 2005, continuous afterwards) 2 = Change of regression coefficient of atomoxetine initiation after Sept 2005 e t = random variation at time t not explained by the model

PAGE 63

63 Joinpoint r egression Finally, Joinpoint regression models were used for an exploratory analysis of other significant change points (joinpoints) in treatment initiation rates of a tomoxetine and stimulants (R esearch question 4 & 5). 96 Joinpoint regression models have been previously used to describe utilization trends of antidepressants in association with public safety concerns of an increased suicidal risk. 12 3,124 Joinpoint regressi on evaluates data over time by determining continuous linear sections, their slope and linking points of signifi cant trend change, also called joinpoints (E quation ( 3 5 )) The sequential testing of a Join point regression model begins with the null hypothesis of no joinpoint (H 0 : k 0 =0) and the alternative hypothesis of four joinpoints (H A : k A = 4). Testing continues with increasing numbers of joinpoints as the null hypothesis is rejected and decreasing numb ers of joinpoints as the alternative hypothesis if the null hypothesis is accepted. As a result, the Joinpoint regression determines the best fitting model, with the least number of joinpoints, and provides estimates of the regression coefficient for the slope/trend for each continuous linear section. Joinpoint s oftware, version 3.5.1 was used for data analysis. 96 Joinpoint regression model: Y i i0 i1 i1 (X i k 1 ) + k (X i k i ) + ( 3 5 ) Where: Y = monthly atomoxetine initiation per person at risk X = study month (e.g., 01 /200 3 = 1, 02 /200 3 = 2,..., 12/2006 = 48 ) i = individual i k = unknown nu mber of joinpoints (change points) 1 = Regression coefficients prior to first joinpoint

PAGE 64

64 Part II : Determinants of Treatment Choice Study Population The cohorts for par t II were assembled from subjects newly treated wit h atomoxetine or stimulants and a diagnosis of mental dis orders commonly treated wi th ADHD pharmacotherapy in the preceding six month ( Figure 3 2 ) Inclusion c riteria Subjects ent ered the cohort at the first dispensed prescription (index date) fo r atomoxetine or stimulants The index date had to be pre ceded by a minimum of six month continuous Medicaid eligibility (baseline) with at least one diagnosis of a mental health disorder c ommonly treated with atomoxetine and stimulants ( Appendix D ). Initial and subsequent treatment choice Two sub cohorts were extracted from the study cohort to separately analyze determinants of initial treatment choice and determin ants of subsequent treatment choice. The first sub cohort included treatment nave subjects initiating treatment for the first time with either atomoxetine or stimulant between 2003 and 2006 (initial treatment cohort) We included only subjects initiating treatment between 2003 and 2006 because atomoxe tine was approved late in 2002. S timulants are the principal ADHD treatment and a tomoxetine is often considered second line therapy, therefore we decided to separately assess instances where atomoxetine was in itiated after initial treatment with stimulants. In order to capture determinants of treatment choice at a later point of subsequent atomoxetine initiation we assembled a second cohort of subsequent atomoxetine initiators matched to subjects still treated with stimulants (subsequent treatment cohort) We matched subsequent atomoxetine initiators by the month since

PAGE 65

65 stimulant initiation with patients exposed to stimulants in the same month since treatment initiation in a one to three ratio. Thus, the index da te for the second cohort is the fill date of the first atomoxetine prescription and the last stimulant prescription filled in the matched month. All baseline covariates were measured i n the preceding six month Although subjects initiating stimulants betwe en 1999 and 2006 were included, they only contributed to subsequent atomoxetine initiators and respective matches after the market introduction of atomoxetine starting in January 2003. We assessed multiple matching strategies, including matching individual patients multiple times at different matching months as well as different matching ratios. Matching patients at multiple times at different months allowed matching ratios of 1:5 and more, but resu lted in loss of unique patients and potential for overlap o f the six month baseline periods. Use of unique matches resulted in 5 598 (8.7%) and 12 401 (19.2%) unmatched subsequent atomoxetine initiators with matching ratios of 1:4 and 1:5, respectively. A ratio of 1:3 resulted in only 72 (0.1%) unmatched atomoxeti ne initiators, represen ting the optimal matching ratio preventing overlap of baseline periods and including the majority of atomoxetine initiators. M atching of subsequent atomoxetine initiators and current stimulant users by month since initial treatment initiation was conducted using R software, version 2.15.1, R Foundation for Statistical Computing, Vienna, Austria 125 Subsequent CNS Stimulant I nitiatio n Origina lly we proposed to also analyze baseline sociodemographic and clinical characteristics associated with subsequent initiation of stimulants after initial atomoxetine treatment compared to those who stay on the original atomoxetine therapy. However, only 24,344 patients in our cohort switched from initial atomoxetine to

PAGE 66

66 subsequent stimulant treatment during our study period. In accordance with this small study sample, the extent of measured covariates, resulting small cells and limited significance t o the principal safety study, we concluded an analysis of this researc h question not feasible (Research question 2). Florida Provider S pecialty For each research question in part two we proposed to analyze the association of provider specialty with treatme nt utilization for a Florida sub sample. Personal communication with CMS and ResDAC suggested incomplete and unreliable provider specialty information in the MAX data with strong variations among different states and study years. However for Florida, the F lorida Agency for Health Care Administration (AHCA) provided us with provider specialty codes during the study period. 126 We determined provider specialty for Florida beneficiarie s from the last mental health care claim before treatment initiation. Provider specialty was categorized as: psychiatry, primary care (including family practice & general practice), pediatrics or other (including unspecified) (R esearch question 1a, 2a & 3a ). However, in the treatment initiator cohort, 35,029 (11.9%) patients were from Florida, but only 17,446 (5.8%) had complete provider specialty information. Therefore, with only half of the patients in Florida Medicaid with valid provider specialty inform ation, we concluded the inclusion of provider specialty even only for Florida as not feasible. Data A nalysis We described the balance of baseline covariates of the two treatment groups with the standardized difference test [SDT] which is a measure of grou p balance independent of sample size. 124 We used a SAS macro developed by Yang and Dalton to calculate the standardized difference. 125 We defined significant imbalance of means

PAGE 67

67 or proportions between the two treatment groups if the absolute value of the st andardized difference was greater than 0.20. 127,128 W e fitted two multivariate logistic regression models to analyze the propensity of treatment ini tiation of atomoxetine compared to stimulants asso ciated with baseline (six month prior to treatment initiation) sociodemographic and clinical characteristics (E quation ( 3 6 ) ) The dependent variables for the two m odels were : initial atomoxetine treatment versus sti mulant initiation (M odel 1) (R esearch question 1) and subsequent initiation of atomoxetine versus current stimulant treatment (M odel 2) (Research question 3 ) ( Figure 3 2 ) In ea c h model, stimulant initiation or c urrent stimulant pharmacotherapy was the reference group. Logistic Regression Model: 1 X 1 2 X 2 3 X 3 4 X 4 5 X 5 6 X 6 7 X 7 8 X 8 3 9 X 3 9 ( 3 6 ) Where: Y: Atomoxetine initiation (Model 1), s ubsequent atomoxetine initiation (M odel 2) compa red to stimulants treatment 1 3 9 : As sociation parameter for the specific variable (x 1 3 9 dummy variables, if categorical variable ) X 1 : Sex (Male/Female) X 2 : Race (White, Black, Hispanic, Other) X 3 : Age at initiation/ matching date X 4 : TANF eligibility during baseline X 5 : Foster care dur ing baseline X 6 : SSI eligibility during baseline X 7 : Calendar year (200 3 200 6 ) at initiation/ matching date X 8 : ADHD with h yperactivity during baseline X 9 : ADHD with out h yperactivity during baseline X 10 : Adjustment reaction during baseline X 11 : Conduc t d isorder during baseline X 12 : Mixed emotional disorder during baseline X 1 3 : Other unspecified emotional disorder during baseline

PAGE 68

68 X 1 4 : Obesity during baseline X 1 5 : Substance use disorder during baseline X 1 6 : Smoking during baseline X 1 7 : Anxiety diagn osis during baseline X 1 8 : Bipolar disorder diagnosis during baseline X 1 9 : Schizophrenia diagnosis during baseline X 20 : Depression diagnosis during baseline X 21 : Autism diagnosis during baseline X 22 : Mental retardation diagnosis during baseline X 23 : Tic dis order diagnosis during baseline X 2 4 : Oppositional defiant disorder diagnosis during baseline X 2 5 : Psychosis diagnosis during baseline X 2 6 : X 2 7 : Number of mental co morbidities during baseline X 2 8 : Antidep ressants exposure during baseline X 2 9 : Antipsychotics exposure during baseline X 30 : Anticonvulsants exposure during baseline X 31 : Anxiolytics (including sedatives and hypnotics) exposure during baseline X 32 : Lithium exposure during baseline X 3 3 : Analgesics (opiates) during baseline X 34 : Alpha agonist exposure during baseline X 3 5 : Number psychotropic medication categories during baseline X 3 6 : Number of non mental health hospitalization during baseline X 3 7 : Number of psychiatric hospitalizations during baseli ne X 3 8 : Suicide a ttempt during baseline X 3 9 : Suicidal ideation during baseline Part III: Risk of Suicide and Suicide Attem pt Associated with Atomoxetine C ompared to CNS Stimulant Treatment Study Population Similar to part two the cohorts for part three w ere assembled from subjects newly treated wit h atomoxetine or stimulants and a diagnosis of mental dis orders commonly treated wi th ADHD pharmacotherapy in the preceding six months ( Figure 3 2 Figure 3 3 ) However we applied more stringent exclusion criteria to the afore mentioned cohorts.

PAGE 69

69 Inclusion c riteria Subjects entered the cohort at the first dispensed prescription (index date) for atomoxetine or stimulants The index date had to be preceded by a mi nimum of six months continuous Medicaid eligibility (baseline) with at least one diagnosis of a mental health disorder commonly treated with atomoxetine and stimulants ( Appendix D ). Exclusion criteria Subjects were excluded from o ur cohort if they ever had a drug claim for pemoline or methamphetamine during the study period because of small numbers and irrelevance to our analysis. Also subjects with any drug claim for Monoamine oxidase Inhibitors (MAO I) were exclude d because of se vere drug drug interactions with atomoxetine and stimulants ( Appendix C ) S evere or terminal diseases alter the baseline risk for suicidality outcomes and are generally considered rare in the study population. 74 Therefore, p atients with any inpatient or outpatient diagnosis for severe or terminal diseases during the study period were excluded. Excluded severe or terminal, non mental diseases comprised any diagnosis of HIV/AIDS (ICD9 CM 042.xx), malign ant neoplasm (140.xx to 208.xx; 209.0x to 209.3 x ; 230.xx to 234.xx), any organ or tissue replaced by transplant (V42.xx, V43.2x, V43.3x, V43.4x, V49.83) or dialysis dependency (V56.0 x ) during the studied eligibility period ( Appendix D ) Furthermore we excluded patients with any inpatient or outpatient diagnosis for pervasive developmental disorders (299.x) or severe (318.1) or profound (318.2) mental retardation. Pervasive developmental disorders include for example autism and Asp generally considered rare in our cohort and associated with a high and difficult to assess suicidality risk. Although it is unclear

PAGE 70

70 whether this risk is caused by these dis orders per se or with associated psychotic symptoms, previous research cautions to estimate suicidality risk based on age or psychiatric co morbidities in affected patients 129 Severe mental retardation characterizes patients with an intelligence quotient (IQ) of 20 34 and profound mental retardation characterizes patients with an IQ under 20. Previous research showed higher rates of suicidality risk factors but lower suicide rates in patients with i ntellectual disability than the general population. 130 Censoring criteria Patients were censored from our study cohort in addition to censoring criteria in the previous study parts ( loss of Medicaid eligibility, 19 th birthday, non suicide death) if any of the following criteria were met: a hospitalization or institutionalization of more than 30 days or pregnancy (V22.x), whatever occurred first If any of such events occurred during t he six month baseline period patients were excluded. Hospitalization longer than 30 days obstruct ed us of measuring actual drug exposure during the hospitalization. A pregnancy can alter suicide risk as well as drug exposure, resulting in avoidable confou nding. Initial and subsequent treatment Two sub cohorts were extracted from the study cohort to separately analyze initial and subsequent treatment (as in part two ). The first sub cohort included subjects initiating treatment for the first time with either atomoxetine or stimulant between 2003 and 2006, prec e ded by a minimum of six months continuous Medicaid eligibility ( initial treatment cohort) For the second cohort, we matched subsequent atomoxetine initiators by the month since stimulant initiation wit h patients exposed to stimulants in the same month

PAGE 71

71 since treatment init iation in a one to three ratio (subsequent treatment cohort) We included only subjects initiating treatment between 2003 and 2006 because atomoxetine was approved late in 2002 We eval uated suicidal risk in the two sub cohorts separately as a result of significant differences in the determinants of initial and subsequent treatment in part two. Time dependent drug exposure For stimulants and atomoxetine w e e stimated the end date for each pharmacy plus a grace period of 25% to incorporate residual s upply as a result of drug holidays S timulants are classifie d as s chedule II controlled substances and are typically restricted to a dispensing ma ximum of 30 131 In our cohorts, approximately 8 5 % of atomoxetine and stimulant claims were for a 3 0 day supply, resulting in 3 8 day exposure p eriods for any of these claims. Periods of current atomoxetine or stimulant exposure were defined by presence of an active prescription (time between dispensing date and end date) for the respective t reatment If active prescriptions for atomoxetine and stimulant s were present, d rug exposure was defined as current atomoxetine exposure and flagged as dual therapy. Periods following current use (without an active prescription for either treatment) were defined as former use thereafter A new prescription dispensing was set to overri de previous exposure assignments without adjustments for potential ove rlap All other drug classes measured as time dependent covariates had more than 90% of prescriptions with days of supply of 31 day s or less and no grace period or former use was impleme nted. We measured time dependent exposure to antidepressants, anticonvulsants, antipsychotics, anxiolytics, alph a agonist and opioid analgesics. However, we did not include time dependent exposure antidepressants and

PAGE 72

72 anticonvulsants in the final models de spite their reported suicidal risk, because their use might be in the direct outcomes pathway. All other time dependent covariates were included in the final risk models if they changed the point estimate of atomoxetine risk by more than 10% as recommended for model building in epidemiologic studies. 132 Days supply was reported for the majority of prescriptions dispensed, except in Ohio (less than 2% of all prescriptions) dispensed amount by t he median daily dose ascertained from all other claims for the or larger than 365 days were imputed in the same fashion. Sensitivity A nalysis of Time dependence of Co mo rbidities We analyzed and considered additional time dependent covariates involving onset of mental comorbidities for inclusion in our statistical models. For each month (1 4 8 ) after treatment initiation we plotted the percent of patients with the respec tive diagnosis per treatment group and month For both sub cohorts, we concluded all analyzed co morbidities during baseline had similar and parallel distributions of diagnoses among treatment groups over time and therefore we did not include these covaria tes as time dependent covariates in the Cox proportional hazard regression models (Initial user cohort: Figure 3 4 Figure 3 5 and Figure 3 6 ; Subsequent user cohort: Figure 3 7 Figure 3 8 and Figure 3 9 ). Study E ndpoints The primary study endpoint was a composite endpoint of t wo majo r categories: completed suicide and su icide attempt requiring hospitalization or emergency department (ED) visits.

PAGE 73

73 Completed suicides were defined as deaths with an International Classification of Diseases, Tenth Revision ( ICD 10 ), code of X60 X84 as classified on the death certificate ( Appendix D ). The range of codes specifies the method of suicides, such as: X60 X69 for intentional self poisoning, X70 (self inflicted suffocation), X71 (self inflicted drowning), X72 X 74 (self inflicted by firearm), X 75 (self i nflicted by explosive), X76 (self inflicted by smoke, fire or flame), X77 (self inflicted by hot vapor or object), X78 (self inflicted by shar p object), X79 (self inflicted by blunt object), X80 X 81 (self inflicted by jumping of high place or in front o f moving object), X82 (self inflicted by crashing motor vehicle), X83 (self inflicted by other specified means) and X84 (self inflicted by other unspecified means). 111 Suicide attempts leading to emergency department (ED) visits or hospitalizations were identi fied with ICD9 CM codes for external cause of injury codes (E codes) of deliberate self harm (E950.x E959.x) in any diagnostic field ICD 9 CM E codes for suicide attempts include E950.x E952.x for intentional self poisoning, E953.x (self inflicted injury by hanging), E954.x (self inflicted injury by drowning), E955.x (self inflicted injury by firearms), E956.x (self inflicted injury by cutting), E957.x (self inflicted injury by jumping from high places), E958.x (other/unspecified self inflicted injury), a nd E959.x (late effects of self inflicted injury). 121 Data A nalysis Propensity s core The propensity score is a common method to control for confounding in observational research. 133,134 The advantage of the propensity score is its ability to summarize numerous covariates as a single composite score, which is the propensity of being assigned to a particular treatment (compared to an other or non treatment)

PAGE 74

74 conditional on the observed covariates 135 Previous research has illustrated the benefit of less biased results using propensity score adj usted estimates especially when the number of observed covariates was large and the number of observed outcomes was small 136,137 For each sub cohort, w e calculated exposure propensity scores utilizing a logistic r egression model to estimate the propensity to receive atomoxetine conditional on baseline covariates compared to stimulant treatment We defined all baseline covariates based on the six month s period prior to tre atment initiation or time of match (index da te) (E quation ( 3 7 ) ) 138 Calculated propensity scores showed good overlap between treatment groups resulting in no need to trim our cohorts ( Figure 3 10 and Figure 3 11 ) Subjects were weighted by the inverse of their exposure propensity score to evaluate the performance of the propensity score in balancing all baseline covariate distributions across treatmen t group s. Propensity Score: Log [p/(1 1 X 1 2 X 2 3 X 3 4 X 4 5 X 5 6 X 6 7 X 7 8 X 8 3 8 X 3 8 ( 3 7 ) Where: Y: Atomoxetine initiation (Model 1), subsequent atomoxetine initiation (M odel 2) compa red to receiving stimulants 1 3 8 : Ass ociation parameter for the specific variable (x 1 3 9 dummy variable s, if categorical variable) X 1 : Sex (Male/Female) X 2 : Race (White, Black, Hispanic, Other) X 3 : Age at initiation/ matching date X 4 : TANF eligibility during baseline X 5 : Foster care duri ng baseline X 6 : SSI eligibility during baseline

PAGE 75

75 X 7 : Calendar year (200 3 200 6 ) at initiation/ matching date X 8 : ADHD with h yperactivity during baseline X 9 : ADHD with out h yperactivity during baseline X 10 : Adjustment reaction during baseline X 11 : Conduct d isorder during baseline X 12 : Mixed emotional disorder during baseline X 1 3 : Other unspecified emotional disorder during baseline X 1 4 : Obesity during baseline X 1 5 : Substance use disorder diagnosis during baseline X 1 6 : Smoking during baseline X 1 7 : Anxi ety diagnosis during baseline X 1 8 : Bipolar disorder diagnosis during baseline X 1 9 : Schizophrenia diagnosis during baseline X 20 : Depression diagnosis during baseline X 2 1 : Mild m ental retardation diagnosis during baseline X 2 2 : Tic disorder diagnosis during b aseline X 2 3 : Oppositional defiant disorder diagnosis during baseline X 2 4 : Psychosis diagnosis during baseline X 2 5 : X 2 6 : Number of mental co morbidities during baseline X 2 7 : Antidepressants exposure during baseline X 2 8 : Antipsychotics exposure during baseline X 29 : Anticonvulsants exposure during baseline X 3 0 : Anxiolytics (including sedatives and hypnotics) exposure during baseline X 3 1 : Lithium exposure during baseline X 3 2 : Analgesics (opiates) during baseli ne X 3 3 : Alpha agonist exposure during baseline X 3 4 : Number psychotropic medication categories during baseline X 3 5 : Number of non mental health hospitalization during baseline X 3 6 : Number of psychiatric hospitalizations during baseline X 3 7 : Suicide a ttempt during baseline X 3 8 : Suicidal ideation during baseline Cox proportional hazard r egression We used discr et e time Cox proportional hazard regression adjusted for exposure propensity score time dependent age and dual therapy t o estimate the risk of suicide and suicide attempt in pediatric patients initiating treatment with atomoxetine compared to stimulants 139 The analys e s used a new user design to detect study endpoints occurring shortly after treatment initiation and time dependent drug exposure definit ions We fitted separate Cox proportional regression models for the two sub cohorts

PAGE 76

76 comparing initial and su bsequent atomoxetine versus stimulant treatment (Research question 1 & 2 ) ( Figure 3 2 ) (Equation ( 3 8 ) ). For computational efficiency, we segmented follow up time after the index date into 15 day increments and assigned exposure status based on the majority of days (eight days or more) attributed to current and former exposure of atomoxetine and stimulants within each 15 day increment For all other time dependent covariates we constituted the same 15 da y increments and majority rule. We used Cox proportional hazard models, because they are more robust than accelerated failure time methods, have excellent capabilities to analyze ti me dependent covariates, handle discrete time dat a, allow for temporary exit from the risk set and are a commonly used method for regression analysis of survival data. 140 Cox model with time dependent covariates: ( 3 8 ) Where: = I empt) = B aseline hazard function at time t 1 : Association parameter for the specific time independent covariate 2 1 0 : Association parameter for the specific time dependent covariate x i 2 1 0 = I time independent covar iate x i2 1 0 (t) = I time dependent covariate at time t Static covariates : X 1 : Exposure propensity s core Possible t ime dependent explanatory covariates : X 2 : Atomoxetine or stimulant current or former exposure X 3 : Dual therapy with atomoxetine and stimulants X 4 : Age category (time dependent) X 5 10 : Time dependent exposure to: a ntidepressants a ntipsychotics a nticonvulsants a nxiolytics (including sedatives and hypnotics) a nalgesics (opiates) a lpha agonist ( all not included)

PAGE 77

77 General Table s and Figures Table 3 1 Patients newly initiating treatment under five years of age Age at Initiation N 0 1 years 255 1 2 years 705 2 3 years 3494 3 4 years 18,734 4 5 years 21,371 Figure 3 2 Overview study design

PAGE 78

78 Figure 3 3 Overview cohort selection

PAGE 79

79 Part II: Tables and Figures Table 3 2 Part II: Inclusion and exclusion criteria with associated cohort size Inclusion/Exc lusion criteria Cohort Size Patients with any claim of atomoxetine or CNS stimulants 1,866,215 Patients with i ndex date after 5 th and before 19 th birthday 1,583,944 Patients with at least six month continuo us Medicaid eligibility before index d ate 68 0,035 Patients with end, index or switch d ate between 2003 and 2006 604,455 Patients with at least one common mental health diagnosis preceding treatment Initiation 437,460 Patients initiating treatment between 2003 and 2006 299,051 Patients newly i nitiating atomoxetine 60,542 Patients switching treatment eligible between 2003 and 2006 and not switching to CNS stimulants 312,428 Patients switching to atomoxetine 64,486

PAGE 80

80 Part III: Tables and Figures Figure 3 4 Part III: Distribution of cohort eligibility, drug exposure, substance use, tic and oppositional defiant disorder by treatment group over time; initial treatment cohort

PAGE 81

81 Figure 3 5 Part III: Distribution of mental disorder d iagnoses commonly treated with atomoxetine or CNS stimulant s by treatment group over time; initial treatment cohort

PAGE 82

82 Figure 3 6 Part III: Distribution of other disorder diagnoses, by treatment group over time; initial trea tment

PAGE 83

83 Figure 3 7 Part III: Distribution of cohort eligibility, drug exposure, substance use, tic and oppositional defiant disorder by treatment group over time; subsequent treatment

PAGE 84

84 Figure 3 8 Part III: Distribution of mental disorder diagnoses commonly treated with atomoxetine or CNS stimulant s by treatment group over time; subsequent treatment cohort

PAGE 85

85 Figure 3 9 Part III: Distribution of other disorder diagno ses, by treatment group over time; subsequent treatment cohort

PAGE 86

86 Figure 3 10 Exposure propensity score distribution of initial user cohort

PAGE 87

87 Figure 3 11 Exposure propensity score of subsequent user cohort

PAGE 88

88 CHAPTER 4 RESULTS Part I: Change(s) in Atomoxetine and CNS Stimulant Utilization S ensitivity Analysis of Maximum F ollow up With a maximum follow up of 12 month the resulting cohort included o nly 67,752 atomoxetine prescription claims, whereas 24 month and 36 month of maximum follow up each roughly double the included atomoxetine prescriptions to 166,571 and 257,599 prescriptions respectively. Atomoxetine t reatment initiations mirror this incr easing trend with 19,267 34,266 and 45,040 treatment initiations with 12, 24 and 36 month of maximum follow up respectively O verall stimulant prescriptions also strongly increase with longer follow up, 1,756,944 to 3,611,850, the proportion of treatment initiations d id not increase so significantly, 323,460 to 3 92,972 (12 and 36 month). F o llow up time without a maximum follow up time was a mean 2.39 ( standard deviation [ SD ]1 .69) years, with a maximum of 36 months 1.92 ( SD 0.99) years and 1.49 ( SD 0.64) years for a maximum of 24 month follow up. Hence, we concluded a maximum of 36 mo nth of follow up as the best balance of optimal sample size and duration of a qua lifying mental health diagnosis ( Table 4 1 ). Baseline Covariates Soc iod emographic s Our dataset included 4,615,177 subjects with at least one common mental health diagnosis associat ed with stimulant and atomoxetine treatment. The previously described inclusion criteria reduced the eligible cohort to 1,097,208 subjects ( Table 4 2 ). The final study cohort consisted of 1,013,556 children and adolescents who contributed 13.54 million eligible patient months, with a relative constant mean of

PAGE 89

89 282,153 eligible patient months per calendar month (SD 11,424) between 2003 and 2006 Overall, subjects were predominantly m ale and Caucasian. The study included 13.1 million patient months at risk of atomoxetine initiation and 9.74 million patient months at risk of stimulant initiation ( Figure 4 1 ). 343,912 (33.9%) patients had at least one claim for atomoxetine or stimulants during follow up Ever atomoxetine or stimulant exposed patients had a younger mean age at index date (8.4 vs. 11.4 years), longer mean follow up (1.92 vs. 1.43 years) and were more often male (69. 1 %). Of the exposed patients, 47,020 (13.7%) had a prescription preceding the index diagnosis by a mean of 108.9 days (S D98.9). Reasons for Medicaid eligibility, such as Foster care (9.1%), receipt of temporary assistance for needy families (TANF 14.0%) and disability ( SSI 1.9%) were relatively evenly distributed in the exposed and non exposed groups ( Table 4 3 ). Treatment initiators and index d iagnosis Atomoxetine and stimulant treatment w as initiated in 45,040 (11.5%) and in 392,972 (88.5%) instances respectively If a patient initiated treatment, he or she filled a mean of 7. 8 (SD9.3) prescriptions for atomoxetine and 13 .0 (SD15.6) prescriptions for stimulants during follow up The cohort consiste d of patients diagnosed with ADHD (38.0%), adjustment reaction (36.1%), disturbance of conduct (14.7%) or other, mixed or unspecified emotional disturbance (12.0%). Treatment e xposed patients had predominantly an index diagnosis of ADHD (75.2%) and less ad justment reaction (10.9%) as index diagnosis ( Table 4 4 ). C o morb idities The most common comorbidi ties in the study cohort were o ppositional defiant disorder [ODD ] (10.3%), d epression (7.6% ) and anxiety (6.7%). All common co

PAGE 90

90 morbid ities were less prevalent in t he ever exposed group, such as o ppositional defia nt disorder (5.2% vs. 12.9%), depression (4.3% vs. 9.3%) and a nxiety (3.0% vs. 8.5%). In contrast, ADHD with hyperactivity (47.5 % vs. 13.6 %) and without h yperactivity (17.5 % vs. 5.5%) w ere more prevalent in the exposed group ( Table 4 5 ). Treatment Initiation and Prevalent Use Atomoxetine treatment initiations increased in the months following FDA approval to maximum s of 2,216 patients in March and August 2003. Subsequently, a slow, but steady decline followed to a low of 286 treatment initiations in December 2006. The number of prevalent users follows this trend with a steady increase following FDA approval early in 2003 to a maximum in March 2004 of 6,7 12 patients filling at least one prescription. The number of prevalent users declined thereafter to a minimum of 1,676 patients filling at least one prescription in December 2006 ( Figure 4 2 ) In contrast, stimulant treatment init iations and the number of prevalent users remained constant over the four study years with a mean of 4,946 (SD1,246) initiations and 44,453 (SD5,221) patients filling at least one prescription per calendar month (Research question 1 & 2) ( Figure 4 3 ) Treatment Initiation s Rates The a tomoxetine treatment initiation rate increased after FDA approval and reached a maximum of 8.52 per 1,000 patient months in August of 2003. Subsequently, the rate declined steadily to 1.18 until Decem ber 2006 ( Figure 4 4 ). In contrast, stimulant treatment was initiated rather consistently across the study period at an average rate of 24.35 ( SD 5.96) pa tients per 1,000 patient months After adjusting for seasonality, mean treat ment initiations rate s were almost identical with a

PAGE 91

91 mean of 24.37. However the variation was significantly lower with a standard deviation of 1.83 ( Figure 4 5 ) Segmented Regression W e fitted a segmented regression model to analy ze the atomoxetine initiation rate from January 2003 to the implementation of the boxed warning in September 2005 (month 1 33) and then from October 2005 to December 2006 (month 34 38). For the time prior to the boxed warning, the intercept was 7.25 ( 95% CI 6.72 7.78 p<0.001) and the monthly trend of the treatment initiation rate (slope) was 0.17 (95% CI 0.20 0.14 p<0.001). The immediate change in the treatment initiation rate (intercept) after the boxed warning implementation in September 200 5 was statistically insignificant, with a magnitude of only 0.06 (95% CI 0.85 0.74 p=0.88). The slope of the monthly treatment initiation rate changed significantly with a level off of 0.15 treatment initiations (95% CI 0.07 0.23 p=0.0003) (R esearch q u estion 3) ( Figure 4 6 ) Joinpoint Regression Joinpoint regression determined two significant joinpoints at month 3 (March 2003, 95% CI month 3 to 4) and month 39 (March 2006, 95% CI month 34 to 44) as the best fitting model. Est imated slopes for atomoxetine initiations per 1,000 patient months were a steep increase of 0.34 (p=0.0022) before the first j oinpoint in March 2003, a decrease of 0.39 (p<0.001) between March 2003 and 2006, and a very slight increase of 0.0049 (p<0.001) thereafter until December 2006 The lower confidence interval for the second j oinpoint was adjacent to the boxed warning implementation supporting the results of the segmented regression with a significantly level off around the time of the boxed warning (Research question 4) ( Figure 4 7 ).

PAGE 92

92 The sequential testing of the Joinpoint regression model for stimulant initiations per 1,000 patient months selected one j oinpoint at month 19 (July 2004, 95% CI month 14 to 24) as the best fit ting mode l. Estimated slopes before the j oinpoint were a slight increase in the treatment initiation rate of 0.00652 (p=0.006) per month a nd a slight decrease after the j oinpoint of 0.014 (p<0.001). These minimal slopes and the visual impression of the gr aph confirm ed the consistency of stimulant initi ations during the study period (Research question 5) ( Figure 4 8 ) Part II: Determinants of Treatment Choice Initial Treatment Choice The study cohort consisted of 299,051 treatment nave children and adolescents who initiated treatment of either atomoxetine or stimulants between 2003 and 2006. Atomoxetine was initiated by 60,542 (20.2%) and stimulant treatment by 238,5 09 (79.8%) ( Table 4 6 ). Compared to sti mulant initiators, patients initiating atomoxetine were about one year older with a higher propensity of being Caucasian, initiating treatment in 2003 or 2004, diagnosed with all measured mental co morbidities, except ADHD with hyperactivity, unspecified e motional dist urbance or mental retardation Also, patients initiating atomoxetine had a higher propensity of expos ure to all measured concomitant medications, except alpha agonists. However, only age, race/ethnicity, gender and index calendar year were sig nificant ly imbalance d among treatment groups with an ab solute standardized difference test [S D T ] greater than 0.20 ( Table 4 6 Table 4 7 )

PAGE 93

93 S ociodemographic s Patients treated with atomoxetine had a higher mean age at the index date ( atomoxetine 9.9 vs. stimulant 9.0 years standard deviation [SD] 3.4 vs. 3.1 ) a slightly longer mean follow up time ( 1.77 vs. 1.52 years SD 1.1 vs. 1.1 ) and accordingly a higher mean age at the end of follow up (11.65 years vs. 10.52 years, SD 3.5 vs. 3.3) Males (66.3 % vs. 68.3% ) and Caucasians (71.3% vs 58.2%) were overrepresented in both groups. Caucasians were further overrepresented in th e atomoxetine initiators, constituting almost three quarters of the atomoxetine tr eated group ( SDT > 0.20) Treatment initiation of stimulants was evenly distributed among the study years (25.3%, 26.3%, 25.1%, and 23.3 %). In contrast, the atomoxetine treatment initiations were constant in 2003 and 2004 (35.8%, 34.5%), but in 2005 and 200 6 initiations dropped to 19.2% and 10.6% respectively ( SDT >0.2) ( Table 4 6 ) Calendar month and state of residence were not associated with significant imbalances (data not shown) Mental d isorders & other c o morbidities The coho rt consisted primarily of patients diagnosed with ADHD ( 88.9 % atomoxetine vs. 91.4% stimulant), adjustment reaction (17. 1% vs. 14.8% ), disturbance of conduct (12.6 % vs. 12.5% ) or other, mixed and unspe cified emotional disturbance (13.1 % vs. 11.8% ). P atient s initiating atomoxetine compared to stimulants, had less frequent diagnoses of ADHD with hyperactivity ( 61.4 % vs. 68.6% SDT =0.15 ) and more frequent ADHD without hyperactivity (27.4% vs. 22.8% SDT =0.11 ). The most common mental co morbidities in the study cohort were o ppositional defiant disorder ( 11.4 % vs. 9.9%), d epression (9 .6% vs. 7.0%) and a nxiety (7.2 % vs. 5.6% ) Hospitalizations were rare and equally distributed among treatment groups and reason for hospitalization ( Table 4 6 ).

PAGE 94

94 Drug e xposure Of the 238,509 patients initiating stimulant treatment, mixed amphetamine salts and methylphenidate w ere initiated by 98,728 (41.4%) and 139,950 (58.7%), respectively Most common overall concomitant drug exposure as well as more frequen t in the atomoxetine group were antidepressants (14.8% vs. 10.5% SDT =0.13 ), antipsychotics (9.9% vs. 7.8%), anticonvulsants (6.4% vs. 4.7%) opioid analgesics (6.6% vs. 5.6%) and anxiolytics (3.8% vs. 3.5%). A lpha agonist s were the only examined drug clas s more common in the stimulant group (4.8% vs. 5.5%) ( Table 4 7 ) Logistic r egression analysis of determinants of initial treatment choice All non Caucasian race s/ethnicities w ere associated with a decreased probability of atomoxe tine initiation (adjusted odds ratio [OR] = between 0.54 to 0. 70, 95% confidence interval [ CI ] = 0.53 to 0.74, all p <0.0001 ). Per calendar year, the propensity of atomoxetine initiation continuously decreased over time to an odds ratio of 0.3 3 (95% CI 0.3 2 0.3 4, p < 0.0001 ) for 200 6 compared to 2003. The propensity of atomoxetine initiation increased continuously with increasing age, from the reference 6 to 8 year olds up to 15 to 19 year olds ( OR = 1.18 to 1 .6 7 95%CI 1. 15 1.74 p <0.0001). Substance use diso rder (OR = 1.5 4 95% CI 1. 41 1.69 p< 0.0001 ), tic disorder (OR = 3. 31 95% CI 2.85 3.83 p< 0.0001 ) d isability (OR =1.35, 95% CI 1.28 1.43, p<0.0001) and schizophrenia (OR = 1. 30, 95% CI 1.07 1.58, p=0.01 ) were the other strong co variate s associated with a significan tly higher propensity of atomoxetine initiation All other co variates had either a small or non significant OR s (Research question 1) ( Table 4 8 ) The Gamma and C statistics for our model were 0.3 29 and 0.663, respectively. 141

PAGE 95

95 Subsequent Treatment Choice The subsequent treatment cohort consisted of 64,414 (25%) patients starting atomoxetine treatme nt after initial stimulants treatment and 193,242 (75%) matched patients still exposed to stimulants ( Table 4 9 ) In general, subsequently initiating atomoxetine had a higher propensity of being Caucasian, male, treated in 2003 or 2004, diagnosed with all measured mental co morbidities except ADHD with hyperactivity and mental retardation. Also, subsequently initiating atomoxetine had a higher propensity of exposure to all measured co medications ( Table 4 9 Table 4 10 ) S ocio d emographics Patients subsequently initiating atomoxetine had a slightly lower mean age at the time of first line initiation compared to those staying on stimulant treatment ( atomoxetine 8. 3 vs. stimulants 8.6 years SD 2.6 vs. 2.8 ), and a half year longer mean follow up time ( 3. 7 vs. 3.1 years SD 1. 8 vs. 1.9 ) Males ( 71.9 % vs 70.4 % ) and Caucasians ( 69.5 % vs. 59.1 %) were overrepresented in both groups, although Caucasians were further overr epresented in the subsequent atomoxetine initiators, constituting more than two thirds of the atomoxetine group ( SDT >0.2) Matched patients currently exposed to stimulant treatment w ere relative evenly distributed among the four study years (28.8%, 21.6%, 22.3%, 27.3 %). In contrast, almost half of the subsequent atomoxetine treatment initiations were in 2003 alone (46%), with a drop to a minimum in 2006 (9.8%) ( Table 4 9 ). Mental d isorders & other c o morbidities The cohort consist ed primarily of patients diagnosed with ADHD ( atomoxetine 89 % vs. stimulants 87.9%), where ADHD with hyperactivity was the most common

PAGE 96

96 ADHD diagnosis (67.4% vs. 67.7% of all ). O ther mental diagnoses commonly associated with subsequent atomoxetine or contin uous stimulant treatment were adjustment reaction (1 2 1% vs. 1 0 .8%), disturbance of conduct ( 8.8 % vs. 7.4 % ) and other, mixed and unspe cified emotional disturbance ( 12.2 % vs. 10.4 % ). Drug e xposure The majority of patients was exposed to any stimulant during the previous six month (84% vs. 91.9% SDT =0.24 ). Stratified by stimulant drug classes, the proportion of patient s exposed to mixed amphetamine salts (41.5% vs. 43 .0 %) and methylphenidate (51.9% vs. 55.5%) were similar. All co ncomitant medications were mo re frequent in the atomoxetine group most common were antidepressants ( 21.0% vs. 16.6 % SDT =0.11 ), followed by antipsychotics ( 16.6% vs. 14.2%), anti convulsants (9.3% vs. 7.8%), opioid analgesics (6.0 % vs. 5.6%) anxiolytics (4.2% vs. 3.8%) and alpha agon ists (14.2% vs. 12.4) ( Table 4 10 ). Logistic r egression analysis of determinants of subsequent treatment choice All non Caucasian race s/ethnicities w ere associated with a decreased probability of subsequent atomoxetine initiation (adjusted odds ratio [OR] = between 0. 6 4 to 0. 8 0 95% confidence interval [ CI ] = 0. 6 2 to 0. 8 4, all p <0. 01 ) compared to current stimulant treatment The propensity of atomoxetine initiation de creased continuously with increasing age, from the reference 6 to 8 year olds up to 15 to 19 year olds ( OR s = between 0.86 to 0.60 95% CI between 0. 58 to 0.88 all p <0.0001). Per calendar year, the propensity of subsequent atomoxetine initiation continuously decreased over time to an odds ratio of 0. 23 (95% CI 0. 22 0.24 p<0.0001 ) for 200 6 compared to 200 3 Substance use disorder (OR= 1.39 95% CI 1. 25 1.56, p<0.0001 ), tic disorder (OR= 2.69 95% CI 2.35 3.08, p<0.0001), prior suicidal ideation (OR= 1. 75, 95% CI 1.11 2.77,

PAGE 97

97 p=0.02 ) more than one distinct mental disorder diagn osis (ORs=1.14 to 1.36, 95% CI 1.10 to 1.53, all p<0.0001) and one or more prio r mental health hospitalization (ORs= 1.31 to 1.36, 95%CI 1.09 to 1.58, all P<0.0041) were the strongest covariate s associated with a significantly higher propensity of atomoxetin e use (Research question 3) ( Table 4 11 ). The Gamma and c statistic for this logistic regression model was 0.35 4 and 0.676. Comparison of Initial and Subsequent Treatment Cohorts I n both sub cohorts atomoxetine treated patients ha d a higher propensity to be Caucasian, being treated in 2003 or 2004 and diagnosed with all measured mental co morbidities except ADHD with hyperactivity. Also, patients initiating atomoxetine had a higher propensity of expos ure to all measured co medicati ons, except for alpha agonists in initial atomoxetine treated patients The age distribution was significant ly di fferent within the treatment groups Initial users of atomoxetine and stimulant s included a higher proportion of younger patients and subsequen t users a higher proportion of older patients. Accordingly, i ncreasing age was strongly associated with higher odds of initial and lower odds of subsequent atomoxetine treatment.

PAGE 98

98 Part III: Risk of Suicide and Suicide Attempts Initial Treatment Starting with an original cohort size of 298,5 46 patients, we excluded a total of 19,23 1 (6.4%) due to the following reasons: a hospitalization longer than 30 days during baseline (3.7%), a diagnosis of pervasive development disorder (2%), neoplasm (0.4%), a prescr iption for pemoline, methamphetamine or MAO inhibitors (0.14% ) pregnancy during baseline (0.08%), HIV/AIDS (0.06% ) organ transplant (0.05%) or severe mental retardation (0.03%) R easons for censoring were : end of Medicaid eligibility ( 94. 9 % ), hospitaliza tion more than 30 days (3.5%), patient th birthday ( 1.1% ) and pregnancy ( 0.5% ) Only 4% of patients were censored prior to end of Medicaid eligibility or their 19 th birthday. A total of 92 ( 0.0 3 % ) children and adolescents died d uring the ir eligibility period of causes other than the study endpoints. Overall, all exclusion and censoring criteria were distributed similarly among the two treatment groups. The final cohort included 297, 315 patients initiating treatment, consisting of 56, 012 (20.1%) and 223, 303 (79.9%) patients initiating atomoxetine and stimulant, respectively ( Table 4 12 ) The distribution of covariates was similar to part two but absolute numbers were slightly lower as a result of more stringent exclusion criteri a in part three Inverse weighting of subjects by their exposure propensity score established balanced groups with less than 0.5% absolute difference in the distribution of covariates among atomoxetine and stimulant users ( Table 4 13 Table 4 14 )

PAGE 99

99 Drug exposure Patients treated with stimulants contributed 190,026 person years (44.4%) of current stimulant exposure and 144,144 person years (33.7%) of former stimulant exposure. Patients initiating atomoxetine treatment contributed 46,929 person years (11.0%) of current atomoxetine exposure and 47,173 person years (11.0%) of former atomoxetine exposure. A total of 24,998 (47%) patients initiating atomoxetine filled at least one prescription for stimulants after a mean of 271 days (median 176). Whereas, 28,659 (13%) patients filled at least one atomoxetine prescription after initial stimulant treatment after a similar mean of 282 days (median 199). However, of the current atomoxetine exposure, only 6,348 person y ears (13.5%) were dual therapy of atomoxetine and stimulants ( Table 4 15 ). Study endpoints We observed 140 suicide events CMS guidelines on reporting small cells prevented us from reporting suicides separately. S u icide was most commonly attempted by self inflicted poisoning or cutting and piecing instruments Of the suicide attempts, the majority was in female s ( 60%). In contrast, the majority of suicides were in males (89%) The average age at a suicide event was 15 .5 years (SD2.7) and occurre d after a mean of 1.12 years (median 0.93 years ) after the index date. Risk of suicide and suicide attempt associated with atomoxetine compared to CNS stimulant treatment The final cohort included 279, 3 15 patients with a tota l of 428,272 person years of follow up. We observed 50 suicide events in current stimulant exposed patients (26.3/100,000 person years), 47 suicide events in former stimulant use (32.6/100,000 person years), 18 suicide events in the current atomoxetine exp osed patients

PAGE 100

100 (38.4/100,000 person years) and 25 suicide events in former atomoxetine use (53.0/100,000 person years). Restriction of follow up time to 3, 6, 12 and 24 month resulted in 23, 47, 70 and 111 suicidal events respectively ( Table 4 16 ). During the first year of follow up, t he unadjusted hazard ratio (HR ) for current atomoxetine use compared to current stimulant use was 1. 51 (95%CI 0. 77 2.95 p=0. 23 ). Besides the propensity score that summarized all baseline covariates, the only included time dependent covariates in the final model were age and dual therapy (current atomoxetine and stimulant exposure ). Compared to current stimulant use, the unadjusted H R for former atomoxetine use was 2.42 (95% CI 1.15 5.11 p= 0.02 ) and f or former stimulant use the H R was 1. 24 (95% CI 0. 67 2. 30 p=0. 50 ). The propensity score dual therapy and time dependent a ge adjusted HR for current atomoxetine use compared to current stimulant use was 0.95 (95% CI 0. 47 1.92 p= 0.88 ). Compared to current stimulant use, the adjusted H R for former atomoxetine use was 1.11 (95% CI 0. 5 2 2.37 p=0. 79 ) and for former stimulant use the H R was 0. 95 (95% CI 0. 51 1. 77 p= 0. 87 ) (Research question 1) ( Table 4 17 ) Subsequent Treatment Prior to matching we excluded all patients with any diagnosis of per vasive development disorder, neoplasm HIV/AIDS, organ transplant, dialysis dependency or a prescription for pemoline, methamphetamine or MAO inhibitors (data not shown) Starting with a matched c ohort size of 241,125 patients, we excluded a total of 20,910 (8.7%) due to the following reasons: hospitalization longer than 30 days ( 4.7 %), a diagnosis of severe mental retardation (0.0 2 %) or pregnancy (0.0 5 %) during baseline Furthermore, we excluded p atients if the exclusion criteria resulted in a matching cluster of less than one to two (3.9%).

PAGE 101

101 R easons for censoring were : end of Medicaid eligibility (9 4.2 %), hospitalization more than 30 days ( 3.9 %), patient th birthday ( 1.4 %) and pregnancy ( 0. 4 %). Only 4.4 % of patients were censored prior to end of Medicaid eligibility or their 19 th birthday. A total of 106 (0.0 5 %) children and adolescents died d uring the ir eligibility period of causes other than the study endpoints. Overall, all exclusion and cens oring criteria were distributed similarly among the two treatment groups. All exclusion and censoring criteria lead to a final cohort of 220,215 patients initially treated with stimulants and eligibility periods extending between 2003 and 2006 Of those, 5 6,948 (25.9%) subsequently initiated atomoxetine, matched to 163,267 patients (74.1%) exposed to stimulants ( Table 4 18 ) In general covariates were similarly distributed among the two treatment groups and c ompared to the cohort in part two with the exception of mean follow up after subsequent initiation/match resulting in significant SDT characterizing imbalances greater than 0.2 However, absolute numbers were slightly lower compared to part two a s a result of more stringent ex clusion criteria. Inverse weighting of subjects by their exposure propensity score established balanced groups with less than 0.5% absolute difference in the distribution of covariates among atomoxetine and stimulant users ( Table 4 19 Table 4 2 0 ) Drug exposure Patients contributed 142,015 person years (47.2%) of current and 78,054 person years (26.0%) of former stimulant exposure. 56,948 patients subsequently initiated atomoxetine resulting in 37,948 per son years (12.6%) of current and 43,344 person years (14.2%) former atomoxetine exposure 36,472 (64%) patients initiating subsequent atomoxetine filled at least one prescription for stimulants after a mean of

PAGE 102

102 138 days (median 43). Whereas, 24,031 (15%) pa tients filled at least one atomoxetine prescription in the continuously stimulant treated sub cohort. Those patients initiated atomoxetine after a mean of 282 days (median 199). In contrast to the init ial treatment cohort, the time until addition or substi tution of stimulants in atomoxetine users was much shorter 9,881 person years (26.0% of all atomoxetine exposure) were dual therapy with a combination of atomoxetine and stimulant exposure indicating more prevalent and quicker addition of stimulant treat ment for dual therapy ( Table 4 21 ) Study endpoints We observed 90 suicide events CMS guidelines on reporting small cells prevented us from reporting suicides separately. Suicide was most commonly attempted by self inflicted pois oning, cutting and piecing instruments Of the suicide attempts, the majority was female ( 60 %). In contrast, the majority of suicides were in males ( 72.7 %). The average age at a suicide event was 14.7 years (SD 2.4) and occurred after a mean of 0.98 years ( median 0.6 years ) after the index date. Risk of suicide and suicide attempt associated with subsequent atomoxetine initiation compared to CNS stimulant treatment The final cohort included 220,215 patients with a total of 30 0 772 person years of follow up We observed 46 events of suicide attempt and suicide in current stimulant exposed patients (32.4/100,000 person years), 17 events in former stimulant use (21.8/100,000), 11 events in the current atomoxetine exposed patients (29.0/100,000) and 16 cases in former atomoxetine use (37.4/100,000). Overall, we observed 90 suicidal events (11 suicides, 12.2%), resulting in a global event rate of 29.9 per 100,000 person years. A follow up time of 3, 6, 12 and 24 month resulted in 20, 29, 55 and 77 suicidal events respectively ( Tab le 4 22 )

PAGE 103

103 During the first year of follow up, t he unadjusted hazard ratio ( H R) for current atomoxetine use compared to current stimulant use was 0. 8 8 (95%CI 0. 40 to 1. 91 p =0. 74 ), 1.25 (9 5% CI 0. 51 to 3.05 p= 0. 63 ) and 0. 82 (95% CI 0. 40 to 1. 72 p=0. 61 ) for former atomoxetine and stimulant use, respectively Besides the propensity score that summarized all baseline covariates, the only included time dependent covariates in the final model were age and dual therapy (current atomoxetine and stimulant exposure ). Compared to current stimulant use, adjusted hazard ratio (H R) for current atomoxetine use was 0. 71 (95%CI 0. 30 to 1.67, p=0.4 3) 0. 66 (95% CI 0. 26 to 1. 65 p= 0. 37 ) and 0.5 4 (95% CI 0.2 6 to 1. 1 3 p= 0. 10 ) for former atomoxetine and stimulant exposure (Research question 2) ( Tabl e 4 23 ).

PAGE 104

104 Part I: Tables and Figures Table 4 1 Part I: Sensitivity analysis of 12, 24 and 36 month of maximu m follow up Included N Exposed N Overall claims New claims (%) Number of Prescriptions per p e rson (SD ) % of all claims Maximum of 36 month follow up 1,013,556 343,912 Atomoxetine 257,599 45,040 (17.5) 7.9 (9.4) 6.66% CNS stimulant 3,611,850 392,9 72 (10.9) 13.5 (16.4) 93.34% Maximum of 24 month follow up 1,013,556 331,010 Atomoxetine 166,571 34,266 (20.6) 7.0 (8.3) 5.46% CNS stimulant 2,886,825 370,861 (12.8) 11.2 (12.2) 94.54% Maximum of 12 month follow up 1,013,556 310 ,289 Atomoxetine 67,572 19,267 (28.5) 3.70% CNS stimulant 1,756,944 323,460 (18.4) 96.30% Table 4 2 Part I : Inclusion and e xclusion criteria with associated cohort size Inclusion/Exclusion criteria Cohort Size Patients with at least one common mental health di sorder associated with CNS stimulant & atomoxetine 4,615,177 Patients with i ndex date after 3 rd and before 19 th birthday 3,456,554 Patients with at least six month continuous Medicaid eligibility before In dex Date 1,487,713 Patients without CNS stimulant use in six month prior to index date 1,390,807 Patients without atomoxetine use i n six month prior to index date, no Patients from bad states/years, index date after 5 th birthday 1,169,129 Patients with 365 or less days between treatm ent initiation preceding first diagnosis 1,097,208

PAGE 105

105 Table 4 3 Part I: Baseline socio demographic characteristics of the study cohort 200 3 2006 Study cohort Unexposed Exposed N 1,013,556 (1 00.0) 669,644 (66.07) 343,912 (33.93) Mean i ndex age [y ea rs] ( SD ) 10.38 (3.76) 11.35 (3.83) 8.75 (2.95) Mean end age [ yea rs ( SD ) 12.16 (3.84) 12.97 (3.97) 11.14 (2.31) Mean end age after max 36 month [ yea rs ] ( SD ) 11.72 (3.77) 12.78 (3.90) 10.67 (3.08) Mean follow up [ yea rs ] ( SD ) 1.73 (1.37) 1.61 (1.38) 2.39 (1.69) Mean follow up after max 36 month [ yea rs ] ( SD ) 1.30 (0.70) 1.43 (0.98) 1.92 (0.99) Male gender (%) 582,792 (57.5) 345,287 (51.56) 237,505 (69.06) Patients without first prescri ption before first d iagnosis (%) 966,536 (95.4) n/a n/a 296,892 (86.3) Proportion of cohort with Rx before Dx [%] 4.6 n/a n/a 13.7 Mean difference between exposure & diagnosis [days] ( SD ) 108.9 (98.9) n/a n/a 108.9 (98.9) Race /ethn icity White (%) 562,170 (55.5) 363,911 (54.3) 198,259 (57.7) Black (%) 300,139 (29.6) 197,389 (29.5) 102,750 (29.9) Hispanic (%) 114,311 (11.3) 83,534 (12.5) 30,777 (9.0) Other (%) 36,936 (3.6) 24,810 (3.7) 12,126 (3.5) Reason for M edicaid eligibility TANF (%) 141,560 (14.0) 97,289 (14.5) 44,271 ( 12.9 ) Foster care (%) 91,778 (9.1) 65,952 (9.9) 25,826 ( 7.5 ) SSI (%) 19,552 (1.9) 13,716 (2.1) 5,836 ( 1.7 ) Rx= first prescription D x = first Diagnosis

PAGE 106

106 Table 4 4 Part I : Baseline treatment initiation and index diagnosis Study cohort Unexposed Exposed N 1,013,556 669,644 (66. 1 ) 343,912 (33.9) Atomoxetine i nitiat ors (%) 45,040 (11.5) n/a n/a 45,040 (11.5) If exposed to atomoxetine, mea n number of Rx ( SD ) 7.78 (9.3) n/a n/a 7.78 (9.3) Time between i ndex d ate and atomoxetine i nitiation [months] ( SD ) 20.5 (15.2) n/a n/a 20.5 (15.2) Stimulant i nitiators (%) 392,972 (88.5) n/a n/a 392,972 (88.5) If exposed to CNS stimulant s, mean number of Rx ( SD ) 13 (15.6) n/a n/a 13 (15.6) Time between i ndex date a nd CNS stimulant i nitiation [months] ( SD ) 3.8 (8.4) n/a n/a 3.8 (8.4) Index d iagnosis ADHD ( % ) 384,710 (38.0) 126,221 (18.9) 258,489 (75.2) Adjustm ent reaction ( % ) 365,641 (36.1) 328,288 (49.0) 37,353 (10.9) Disturbance of conduct ( % ) 149,001 (14.7) 116,503 (17.4) 32,498 (9.5) Other or mixed emotional disturbances ( % ) 122,060 (12.0) 98,976 (14.8) 23,084 (6.7) Unspecified emotional disturbance ( % ) 12,894 (1.3) 10,486 (1.6) 2,408 (0.7)

PAGE 107

107 Table 4 5 Part I: Baseline co morbidities Index year >=2000 Unexposed Exposed N 1,013,556 669,644 (66. 1 ) 343,912 (100.0) ADHD with h yperactivity (%) 218,708 (25.1) 78,595 (13.6) 1 40,113 (47.5) ADHD without h yperactivity (%) 83,519 (9.6) 31,970 (5.5) 51,549 (17.5) Other co morbidities: Mental health h ospitalization (%) 14,563 (1.7) 12,450 (2.2) 2,113 (0.7) Substance use disorder (%) 17,517 (2.0) 16,242 (2.8) 1,275 (0.4) Smoking (%) 2,391 (0.3) 2,218 (0.4) 173 (0.1) Anxiety (%) 58,062 (6.7) 49,149 (8.5) 8,913 (3.0) Bipolar disorder (%) 9,221 (1.1) 6,785 (1.2) 2,436 (0.8) Schizophrenia (%) 1,699 (0.2) 1,466 (0.3) 233 (0.1) Depression (%) 66,231 (7.6) 53,607 (9.3) 12,624 (4.3) Autism (%) 5,424 (0.6) 3,528 (0.6) 1,896 (0.6) Mental retardation (%) 3,543 (0.4) 2,665 (0.5) 878 (0.3) Tic disorder (%) 1,451 (0.2) 957 (0.2) 494 (0.2) Oppositional defian t disorder (%) 89,712 (10.3) 74,419 (12.9) 15,293 (5.2) Psychosis (%) 4,305 (0.5) 3,484 (0.6) 821 (0.3) Other mental health d iagnosis (%) 108,307 (12.4) 69,523 (12.0) 38,784 (13.1) Number of distinct mental health diagnoses during baseline 1 (%) 563,864 (64.6) 363,724 (62.9) 200,140 (67.8) 2 (%) 207,923 (23.8) 156,301 (27.0) 51,622 (17.5) 3 (%) 52,585 (6.0) 41,612 (7.2) 10,973 (3.7) 4 or more (%) 19,198 (2.2) 16,347 (2.8) 2,851 (1.0)

PAGE 108

108 Figure 4 1 Eligible and at risk patients o f treatment initiation per calendar month and treatment g roup, 2003 2006

PAGE 109

109 Figure 4 2 Atomoxetine treatment initiation and prevalent utilization per c alend ar m onth, 2003 2006

PAGE 110

110 Figure 4 3 CNS stimulant treatment initiation and prevalent utilization per calendar m onth, 2003 2006

PAGE 111

111 Figure 4 4 At omoxetine treatment i nitiations/1,0 00 patient months 2003 2006 seasonal ity adjusted

PAGE 112

112 Figure 4 5 CNS stimulant t r eatment initiations/1,000 patient months 2003 2006 seasonality adjusted

PAGE 113

113 Figure 4 6 Segmented linear regression analysis of atomoxetine t r eatment i nitiations/1 000 patient months 2003 2006 Boxed warning

PAGE 114

114 Figure 4 7 Joinpoint analysis of atomoxetine treatment initiations/1,000 patient months 2003 2006

PAGE 115

115 Figure 4 8 Joinpoint a nalysi s of CNS stimulant treatment initiations/1,000 patient months, 2003 200 6

PAGE 116

116 Part II: Tables and Figures Table 4 6 Part II: Baseline sociodemographic and clinical characteristics initial treatment cohort Atomoxetine CNS st imulant Standardized difference* N (%) 60,542 (20.2) 238,509 (79.8) Mean i ndex age [years] ( SD ) 9.88 (3.4) 9.00 (3.1) 0.27 Mean end age [years] ( SD ) 11.65 (3.5) 10.52 (3.3) 0.33 Mean follow up [years] ( SD ) 1.77 (1.1) 1.52 (1.1) 0.22 Male gender ( %) 40,119 (66.3) 162,922 (68.3) 0.04 Age 5 years (%) 5,944 (9.8) 38,476 (16.1) 0.19 6 8 years (%) 23,582 (39.0) 103,502 (43.4) 0.34 9 11 years (%) 14,774 (24.4) 52,527 (22.0) 0.19 12 14 years (%) 9,975 (16.5) 29,752 (12.5) 0.09 15 18 years (%) 6,267 (10.4) 14,252 (6.0) 0.16 Race /ethnicity Caucasian (%) 43,165 (71.3) 138,820 (58.2) 0.28 Black (%) 11,894 (19.7) 68,813 (28.9) 0.22 Hispanic (%) 3,814 (6.3) 23,403 (9.8) 0.13 Other (%) 1,669 (2.8) 7,473 (3.1) 0.02 Reason for Medicaid eligibility TANF (%) 5,498 (9.1) 21,756 (9.1) 0.001 Foster care (%) 5,779 (9.6) 23,922 (10.0) 0.02 SSI (%) 1,776 (2.9) 4,231 (1.8) 0.07 Calendar year 2003 (%) 21,656 (35.8) 60,262 (25.3) 0.23 2004 (%) 20,888 ( 34.5) 62,793 (26.3) 0.18 2005 (%) 11,609 (19.2) 59,902 (25.1) 0.14 2006 (%) 6,389 (10.6) 55,552 (23.3) 0.34

PAGE 117

117 Table 4 6 C ontinued Atomoxetine CNS stimulant Standardized difference* Index d iagnosis: ADHD (%) 53,805 (88.9) 217,926 (91.4) 0 .15 Adjustment reaction (%) 10,379 (17.1) 35,241 (14.8) 0.06 Disturbance of conduct (%) 7,634 (12.6) 29,823 (12.5) 0.003 Other or mixed emotional disturbances (%) 7,426 (12.3) 26,062 (10.9) 0.04 Unspecified emotional disturbance (%) 452 (0.8) 2,186 (0. 9) 0.02 Other co morbidities ADHD with h yperactivity (%) 37,193 (61.4) 163,580 (68.6) 0.15 ADHD without h yperactivity (%) 16,612 (27.4) 54,346 (22.8) 0.11 Substance use disorder (%) 1,177 (1.9) 2,053 (0.9) 0.09 Anxiety (%) 4,336 (7.2) 13 ,328 (5.6) 0.06 Bipolar disorder (%) 1,747 (2.9) 5,187 (2.2) 0.05 Schizophrenia (%) 162 (0.3) 360 (0.2) 0.03 Depression (%) 5,786 (9.6) 16,779 (7.0) 0.09 Autism (%) 397 (0.7) 1,520 (0.6) 0.002 Mental retardation (%) 184 (0.3) 752 (0.3) 0.002 Tic dis order (%) 402 (0.7) 423 (0.2) 0.08 Oppositional defiant disorder (%) 6,880 (11.4) 23,604 (9.9) 0.05 Psychosis (%) 407 (0.7) 1,203 (0.5) 0.02 Other mental h e alth d iagnosis (%) 9,718 (16.1) 40,180 (16.9) 0.02 Distinct mental health disorders 1 (%) 31,630 (52.2) 131,931 (55.3) 0.06 2 (%) 16,233 (26.8) 62,257 (26.1) 0.02 3 (%) 7,368 (12.2) 27,121 (11.4) 0.02 4 or more (%) 5,311 (8.8) 17,200 (7.2) 0.06

PAGE 118

118 Table 4 6 C ontinued Atomoxetine CNS stimulant Standardized difference* Other co morbidities Obesity (%) 648 (1.1) 2,769 (1.2) 0.01 Smoking (%) 178 (0.3) 304 (0.1) 0.04 Suicidal i deation (%) 33 (0.055) 151 (0.063) 0.0036 Suicide a ttempt (%) 39 (0.064) 90 (0.038) 0.0118 Non mental health hospitalization 0 59,833 (98.8) 235,788 (98.9) 0.0028 1 642 (1.1) 2,450 (1.0) 0.0033 2 or more 67 (0.1) 271 (0.1) 0.0090 Mental health hospitalization 0 59,432 (98.2) 234,996 (98.5) 0.0283 1 936 (1.5) 3,055 (1.3) 0.0225 2 or more 174 (0.3) 458 (0.2) 0.0195 *Standardized difference = significant imbalance defined as absolute value greater than 0.20

PAGE 119

119 Table 4 7 Part II: Baseline drug exposure initial treatment cohort Atomoxetine CNS stimulant Standardized difference* N (%) 60,5 42 (20.24) 238,509 (79.76) Methylphenidate (%) 139,950 (58.7) Mixed amphetamine salts (%) 98,559 (41.3) Concomitant drug use Antidepressant (%) 8,937 (14.8) 25,050 (10.5) 0.1258 Antipsychotic (%) 5,993 (9.9) 18,654 (7.8) 0.07 32 Anticonvulsant (%) 3,885 (6.4) 11,296 (4.7) 0.0733 Anxiolytic (in c l. sedatives & hypnotics ) (%) 2,295 (3.8) 8,350 (3.5) 0.0155 Lithium (%) 250 (0.4) 703 (0.3) 0.0199 Alpha agonist (%) 2,900 (4.8) 13,173 (5.5) 0.0331 Opioid analgesics (%) 4,004 (6. 6) 13,434 (5.6) 0.0409 Number of concomitant drug classes 0 (%) 43,159 (71.3) 182,393 (76.5) 0.1182 1 (%) 11,559 (19.1) 40,073 (16.8) 0.0597 2 (%) 4,053 (6.7) 11,586 (4.9) 0.1182 3 (%) 1,420 (2.4) 3,685 (1.6) 0.0580 4 or more (%) 351 (0 .6) 772 (0.3) 0.0382 *Standardized difference = significant imbalance defined as absolute value greater than 0.20

PAGE 120

120 Table 4 8 Part II: Determinants of atomoxetine versus CNS stimulant initiation Adjusted Odds Ratio [OR] 95 % Confidence Intervals [CI] p value Gender Male 1.00 Female 1.04 1.02 1.06 <.0001 Race/e thnicity Caucasian 1.00 Black 0.57 0.56 0.59 <.0001 Hispanic 0.54 0.53 0.57 <.0001 Other 0.70 0.67 0.74 <.0001 Age 5 years 0.69 0.67 0.72 <.0001 6 8 years 1.00 9 11 years 1.18 1.15 1.21 <.0001 12 14 years 1.36 1.32 1.40 <.0001 15 18 years 1.67 1.61 1.74 <.0001 Medicaid eligibility TANF 1.01 0.9 8 1.04 0.73 Foster care 0.93 0.90 0.96 <.0001 SSI 1.35 1.28 1.43 <.0001 Calendar year 2003 1.00 2004 0.93 0.91 0.95 <.0001 2005 0.54 0.53 0.56 <.0001 2006 0.33 0.32 0.34 <.0001

PAGE 121

121 Table 4 8 C ontinued Adjusted Odds Ratio [OR] 95% Confidence Intervals [CI] p value Index d ia gnosis ADHD with hyperactivity 0.89 0.85 0.92 <.0001 ADHD without hyperactivity 1.12 1.07 1.16 <.0001 Adjustment reaction 1.08 1.03 1.13 0.0014 Conduct Disorder 1.03 0.98 1.07 0.27 Mixed emotional disorder 0.90 0.81 0.99 0.03 Other unspecified em otional disorder 0.85 0.76 0.95 0.0044 Other co morbidities Obesity 0.83 0.76 0.90 <.0001 Substance use 1.54 1.41 1.69 <.0001 Smoking 0.98 0.80 1.21 0.87 Anxiety 1.09 1.03 1.15 0.00 Bipolar 0.98 0.91 1.05 0.53 Schizophrenia 1.30 1.07 1.58 0.01 Depression 1.02 0.97 1.08 0.41 Autism 1.15 1.02 1.30 0.03 Mental retardation 0.94 0.79 1.11 0.46 Tic disorder 3.31 2.85 3.83 <.0001 ODD 1.21 1.09 1.35 0.0003 Psychosis 1.18 1.04 1.33 0.01 Other mental diagnosis 0.91 0.87 0.95 <.0001 Suicidal Ideation 1.17 0.80 1.73 0.42 Suicide attempt 0.89 0.60 1.31 0.55 Distinct mental health disorders 1 1.00 2 1.07 1.02 1.11 0.0029 3 1.04 0.96 1.12 0.34 4 or more 1.04 0.91 1.18 0.57

PAGE 122

122 Table 4 8 C ontinued Adjusted Odds Ratio [OR] 95 % Confidence Intervals [CI] p value Concomitant drug use Antidepressant 0.76 0.51 1.13 0.18 Antipsychotic 0.72 0.55 1.23 0.34 Anticonvulsant 0.83 0.55 1.24 0.36 Anxiolytic 0.74 0.50 1.11 0.14 Lithium 0.67 0.44 1.02 0.06 Opioid a nalgesic 0.74 0.5 0 1.11 0.15 Alpha agonist 0.86 0.82 0.89 <.0001 Number of concomitant drug classes 0 1.00 1 1.38 0.92 2.07 0.18 2 1.86 0.83 4.15 0.34 3 2.30 0.69 7.67 0.36 4 or more 3.40 0.65 17.71 0.14 Non mental health hospitalization 0 1.00 1 0.96 0.88 1.05 0.36 2 or more 0.83 0.63 1.10 0.20 Mental health hospitalization 0 1.00 1 0.86 0.79 0.93 0.004 2 or more 0.91 0.75 1.11 0.36 Significant at 0.05 level

PAGE 123

123 Table 4 9 Part I I: Baseline characteristics of subsequent atomoxetine initiators a nd matched CNS stimulant users Subsequent atomoxetine Current CNS stimulant Standardized difference* N (%) 64,414 (25.0) 193,242 (75.0) Mean age at CNS treatment initiation [years] (SD ) 8.31 (2.61) 8.61 (2.78) 0.11 Mean atomoxetine match age [years] (SD) 9.92 (2.86) 10.22 (2.98) 0.10 Mean end age [years] (SD) 11.98 (3.04) 11.71 (3.23) 0.09 Mean follow up [years] (SD) 3.67 (1.77) 3.10 (1.86) 0.32 Mean time until match [years] ( SD) 1.61 (1.33) 1.61 (1.34) 0.001 Mean follow up after match [years] (SD) 2.06 (1.15) 1.48 (1.16) 0.49 Male gender (%) 46,284 (71.85) 136,126 (70.44) 0.03 Age 5 years (%) 2,738 (4.3) 7,393 (3.8) 0.02 6 8 years (%) 25,411 (39.5) 70,434 (3 6.5) 0.06 9 11 years (%) 21,520 (33.4) 64,492 (33.4) 0.04 12 14 years (%) 10,705 (16.6) 35,191 (18.2) 0.04 15 18 years (%) 4,040 (6.3) 15,732 (8.1) 0.07 Race/ e thnicity Caucasian (%) 44,779 (69.5) 114,161 (59.1) 0.22 Black (%) 13,356 (2 0.7) 56,140 (29.1) 0.19 Hispanic (%) 3,963 (6.2) 15,659 (8.1) 0.08 Other (%) 2,316 (3.6) 7,282 (3.8) 0.01 Reason for Medicaid eligibility TANF (%) 6,371 (9.9) 15,925 (8.2) 0.058 Foster care (%) 7,797 (12.1) 25,344 (13.2) 0.03 SSI (%) 1 ,382 (2.2) 3,391 (1.8) 0.03 Calendar year of subsequent atomoxetine /match 2003 (%) 29,629 (46.0) 55,641 (28.8) 0.36 2004 (%) 19,046 (29.6) 41,782 (21.6) 0.18 2005 (%) 9,452 (14.7) 43,028 (22.3) 0.20 2006 (%) 6,287 (9.8) 52,791 (27.3) 0.4 6

PAGE 124

124 Table 4 9 C ontinued Subsequent atomoxetine Current CNS stimulant Standardized difference* Index d iagnosis at atomoxetine /match ADHD with h yperactivity (%) 43,432 (67.4) 130,759 (67.7) 0.01 ADHD without h yperactivity (%) 13,934 (21.6) 39,043 (20.2) 0.04 Adjustment reaction (%) 7,790 (12.1) 20,938 (10.8) 0.040 Disturbance of conduct (%) 5,652 (8.8) 14,378 (7.4) 0.05 Other or mixed emotional disturbances (%) 7,392 (11.5) 19,008 (9.8) 0.05 Unspecified emotional disturbance (%) 476 (0.7) 1,17 7 (0.6) 0.02 Other mental co morbidities Substance use disorder (%) 625 (1.0) 1,421 (0.7) 0.03 Anxiety (%) 4,063 (6.3) 9,967 (5.2) 0.05 Bipolar disorder (%) 2,811 (4.4) 6,063 (3.1) 0.06 Schizophrenia (%) 204 (0.3) 447 (0.2) 0.02 Depressio n (%) 5,249 (8.1) 13,015 (6.7) 0.05 Autism (%) 532 (0.8) 1,186 (0.6) 0.03 Mental retardation (%) 318 (0.5) 920 (0.5) 0.003 Tic disorder (%) 531 (0.8) 472 (0.2) 0.080 Oppositional defiant disorder (%) 6,784 (10.5) 17,280 (8.9) 0.05 Psychosis (%) 533 (0 .8) 1,061 (0.5) 0.03 Other m ental h ealth d iagnosis (%) 11,868 (18.4) 31,737 (16.4) 0.05 Distinct mental health disorders 0 (%) 1,314 (2.0) 4,138 (2.1) 0.01 1 (%) 9,823 (15.2) 33,105 (17.1) 0.10 2 (%) 29,390 (45.6) 95,222 (49.3) 0.03 3 (%) 13,300 (20.6) 35,907 (18.6) 0.05 4 or more (%) 10,587 (16.4) 24,870 (12.9) 0.09

PAGE 125

125 Table 4 9 C ontinued Subsequent atomoxetine Current CNS stimulant Standardized difference* Other co morbidities Obesity (%) 492 (0.76) 1,672 (0.87) 0.01 Smoking (%) 67 (0.10) 184 (0.10) 0.003 Suicidal i deation (%) 25 (0.0388) 84 (0.0435) 0.0023 Suicide a ttempt (%) 34 (0.0528) 58 (0.0300) 0.0112 Non mental health hospitalization 0 63,669 (98.8) 191,397 (99.0) 0.0197 1 671 (1.0) 1,679 (0.9) 0.017 8 2 or more 74 (0.1) 166 (0.1) 0.0092 Mental health hospitalization 0 63,025 (97.8) 191,042 (98.9) 0.0800 1 1,148 (1.8) 1,836 (1.0) 0.0717 2 or more 241 (0.4) 364 (0.2) 0.0351 *Standardized difference = significant imbalance defined as absolute value greater than 0.20

PAGE 126

126 Table 4 10 Part II: Baseline drug exposure of subsequent atomoxetine initiators and matched CNS stimulant users Subsequent atomoxetine Current CNS stimulant Standardized difference* N (%) 6 4,414 (25.0) 193,242 (75.0) Mixed amphetamine salts (%) 26,704 (41.5) 83,019 (43.0) (0.031) Methylphenidate (%) 33,446 (51.9) 107,206 (55.5) (0.071) Any CNS stimulant (%) 54,121 (84.0) 177,506 (91.9) (0.242) Concomitant drug use Antidepressant (%) 13,556 (21.0) 31,984 (16.6) 0.1152 Antipsychotic (%) 10,722 (16.6) 27,519 (14.2) 0.0666 Anticonvulsant (%) 6,022 (9.3) 15,020 (7.8) 0.0688 Anxiolytic (incl. sedatives & hypnotics ) (%) 2,694 (4.2) 7,384 (3.8) 0.0288 Lithium (% ) 523 (0.8) 1,092 (0.6) 0.0318 Alpha agonist (%) 9,119 (14.2) 23,938 (12.4) 0.0746 Opioid analgesics (%) 3,887 (6.0) 10,839 (5.6) 0.0254 Number of concomitant drug classes 0 (%) 39,616 (61.5) 126,394 (65.4) 0.1115 1 (%) 15,422 (2 3.9) 43,877 (22.7) 0.0498 2 (%) 6,556 (10.2) 16,691 (8.6) 0.0668 3 (%) 2,341 (3.6) 5,309 (2.7) 0.0601 4 or more (%) 479 (0.7) 971 (0.5) 0.0360 *Standardized difference = significant imbalance defined as absolute value greater than 0.20

PAGE 127

127 Table 4 11 Part II: Determinants of subsequent atomoxetine versus matched CNS stimulant use Adjusted Odds Ratio [OR] 95% Confidence Intervals [CI] p value Gender Male 1.00 Female 0.94 0.92 0.96 <.0001 Race/ e thn icity Caucasian 1.00 Black 0.64 0.6 2 0.6 5 <.0001 Hispanic 0.68 0.65 0.70 <.0001 Other 0. 80 0.7 6 0.8 4 <.0001 Age 5 years 1.0 6 1.01 1.1 2 0.0 1 1 6 8 years 1.00 9 11 years 0.86 0.84 0.88 <.0001 12 14 years 0.7 8 0.7 6 0. 80 <.0001 15 18 years 0. 60 0.5 8 0.6 3 <.0001 Medicaid eligibility TANF 1.0 9 1.05 1.12 <.0001 Foster care 0.9 4 0.9 1 0.96 <.0001 SSI 1.16 1.09 1. 24 <.0001 Calendar year 2003 1.00 2004 0.8 8 0.8 6 0. 90 <.0001 2005 0. 42 0. 4 1 0.4 4 <.0001 2006 0.2 3 0.22 0.24 <.0001

PAGE 128

128 Table 4 11 C ontinued Adjusted Odds Ratio [OR] 95% Confidence Intervals [CI] p value Index d iagnosis ADHD with hyperactivity 0.97 0.93 1.01 0.56 ADHD without hyperactivity 1.01 0.97 1.05 0.13 Adjust ment reaction 0.97 0.93 1.02 0.25 Conduct d isorder 1.06 1.01 1.11 0.02 Mixed emotional disorder 0.97 0.88 1.08 0.609 Other unspecified emotional disorder 1.09 0.96 1.22 0.169 Other co morbidities Obesity 0.93 0.84 1.04 0.196 Substance use 1.39 1.25 1.56 <.0001 Smoking 0.69 0.50 0.94 0.018 Anxiety 1.00 0.95 1.06 0.910 Bipolar 1.13 1.06 1.20 <.0001 Schizophrenia 1.14 0.95 1.36 0.153 Depression 1.01 0.96 1.06 0.017 Autism 1.15 1.03 1.29 0.001 Mental retardation 0.90 0.78 1.03 0.119 Tic disorder 2.69 2.35 3.08 <.0001 ODD 0.99 0.90 1.11 0.917 Psychosis 1.21 1.08 1.36 0.001 Other mental diagnosis 0.99 0.95 1.03 0.63 Suicidal i deation 1.75 1.11 2.77 0.02 Suicide attempt 1.24 0.79 1.94 0.347 Distinct mental health disorders 0 1.08 1.03 1.13 0.00 34 1 1.00 2 1.14 1.10 1.19 <.0001 3 1.27 1.18 1.36 <.0001 4 or more 1.36 1.20 1.53 <.0001

PAGE 129

129 Table 4 11 Continued Adjusted Odds Ratio [OR] 95% Confidence Intervals [CI] p value Concomitant drug use Antidepressant 1.1 2 0.79 1.60 0.52 Antipsychotic 1.04 0.73 1.49 0.82 Anticonvulsant 1.07 0.75 1.53 0.71 Anxiolytic 1.06 0.75 1.52 0.73 Lithium 1.11 0.77 1.59 0.58 Opioid a nalgesic 1.08 0.76 1.54 0.75 Alpha agonist 1.13 1.10 1.16 <.0001 Number of concomitant dr ug classes 0 1.00 1 1.01 0.71 1.43 0.98 2 1.00 0.49 2.02 0.99 3 0.98 0.34 2.83 0.97 4 or more 0.99 0.23 4.23 0.99 Non mental health hospitalization 0 1.00 1 1.11 1.01 1.22 0.03 2 or more 1.31 0.98 1.74 0.07 Mental health hospitalization 0 1.00 1 1.36 1.24 1.48 <.0001 2 or more 1.31 1.09 1.58 0.0041 Significant at 0.05 level

PAGE 130

130 Part III: Tables and Figures Table 4 12 Part III: Distribution of exclusion and inclusion crit eria initial treatment cohort Sum of patients Percent Initial atomoxetine Percent Initial CNS stimulant Percent Full cohort 298,5 46 60,422 (20.2) 238,12 4 (79.8) Pregnancy during baseline 239 (0.08) 59 (0.10) 180 (0.08) Hospitalization of mor e than 30 days 11,035 (3.70) 2,623 (4.34) 8,412 (3.53) O rgan transplant status 136 (0.05) 23 (0.04) 113 (0.05) Malign neoplasm 1,196 (0.40) 240 (0.40) 956 (0.40) HIV/AIDS 180 (0.06) 21 (0.03) 159 (0.07) P ervasive development disorder 5,948 (1 .99) 1,263 (2.09) 4,685 (1.97) Severe or profound mental retardation 77 (0.03) 18 (0.03) 59 (0.02) Pemoline, methamphetamine or MAO inhibitors exposure 420 (0.14) 163 (0.27) 257 (0.11) Total excluded 19,235 (6.4) 4,410 (7.3) 14,825 (6 .2) Censoring criteria Full c ohort 279,315 (93.6) 56,012 (20.1) 223,303 (79.9) Medicaid e ligibility 265,160 (94.9) 52,456 (93.7) 212,704 (95.3) 19 th birthday 2,984 (1.1) 1,097 (2.0) 1,887 (0.8) Pregnancy 1,284 (0.5) 411 (0.7) 873 (0.4) Hospitalization of more than 30 days 9,655 (3.5) 1,982 (3.5) 7,673 (3.4) Non suicide death 92 (0.03) 22 (0.04) 70 (0.03) Suicide events 140 (0.05 3 ) 4 4 (0.1 1 ) 9 6 (0.04 3 )

PAGE 131

131 Table 4 13 Part III: Baseline sociod emographic and clinical characteristics initial treatment cohort Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweighted N (%) 56,012 (20.1) 223,303 (79.9) Mean i ndex age [yea rs] (SD) 9.82 (3.4) 9.2 8.97 (3.1) 9.1 0.26 Mean end age [years] (SD) 11.58 (3.5) 10.7 10.49 (3.3) 10.7 0.33 Mean follow up [years] (SD) 1.75 (1.1) 1.5 1.51 (1.1) 1.5 0.22 Male gender (%) 36,919 (65.9) (67.6) 152,098 (68.1) (67.7) 0.05 Age 5 years (%) 5,489 (9.8) (14.8) 35,759 (16.0) (14.8) 0.19 6 8 years (%) 22,113 (39.5) (42.6) 97,591 (43.7) (42.8) 0.09 9 11 years (%) 13,746 (24.5) (22.6) 49,584 (22.2) (22.7) 0.06 12 14 years (%) 9,126 (16.3) (13.3) 27,551 (12.3) (13.2) 0.11 15 18 years (%) 5,538 (9.9) (6.7) 12,818 (5.7) (6.6) 0.16 Race /ethnicity Caucasian (%) 39,911 (71.3) (60.8) 129,691 (58.1) (60.7) 0.28 Black (%) 11,197 (20.0) (26.9) 65,124 (29.2) (27.3) 0.21 Hispanic (%) 3,583 (6.4) (9.4) 22,21 9 (10.0) (9.2) 0.13 Other (%) 1,418 (2.5) (2.9) 6,550 (2.9) (2.9) 0.02 Reason for Medicaid eligibility TANF (%) 5,256 (9.4) (9.4) 20,771 (9.3) (9.3) 0.003 Foster care (%) 4,922 (8.8) (9.4) 20,965 (9.4) (9.3) 0.02 SSI (%) 1,711 (3.1) (2.2) 4,113 (1.8) (2.1) 0.08 Calendar year 2003 (%) 19,778 (35.3) (27.3) 55,874 (25.0) (27.1) 0.23 2004 (%) 19,378 (34.6) (28.2) 58,939 (26.4) (28.1) 0.18 2005 (%) 10,884 (19.4) (23.9) 56,490 (25.3) (24.1) 0.14 2006 (%) 5,962 (10.6) (20.6) 52,000 (23.3) (20.7) 0.34

PAGE 132

132 Table 4 1 3 C ontinued Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweighted Index d iagnosis: ADHD with h yperactivity (%) 34,543 (6 1.7) (66.9) 153,696 (68.8) (67.3) 0.15 ADHD without h yperactivity (%) 15,653 (27.9) (24.4) 51,680 (23.1) (24.1) 0.11 Adjustment reaction (%) 9,392 (16.8) (15.2) 32,141 (14.4) (14.9) 0.07 Disturbance of conduct (%) 6,680 (11.9) (11.8) 27,036 (12.1) (12. 1) 0.006 Other or mixed emotional disturbances (%) 6,619 (11.8) (10.7) 23,460 (10.5) (10.8) 0.04 Unspecified emotional disturbance (%) 359 (0.6) (0.7) 1,769 (0.8) (0.8) 0.02 Other mental co morbidities Substance use disorder (%) 987 ( 1.8) (1.0) 1,758 (0.8) (1.0) 0.09 Anxiety (%) 3,791 (6.8) (5.7) 11,863 (5.3) (5.6) 0.06 Bipolar disorder (%) 1,458 (2.6) (2.1) 4,497 (2.0) (2.1) 0.04 Schizophrenia (%) 121 (0.2) (0.2) 297 (0.1) (0.2) 0.02 Depression (%) 5,184 (9.3) (7.4) 15,004 (6.7) ( 7.3) 0.09 Mild mental retardation (%) 71 (0.1) (0.1) 331 (0.1) (0.1) 0.0058 Tic disorder (%) 351 (0.6) (0.3) 350 (0.2) (0.3) 0.08 Oppositional defiant disorder (%) 6,160 (11.0) (9.8) 21,358 (9.6) (9.9) 0.05 Psychosis (%) 341 (0.6) (0.5) 997 (0.4) (0.5 ) 0.02 Other m ental h ealth d iagnosis (%) 7,562 (13.5) (14.0) 32,618 (14.6) (14.4) 0.03 Distinct mental health disorders 1 (%) 30,632 (54.7) (57.4) 128,456 (57.5) (57.0) 0.06 2 (%) 14,626 (26.1) (25.2) 56,719 (25.4) (25.5) 0.02 3 (%) 6,368 (11.4) (10.7) 23,653 (10.6) (10.8) 0.03 4 or more (%) 4,386 (7.8) (6.7) 14,475 (6.5) (6.8) 0.06

PAGE 133

133 Table 4 1 3 C ontinued Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweight ed Other co morbidities Obesity (%) 585 (1.0) (1.1) 2,541 (1.1) (1.1) 0.009 Smoking (%) 150 (0.3) (0.2) 257 (0.1) (0.2) 0.03 Suicidal i deation (%) 26 (0.0464) (0.05) 125 (0.0560) (0.05) 0.0042 Suicide a ttempt (%) 34 (0.0607) (0.04) 78 (0.0349 ) (0.05) 0.0118 Non mental health hospitalization 0 55,430 (99.0) (99.0) 221,028 (99.0) (99.0) 0.0020 1 535 (1.0) (1.0) 2,086 (0.9) (0.9) 0.0022 2 or more 47 (0.1) (0.1) 189 (0.1) (0.1) 0.0004 Mental health hospitalization 0 55,064 (98.3) (98.7) 220,315 (98.7) (98.6) 0.029 1 802 (1.4) (1.2) 2,630 (1.2) (1.2) 0.022 2 or more 146 (0.3) (0.2) 358 (0.2) (0.2) 0.022 Standardized difference = significant imbalance defined as ab solute value greater than 0.20

PAGE 134

134 Table 4 14 Part II I : Baseline concomitant drug exposure initial treatment cohort Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweighted N (%) 56,012 (20.1) 223,303 (79.9) Methylphenidate (%) 131,105 (58.7) Mixed amphetamine salts (%) 92,360 (41.4) Concomitant drug use Antidepressant (%) 7,829 (14.0) (10.9) 22,146 (9.9) (10.8) 0.1254 Antipsychotic (%) 4,929 (8.8) (7 .5) 15,603 (7.0) (7.4) 0.0673 Anticonvulsant (%) 3,194 (5.7) (4.6) 9,448 (4.2) (4.5) 0.0678 Anxiolytic (incl. sedatives & hypnotics ) (%) 2,034 (3.6) (3.5) 7,590 (3.4) (3.5) 0.0126 Lithium (%) 194 (0.3) (0.3) 580 (0.3) (0.3) 0.0158 Alpha agonist (%) 2,4 10 (4.3) (5.0) 11,632 (5.2) (5.0) 0.0426 Opioid analgesics (%) 3,694 (6.6) (5.8) 12,603 (5.6) (5.8) 0.0397 Number of concomitant drug classes 0 (%) 40,743 (72.7) (76.3) 173,143 (77.5) (76.5) 0.1112 1 (%) 10,383 (18.5) (17.0) 36,640 (1 6.4) (16.9) 0.0559 2 (%) 3,462 (6.2) (4.8) 9,922 (4.4) (4.8) 0.0778 3 (%) 1,154 (2.1) (1.5) 2,988 (1.3) (1.5) 0.0554 4 or more (%) 270 (0.5) (0.3) 610 (0.3) (0.3) 0.0348 Standardized difference = significant imbalance defined as ab solute value greater than 0.20 Table 4 15 Part III: Distribution of mono and dual therapy initial treatment cohort Initial therapy Exposure [person years] Atomoxetine and CNS stimulant exposure Current CNS stimulant (%) 190,026 ( 44.37 ) Cur rent atomoxetine (%) 46,929 ( 10.96 ) 6,348 (13.53) Former atomoxetine (%) 47,173 ( 11.01 ) Former CNS stimulant (%) 144,144 ( 33.66 )

PAGE 135

135 Table 4 16 Part III: Drug exposure, suicide & suicide attempts after 3, 6, 12, 24 month and overall initial treatment cohort Current CNS stimulant (%) Current atomoxetine (%) Former atomoxetine (%) Former CNS stimulant (%) Sum Exposure 3 months [person years] 43,184 (65.8) 11,445 (17.4) 2,210 (3.4) 8,821 (13.4) 65,661 Events after 3 months 23 Events/100,000 person years 35.03 Exposure 6 months [person years] 69,760 (56.6) 18,597 (15.1) 7,354 (6.0) 27,600 (22.4) 123,311 Events after 6 months 47 Events/100,000 person years 3 8.12 Exposure 12 months [person years] 110,194 (50.1) 29,018 (13.2) 18,005 (8.2) 62,551 (28.5) 219,769 Events after 12 months 70 Events/100,000 person years 31.85 Exposure 24 months [person years] 160,1 85 (46.1) 40,860 (11.8) 35,032 (10.1) 111,468 (32.1) 347,544 Events after 24 months 43 (38.7) 16 (14.4) 19 (17.1) 33 (29.7) 111 Events/100,000 person years 26.84 39.16 54.24 29.60 31.94 Exposure full cohort [person years] 190,026 (44.4) 4 6,929 (11.0) 47,173 (11.0) 144,144 (33.7) 428,272 Events full follow up 50 (35.7) 18 (12.9) 25 (17.9) 47 (33.6) 140 Events/100,000 person years 26.31 38.36 53.00 32.61 32.69

PAGE 136

136 Table 4 17 Part III: Hazard ratios (H R) for suicide and suicide attempt initial treatment cohort Exposure Unadjusted Adjusted for time dependent age, dual therapy and propensity score Events Hazard ratio 95 % Confidence Interval p value Hazard ratio 95 % Confidence Interval p value Ful l follow up Current CNS stimulant 50 1.00 1.00 Current atomoxetine 18 1.45 0.85 2.48 0.18 0.88 0.50 1.56 0.66 Former atomoxetine 25 2.19 1.33 3.60 0.002 0.88 0.53 1.46 0.62 Former CNS stimulant 47 1.35 0.89 2.04 0.16 0.87 0.56 1.33 0.520 24 month follow up Current CNS stimulant 43 1.00 1.00 Current atomoxetine 16 1.45 0.82 2.57 0.21 0.94 0.52 1.72 0.85 Former atomoxetine 19 2.25 1.28 3.93 0.005 0.95 0.54 2.68 0.86 Former CNS stimulant 3 3 1.22 0.76 1.96 0.41 0.83 0.52 1.35 0.46 12 month follow up 70 Current CNS stimulant 1.00 1.00 Current atomoxetine 1.51 0.77 2.95 0.23 0.95 0.47 1.92 0.88 Former atomoxetine 2.42 1.15 5.11 0.02 1.11 0.52 2.37 0.79 Former CNS stimulant 1.24 0.67 2.30 0.50 0.95 0.51 1.77 0.87 6 month follow up 47 Current CNS stimulant 1.00 1.00 Current atomoxetine 1.63 0.77 3.42 0.19 1.07 0.50 2.28 0.86 Former atomoxetine 1.77 0.5 9 5.31 0.31 0.86 0.28 2.60 0.78 Former CNS stimulant 1.18 0.53 2.61 0.68 1.00 0.45 2.21 0.99 3 month follow up 23 Current CNS stimulant 1.00 1.00 Current atomoxetine 1.60 0.62 4.20 0.33 1.00 0.38 2.66 0.990 F ormer atomoxetine 1.93 0.24 15.90 0.54 0.89 0.11 7.43 0.92 Former CNS stimulant 0.96 0.20 4.65 0.96 0.83 0.17 4.07 0.810

PAGE 137

137 Table 4 18 Part III: Distribution of e xclusion and inclusion criteria subsequent treatment cohor t Sum of patients Percent Subsequent atomoxetine initiation Percent Continuous CNS stimulant use Percent Exclusion criteria Full c ohort 241,125 60,276 (25.0) 180,849 (75.0) Pregnancy during baseline 126 (0.05) 30 (0.05) 96 (0.05) Hospita lization of more than 30 days 11,317 (4.69) 2,815 (4.67) 8,502 (4.70) S evere or profound mental retardation 52 (0.02) 12 (0.02) 40 (0.02) Incomplete matched cluster 9,415 (3.90) 471 (0.78) 8,944 (4.95) Total excluded 20,910 (8.67) 3,328 (5.5) 17 ,582 (9.7) Censoring criteria Full c ohort 220,215 56,948 (25.9) 163,267 (74.1) Medicaid e ligibility 207,391 (94.18) 52,978 (93.8) 154,413 (94.3) 19 th birthday 3,070 (1.39) 724 (1.3) 2,346 (1.43) Pregnancy 940 (0.43) 230 (0.4) 710 (0.43) Hospitalization of more than 30 days 8,618 (3.91) 2,963 (5.2) 5,655 (3.45) Non suicide death 106 (0.05) 28 (0.05) 78 (0.05) Suicide attempt 90 (0.041) 27 (0.048) 63 (0.038)

PAGE 138

138 Table 4 19 Part III: Baseline sociodemograp hic and clinical characteristics subsequent treatment cohort Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweighted N (%) 56,948 (25.9) 163,267 (74.1) Mean age at initial tr eatment (SD) 8.31 (2.6) 8.46 8.58 (2.8) 8.53 0.10 Mean atomoxetine start/ match age (SD) 9.88 (2.8) 10.07 10.14 (3.0) 10.07 0.09 Mean end age [years] (SD) 11.82 (3.0) 11.71 11.55 (3.2) 11.58 0.09 Mean follow up [years] (SD) 3.51 (1.7) 3.25 2.97 (1. 8) 3.05 0.30 Mean time to switch / match [y ea rs] (SD) 1.57 (1.3) 1.56 1.56 (1.3) 1.55 0.006 Mean follow up after switch/match (SD) 1.94 (1.1) 1.65 1.41 (1.1) 1.50 0.47 Male gender (%) 40,657 (71.4) (70.4) 114,426 (70.1) (70.4) 0.03 Age 5 years (%) 2,453 (4.3) (4.2) 6,396 (3.9) (4.1) 0.02 6 8 years (%) 22,697 (39.9) (37.4) 60,744 (37.2) (37.9) 0.06 9 11 years (%) 19,108 (33.6) (33.5) 54,866 (33.6) (33.5) 0.001 12 14 years (%) 9,305 (16.3) (17.6) 28,819 (17.7) (17.3) 0.03 15 18 years (%) 3,3 85 (5.9) (7.4) 12,442 (7.6) (7.2) 0.07 Race /ethnicity Caucasian (%) 39,564 (69.5) (61.5) 95,877 (58.7) (61.5) 0.23 Black (%) 12,035 (21.1) (27.2) 48,086 (29.5) (27.3) 0.19 Hispanic (%) 3,494 (6.1) (7.8) 13,590 (8.3) (7.8) 0.08 Other (%) 1,85 5 (3.3) (3.4) 5,714 (3.5) (3.4) 0.01 Reason for Medicaid eligibility TANF (%) 5,871 (10.3) (9.0) 13,786 (8.4) (9.0) 0.064 Foster care (%) 6,462 (11.3) (11.9) 19,991 (12.2) (11.9) 0.03 SSI (%) 1,304 (2.3) (2.0) 3,026 (1.9) (2.0) 0.03 Calendar year 2003 (%) 25,839 (45.4) (32.6) 45,830 (28.1) (32.6) 0.37 2004 (%) 17,022 (29.9) (23.9) 35,602 (21.8) (23.9) 0.19 2005 (%) 8,487 (14.9) (20.6) 37,238 (22.8) (20.8) 0.20 2006 (%) 5,600 (9.8) (22.8) 44,597 (27.3) (22.8) 0.46

PAGE 139

139 Table 4 1 9 C o ntinued Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweighted Index d iagnosis: ADHD with h yperactivity (%) 38,399 (67.4) (68.1) 110,486 (67.7) (67.8) 0.005 ADHD with out h yperactivity (%) 12,508 (22.0) (21.2) 33,390 (20.5) (20.9) 0.04 Adjustment reaction (%) 6,720 (11.8) (10.9) 17,208 (10.5) (10.9) 0.04 Disturbance of conduct (%) 4,678 (8.2) (7.5) 11,514 (7.1) (7.4) 0.04 M ixed emotional disturbances (%) 6,243 (11.0) (9.9) 15,170 (9.3) (9.7) 0.06 Unspecified emotional disturbance (%) 327 (0.6) (0.5) 813 (0.5) (0.5) 0.01 Other mental co morbidities Substance use disorder (%) 521 (0.9) (0.8) 1,098 (0.7) (0.7) 0.02 Anxiety (%) 3,358 (5.9) (5.2) 7,811 (4.8) (5.1) 0.05 Bipolar disorder (%) 2,194 (3.9) (3.1) 4,536 (2.8) (3.1) 0.06 Schizophrenia (%) 154 (0.3) (0.2) 316 (0.2) (0.2) 0.02 Depression (%) 4,440 (7.8) (6.9) 10,508 (6.4) (6.8) 0.05 Mild mental retardation (%) 116 (0.2) (0.2) 398 (0.2) (0.2) 0.01 Tic disorder (%) 429 (0.8) (0.4) 378 (0.2) (0.4) 0.08 Oppositional defiant disorder (%) 5,793 (10.2) (9.1) 13,976 (8.6) (9.0) 0.06 Psychosis (%) 412 (0.7) (0.6) 803 (0.5) (0.6) 0.03 Other mental health d iagnosis (%) 8,620 (15.1) (14.3) 22,552 (13. 8) (14.2) 0.04 Distinct mental health disorders 0 (%) 5,722 (10.0) (10.0) 16,540 (10.1) (10.1) 0.003 1 (%) 26,680 (46.8) (50.7) 84,545 (51.8) (50.6) 0.10 2 (%) 13,481 (23.7) (22.9) 36,628 (22.4) (22.8) 0.03 3 (%) 6,312 (11.1) (9.9) 1 5,472 (9.5) (9.9) 0.05 4 or more (%) 4,753 (8.3) (6.9) 10,082 (6.2) (6.8) 0.08

PAGE 140

140 Table 4 1 9 Continued Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweighted Other co morbidities Obesity (%) 415 (0.7) (0.8) 1,368 (0.8) (0.8) 0.01 Smoking (%) 56 (0.1) (0.1) 126 (0.1) (0.1) 0.002 Suicidal i deation (%) 20 (0.035) (0.04) 66 (0.040) (0.04) 0.0142 Suicide a ttempt (%) 29 (0.051) (0.04) 47 (0.029) (0.04) 0.0104 Non m ental health hospitalization 0 56,345 (98.9) (99.1) 161,860 (99.1) (99.1) 0.020 1 556 (1.0) (0.9) 1,286 (0.8) (0.8) 0.020 2 or more 47 (0.1) (0.1) 121 (0.1) (0.1) 0.003 Mental health hospitalization 0 55,831 (98.0) (98.8) 161, 637 (99.0) (98.8) 0.0790 1 930 (1.6) (1.0) 1,376 (0.8) (1.0) 0.0715 2 or more 187 (0.3) (0.2) 254 (0.2) (0.2) 0.0352 Standardized difference = significant imbalance defined as ab solute value greater than 0.20

PAGE 141

141 Table 4 2 0 Part II I : Baseline drug exposure su bsequent t reatment cohort Atomoxetine CNS stimulant Standardized difference* unweighted Inverse PS weighted unweighted Inverse PS weighted unweighted N (%) 56,948 (25.9) 163,267 (74.1) Mixed amph etamine salts (%) 23,705 (41.6) (42.7) 69,968 (42.9) (42.6) 0.0249 Methylphenidate (%) 29,724 (52.2) (54.7) 90,667 (55.5) (54.7) 0.0670 Any CNS stimulant (%) 48,101 (84.5) (89.8) 149,924 (91.8) (89.9) 0.2293 Concomitant drug use Anti depressant (%) 11,296 (19.8) (16.6) 25,390 (15.6) (16.6) 0.1124 Antipsychotic (%) 8,343 (14.7) (13.2) 20,847 (12.8) (13.2) 0.0547 Anticonvulsant (%) 4,627 (8.1) (6.9) 10,659 (6.5) (6.9) 0.0613 Anxiolytic (%) 2,258 (4.0) (3.6) 5,606 (3.4) (3.5) 0.0282 L ithium (%) 390 (0.7) (0.5) 759 (0.5) (0.5) 0.0291 Alpha agonist (%) 7,568 (13.3) (11.7) 17,861 (10.9) (11.5) 0.0721 Opioid analgesics (%) 3,469 (6.1) (5.8) 9,000 (5.5) (5.7) 0.0248 Number of concomitant drug classes 0 (%) 36,347 (63.8) ( 67.5) 112,288 (68.8) (67.5) 0.1049 1 (%) 13,304 (23.4) (21.8) 34,596 (21.2) (21.8) 0.0522 2 (%) 5,203 (9.1) (7.9) 12,172 (7.5) (7.9) 0.0610 3 (%) 1,748 (3.1) (2.4) 3,589 (2.2) (2.4) 0.0544 4 or more (%) 346 (0.6) (0.4) 622 (0.4) (0.4) 0.0323 Standar dized difference = significant imbalance defined as ab solute value greater than 0.20

PAGE 142

142 Table 4 21 Part III: Distribution of mono and dual therapy, subsequent treatment cohort Subsequent therapy Exposure [person years] A tomoxetine and CNS stimulant exposure Current CNS stimulant (%) 142,015 ( 47.22 ) Current atomoxetine (%) 37,948 ( 12.62 ) 9,881 (26.04) Former atomoxetine (%) 42,756 ( 14.22 ) Former CNS stimulant (%) 78,054 ( 25.95 )

PAGE 143

143 Tab le 4 22 Part III: Drug exposure, suicide & suicide attempts after 3, 6, 12, 24 months & overall, subsequent treatment Current CNS stimulant (%) Current atomoxetine (%) Former atomoxetine (%) Former CNS stimulant (%) Sum Expos ure 3 months [person yea rs] 32,677 (64.4) 10,927 (21.5) 1,755 (3.5) 5,373 (10.6) 50,731 Events after 3 months 20 Events/100,000 person y ea rs 39.42 Exposure 6 months [person yea rs] 54,314 (58.3) 16,909 (18.2) 5,731 (6.2) 16,163 (17.4) 93,117 Events after 6 months 29 Events/100,000 person y ea rs 31.14 Exposure 12 months [person yea rs] 85,676 (53.6) 24,873 (15.6) 14,082 (8.8) 35,113 (22.0) 159,744 Events after 12 months 55 Events/100,000 person y ea rs 34.43 Exposure 24 months [person yea rs] 121,151 (49.5) 33,371 (13.6) 29,128 (11.9) 60,950 (24.9) 244,600 Events after 24 months 77 Events/100,000 person yrs 31.48 Exposure full cohort [person yea rs] 142,015 (47.2) 37,948 (12.6) 42,756 (14.2) 78,054 (26.0) 300,772 Events full follow up 46 (51.1) 11 (12.2) 16 (17.8) 17 (18.9) 90 Events/100,000 person y ea rs 32.39 28.99 37.42 21.78 29.92

PAGE 144

144 Tabl e 4 23 Part III: Hazard ratios (HR) for suicide and suicide attempt subsequent treatment cohort Exposure Unadjusted Adjusted for time dependent age, dual therapy and propensity score Suicide e vents Hazard ratio 95 % Confid ence Interval p value Hazard ratio 95 % Confidence Interval p value Full follow up Current CNS stimulant 47 1.00 1.00 Current atomoxetine 11 0.87 0.45 1.69 0.68 0.65 0.31 1.36 0.25 Former atomoxetine 19 1.30 0.72 2.35 0.380 0. 67 0.36 1.24 0.20 Former CNS stimulant 23 0.72 0.41 1.28 0.27 0.44 0.25 0.78 0.005 24 month follow up 77 Current CNS stimulant 1.00 1.00 Current atomoxetine 0.68 0.32 1.44 0.31 0.57 0.25 1.30 0.18 Former atomoxe tine 1.32 0.68 2.56 0.410 0.70 0.35 1.39 0.31 Former CNS stimulant 0.77 0.42 1.40 0.39 0.47 0.25 0.86 0.02 12 month follow up 55 Current CNS stimulant 1.00 1.00 Current atomoxetine 0.88 0.40 1.91 0.74 0.71 0 .30 1.67 0.43 Former atomoxetine 1.25 0.51 3.05 0.63 0.66 0.26 1.65 0.37 Former CNS stimulant 0.82 0.40 1.72 0.61 0.54 0.26 1.13 0.10 6 month follow up 29 Current CNS stimulant 1.00 1.00 Current atomoxetine 0.30 0.07 1.27 0.10 0.28 0.06 1.25 0.09 Former atomoxetine 0.55 0.07 4.27 0.57 0.31 0.04 2.50 0.27 Former CNS stimulant 0.96 0.34 2.73 0.94 0.70 0.25 1.99 0.50 3 month follow up 20 Current CNS stimulant 1.00 1.00 Current atomoxetine 0.40 0.09 1.73 0.22 0.37 0.08 1.71 0.200 Former atomoxetine 1.32 0.16 10.67 0.80 0.74 0.09 6.22 0.78 Former CNS stimulant 0.85 0.17 4.13 0.84 0.63 0.13 3.08 0.570

PAGE 145

145 CHAPTER 5 DISCUSSION Part I Our analysis of atomoxetin e utilization showed atomoxetine initiation rate s increased to about a third of the corresponding stimulant initiation rate (8/1,000 vs. 24/1,000) immediately after atomoxetine approval However, soon thereafter initiation rate s decline d until 2005, around the time of the implementation of the boxed warning, where a significant level off of the decline occurred. In contrast, stimulant treatment initiation rates were remarkably constant It is important to note that our study analyzed only atomoxetine utiliz ation immediately following market introduction until one year after the boxed warning with only stimulants as a descriptive comparator. However, utilization of other ADHD treatment alternatives (clonidine, guanfacine, and bupropion) have been reported con stant during our study period as well. 33 This suggests that changes in atomoxetine utilization were caused by effects specific to atomoxetine, and not overall changes in psychotropic medication use or treatment of ADHD. As in previous studies in other drugs with emerging safety concerns, downward trends in utilization followed soon after the first publicized safety concerns and preceded regulatory action such as boxed warnings by a year or more. 65 67,124,142,143 Of particular intere st are parallel reports o f safety concerns of antidepressants and declining antidepressant use, because a tomoxetine is structurally similar to the SSRI fluoxetine. The public safety concerns on antidepressants emerged in October 2003 with the FDA issuing a public health advisory reporting the possibility of an increased risk of suicidality in pediatric patients treated with antidepressant s. As in our

PAGE 146

146 study, downward trends in utilization of antidepressants began in 2003, before the boxed warnings were inclu ded on antidepressant prescriptions in 2005. 11,45 Publicized safety concerns for atomoxetine and antidepressants were conceivably associated with reduced atomoxetine utilization. However, the strengths of this assoc iation cannot be established. In particular, it is impossible to determine to what extend declines in atomoxetine utilization are the result of publicized safety concerns (for atomoxetine and /or antidepressants), an imminent boxed warning, the boxed warnin g itself, atomoxetine effectiveness or the atomoxetine safety profile in general For instance clinical trials suggested smaller effect sizes of atomoxetine compared to stimulants and less long term safety data is available for atomoxetine, which may have option. 44 Although we cannot determine the strengths of an association of atomoxetine utilization decline and the boxed warning, the imminent low utilization of ato moxetine in 2005 and 2006, combined with the infrequency of suicidal ideation suggests a low potential risk of increased suicidality. Moreover, the low potential risk of suicidality suggests no current need for further FDA regulatory action. Part II Althou gh of small magnitude, we noted significant differences in patients who were treated with atomoxetine versus stimulants. A tomoxetine users were more likely to be Caucasian, emphasizing that the choice between ADHD treatments is also influenced by sociodemo graphic characteristics consistent with determinants of any ADHD pharmacotherapy. 15,144

PAGE 147

147 Tic disorder and substance use disorder were the clinical characteristics associated with the strongest preference for atomoxe tine, in concordance with current guidelines. 27,30 However, both disorders were rare (0.8% 1%) resulting in limited predictive ability of treatment choice Fur thermore, atomoxetine users were more likely diagnosed with all other measured mental co morbidities ( except ADHD with hyperactivity and mild mental retardation ) and more likely exposed to all measured co ncomitant medications ( except alpha agonists ). Howev er, the resulting ORs were small or insignificant. Increasing age and alpha agonist exposure were associated with opposing odds of initial and subsequent atomoxetine treatment indicating differences in treatment choice when initial or subsequent therapy i s considered. We conducted a sensitivity analysis requiring at least two diagnoses determining baseline co morbid conditions to explore potential effects of measurement bias The distribution of co morbidities among atomoxetine and st imulant initiators wer e similar, although absolute numbers were smaller, compared to requiring only one diagnosis to define presents of co morbidities with no significant effects on the original prediction model. The sociodemographic and clinical characteristics analyzed from t he automated healthcare claims data available to us might not reflect all information available to a prescribing clinician. For instance, we were unable to measure and subsequently analyze co morbidities and concomitant medication use not present or coded in our data, ADHD and co morbidity severity (e.g. DSM IV score), pharmacotherapy adherence and effectiveness, patient /care taker preferences of treatment, prescriber

PAGE 148

148 specialty, any family history of mental disorder or substance abuse, violence or other str essful life events. Furthermore, t he validity of Medicaid claims data with respect to ADHD diagnosis accuracy, severity or accuracy of atomoxetine and stimulant treatment has not been analyzed so far. Thus, other determinants of treatment choice might exis t, but were immeasurable in our analysis. Part III Our study did not observe a statistically or clinically meaningful difference in the risk of suicide and suicide attempts associated with i nitial or subsequent treatment of youths with atomoxetine compared to stimulants. In newly treated patients, all point estimates were close to one, consistent over time and consistent among current and former use, indicating no excess immediate (within six months of index date) or intermediate (up to two years) risk o f s uicide and suicide attempts. In subsequent treated patients, all comparisons to current stimulant treatment showed a slight protective effect Because such an effect is clinically implausible, we atomoxetine users may have dissipating ADHD and co morbidity severity, resulting in preference of the less effective treatment option. This effect may have been more pronounced in the subsequent atomoxetine users as a result of more and fa ster treatment switching back to stimulant. Additionally, similar protective effects in former atomoxetine and stimulant use reduce concerns of a healthy user effect masking an increased risk of suicidal events associated with atomoxetine. Alternatively, o ur (as well have been limited, which likely is associated with both increased suicidal risk and a patient preference for stimulants.

PAGE 149

149 Although the confidence interva ls of all hazard rations were wide, the upper confidence limits suggest a maximum increased risk of 40 % to 70%. Considering the baseline incidence rate (during stimulant use) of 30 suicide events per 100 000 person years, the unlikely increase of the absol ute risk would be relatively small. Even if general concerns about bias in observational studies are considered, the observed incidence rates provide assurance for minimal absolute risk and thus limited potentials for significant ly increased risk differenc es. Following is a discussion of potential biases in the context of our findings. Covariates C ovariates among the treatment groups were mostly balanced and i nverse weighting of subjects by their exposure propensity score established balanced groups with le ss than 0.5% of absolute difference Reflecting the finding of limited measurable confounding, propensity scores were overlapping Publicized safety concerns conceivably led to increased awareness of suicidality in patients treated with atomoxetine, their care takers and public health professionals. This might result in more thorough analytic work up and disproportioned screening prior to treatment initiation of atomoxetine compared to stimulants, resulting in biased assessment of baseline covariates. Howev er, baseline incidence rates of suicidal ideation and suicide attempt prior t o atomoxetine initiation were higher indicating absence of measurement bias. Furthermore, a sensitivity analysis requiring at least two diagnoses of mental disorders showed lower but not significantly different distributions of covariates among the treatment groups. An additional sens itivity analysis excluding high risk patients with suicidal ideation or suicide attempt during baseline also show ed no significantly different HRs.

PAGE 150

150 S uicidality safety concerns of atomoxetine and antidepressants were actively communicated during most of the study period. Prescribing clinicians undoubtedly adjusted their treatment patterns and patient monitoring accordingly, for instance by excluding hig h risk patients from atomoxetine exposure. Hence, all our inferences on the hazard of suicidality or the lack thereof, are only generalizable to current practice and precautions. We included patients initiating ADHD pharmacotherapy and required only one as sociated mental health diagnosis. Although we accept a higher potential for a misclassification of a commonly associated mental health disorder by requiring only one diagnosis, differential misclassification is much more likely in the comparison of an expo sed and unexposed group. In that case, misclassified patients will accumulate in the unexposed group, because most mis classified patient s will not receive treatment Exposure We cannot exclude that treatment classified as initial treatment was preceded by treatment episodes prior to Medicaid eligibility, especially in older age groups compromising a new user design. Drug utilization was measured from pharmacy prescription claims, which capture only dispensed prescriptions, but not actual drug utilization resulting in some risk of misclassification S timulants are controlled substances and most states permit maximum dispensing of only 30 substance e potential of misclassification However, we observed similar distributions of atomoxetine and stimulants with 9 5 % of resulting in frequent exposure status updates based on prescription refills. Finally, with

PAGE 151

151 a higher abuse potential, stimulants might have a higher propensity of exposure misclassification due to diversion. 145 Because stimulant use is thought to have a neutral effect on suicidal events, expo sure misclassification of stimulants is expected to have less impact on our study validity. It is important to note that our ability to assess an atomoxetine effect was further limited by high rates of treatment discontinuation and switching. In initially and subsequently atomoxetine treated patients, 45% & 64% filled at least one prescription for stimulants after a mean of 271 days (median 176 ) and 138 days (median 43), respectively. In combination with rare suicide events our study struggled with a limite d amount of follow up and events, despite a large sample This resulted in wide confidence intervals and limited power to conduct stratified analyses of high risk groups. Despite sample size constraints our study was able to show significant effects of ant iconvulsants, antipsychotics and antidepressants (not included in final models) on the risk of suicidal events. 80 82 146 Of note, our study design does not allow inferences for any of these drugs as it does not adjust for respective confounder s, but presence of statistically significant associations (even if confounded) supports the general ability to investigate suicidal risk in our sample. Study Endpoints In our study cohort, the majority of suicide events occurred within one year (median 0.9 3 & 0.6 years) In contrast in the Bangs et al. meta analysis all outcomes occurred within 3 2 days. 8 O ther observational studies investigating suicide and suicide attempts associated with antidepressants, observed the majority of events in the first six month after treatment initiation 122

PAGE 152

152 While the positive predictive value of ICD9 CM codes for suicides and suicide attempts is expected to be high, concerns about sensitivity are important. Low sensitivity would bias our results towards the null hypothesis, thus compromising our ability to detect an increased risk of suicidal events associated with atomoxetine. However, o bserved su icide and suicide attempt rates consistent with reported estimates support the validity of our outcome measurement. 71 73 Also, gender distributions among suicides (more likely in males) and suicide attempts ( more likely in females ) and the distribution of method s to commit suicidal events were consistent with previous research. 72 Increased provider attribution of suicidal intent to injury following the publicized safety concerns might have introduced misclassification bias. However, we minimized this bias by including only suicides verified by death certificates and severe suicide attempts requiring hospitalization or ED visits. Hence, our analysis d id not include suicidal ideation, ambiguous events or suicidal attempts without physical harm. 74,147 149 Strength and Limitations Major strengths of our study are a large population based sample size, diverse geogra phic locations a new user design and balanced treatment groups resulting from propensity score adjustment 150 152 In particular, w e analyzed a larger and more generalizable population with longer follow up than the clinical trial populations in the meta analysis leading to the FDA boxed warning. 8 O ur study include d comprehensive administrative data from four years and the 26 U.S. states with the largest fee for service Medica id populations, and is t o our knowledge, the largest study so far to analyze atomoxetine utilization determinants of treatment choice and safety compared to stimulant treatment. Moreover, included

PAGE 153

153 Medicaid patient s overrepresented especially v ulnerable pe diatric populations with complex psychiatric needs and minorities. W e utiliz ed a new user design because the meta analysis leading to the boxed warning suggested immediate effects with time to onset of the suicide related events within 32 days. 8 A new user design reduces the probability of missing early events and over treatment. Lastly, we established balanced treatment groups with requiring an additional mental health diagnosis to minimize confounding by indication and c omprehensive coverage of health care services permitted a broad selection of covariates for propensity score adjustment. Several limitations should be considered in interpreting these findings O ur study validity is limited by its data source and the nature of administrative health data. It is only representative of the Medicaid population, and thus, skewed towards low income groups and minorities, although a large proportion of the investigated age group s are enrolled in Medicaid. 100,150 Furthermore, our study is based on a utomated healthcare claims data intended for reimbursement and not as electronic medical records. Therefore, ICD 9CM codes used for billing practices m ight not accurately reflect the clinical diagnoses and diagnoses were not verified by medical chart review or other clinical assessments D rug utilization was measured from pharmacy prescription claims, which capture only dispensed prescriptions, but not actual drug utilization. Though Medicaid pharmacy claims data have been validated to define psychotropic drug exposure with a positive pr edictive value and negative predictive value of over 85%. 102

PAGE 154

154 Some important covariates influencing treatment decisions and/or suicidality risk cannot be measured with administrative health data. Examples of such covariates are general socioeconomic status, clinical disease severit y (e.g. DSM IV score), over the counter (OTC) drug use, co morbidities not coded or miscoded in our data, pharmacotherapy adherence and effectiveness, prescriber specialty, firearms in the home, hopelessness social difficulties, any family history of ment al disorder or substance abuse, violence ( including physical or sexual abuse ) or other stressful life events. Although this observational research cannot establish nor dismiss a causal relationship between atomoxetine and an increased risk of suicide event s, a randomized controlled trial with enough power to do so is not feasible. Conclusion Quickly after market introduction, atomoxetine treatment initiation rates reach ed a maximum of one third of stimulant initiations Starting late in 2003 atomoxetine ut ilization significant ly decline d until the end of 2006. Concomitantly, utilization of stimulants remained almost constant. The decline in atomoxetine utilization significantly preceded the boxed warning, but corresponded with non regulatory publicized safe ty concerns of atomoxetine and closely related antidepressants. At the time of the actual boxed warning implementation, utilization reached already a decline level off. Clinical practice acts more than one year ahead of regulatory actions. However, it rema ins unknown, if atomoxetine utilization declines were associated with published safety concerns, the boxed warning or effectiveness. In any case, resulting low atomoxetine utilization implies a low potential risk of increased suicidal ideation and no curre nt need for further regulatory action.

PAGE 155

155 Sociodemographic and clinical characteristics were significant determinants of ADHD pharmacotherapy choice T he strongest predictors of atomoxetine initiation were b eing Caucasian, t ic disorder and substance abuse In creasing age and alpha agonist exposure were associated with opposing odds of initial and subsequent atomoxetine treatment, indicating that treatment decisions are somewhat different. Other determinants of treatment choice might be of importance but were n ot quantifiable with our administrative health data. Initial and subsequent treatment of youths with atomoxetine compared to stimulants w as not significantly associated with an increased risk of suicide attempts and suicides in current practice. Current p ractice, precautions and warnings concerning suicidal ideation seem to be sufficient and no extended boxed warning concerning suicidal events for atomoxetine seem s necessary. The small incidence of suicidal events resulted in wide confidence intervals and did not allow stratified analysis of high risk groups.

PAGE 156

156 APPENDIX A M AX DATA The Medicaid Analytic eXtract (MAX) data is organized in five file categories: Inpatient Records (IP), Long Term Care Records (LT), Other Services records (OT), Drug Records (Rx) and Personal Summary Records (PS) files. 98 Files for different states and calendar years were combined to one master file for each category with covariate s indicating the originating state (two letter state abbreviations, see Appendix B) and calendar year of each record. MAX data was restricted to enrollees in fee for service program s to ensure comprehensive availability of all claims. All Medicaid beneficiaries, State Child Health Insurance Program (SCHIP) and Medicaid expansion SCHIP Program (MCHIP) enrollees are included. In case of contradicting race/ethnicity information per patient, the most frequent race/ethnicity category of all eight study years were considered with the exception of race other n this case, if a more specific value was present (e.g. white or black), we selected the more speci fic race/ethnicity information.

PAGE 157

157 APPENDIX B 2 6 LARGEST M EDICAID FEE FOR SERVICE STATES Table B 1 State a bbreviations State Abbreviat ion Arkansas AR Florida FL Georgia GA Idaho ID Illinois IL Indiana IN Iowa IA Kansas KS Louisiana LA Massachusetts MA Minnesota MN Mississippi MS Missouri MO New Hampshire NH New York NY North Carolina NC Ohio OH Pennsylvania PA South Ca rolina SC Tennessee TN Texas TX Vermont VT Virginia VA West Virginia WV Wisconsin WY New Hampshire NH

PAGE 158

158 Table B 2 2 6 largest fee for service s tate e nrollees in 1999 & 2000 Number of all Medicaid beneficiaries (PS), p ercent fee for service enrollees (% FFS) and overall fee for service enrollees (FFS) by state 1999 200 0 Year 1999 2000 State PS % FFS FFS PS % FFS FFS Arkansas 493,696 13.98% 69,038 530,978 14.34% 76,123 Florida 2,135,116 23.47% 501,045 2,334,173 22.48 % 524,699 Georgia 1,330,992 21.86% 290,971 1,520,645 20.07% 305,255 Idaho 134,691 52.44% 70,638 158,350 56.84% 90,001 Illinois 1,729,208 68.34% 1,181,700 1,786,132 72.13% 1,288,382 Indiana 696,978 26.82% 186,899 776,394 23.69% 183,935 Iowa 315,004 15. 15% 47,729 320,120 14.21% 45,487 Kansas 260,006 31.23% 81,189 273,392 27.85% 76,145 Louisiana 824,147 80.36% 662,267 869,535 77.56% 674,431 Massachusetts 1,074,232 20.55% 220,714 1,131,670 22.20% 251,229 Minnesota 596,416 29.40% 175,324 625,397 28.36% 177,368 Mississippi 553,018 47.99% 265,386 610,253 41.73% 254,659 Missouri 898,767 49.76% 447,185 1,008,624 47.33% 477,359 Nebraska 227,661 20.91% 47,593 239,639 20.46% 49,023 New Hampshire 107,127 79.02% 84,652 109,680 78.91% 86,547 New York 3,403,88 8 60.10% 2,045,683 3,425,582 58.18% 1,992,994 North Carolina 1,253,691 25.73% 322,614 1,326,155 24.20% 320,879 Ohio 1,393,580 59.45% 828,441 1,472,784 64.48% 949,596 Pennsylvania 1,696,058 27.24% 462,090 1,683,650 28.43% 478,693 South Carolina 758,480 81.65% 619,270 806,240 79.71% 642,671 Tennessee 1,613,135 0.71% 11,454 1,607,688 0.64% 10,267 Texas 2,741,035 58.94% 1,615,651 2,800,108 51.48% 1,441,414 Vermont 143,603 33.61% 48,261 151,532 38.28% 58,004 Virginia 713,629 55.62% 396,927 742,867 50.75% 377,026 West Virginia 364,904 66.88% 244,063 358,208 50.29% 180,155 Wisconsin 580,098 37.67% 218,548 631,041 35.82% 226,035 Sum of beneficiaries & average of % 27,605,786 41 % 11,593,663 29,009,562 39.1 % 11,631,881

PAGE 159

159 Table B 3 2 6 largest fee for s erv ice s tate e nrollees in 2001 & 200 2 Number of all Medicaid beneficiaries (PS), percent fee for service enrollees (% FFS) and overall fee for service enrollees (FFS) by state 2001 200 2 Year 2001 2002 State PS % FFS FFS PS % FFS FFS Arkansas 587,684 13.12% 77,092 648,072 15.22% 98,632 Florida 2,582,010 22.37% 577,682 2,799,154 20.09% 562,455 Georgia 1,722,114 18.51% 318,692 1,812,987 21.30% 386,220 Idaho 184,566 55.88% 103,134 199,892 40.74% 81,430 Illinois 1,882,237 73. 56% 1,384,609 2,132,554 68.92% 1,469,788 Indiana 855,232 23.04% 197,068 931,109 22.39% 208,487 Iowa 338,145 13.27% 44,857 363,677 12.20% 44,381 Kansas 297,213 27.02% 80,297 314,827 26.26% 82,689 Louisiana 932,216 71.51% 666,614 1,065,810 51.71% 551,112 Massachusetts 1,166,242 25.61% 298,666 1,228,684 26.86% 330,062 Minnesota 677,329 25.07% 169,822 725,283 23.10% 167,515 Mississippi 695,868 37.84% 263,347 723,179 48.04% 347,389 Missouri 1,085,041 45.73% 496,201 1,130,756 46.57% 526,645 Nebraska 254, 831 20.25% 51,600 268,953 57.83% 155,546 New Hampshire 112,602 77.38% 87,127 126,754 71.94% 91,183 New York 3,774,727 55.19% 2,083,388 4,546,162 44.78% 2,035,580 North Carolina 1,473,871 24.56% 361,961 1,505,412 22.95% 345,493 Ohio 1,749,855 61.95% 1,0 83,974 1,839,178 61.31% 1,127,635 Pennsylvania 1,668,793 29.17% 486,746 1,738,198 15.81% 274,866 South Carolina 904,335 77.61% 701,871 933,817 76.52% 714,537 Tennessee 1,693,143 0.56% 9,546 1,730,087 2.97% 51,316 Texas 2,952,397 49.58% 1,463,691 3,376, 384 47.49% 1,603,349 Vermont 159,555 29.77% 47,497 163,438 27.74% 45,341 Virginia 765,098 31.89% 243,986 799,254 26.34% 210,489 West Virginia 359,170 45.48% 163,353 381,079 41.63% 158,662 Wisconsin 690,762 35.13% 242,660 817,420 37.02% 302,627 Sum of beneficiaries & average of % 31,421,841 37 % 12,095,802 34,323,315 36 % 12,380,985

PAGE 160

160 Table B 4 2 6 largest fee for service state e nrollees in 2003 & 2004 Number of all Medicaid beneficiaries (PS), percent fee for service enr ollees (% FFS) and overall fee for service enrollees (FFS) by state 2003 200 4 Year 2003 2004 State PS % FFS FFS PS % FFS FFS Arkansas 688,523 30.87% 212,514 725,408 30.32% 219,918 Florida 2,920,400 18.80% 549,093 2,936,906 17.81% 523,114 Georgia 1,88 7,197 23.45% 442,545 2,016,379 21.21% 427,650 Idaho 211,920 23.21% 49,195 225,025 33.55% 75,491 Illinois 2,246,388 69.42% 1,559,478 2,440,929 71.12% 1,735,892 Indiana 967,366 18.31% 177,088 1,026,274 16.71% 171,518 Iowa 382,894 11.63% 44,520 401,458 11 .08% 44,482 Kansas 367,051 24.78% 90,941 382,505 29.73% 113,724 Louisiana 1,130,663 20.98% 237,165 1,215,412 17.56% 213,431 Massachusetts 1,197,415 22.38% 267,943 1,186,540 22.65% 268,764 Minnesota 762,630 21.14% 161,227 776,857 21.05% 163,551 Mississ ippi 768,803 85.91% 660,491 801,071 84.74% 678,848 Missouri 1,186,348 46.79% 555,112 1,243,452 46.77% 581,590 Nebraska 265,606 56.57% 150,243 264,006 56.54% 149,259 New Hampshire 136,134 72.38% 98,533 142,071 82.41% 117,079 New York 4,698,972 32.25% 1, 515,404 4,967,110 28.42% 1,411,600 North Carolina 1,588,095 22.96% 364,700 1,663,192 22.47% 373,711 Ohio 1,961,986 58.68% 1,151,321 2,044,300 58.16% 1,188,884 Pennsylvania 1,848,912 14.32% 264,790 1,987,856 13.91% 276,554 South Carolina 1,034,679 73.20 % 757,400 1,008,133 73.82% 744,181 Tennessee 1,610,768 1.17% 18,823 1,633,614 0.66% 10,712 Texas 3,746,761 45.80% 1,716,116 2,992,874 42.90% 1,283,825 Vermont 165,479 32.42% 53,656 168,697 86.34% 145,660 Virginia 825,879 26.03% 214,957 876,388 23.79% 2 08,468 West Virginia 388,420 41.29% 160,379 396,039 45.89% 181,758 Wisconsin 926,043 40.41% 374,244 990,431 41.81% 414,138 Sum of beneficiaries & average of % 35,967,433 34. 9 % 12,266,416 36,575,779 37. 2 % 12,126,225

PAGE 161

161 Table B 5 2 6 largest fee for service state enrollees in 2005 & 2006 Number of all Medicaid beneficiaries (PS), percent fee for service enrollees (% FFS) and overall fee for service enrollees (FFS) by state 2005 200 6 Year 2005 2006 State PS % FFS F FS PS % FFS FFS Arkansas 762,193 30.14% 229,711 771,587 29.3 % 225,744 Florida 3,059,524 17.53% 536,456 3,042,834 16.0 % 486,405 Georgia 2,107,826 5.17% 109,078 2,101,713 5.0 % 104,047 Idaho 233,765 39.93% 93,346 230,815 10.4 % 23,972 Illinois 2,612,7 09 70.87% 1,851,515 2,602,268 68.0 % 1,768,145 Indiana 1,060,392 16.13% 171,060 1,067,016 14.6 % 155,659 Iowa 430,985 10.41% 44,880 468,572 13.6 % 63,725 Kansas 361,325 35.50% 128,258 361,498 35.6 % 128,850 Louisiana 1,244,886 18.60% 231,491 1,273,978 21. 3 % 270,987 Massachusetts 1,255,313 22.44% 281,662 1,315,266 22.1 % 290,269 Minnesota 792,366 20.67% 163,767 809,009 22.1 % 178,662 Mississippi 787,537 84.28% 663,761 777,471 79.8 % 620,180 Missouri 1,219,756 29.81% 363,605 1,115,164 26.2 % 292,016 Ne braska 261,841 56.20% 147,152 262,403 55.7 % 146,035 New Hampshire 145,834 82.42% 120,197 148,759 82.7 % 122,988 New York 5,118,695 26.54% 1,358,482 5,123,013 25.1 % 1,284,085 North Carolina 1,729,575 21.98% 380,079 1,782,760 22.4 % 399,873 Ohio 2,127, 176 53.82% 1,144,818 2,157,415 37.1 % 800,902 Pennsylvania 2,038,176 8.76% 178,468 2,111,259 7.9 % 166,324 South Carolina 1,012,763 68.71% 695,864 949,921 63.2 % 600,333 Tennessee 1,607,615 0.15% 2,445 1,499,759 0.3 % 5,082 Texas 4,164,091 25.20% 1,049 ,203 4,184,402 19.6 % 821,411 Vermont 165,860 87.71% 145,468 164,682 88.2 % 145,249 Virginia 928,642 20.85% 193,621 956,145 20.5 % 195,509 West Virginia 393,616 40.05% 157,637 393,607 35.4 % 139,432 Wisconsin 1,034,583 41.35% 427,797 1,051,903 38.8 % 40 8,365 Sum of beneficiaries & average of % 38,755,331 34.8 % 11,272,584 38,889,756 32 % 10,237,129

PAGE 162

162 APPENDIX C N ATIONAL DRUG CODES Table C 1 Atomoxetine active ingredient Active i ngredient Drug clas s atomoxetine hydrochloride A tomoxetine Table C 2 CNS stimulant active ingredients Active i ngredient Drug class amphetamine aspartate, amphetamine sulfate CNS stimulants dexmethylphenidate hydrochloride CNS stimulants dextroa mphetamine sulfate CNS stimulants lisdexamfetamine dimesylate CNS stimulants methylphenidate hydrochloride CNS stimulants

PAGE 163

163 Table C 3 Antidepressant active ingredients Active i ngredient Drug class amitriptyline hydrochloride tri cyclic antidepressants amitriptyline hydrochloride, chlordiazepoxide psychotherapeutic combinations amitriptyline hydrochloride, perphenazine psychotherapeutic combinations amoxapine tricyclic antidepressants bupropion hydrobromide miscellaneous antide pressants, smoking cessation agents bupropion hydrochloride miscellaneous antidepressants, smoking cessation agents citalopram (as citalopram hydrobromide) SSRI antidepressants clomipramine hydrochloride tricyclic antidepressants desipramine hydrochlor ide tricyclic antidepressants desvenlafaxine (as succinate) SSNRI antidepressants doxepin hydrochloride miscellaneous anxiolytics, sedatives and hypnotics, tricyclic antidepressants duloxetine hydrochloride SSNRI antidepressants escitalopram oxalate SS RI antidepressants fluoxetine hydrochloride SSRI antidepressants fluoxetine, olanzapine psychotherapeutic combinations fluvoxamine maleate SSRI antidepressants imipramine hydrochloride tricyclic antidepressants imipramine pamoate tricyclic antidepress ants maprotiline hydrochloride tetracyclic antidepressants mirtazapine tetracyclic antidepressants nefazodone hydrochloride phenylpiperazine antidepressants nortriptyline hydrochloride tricyclic antidepressants paroxetine hydrochloride SSRI antidepres sants paroxetine mesylate SSRI antidepressants protriptyline hydrochloride tricyclic antidepressants sertraline hydrochloride SSRI antidepressants trazodone hydrochloride phenylpiperazine antidepressants trimipramine maleate tricyclic antidepressants venlafaxine hydrochloride SSNRI antidepressants

PAGE 164

164 Table C 4 Anticonvulsant active ingredients Active i ngredient Drug class carbamazepine dibenzazepine anticonvulsants clonazepam benzodiazepine anticonvulsants, benzodiazepines cl orazepate dipotassium benzodiazepines diazepam benzodiazepine anticonvulsants, benzodiazepines divalproex sodium fatty acid derivative anticonvulsants divalproex sodium (as valproic acid) fatty acid derivative anticonvulsants ethosuximide succinimide a nticonvulsants ethotoin hydantoin anticonvulsants felbamate carbamate anticonvulsants fosphenytoin (as phenytoin equivalent) hydantoin anticonvulsants gabapentin gamma aminobutyric acid analogs lacosamide miscellaneous anticonvulsants lamotrigine tri azine anticonvulsants levetiracetam pyrrolidine anticonvulsants lorazepam benzodiazepine anticonvulsants, benzodiazepines mephenytoin hydantoin anticonvulsants mephobarbital barbiturate anticonvulsants, barbiturates methsuximide succinimide anticonvul sants oxcarbazepine dibenzazepine anticonvulsants paramethadione oxazolidinedione anticonvulsants phenacemide urea anticonvulsants phenobarbital barbiturate anticonvulsants, barbiturates phenobarbital sodium barbiturate anticonvulsants, barbiturates phensuximide succinimide anticonvulsants phenytoin group I antiarrhythmics, hydantoin anticonvulsants phenytoin sodium group I antiarrhythmics, hydantoin anticonvulsants pregabalin gamma aminobutyric acid analogs primidone barbiturate anticonvulsants rufinamide dibenzazepine anticonvulsants tiagabine hydrochloride gamma aminobutyric acid reuptake inhibitors topiramate carbonic anhydrase inhibitor anticonvulsants trimethadione oxazolidinedione anticonvulsants valproate sodium fatty acid derivative a nticonvulsants valproic acid fatty acid derivative anticonvulsants zonisamide carbonic anhydrase inhibitor

PAGE 165

165 Table C 5 Antipsychotic active ingredients Active i ngredient Drug class amitriptyline hydrochloride, chlordiazepoxide psy chotherapeutic combinations amitriptyline hydrochloride, perphenazine psychotherapeutic combinations aripiprazole atypical antipsychotics asenapine atypical antipsychotics chlorpromazine phenothiazine antiemetics, phenothiazine antipsychotics chlorpro mazine hydrochloride phenothiazine antiemetics, phenothiazine antipsychotics clozapine atypical antipsychotics fluoxetine, olanzapine psychotherapeutic combinations fluphenazine decanoate phenothiazine antipsychotics fluphenazine enanthate phenothiazin e antipsychotics fluphenazine hydrochloride phenothiazine antipsychotics haloperidol miscellaneous antipsychotic agents haloperidol decanoate miscellaneous antipsychotic agents haloperidol lactate miscellaneous antipsychotic agents lamotrigine triazin e anticonvulsants loxapine hydrochloride miscellaneous antipsychotic agents loxapine succinate miscellaneous antipsychotic agents mesoridazine besylate phenothiazine antipsychotics methotrimeprazine phenothiazine antipsychotics molindone hydrochloride miscellaneous antipsychotic agents olanzapine atypical antipsychotics paliperidone atypical antipsychotics paliperidone palmitate atypical antipsychotics perphenazine phenothiazine antiemetics, phenothiazine antipsychotics pimozide miscellaneous anti psychotic agents prochlorperazine phenothiazine antiemetics, phenothiazine antipsychotics prochlorperazine edisylate phenothiazine antiemetics, phenothiazine antipsychotics prochlorperazine maleate phenothiazine antiemetics, phenothiazine antipsychotics promazine hydrochloride phenothiazine antiemetics, phenothiazine antipsychotics quetiapine fumarate atypical antipsychotics risperidone atypical antipsychotics thioridazine phenothiazine antipsychotics thioridazine hydrochloride phenothiazine antipsy chotics thiothixene thioxanthenes thiothixene hydrochloride thioxanthenes

PAGE 166

166 Table C 5. Continued Active i ngredient Drug class trifluoperazine hydrochloride phenothiazine antipsychotics triflupromazine hydrochloride phenothiazine antiemetics, phenothia zine antipsychotics ziprasidone hydrochloride monohydrate atypical antipsychotics ziprasidone mesylate atypical antipsychotics

PAGE 167

167 Table C 6 A nxiolytic (including sedatives and hypnotics) active ingredients Active i ngredient Drug cla ss alprazolam benzodiazepines amobarbital sodium barbiturates amobarbital sodium, secobarbital sodium barbiturates buspirone hydrochloride miscellaneous anxiolytics, sedatives and hypnotics butabarbital sodium barbiturates butalbital barbiturates ch loral hydrate miscellaneous anxiolytics, sedatives and hypnotics chlordiazepoxide benzodiazepines chlordiazepoxide hydrochloride benzodiazepines clorazepate dipotassium benzodiazepines dexmedetomidine hydrochloride miscellaneous anxiolytics, sedatives and hypnotics diazepam benzodiazepine anticonvulsants, benzodiazepines estazolam benzodiazepines eszopiclone miscellaneous anxiolytics, sedatives and hypnotics flurazepam hydrochloride benzodiazepines halazepam benzodiazepines hydroxyzine hydrochlori de antihistamines, miscellaneous anxiolytics, sedatives and hypnotics hydroxyzine pamoate antihistamines, miscellaneous anxiolytics, sedatives and hypnotics meprobamate miscellaneous anxiolytics, sedatives and hypnotics midazolam benzodiazepines midazo lam hydrochloride benzodiazepines midazolam hydrochloride, sodium chloride benzodiazepines oxazepam benzodiazepines pentobarbital barbiturates pentobarbital sodium barbiturates quazepam benzodiazepines ramelteon miscellaneous anxiolytics, sedatives a nd hypnotics secobarbital sodium barbiturates temazepam benzodiazepines triazolam benzodiazepines zaleplon misce llaneous anxiolytics, sedatives and hypnotics zolpidem tartrate miscellaneous anxiolytics, sedatives &

PAGE 168

168 Table C 7 Alpha agonist active ingredients Active i ngredient Drug class clonidine antiadrenergic agents, centrally acting clonidine hydrochloride antiadrenergic agents, centrally acting guanfacine hydrochloride antiadrenergic agents, centrally acting Table C 8 Lithium active ingredients Active i ngredient Drug class lithium carbonate miscellaneous antipsychotic agents lithium citrate miscellaneous antipsychotic agents

PAGE 169

169 APPENDIX D O PERATIONAL DEFINITIO NS Table D 1 Index d ate Covariate Operational d efinition Attention Deficit/Hyperactivity Disorder At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis: 314, 314.0, 314.00, 314.01 E xc lusion: 314.1, 314.2, 314.8, 314 .9 Atomoxetine exposure At least one pharmacy claim with NDC for active ingredient: atomoxetine HCl CNS stimulant exposure At least one pharmacy claim with NDC for active ingredient: amphetamine aspartate/ amphetamine sulfate, dexmethylphenidate hydroch loride, dextroamphetamine sulfate, lisdexamfetamine dimesylate, methylphenidate hydrochloride Note: National Drug Code (NDC) for active ingredient was translated with Multum Lexicon database Table D 2 M ental health diagnoses assoc iated with ADHD pharmacotherapy Variable Operational d efinition Adjustment reaction Disturbance of conduct, not elsewhere classified Other or mixed emotional disturbances of childhood or adolescence Unspecified emotional disturbance of childh ood or adolescence At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: 309, 309.0, 309.00, 309.1, 309.2x, 309.3, 309.4. 309.8x, 309.9 At least one of the following ICD9 CM code at any in patient or outpatient d iagnosis field: 312, 312.0, 312.00, 312.1, 312.2, 312.4, 312.8x, 312.9 not 312.3x (312.3x were not included, b ecause this include d : pathological gambling, kl eptomania, pyromania At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis f ield: 313.8, 313.81, 313.82, 313.83, 313.89 At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: 313.9

PAGE 170

170 Table D 3 Exclusion c riteria Covariate Operational d efinition Pemoline exposure At least one pharmacy claim with NDC for active ingredient: pemoline Methamphetamine HCl exposure At least one pharmacy claim with NDC for active ingredient: Methamphetamine HCl MAO I exposure At least one pharmacy claim with NDC for active ingredient: any MAO I Human Immunodeficiency Virus (HIV) Infection At least one inpatient or outpatient with the following ICD9 CM code at any d iagnosis field: 042.xx Malignant neoplasm At least one inpatient or outpatient with the following ICD9 CM code at any d iagnosis fiel d: 140.xx to 208.xx; 209.0x to 209.3; 230.xx to 234.xx Any organ or tissue replaced by transplant At least one inpatient or outpatient claims with the following ICD9 CM V code a at any d iagnosis field: V42, V42.0,V42.1, V42.2, V42.3, V42.4, V42.5, V42.6, V 42.7, V42.8, V42.9 Any dialysis dependency At least one inpatient or outpatient claims with the following ICD9 CM V code a at any d iagnosis field: (V56.0 x ) a S upplementary classification of factors influencing health status and contact with health service s Note: National Drug Code (NDC) for active ingredient was translated with Multum Lexicon database Table D 4 Censoring (exclusion during baseline) Covariate Operational d efinition Hospitalization > 30 days Any Inpatient hospitali zation if rending date of service admission date >30 days Death, not study endpoint Any ICD 10 code NOT: X60 X84, Y10 34, S00 T78, V01 V99, W00 X59 Pregnancy At least one inpatient or outpatient claims with the following ICD9 CM V code a at any d iagno sis field:V22.x a a S upplementary classification of factors influencing health status and contact with health services

PAGE 171

171 Table D 5 Study drug e xposure Covariate Operational d efinition Atomoxetine exposure At least one pharmacy claim with NDC for active ingredient: atomoxetine HCl CNS stimulant exposure At least one pharmacy claim with NDC for active ingredient: amphetamine aspartate/ amphetamine sulfate, dexmethylphenidate hydrochloride, dextroamphetamine sulfate, lisdexamfetamine d imesylate, methylphenidate hydrochloride Note: National Drug Code (NDC) for active ingredient was translated with Multum Lexicon database Table D 6 Study e ndpoints Covariate Operational d efinition Suicide Any ICD 10 code X60 X8 4 Suicide attempt Any ICD 9 CM E code E950.x E959.x

PAGE 172

172 Table D 7 Potential confounders and other covariate s Covariate Operational d efinition Age Year of age Gender Male, Female Race/Ethnicity White, Hispanic, Black, Other Calenda r year Year of index date Calendar month Month of index date State State of Medicaid eligibility at index date ADHD with h yperactivity At least one inpatient or outpatient claim with the following ICD9 CM code at any d iagnosis field: 314.01 ADHD wit h out h yperactivity At least one inpatient or outpatient claim with the following ICD9 CM code at any d iagnosis field: 314.0 0 Substance use disorder At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: 291xx 292xx, 303 xx 305xx Anxiety At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: 300.0, 300.2, 300.3, 308.XX, 309.21, 309.81 Bipolar disorder At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis f ield: 296.XX, 301.13 Schizophrenia At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: 295.XX Depression At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: 296.2 296.3, 300.4, 311.xx Pervasive development disorder At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: 299.xx Mild mental retardation At least one of the following ICD9 CM code at any inpatient or outpatient diagnosis field: 31 8.0 (any=318.0 2) Autism At least one inpatient or two outpatient claims with the following ICD9 CM code at any Diagnosis field: 299.xx Severe mental retardation At least one of the following ICD9 CM code at any inpatient or outpatient diagnosis field: 3 18.1, 318.2

PAGE 173

173 Table D 7. Continued Covariate Operational d efinition Oppositional defiant disorder At least one of the following ICD9 CM code at any inpatient or outpatient diagnosis field: 313.8 Psychosis At least one of the following ICD9 CM code at a ny inpatient or outpatient d iagnosis field: 297.xx, 298.xx At least one of the following ICD9 CM code at any inpatient or outpatient d iagnosis field: not mentioned before within 290.xx 319.xx Number of mental co morbid ities At least one pharmacy claim with NDC for active ingredient Antidepressants exposure At least one pharmacy claim with NDC for active ingredient Anticonvulsants exposure At least one pharmacy claim with NDC for active ingredient Antipsychotics expos ure At least one pharmacy claim with NDC for active ingredient Anxiolytics (including sedatives and hypnotics) exposure At least one pharmacy claim with NDC for active ingredient Alpha agonists exposure At least one pharmacy claim with NDC for active ing redient Lithium exposure At least one pharmacy claim with NDC for active ingredient Opiate a nalgesics At least one pharmacy claim with NDC for active ingredient Number psychotropic medication categories Number of distinct psychotropic drug classes with at least one pharmacy claim with NDC for active ingredient Reason for Medicaid eligibility TANF Any TANF eligibility during baseline Foster care Any Foster care eligibility during baseline SSI Any SSI eligibility during baseline Number of non mental health hospitalization Number of hospitalizations with non mental health diagnoses Number of mental health hospitalizations Number of hospitalizations with mental health diagnoses Prior Suicide Attempt Any inpatient or ED ICD9 CM E code E950.x E959.x P rior Suicidal ideation Any inpatient or outpatient claims with ICD9 CM diagnosis between cohort entry and index date: V62.84 or 300.9

PAGE 174

174 LIST OF REFERENCES 1. Guevara J, Lozano P, Wickizer T. Utilization and cost of health careservices f or children with attention deficit/hyperactivity disorder. Pediatrics 2001:108:71 8. 2. Centers for Disease Control and Prevention. Increasing prevalence of parent reported attention deficit/hyperactivity disorder among children United States, 2003 and 2 007. MMWR Morb Mortal Wkly Rep. United States2010:1439 43. 3. Bhatara V, Aparasu R. Pharmacotherapy with atomoxetine for US children and adolescents. Ann Clin Psychiatry 2007;19(3). 4. Report of the committee for medicinal products for human use: Increase d risk of suicidal thoughts and behavior in children treated with atomoxetine resulting in updated warnings added to product information. European Medicines Agency (EMEA), 2005. (Accessed 2012, September, at http://www.ema.europa.eu/docs/en_GB/document_library/Referrals_document/S SRI_31/WC500013126.pdf. ) 5. Public Health Advisory Suicidal thinking in children and adolescents being treated with Strattera (atomoxetine). Center for Drug Evaluation and Research, September 2005. (Accessed July, 2012, at http://www.fda.gov/Safety/Me dWatch/SafetyInformation/SafetyAlertsforHumanMe dicalProducts/ucm152628.htm. ) 6. Postmarket Drug Safety Information for Patients and Providers Atomoxetine (marketed as Strattera) Information. Eli Lilly and Company & Center for Drug Evaluation and Resear ch, 2009. (Accessed March, 2012, at http://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatien tsandProviders/ucm10791 2.htm http://www.accessdata.fda.gov/drugsatfda_docs/label/2008/021411s024s025s02 6lbl.pdf. ) 7. Clinical Pharmacology Online Atomoxetine. Elsevier / Gold Standard, 2009. (Accessed September, 2012, at http://www.clinicalpharmacology ip.com/Forms/Monograph/monograph.aspx?cpnum=2730&sec=mondesc. ) 8. B angs M, Tauscher Wisniewski S, Polzer J, et al. Meta analysis of suicide related behavior events in patients treated with atomoxetine. J Am Acad Child Adolesc Psychiatry 2008;47:209 18. 9. Public Health Advisory Suicidality in Children and Adolescents Be ing Treated With Antidepressant Medications. Center for Drug Evaluation and Research, 2004. (Accessed February, 2012, at http://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatien tsandProviders/DrugSafetyInformationforHeathcareProfessionals/PublicHealthAd visories/ucm161679.htm. )

PAGE 175

175 10. Hammad T, Laughre n T, Racoosin J. Suicidality in pediatric patients treated with antidepressant drugs. Arch Gen Psychiatry 2006;63(3). 11. Banaschewski T, Roessner V, Dittmann R, Santosh P, Rothenberger A. Non stimulant medications in the treatment of ADHD. Eur Child Adole sc Psychiatry 2004;13 Suppl 1:I102 16. 12. Injury Suicide Prevention. Centers for Disease Control and Prevention, 2009. (Accessed October, 2011, at http://www.cdc.gov/violencep revention/pub/youth_suicide.html. ) 13. Mula M, Hesdorffer D. Suicidal behavior and antiepileptic drugs in epilepsy: analysis of the emerging evidence. Drug Healthc Patient Saf 2011;3:15 20. 14. Beck A, Steer R, Kovacs M, Garrison B. Hopelessness and event ual suicide: a 10 year prospective study of patients hospitalized with suicidal ideation. Am J Psychiatry 1985;142:559 63. 15. Winterstein A, Gerhard T, Shuster J, et al. Utilization of Pharmacologic Treatment in Youths with ADHD in Florida Medicaid 1995 2 004. Psychiatry 2008:24 31. 16. Attention Deficit / Hyperactivity Disorder (ADHD): Data & Statistics in the United States. Centers for Disease Control and Prevention, 2010. (Accessed September, 2012, at http://www.cdc.gov/ncbddd/adhd/data.html. ) 17. Harris G. FDA Requiring Suicide Studies in Drug Trials. The New York Times 2008;January 24:A1. 18. Mundy A. Drugs' Links To Suicide Risk Draw Concern. Wall Street Journal 2008;July 9th. 19. Iribarren C, Sidney S, Jacobs D, Weisner C. Hospitalization for suicide attempt and completed suicide. Soc Psychiatry Psychiatr Epidemiol 2000;35:288 96. 20. Lang H, Scheffler R, Hu T. The discrepancy in attention deficit hyperactivity disorder (ADHD) medications diff usion: 1994 2003 -a global pharmaceutical data analysis. Health Policy 2010;Sep;97(1). 21. Polanczyk G, de Lima M, Horta B, Biederman J, Rohde L. The worldwide prevalence of ADHD: a systematic review and metaregression analysis. Am J Psychiatry 2007;164. 2 2. Fayyad J, De Graaf R, Kessler R, et al. Cross national prevalence and correlates of adult attention deficit disorder. Br J Psychiatry 190 2007;190.

PAGE 176

176 23. American Academy of Child & Adolescent Psychiatry (AACAP). Practice Parameter for the Assessment and Treatment of Children and Adolescents With Attention Deficit/Hyperactivity Disorder. Am Acad Child Adolesc Psychiatry 2007;46(7). 24. The Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM IV). American Psychiatric Association, 201 2. at http://www.psych.org/practice/dsm. ) 25. Elia J AP, Rapoport JL,. Treatment of Attention Deficit Hyperactivity Disorder. N Engl J Med 1999;340:780 8. 26. DiScala C, Lescohier I, Barthel M, Li G. Inju ries to children with attention deficit hyperactivity disorder. Pediatrics 1998;102:1415 21. 27. American Academy of Pediatrics. Clinical Practice Guideline: Treatment of the school aged child with ADHD. Pediatrics 2001:108(4):1033 44. 28. American Academy of Child & Adolescent Psychiatry (AACAP). Practice Parameter for the Assessment and Treatment of Children and Adolescents With Attention Deficit/Hyperactivity Disorder. Am Acad Child Adolesc Psychiatry, 2007;46(7). 29. Faraone S, Biederman J, Weber W, Rus sell R. Psychiatric, neuropsychological, and psychosocial features of DSM IV subtypes of attention deficit/hyperactivity disorder: results from a clinically referred sample. J Am Acad Child Adolesc Psychiatry 1998;37:185 93. 30. American Academy of Child & Adolescent Psychiatry (AACAP). Practice parameters for the assessment and treatment of children, adolescents, and adults with attention deficit/hyperactivity disorder. J Am Acad Child Adolesc Psychiatry 1997;36(10suppl):85S 121S. 31. Hinshaw S, Elliott G, Conners C, et al. A 14 month randomized clinical trial of treatment strategies for attention deficit/hyperactivity disorder. The MTA Cooperative Group. Multimodal Treatment Study of Children with ADHD. Arch Gen Psychiatry 1999;56:1073 86. 32. Prasad S, St eer C. Switching from neurostimulant therapy to atomoxetine in children and adolescents with attention deficit hyperactivity disorder : clinical approaches and review of current available evidence. Paediatr Drugs 2008;10:39 47. 33. Garfield C, Dorsey E, Zh u S, et al. Trends in Attention Deficit Hyperactivity Disorder Ambulatory Diagnosis and Medical Treatment in the United States, 2000 2010. 2012;12:110 6.

PAGE 177

177 34. Lerner M, Wigal T. Long term safety of stimulant medications used to treat children with ADHD. J P sychosoc Nurs Ment Health Serv 2008;46:38 48. 35. King S, Griffin S, Hodges Z, et al. A systematic review and economic model of the effectiveness and cost effectiveness of methylphenidate, dexamfetamine and atomoxetine for the treatment of attention defici t hyperactivity disorder in children and adolescents. Health Technol Assess 2006;10:iii iv, xiii 146. 36. Brown R, Amler R, Freeman W, et al. Treatment of Attention Deficit/Hyperactivity Disorder: Overview of the Evidence. Pediatrics 2005;115:749 57. 37. H ealthCanada. Health Canada Endorsed Important Safety Information on Adderall XR. 2005. 38. HealthCanada. Report of the "Adderall XR New Drug Committee". 2005. 39. HealthCanada. Health Canada Important Safety Information on ADDERALL XR TM (amphetamine salts ). 2005. 40. Nissen SE. ADHD Drugs and Cardiovascular Risk. N Engl J Med 2006;354:1445 8. 41. Newsbrief. MT. FDA Pediatric Advisers Say Nay to ADHD Drug Black Box. 2006. 42. Kernan WN, Viscoli CM, Brass LM. Phenylpropanolamine and the risk of hemorrhagic s troke. N Engl J Med 2000;343:1826 32. 43. FDA. FDA Directs ADHD Drug Manufacturers to Notify Patients about Cardiovascular Adverse Events and Psychiatric Adverse Events 2007. 44. Gibson A, Bettinger T, Patel N, Crismon M. Atomoxetine Versus Stimulants for Treatment of Attention Deficit/Hyperactivity Disorder. 2006. 45. Michelson D, Adler L, Spencer T, et al. Atomoxetine in adults with ADHD: two randomized, placebo controlled studies 2003;53:112 20. 46. Public Health Advisory Worsening Depression and Su icidality in Patients Being Treated With Antidepressants. Center for Drug Evaluation and Research, 2004. (Accessed July, 2012, at http://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatien tsandProviders/DrugSafetyInformationforHeathcareProfessionals/PublicHealthAd visories/ucm161696.htm. ) 47. Paxton G Cranswick N. Acute suicidality after commencing atomoxetine. J Paediatr Child Health 2008;Oct;44 (10). 48. Markx S, Kahn D. An 18 year old woman with new onset suicidal ideation while being treated with atomoxetine. J Psychiatr Pract 2008;Jan;14 (1).

PAGE 178

178 49. Graham J, Banaschewski T, Buitelaar J, et al. European guidelines on managing adverse effects of medication for ADHD. Eur Child Adolesc Psychiatry 2011;20:17 37. 50. Meltzer H, Harrington R, Goodman R, Jenkins R. Children and adolescents who try to harm, hurt or kill themselves. A report of further analysis from the national survey of the mental health of children and adolescents in Great Britain in 1999. National Statistics 1999;2012:15 49. 51. McCarthy S, Cranswick N, Potts L, Taylor E, Wong I. Mortali ty Associated with Attention Deficit Hyperactivity Disorder (ADHD) Drug Treatment: A Retrospective Cohort Study of Children, Adolescents and Young Adults Using the General Practice Research Database. Drug Safety 2009:Volume 32 Issue 11 pp 1089 96. 52. Centers for Disease Control and Prevention. Mental health in the United States. Prevalence of diagnosis and medication treatment for attention deficit/hyperactivity disorder United States, 2003. MMWR Morb Mortal Wkly Rep 2005;54:842 7. 53. Chen C, Gerhard T, Winterstein A. Determinants of initial pharmacological treatment for youths with attention deficit/hyperactivity disorder. J Child Adolesc Psychopharmacol 2009;19:187 95. 54. Stevens J, Harman J, Kelleher K. Race/ethnicity and insurance status as factor s associated with ADHD treatment patterns. J Child Adolesc Psychopharmacol 2005;15:88 96. 55. Zarin D, Suarez A, Pincus H, Kupersanin E, Zito J. Clinical and treatment characteristics of children with attention deficit/hyperactivity disorder in psychiatric practice. J Am Acad Child Adolesc Psychiatry 1998;37:1262 70. 56. Cox D, Motheral B, Henderson R, Mager D. Geographic variation in the prevalence of stimulant medication use among children 5 to 14 years old: results from a commercially insured US sample. Pediatrics 2003;111:237 43. 57. Miller A, Lalonde C, McGrail K. Children's persistence with methylphenidate therapy: a population based study. Can J Psychiatry 2004;49:761 8. 58. Thiruchelvam D, Charach A, Schachar R. Moderators and mediators of long term adherence to stimulant treatment in children with ADHD. J Am Acad Child Adolesc Psychiatry 2001;40:922 8. 59. Gau S, Chen S, Chou W, et al. National survey of adherence, efficacy, and side effects of methylphenidate in children with attention deficit/hyper activity disorder in Taiwan. J Clin Psychiatry 2008;69:131 40.

PAGE 179

179 60. Faraone S, Biederman J, Zimmerman B. An analysis of patient adherence to treatment during a 1 year, open label study of OROS methylphenidate in children with ADHD. J Atten Disord 2007;11:15 7 66. 61. dosReis S, Zito J, Safer D, Soeken K. Mental health services for youths in foster care and disabled youths. Am J Public Health 2001;91:1094 9. 62. Olfson M, Marcus S, Zhang H, Wan G. Continuity in methylphenidate treatment of adults with attentio n deficit/hyperactivity disorder. J Manag Care Pharm 2007;13:570 7. 63. Marcus S, Wan G, Kemner J, Olfson M. Continuity of methylphenidate treatment for attention deficit/hyperactivity disorder. Arch Pediatr Adolesc Med 2005;159:572 8. 64. Guo J, Curkendal l S, Jones J, Fife D, Goehring E, She D. Impact of cisapride label changes on codispensing of contraindicated medications. Pharmacoepidemiol Drug Saf 2003;12:295 301. 65. Olfson M, Marcus S, Druss B. Effects of Food and Drug Administration warnings on anti depressant use in a national sample. Arch Gen Psychiatry 2008;65:94 101. 66. Valuck R, Libby A, Orton H, Morrato E, Allen R, Baldessarini R. Spillover effects on treatment of adult depression in primary care after FDA advisory on risk of pediatric suicidal ity with SSRIs. Am J Psychiatry 2007;164:1198 205. 67. Libby A, Brent D, Morrato E, Orton H, Allen R, Valuck R. Decline in treatment of pediatric depression after FDA advisory on risk of suicidality with SSRIs. Am J Psychiatry 2007;164:884 91. 68. Hersh A, Stefanick M, Stafford R. National use of postmenopausal hormone therapy: annual trends and response to recent evidence. JAMA 2004;291:47 53. 69. Starner C, Schafer J, Heaton A, Gleason P. Rosiglitazone and pioglitazone utilization from January 2007 throug h May 2008 associated with five risk warning events. J Manag Care Pharm 2008;14:523 31. 70. Posner K, Oquendo M, Gould M, Stanley B, Davies M. Columbia Classification Algorithm of Suicide Assessment (C CASA): classification of suicidal events in the FDA's pediatric suicidal risk analysis of antidepressants. Am J Psychiatry 2007;164:1035 43. 71. Reducing suicide: a national imperative. Washington (DC): National Academy Press, 2002. (Accessed September, 2012, at http://www.nap.edu/openbook.php?record_id=10398&page=R1. )

PAGE 180

180 72. Injury Prevention & Control: Data & Statistics Web based Injury Statistics Query and Reporting System (WISQARS). Centers for Disease Control and Prevention, 2002 200 6. (Accessed September, 2012, at http://www.cdc.gov/injury/wisqars/fatal.html. ) 73. James A, Lai F, Dahl C. Attention deficit hyperactivity disorder and suicide: a review of possible associatio ns. Acta Psychiatr Scand 2004;110(6). 74. Moscicki E. Epidemiology of completed and attempted suicide: toward a framework for prevention. Clinical Neuroscience Research 2001;1:310 23. 75. Suicide in the U.S.: Statistics and Prevention. National Institute o f Mental Health, 2010. (Accessed September, 2012, at http://www.nimh.nih.gov/health/publications/suicide in the us statistics and preventio n/index.shtml. ) 76. Schneeweiss S, Patrick A, Solomon D, et al. Variation in the risk of suicide attempts and completed suicides by antidepressant agent in adults: a propensity score adjusted analysis of 9 years' data. Arch Gen Psychiatry 2010;67:497 506. 77. Daviss W. A review of co morbid depression in pediatric ADHD: etiology, phenomenology, and treatment. J Child Adolesc Psychopharmacol 2008:Dec;18(6):565 71. 78. Olfson M, Marcus S, Weissman M, Jensen P. National trends in the use of psychotropic medic ations by children. J Am Acad Child Adolesc Psychiatry 2002;41:514 521. 79. Public Health Advisory Reports of Suicidality in Pediatric Patients Being Treated with Antidepressant Medications for Major Depressive Disorder (MDD). Center for Drug Evaluation and Research, 2003. (Accessed July, 2012, at http://www.fda.gov/ Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatien tsandProviders/DrugSafetyInformationforHeathcareProfessionals/PublicHealthAd visories/ucm168828.htm. ) 80. Psychopharmacologic Drugs Advisory CommitteeMeeting Review of results analy sis on antidepressants and suicidality in adult patients and for Drug Evaluation and Research, December 2006. (Accessed September, 2012, at http://www.fda.gov/ohrms/dockets/ac/06/briefing/2006 4272b1 01 FDA.pdf. ) 81. Gibbons R, Brown C, Hur K, Davis J, Mann J. Suicidal thoughts and behavior with antidepressant treatment: reanalysis of the ra ndomized placebo controlled studies of fluoxetine and venlafaxine. Arch Gen Psychiatry 2012;69:580 7.

PAGE 181

181 82. Postmarket Drug Safety Information for Patients and Providers Information for Healthcare Professionals: Suicidal Behavior and Ideation and Antiepil eptic Drugs. Center for Drug Evaluation and Research, 2008. at http://www.fda.gov/Drugs/DrugSafety/PostmarketDrugSafetyInformationforP atien tsandProviders/ucm100192.htm. ) 83. International Classification of Diseases, Ninth Revision (ICD 9) Classification of Diseases, Functioning, and Disability. Centers for Disease Control and Prevention \ National Center for Health Statistics, 2009. (Accessed September, 2012, at http://www.cdc.gov/nchs/icd/icd9.htm. ) 84. Patorno E, Bohn L, Wahl P, et al. Anticonvulsant Medications and the Risk of Suicide, Attempted Suicide, or Violent Death. JAMA 2010;303:1401 9. 85. VanCott A, Cramer J, Copeland L, Zeber J, Steinman M, Dersh J. Suicide related behaviors in older patients with new anti epileptic drug use: data from the VA hospital system. BMC Med 2010;Jan 11. 86. Moyer L, Boyle C, Pollock D. Valid ity of death certificates for injury related causes of death. AM J Epidemiology 1989;130(5). 87. Gjertsen F, Bruzzone S, Vollrath M, Pace M, Ekeberg O. Comparing ICD 9 and ICD 10: The impact on intentional and unintentional injury mortality statistics in I taly and Norway. Injury 2012. 88. Ray W, Hall K, Meador C. Racial differences in antidepressant treatment preceding suicide in a Medicaid population. Psychiatr Serv 2007:58:1317 23. 89. Ohlberg A, Lonnquist J. Suicides Hidden among undetermined deaths. Act a Psychiatr Scand 1998;98(3). 90. McKenzie K, Enraght Moony E, Walker S, McClure R, Harrison J. Accuracy of external cause of injury coding in hospital records. Inj Prev 2009;Feb;15(1). 91. Healthcare Cost and Utilization Project (HCUP)E Code Evaluation Re port, Methods Series Report # 2004 06. Agency for Healthcare Research and Quality (AHRQ), 2004. (Accessed March, 2012, at http://www.hcup us.ahrq.gov/reports/methods/2004_06Ad dendum.pdf. ) 92. MINI SENTINEL systematic evaluation of health outcome of interest definitions for studies using administrative data Suicide Report. 2010. (Accessed April, 2012, at http://www.mini sentinel.org/work_products/HealthOutcomes/MS_HOI_SuicideReport.pdf. ) 93. SAS | Business Analytics and Business Intelligence Software, Version 9.2. SAS Institute, 2012. (Accessed September, 2012, at http://www.sas.com/. )

PAGE 182

182 94. Microsoft Excel, Version 2010. Microsoft Corporation, 2012. (Accessed September, 2012, at http://office.microsoft.com/en us/. ) 95. Nobre F, Monteiro A, Telles P, Williamson G. Dynamic linear model and SARIMA: a comparison of their forecasting performance in epidemiology. Stat Med 2001;20:3051 69. 96. Joinpoint Regression Software, software version 3.5.1. National Cancer Institute, 2011. at http://surveillance.cancer.gov/joinpoint/. ) 97. Ray W. Evaluating medication effects outside of clinical trials: new user designs. Am J Epidemiol 2003;158:915 20. 98. Medicaid Analytic eXtract (MAX) General Inf ormation. US Department of Health and Human Services, 2011. (Accessed December, 2011, at http://www.cms.gov/MedicaidDataSourcesGenInfo/07_MAXGeneralInformation.a sp. ) 99. ChildStats.gov. Federal Interagency Forum on Child and Family Statistics, 2011. (Accessed August, 2011, at http://www.childstats.gov/forum/. ) 100. Kaiser State Health Facts. Kaiser Family Foundat ion / statehealthfacts.org, 2007. (Accessed August, 2011, at http://www.statehealthfacts.org/. ) 101. "Map Maker" Utility. Do it yourself Maps, 2012. (Accessed September, 2012, at http://diymaps.net. ) 102. McKenzie D, Semradek J, McFarland B, Mullooly J, McCamant L. The validity of medicaid pharmacy claims for estimating drug use among elderly nursing home residents: The Oregon experience. Dec 2000. J Clin Epidemiol 2000; 53(12). 103. Winterstein A, Gerhard T, Kubilis P, et al. Cardiovascular safety of central nervous system stimulants in children and adolescents: population based cohort study. BMJ 2012;345:e4627. 104. Medicaid Pharmaceutical Plan Resources. National Pharma ceutical Council, 2011. (Accessed September, 2012, at http://www.npcnow.org/Public/Issues/i_rel_research/Medicaid_Pharmaceutical_P lan_Resources. aspx. ) 105. Multum National Drug Code (NDC) database. Cerner Multum, 2012. (Accessed October, 2012, at http://www.multum.com/. ) 106. Social Security's Public Death Master File (DMF). U.S. Social Security Administra tion, 2010. (Accessed July, 2011, at http://www.ntis.gov/products/ssa dmf.aspx. )

PAGE 183

183 107. Hill M, Rosenwaike I. The Social Security Administration's Death Master File: the completeness of death report ing at older ages. Soc Secur Bull 2001;64:45 51. 108. Schisterman E, Whitcomb B. Use of the Social Security Administration Death Master File for ascertainment of mortality status. Popul Health Metr 2004;2:2. 109. Boyle C, Decoufle P. National sources of vi tal status information: extent of coverage and possible selectivity in reporting. Am J Epidemiol 1990;131:160 8. 110. National Death Index. Centers for Disease Control and Prevention/ National Center for Health Statistics, 2011. (Accessed August, 2011, at http://www.cdc.gov/nchs/ndi.htm. ) 111. International Statistical Classification of Diseases, 10th Revision (ICD 10). World Health Organization, 1992. (Accessed September, 2012, at http://www.who.int/classifications/icd/en/. ) 112. Curb J, Ford C, Pressel S, Palmer M, Babcock C, Hawkins C. Ascertainment of vital status through the National Death Index and the Social Security Administration. Am J Epid emiol 1985;121:754 66. 113. Kraut A, Chan E, Landrigan P. The costs of searching for deaths: National Death Index vs Social Security Administration. Am J Public Health 1992;82:760 1. 114. Cowper DC, Kubal JD, Maynard C, Hynes DM. A primer and comparative r eview of major US mortality databases. Ann Epidemiol 2002;12:462 8. 115. Farmer E, Burns B, Chapman M, Phillips S, Angold E, Costello E. Use of Mental Health Services by Youth in Contact with Social Services. The Social Service Review 2001;75. 116. Taussig H. Risk Behaviors in Maltreated Youth Placed in Foster Care: A Longitudinal Study of Protective and Vulnerability Factors. Child Abuse and Neglect: The International Journal 2002;26:1179 99. 117. Medicaid Spending on Foster Children. THE URBAN INSTITUTE, 2012. (Accessed July, 2012, at http://www.urban.org/UploadedPDF/311221_medicaid_spending.pdf. ) 118. Social Security's Supplemental Security Income (SSI) Program,. U.S. Social S ecurity Administration,, 2012. (Accessed September, 2012, at http://www.ssa.gov/pgm/ssi.htm. ) 119. Temporary Assistance for Needy Families (TANF). Florida Department of Children and Families, 2012. (Accessed September, 2012, at http://www.dcf.state.fl.us/programs/access/docs/TANF%20101%20final.pdf. )

PAGE 184

184 120. Chen G, Khan N, Walker R, Quan H. Validating ICD coding algorithms f or diabetes mellitus from administrative data. Diabetes Res Clin Pract 2010;89:189 95. 121. International Classification of Diseases, Ninth Revision, Clinical Modification (ICD 9 CM). Centers for Disease Control and Prevention, 1988. (Accessed September, 2 012, at http://www.cdc.gov/nchs/icd/icd9cm.htm. ) 122. Schneeweiss S, Patrick A, Solomon D, et al. Comparative safety of antidepressant agents for children and adolescents regarding suicidal acts. Ped iatrics 2010;125:876 88. 123. Martin R, May M, Gunnell D. Did intense adverse media publicity impact on prescribing of paroxetine and the notification of suspected adverse drug reactions? Analysis of routine databases, 2001 2004. Br J Clin Pharmacol 2006;6 1(2):224 228. 124. Nemeroff C, Kalali A, Keller M, et al. Impact of publicity concerning pediatric suicidality data on physician practice patterns in the United States. Arch Gen Psychiatry 2007;64:466 72. 125. R: A language and environment for statistical computing. R Foundation for Statistical Computing, 2012. (Accessed September, 2012, at http://www.r project.org/. ) 126. Better Health Care for all Floridians. Florida Agency for Health Care Administration, 2011. (Accessed October, 2011, at http://ahca.myflorida.com/. ) 127. Cohen J. Statistical Power Analysis for the Behavioral Sciences, Second Edition (2nd ed). Hillsdale, NJ: Lawrence Erlbaum Associates; 1988. 128. Yang D, Dalton J. A unified approach to measuring the effect size between two groups using SAS. Statistics and Data Analysis/SAS Global Forum 2012;Paper 335. 129. Raja M, Azzoni A, Frustaci A. AUTISM Spectrum Disorders and Suicidality. Clin Pract Epidemiol Me nt Health 2011;7:97 105. 130. Giannini M, Bergmark B, Kreshover S, Elias E, Plummer C, O'Keefe E. Understanding suicide and disability through three major disabling conditions: Intellectual disability, spinal cord injury, and multiple sclerosis. Disabil He alth J 2010;3:74 8. 131. Controlled Substance Schedule (CSA), DEA Office of Diversion Control. US Dept of Justice Drug Enforcement Administration, 2010. (Accessed September, 2012, at http://www.deadiversion.usdoj.gov/schedules/orangebook/e_cs_sched.pdf. )

PAGE 185

185 132. Maldonado G, Greenland S. Simulation study of confounder selection strategies. Am J Epidemiol 1993;138:923 36. 133. Rosenbaum P, Rubin D. The central role of the pr opensity score in observational studies for causal effects. Biometrika 1983;70:41 55. 134. Glynn R, Schneeweiss S, Sturmer T. Indications for propensity scores and review of their use in pharmacoepidemiology. Basic Clin Pharmacol Toxicol 2006;98:253 9. 135 Rubin D. Estimating Causal Effects from Large Data Sets Using Propensity Scores. Annals of Internal Medicine 1997;127:757 63. 136. Cepeda M, Boston R, Farrar J, Strom B. Comparison of logistic regression versus propensity score when the number of events is low and there are multiple confounders. Am J Epidemiol 2003;158:280 7. 137. Sturmer T, Schneeweiss S, Brookhart M, Rothman K, Avorn J, Glynn R. Analytic strategies to adjust confounding using exposure propensity scores and disease risk scores: nonsteroi dal antiinflammatory drugs and short term mortality in the elderly. Am J Epidemiol 2005;161:891 8. 138. Schneeweiss S, Patrick A, Sturmer T, et al. Increasing levels of restriction in pharmacoepidemiologic database studies of elderly and comparison with ra ndomized trial results. Med Care 2007;45:S131 42. 139. Cox D. Regression Models and Life Tables. Journal of the Royal Statistical Society Series B (Methodological) 1972;34:187 220. 140. Allison PD. Survival Analysis using SAS A practical guide: Second Edi tion. Cary, NC, USA: SAS Institute Inc.; 2010. 141. Peng C, Lee K, Ingersoll G. An Introduction to Logistic Regression Analysis and Reporting. The Journal of Educational Research 2002;96:1 13. 142. Harman J, Edlund M, Fortney J. Trends in antidepressant ut ilization from 2001 to 2004. Psychiatr Serv 2009;60:611 6. 143. Gibbons R, Brown C, Hur K, et al. Early evidence on the effects of regulators' suicidality warnings on SSRI prescriptions and suicide in children and adolescents. Am J Psychiatry 2007;164:1356 63. 144. Radigan M, Lannon P, Roohan P, Gesten F. Medication patterns for attention deficit/hyperactivity disorder and comorbid psychiatric conditions in a low income population. J Child Adolesc Psychopharmacol 2005;15:44 56.

PAGE 186

186 145. Wilens TE, Adler LA, Adams J, et al. Misuse and diversion of stimulants prescribed for ADHD: a systematic review of the literature. J Am Acad Child Adolesc Psychiatry 2008;47:21 31. 146. Rissanen I, Jskelinen E, Isohanni M, et al. Use of antipsycho tic medication and suicidality the Northern Finland Birth Cohort 1966. Human Psychopharmacology: Clinical and Experimental 2012;27:476 85. 147. Monk M. Epidemiology of suicide. Epidemiol Rev 1987;9:51 69. 148. Brent DA, Perper JA, Allman CJ. Alcohol, firea rms, and suicide among youth. Temporal trends in Allegheny County, Pennsylvania, 1960 to 1983. JAMA 1987;257:3369 72. 149. 1995;25:22 35. 150. Census 2000 data. U.S. Census Bureau, 2 000. (Accessed September, 2012, at http://www.census.gov/main/www/cen2000.html. ) 151. Bright R, Avorn J, Everitt D. Medicaid data as a resource for epidemiologic studies: strengths and limitatio ns. J Clin Epidemiol 1989;42:937 45. 152. Crystal S, Akincigil A, Bilder S, Walkup J. Studying prescription drug use and outcomes with medicaid claims data: strengths, limitations, and strategies. Med Care 2007;45:S58 65.

PAGE 187

187 BIOGRAPHICAL SKETCH Stephan Lin den was born and raised in Dsseldorf, Germany. He received his Wilhelmina University, Braunschweig, Germany in 2005. After graduation he worked as a registered pharmacist in community pharmacies in Dsseldorf, Germany. In 2007, he joined the department of Pharmaceutical Outcomes & Polic y at the University of Florida. Stephan won the annual Levitt Oral Competition at the College of Pharmacy in two consecutive years, 2010 & 2011 In addition, he wa s the department and College of Pharmacy Graduate Student Representative from 2009 to 2011. In 2010, h e was recipient of the DuBow Family Fellowship for Pharmaceutical Outcomes research and recipient of the Mary K. Owens fellowship for Healthcare Innovatio n in 2011. Stephan received a certificate for outstanding academic achievement in 2010 from the University of Florida. For his contribution to the College of Pharmacy teaching mission he was honored by the University of Florida with the Teaching Assistant of the Year Award and voted as the most influential Teaching Assistant in the Pharmacy Curriculum by the PharmD graduating class of 2011. H e received the American Pharmacist Association (APhA) Wiederholt Prize in 2013. Stephan is a n invited member of the D Adv isory Board. Stephan has authored and coauthored several peer reviewed publications and presented his research at national and international conferences. He was an ISPE invited speaker at the annual Meeting of the Asociacion Mexicana de F armacologia and the Western Pharmacology Society in Mexico City, Mexico, in 2011. Stephan has been trained as a pharmacoepidemiologist and his research focused on drug safety, drug effectiveness and drug utilization.