Citation
Combining Adverse Perinatal Outcomes for Women Exposed to Antiepileptic Drugs during Pregnancy, Using a Latent Trait Model

Material Information

Title:
Combining Adverse Perinatal Outcomes for Women Exposed to Antiepileptic Drugs during Pregnancy, Using a Latent Trait Model
Creator:
Wen, Xuerong
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (175 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Pharmaceutical Sciences
Pharmaceutical Outcomes and Policy
Committee Chair:
HARTZEMA,ABRAHAM G
Committee Co-Chair:
DELANEY,JOSEPH
Committee Members:
SEGAL,RICHARD
BRUMBACK,BABETTE A
MEADOR,KIMFORD J
Graduation Date:
12/13/2013

Subjects

Subjects / Keywords:
Antibiotics ( jstor )
Anticonvulsants ( jstor )
Antihypertensives ( jstor )
Antineoplastics ( jstor )
Hospitals ( jstor )
Hydrochlorides ( jstor )
Infants ( jstor )
Medicaid ( jstor )
Neonatal disorders ( jstor )
Pregnancy ( jstor )
Pharmaceutical Outcomes and Policy -- Dissertations, Academic -- UF
aed -- birth -- combining -- defects -- ltm -- outcomes
Genre:
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Pharmaceutical Sciences thesis, Ph.D.

Notes

Abstract:
Application of Latent Variable Models (LVMs) in medical research is becoming increasingly popular. The objective of this study was to employ an LVM in a drug safety study to produce an overall index of adverse perinatal and pregnancy outcome (APO) from four observed component outcomes, including birth defects (BD), abnormal conditions of new borns (ACNB), pregnancy and obstetric complications (PC/OC), and low birth weights (LBW). The validity and reliability of the combined outcome were assessed in accordance with the psychometric theory and the assumptions of the latent variable. Unidimentionality, local independence, internal homogeneity, and construct validity were evaluated using factor analysis, Q3 statistics, and Spearman correlation. The association between infant birth defects and maternal exposure to Antiepileptic drugs (AEDs) or Vaproic Acid (VPA) has been widely investigated. Previous studies reported the significant adverse effects of maternal exposure to AEDs during pregnancy on all four component outcomes, whereas, the significant adverse effects of VPA mainly on one component, BD, specifically neural tube defects. Therefore, AED and VPA, which are representations of two different scenarios of relationships with the component oucomes, were employed in this study to validate the component selection for the combined outcome.  The change of utilization trends for AED polytherapy in 2007 possibly caused by the FDA black box warning for lamortrigine and birth defects was identified in this study. Our findings suggest that combining BD, ACNB, PC/OC, and LBW can generate a valid assessment of overall severity of adverse perinatal and pregnant outcome and the overall adverse effects of AEDs on both mothers and infants. However, caution needs to be taken for combining outcomes. Adding a component outcome that is not associated with the specific drug exposure may mask the relationship between the combined outcome and drug exposure. Therefore, evaluation of the selected components is essential to ensure the validity of the combined outcome.   Future studies are warranted for utilizing the latent variable model with more components, and weighted severity for each component. ( en )
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.
Thesis:
Thesis (Ph.D.)--University of Florida, 2013.
Local:
Adviser: HARTZEMA,ABRAHAM G.
Local:
Co-adviser: DELANEY,JOSEPH.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2014-12-31
Statement of Responsibility:
by Xuerong Wen.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Embargo Date:
12/31/2014
Classification:
LD1780 2013 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

1 C OMBIN ING ADVERSE PERINATAL OUTCOME S FOR WOMEN EXPOSED TO ANTIEPILEPTIC DRUGS DURING PREGNANCY USING A LATENT TRAIT MODEL By XUERONG WEN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PAR TIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 3

PAGE 2

2 2013 Xuerong Wen

PAGE 3

3 To Yang Jessica, and my parents

PAGE 4

4 ACKNOWLEDGMENTS I would like to express my deepest gratitude to my advisor, Dr. Abraham Hartzema, for his continuous support and guidance throughout my PhD study. He has been an excellent mentor who is patient, wise, and provides strong support on my research work He trained me to grow as an independent thinker and researcher. I would like to gratefully thank Dr. Kimford Meador, who directed me to the medical research field, encouraged me to continue my PhD study and supported my dissertation work in many aspects I would also like to thank Dr. Rich Segal, who is a great source of inspiration, very reliable and always provides advice throughout my PhD study I would like to thank my other supervisory committee members Dr. B abette Brumback and Dr. Joseph Delaney, for their advice and encouragement on my dissertati on My gratitude further extends to all faculty members, staff, and students in the Department of Pharmaceutical Outcomes and Policy for their support and friendship. Special thanks go to Dr.Carole Kimberlin, Dr. Almut Winterstein, Dr. Thomas Smith, Dr. Ea rlene Lipowski Wei Liu, Irene Murimi and Paul Kubilis for their input and discussions on my research work I would like to thank Dr. Xuefeng Liu, who originally developed this Latent Variable Model and spent a lot of his valuable time to teach and disc uss with me about the model. I thank Dr. Robert Egerman for his expertise and advice on my research. I would like to acknowledge Dr. Jefferey Roth and Family Data Center from the Department of Pediatrics for the provision of high quality data and help on o utcome assessment. Finally, I thank my daughter, my husband, my parents, and my in laws for their support and encouragement throughout my PhD study.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF ABBREVIATIONS ................................ ................................ ........................... 11 ABSTRACT ................................ ................................ ................................ ................... 13 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 15 Background ................................ ................................ ................................ ............. 15 Need for Study ................................ ................................ ................................ 16 Purpose of Study ................................ ................................ .............................. 18 Study Significance ................................ ................................ ............................ 19 Research Questions and Hypotheses ................................ ................................ ..... 20 Summary ................................ ................................ ................................ ................ 25 2 LITERATURE REVIEW ................................ ................................ .......................... 27 Definition and Epidemiology of Adverse Perinatal Outcomes (AP Os) .................... 27 Birth Defects (BDs) ................................ ................................ ........................... 27 Major Congenital Malformation (MCM) ................................ ............................. 28 Minor Anomaly (MA) ................................ ................................ ......................... 29 Abnormal Conditions of the Newborn (ACNB) ................................ .................. 29 Low Birth Weight (LBW) ................................ ................................ ................... 29 Pregnancy and Obstetric Complication (PC/OC) ................................ .............. 30 Correlation between Component Outcomes ................................ ..................... 32 Impo rtant Teratogens ................................ ................................ .............................. 33 Antiepileptic Drugs (AEDs) ................................ ................................ ............... 33 Valproic Acid (VPA) ................................ ................................ .......................... 34 Theoretical Framework for Latent Variable Modeling ................................ ............. 35 Latent Variable Model (LVM) ................................ ................................ ............ 35 Latent Trait Analysis (LTA) ................................ ................................ ............... 35 Latent Variable in Latent Trait Setting ................................ .............................. 36 Evaluation of Combined Outcome ................................ ................................ .......... 39 Assumptions for Composite Outcome ................................ .............................. 39 Unidimensionality ................................ ................................ ............................. 39 Local Independence ................................ ................................ ......................... 40 Construct Validity ................................ ................................ ............................. 41 Reliability ................................ ................................ ................................ .......... 41 3 METHODOLOGY ................................ ................................ ................................ ... 52

PAGE 6

6 Data Sources ................................ ................................ ................................ .......... 52 Florida Medicaid Claims Dataset and National Drug Code (NDC) ................... 52 Florida Birth Vital Statistics ( BVS) ................................ ................................ .... 53 AHCA Florida Hospital Discharge Data ................................ ............................ 54 Florida Birth Anomaly ................................ ................................ ....................... 55 Data Linkage ................................ ................................ ................................ .... 55 Study Population ................................ ................................ ................................ ..... 56 Study Cohort Inclusion Criteria ................................ ................................ ....... 56 Study Cohort Exclusion Criteria ................................ ................................ ..... 56 Overall Study Design ................................ ................................ ........................ 57 Assessment of Exposure ................................ ................................ ........................ 58 Drug Exposure ................................ ................................ ................................ 58 Assessment of Component Outcomes ................................ ................................ .... 59 Major Congenital Malformation (MCM ) and Minor Anomaly (MA) .................... 59 Abnormal Condition of New Born (ACNB) ................................ ........................ 60 Low Birth Weight (LBW) ................................ ................................ ................... 60 Pregnancy and Obstetric Complications (PC/OCs) ................................ .......... 61 Assessment of Controls and Covariates ................................ ................................ 62 Contro l Group ................................ ................................ ................................ ... 62 Covariates ................................ ................................ ................................ ........ 62 Propensity Score ................................ ................................ .............................. 63 Descriptive Analysi s ................................ ................................ ................................ 66 Study Design and Statistical Analysis for Three Specific Aims ............................... 68 Part 1: Utilization of AEDs in Pregnant Women ................................ ................ 69 Part 2: Application of Latent Variable Model ................................ ..................... 69 Part 3: Evaluation of the Combined Outcome. ................................ ................. 70 Aim 3 1 Assess the Unidimensionality of the Combined Outcomes ........... 71 Aim 3 2 Examine the Local Independence of the Four Component Outcomes ................................ ................................ ............................... 71 Aim 3 3 Assess the Internal Homogeneity of the Combined Outcome ...... 72 Aim 3 4 Evaluate the Convergent Validity of the Combined Outcome ....... 72 Aim 3 5 Evaluate the Discriminant Validity of the Combined Outcome ...... 72 Aim 3 6 Evaluate the Discriminative Validity of the Combined Outcome ... 73 Sensitivity Analysis ................................ ................................ ................................ 73 4 RESULTS ................................ ................................ ................................ ............... 79 Characteristics of Study Population ................................ ................................ ........ 79 Descriptive Analysis of Observed Outcomes ................................ .......................... 80 Part 1: Utilization of AEDs in Pregnant Women Included in the Study .................... 83 Specific Aim 1 1: Assess the secular trend for the utilization of AEDs. ............ 83 Specific Aim 1 2: Assess the secular trend for the utilization o f VPA. .............. 83 Specific Aim 1 3: Compare the utilization of AEDs from 1999 to 2009 in monotherapy and polytherapy in the pregnant women included in the study. ................................ ................................ ................................ ............ 84

PAGE 7

7 Specific Aim 1 4: Compare the utilization of VPA from 1999 to 2009 in monotherapy and polytherapy in the pregnant women included in the study. ................................ ................................ ................................ ............ 84 Part 2: Apply a L atent Variable Model to Combine the Four Birth Defect Outcomes ................................ ................................ ................................ ............ 85 Specific Aim 2: Establish an LVM to combine BD, ACNB, LBW, and PC/OC into one latent variable APO, for the studied mother infa nt pairs. ................. 85 Part 3: Evaluate the Validity and Reliability of the Combined Outcome .................. 85 Specific Aim 3 1: Assess the unidimensiona lity of the combined outcome ...... 85 Specific Aim 3 2: Examine the local independence of the four component outcomes ................................ ................................ ................................ ....... 8 7 Specific Aim 3 3: Assess the internal homogeneity of the combined outcome ................................ ................................ ................................ ........ 88 Specific Aim 3 4: Evaluate the convergent validity of the combined outcome .. 88 Specific Aim 3 5: Evaluate the discriminant validity of the combined outcome ................................ ................................ ................................ ........ 89 Specific Aim 3 6: Evaluate the discriminative validity of the combined outcome ................................ ................................ ................................ ........ 89 5 DISCUSSION ................................ ................................ ................................ ....... 116 Descriptive Analysis ................................ ................................ .............................. 116 Part 1: Utilization of AEDs in Pregnan t Women ................................ .................... 119 Part 2: Application of Latent Variable Model ................................ ......................... 120 Part 3: Evaluation of the Combined Outcome ................................ ....................... 121 Unidimensionality and Local Independence ................................ ................... 121 Reliability ................................ ................................ ................................ ........ 122 Validity ................................ ................................ ................................ ............ 123 Sensitivity Study ................................ ................................ ............................. 124 Generalizability ................................ ................................ ................................ ..... 125 Study Limitations ................................ ................................ ................................ .. 126 Future Studies ................................ ................................ ................................ ...... 128 Summary and Conclusions ................................ ................................ ................... 129 APPENDIX ................................ ................................ ................................ .................. 134 A: ACTIVE INGREDIENTS AND DRUG CATEGORIES ................................ ............. 134 B: OPERATIONAL DEFINITIONS ................................ ................................ ............... 150 LIST OF REFERENCES ................................ ................................ ............................. 157 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 175

PAGE 8

8 LIST OF TABLES Table page 1 1 FDA Reg ulatory Actions on VPA Associated with BDs ................................ ...... 26 2 1 Commonly Seen Minor Anomalies (MAs). ................................ .......................... 44 2 2 Definitions and Symptoms of Abnor mal Conditions for the Newborns. ............... 46 2 3 Four Types of Latent Variable Models. ................................ ............................... 48 3 1 PPVs of overall and specific birth defects identified using birth certificates, 1985 2000. ................................ ................................ ................................ ......... 77 3 2 ICD 9 codes for Abnormal Conditions of the Newborn. ................................ ...... 78 4 1 Demographic Ch aracteristics of Study Participants. Obtained from BVS. .......... 92 4 2 Clinical Characteristics of Study Participants. Obtained from Medicaid Claim Data. ................................ ................................ ................................ ................... 94 4 3 Spearman Correlation of Four Observed Outcomes with Length of Hospital Stay During Delivery. ................................ ................................ .......................... 97 4 4 Effects of AEDs or VPA on Four Observed Outcomes. AED users include the patients using VPA. ................................ ................................ ............................ 98 4 5 Incident Rate in Each Category of Observed Outcomes. ................................ 106 4 6 Estimates of Parameters in LTM. ................................ ................................ ..... 106 4 7 Observed Frequency (OBFREQ), Expected Frequency (EXFREQ), Observed Percents (OB%), and Expected Percents (EX%) by Combinations of Four Observed Outcomes. ................................ ................................ ........................ 107 4 8 Estimates of Posterior Mean of the Latent Variable and the APO. ................ 108 4 9 Results of Unidimentionality Assessment Using SAS Proc Factors with the Method of Maximum Likelihood. ................................ ................................ ....... 109 4 10 CFA Results for Assessing Local Independence of the Observed Outcomes. 110 4 11 Spearman Correlation Statistics for Four Observed Outcomes and One Combined Outcome. ................................ ................................ ......................... 112 5 1 Demographic Characteristics of Study Participants. Obtained from BVS. ........ 131

PAGE 9

9 LIST OF FIGURES page 2 1 Types of Birth Defect and Time Window of Exposure to the Teratogens. .......... 43 2 2 Log ................................ ...... 49 2 3 Unidimentionality measurement Models.. ................................ ........................... 50 3 1 Schematic Diagram of Data Linkage. ................................ ................................ 75 3 2 A Schematic Plot for the St udy Design and Critical Time Points. ....................... 76 4 1 Flow Chart of Study Design. ................................ ................................ ............... 91 4 2 Frequencies of Four Observed Outcomes in the Stud y Population. ................... 97 4 3 PC, OC, Preterm Born, and PCOC in AED, VPA users, and Healthy Controls. AED users include the patients using VPA. ................................ ........................ 99 4 4 Birth Defects (Major Congenital Malformation and Minor Congenital Malformation) and Abnormal Condition in AED and VPA Users, and Healthy Controls. AED users include the patients using VPA. ................................ ....... 100 4 5 Distribution of Four BW Categories in AED and VPA Users, and Healthy Controls. AED users include the patients using VPA. ................................ ....... 101 4 6 Percentage of AED Use in Study Population fr P=0.28. AED users include the patients using VPA. ................................ ........ 102 4 7 Percentage of VPA Use in Study Population from 2000 to 2009. = 0.003, P=0.017. ................................ ................................ ................................ ........... 103 4 8 Percentage of AED Use in Polytherapy in Study Population from 2000 to 2009. =1.90, P 1 =0.20; = 2.61, P 2 =0.52. AED users include the patients using VPA. ................................ ................................ ................................ ........ 104 4 9 Percentage of VPA Use in Polytheray in Study Population from 2000 to 2009. = 0.68, P=0.001. ................................ ................................ ............................. 105 4 10 Correlation between Combined Outcome APO and Total Length of Hospital Stay d uring Delivery. ................................ ................................ ........................ 113 4 11 Correlation between Combined Outcome APO and Infant Breast Fed Status. 114 4 12 C omparison of APO Score s ................................ ................................ ............. 115

PAGE 10

10 5 1 Frequencies of Component Outcomes in Medicaid Enrollees and General Population. ................................ ................................ ................................ ........ 133

PAGE 11

11 LIST OF ABBREVIATION S AC NB Abnormal Condition of Newb o rn AED Anti epileptic Drugs (also called Anticonvulsant Drugs) AHCA Florida Agency for Health Care Administration ANOVA Analysis of Variance APO Adverse Perinatal Outcome BD Birth Defect BVS Birth Vital Statistics CDC Centers for Disease Control and Preven tion CFA Confirmatory Factor Analysis CI Confidence Interval CMS Centers for Medicare & Medicaid Services CPT Current Procedural Terminology DDD Developmental Delay & Disorder DoH Department of Health EFA Exploratory F act or Analysis EGNLS Estimated Genera lized Nonlinear Least Squares ELBW Extremely Low Birth Weight FDA Food and Drug Administration HIP Hospital Inpatient Discharge Data HOP Hospital Out patient Discharge Data ICD 9CM International Classification of Diseases Ninth Revision, clinical modificat ion LBW Low Birth Weight LMP Last Menstrual Period LTM Latent Trait Model

PAGE 12

12 LVM Latent Variable Model MA Minor Anomaly MCHERDC Maternal and Child Health Education and Data Center MCM Major Congenital Malformation NBW Normal Birth Weight NDC National Drug Co de OR Odds Ratio PC/OC Pregnancy and O bstetric C omplication PMF Probability Mass Function PPV Positive Predictive Value VLBW Very Low Birth Weight VPA Valproic Acid WWE Women with Epilepsy

PAGE 13

13 Abstract of Dissertation Presented to the Graduate School of t he University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy COMBIN ING ADVERSE PERINATAL OU TCOMES FOR WOMEN EXP OSED TO ANTIEPILEPTIC DRUGS DURING PREGNANCY USING A LATENT TRAIT MODEL By Xuer ong Wen December 2013 Chair: Abraham Hartzema Major: Pharmaceutical Sciences Applicatio n of Latent Variable M odels (LVMs) in medical research is becoming increasingly popular. The objective of this study was to employ an LVM in a drug safety study to prod uce an overall index of adverse perinatal and pregnancy outcome (APO) from four observed component outcome s, including birth defect s (BD), abnormal condition s of new born s (ACNB), pregnancy and obstetric complication s (PC/OC), and low birth weight s (LBW). The validity and reliability of the combined outcome were assessed in accordance with the psychometric theory and the assumptions of the latent variable Unidimentionality, local independence, internal homogeneity, and constr uct validity were evaluated usi ng factor analysis, Q3 statistics, and Spearman correlation. The association between infant birth defects and maternal exposure to Antiepileptic drugs ( AEDs ) or Vaproic Acid ( VPA ) has been widely investigated. Previous studies reported the significant adv erse effects of maternal exposure to AEDs during pregnancy on all four component outcomes, whereas, the significant adverse effects of VPA mainly on one component, BD specifically neural tube defect s

PAGE 14

14 Therefore, AED and VPA, which are representations of t wo different scenarios of relationship s with the component oucomes were employed in this study to validate the component selection for the combined outcome. T he change of utilization trends for AED polytherapy in 2007 possibly caused by the FDA black box warning for lamortrigine and birth defects was identified in this study. Our findings suggest that combining BD, ACNB, PC/OC, and LBW can generate a valid assessment of overall severity of adverse perinatal and pregnant outcome and the overall adverse eff ects of AEDs on both mothers and infants. However, caution needs to be taken for combining outcomes A dding a component outcome that is not associated with the specific drug exposure may mask the relationship between the combined outcome and drug exposure. Therefore evaluation of the selected component s is essential to ensure the validity of the combined outcome. Future studies are warrant ed for utilizing the latent variable model with more compo nents, and weigh ted severity for each component.

PAGE 15

15 CHAP T ER 1 INTRODUCTION Background In the United States, birth defects (BDs) involving major congenital malformation (MCM) and minor anomaly ( MA ) are the leading causes of infant mortality morbidity, and years of potential life lost. 1 Pregnancy and o bstetric complications ( PC/OCs ) have been reported ly associate d with low birth weight ( LBW ) abnormal condition of new born (ACNB), preterm birth and childhood seizure s 2,3,4,5 The association of infant BDs and PC/OCs with maternal exposure to antiepilept ic drugs (AEDs) ha s been investigated extensively. 6,7 However, the rare occurrence of BDs ACNB s and PC/OCs limit s the power of most published studies and make s some studies inconclusive. 8,9,10 A joint model for combining individual outcomes is suggested to improve the efficiency and power of BD studies. 11 Latent variable models ( LVM s) have become increasingly popular in medical research, including the measurement of quality of life, diagnostic testing, survival analysis, and joint modeling for longitudin al data. 12 A n LVM in a latent trait setting was applied to combine individual BD outcomes and develop an infant morbidity index. 13 The an independence of ma nifest outcomes conditioned on latent variable s 14 Estimated Generalized Nonlinear Least Squares (EGNLS) estimation w as used to obtain the parameters involved in the latent trait model. 15 Application of this model to drug safety studies increases statistical power and improves the efficiency of the st udy. There is a debate over the use of combined or individual outcome s in a BD study. Although a c ombined outcome increases statistical efficiency, it may mislead the

PAGE 16

16 results and threaten the validity of the study if the components are selected inappropri ately. 16,17 Therefore, the combined outcome must be evaluated in terms of conceptualization of the composite outcome. 16 Three important criteria for selecting the components of a composite outcome are: 1) similar importance to the patient; 2) similar event rate, and 3) similar treatment effects across all components. 16 These three assumptions need to be verified to ensure the valid selection of the components. Furthermore, a combined outcome of this study entails an unobservable latent variable where eval uation of the unidimensionality, local independence, construct validity, and reliabi lity of the latent variable become s important 5 To establish construct validity many different types of validity, including convergent validity, discriminant validity, and discrim inative validities are usually operationally defined and assessed. 18 This validation process examines the accuracy of the representation of the overall adverse outcomes for both mothers and infants. Need for Study In the United States, about 120,0 00 infants annually are born with a BD of which 8,000 die within the first year of life. 1 According to the Nat ional Hospital Discharge Survey, more than 22% of hospitalization s of pregnant women are due to pregnancy complication s and loss, which costs ove r 2 million hospital days of care and 1 billion dollars per year in the USA. 19 Approximately 2 % to 3% of BDs are related to teratogen induced malformations. 20,21 AEDs are a group of significant teratogens that are relate d to several BDs and PC/OCs 22 Previ ous studies have discovered that some BDs and PC/OCs are associated with in utero exposure to first generation AEDs. 23,24 Valproic acid (VPA) has been identified as an important teratogen related to the high risk of neural tube defects

PAGE 17

17 (shown in T able 1 1) 6,25 ,26 The FDA warned of the risk of cleft lip and palate with in utero exposure to topiramide and lamotrigine; 2 7 however, the association between BDs and second generation AEDs, either as a class or an individual has not been established conclusively. Previous research was not adequately powered in identify ing a clinically significant risk of BD or pregnancy complication from in utero exposure to specific second generation AEDs. 8,28 ,2 9 More relevant clinical data on levetiracetam, lamotrigine, topirama te, tiagabine, oxcarbazepine, pregabalin, and zonisamide is needed to identify the safest AED in pregnancy and to understand the physiological pathways of their teratogenic effects. The results from previous studies on the teratogenicity of the specific se cond generation AEDs are controversial due to a deficiency of statistical power caused by rare outcomes and exposure s 30 In addition, the combined outcomes used in previous studies were simply any of the components 3 1 The occurrence rates of the components were not taken into consideration The latent trait model proposed in this study combines four specific outcomes into a continuous index of adverse perinatal outcome (APO) based on the occurre nce rate of each of the components. The model will be evaluated for component selection and construct validity. This model may be important to assist pharmacoepidemiologists in conducting safety studies of multiple rare outcomes with different occurrence r ates The s ystematic study of APO with adequate statistical power is essential for the drug safe use by pregnant women. Our study can contribute to the prevention of BDs in the offspring of women using AEDs, and ultimately improve health outcomes in soc iety at large

PAGE 18

18 Purpose of Study The basic purpose of this research is to explore an innovative method to achieve greater power and higher efficiency for drug safety studies of multiple rare outcomes. In this study, a latent trait model is used to combine pregnancy complications and BDs and the combined outcome is evaluated for validity and reliability By establishing a validated combined outcome, our research can provide pharmacoepidemiologists with methodological options that can help them make better informed decisions in the study of rare outcomes. A combined outcome with content and construct validity can maximize the identification of teratogenic signal s increase the likelihood of a statistical ly significant adverse effect, and comprehensively asse ss the adverse perinatal and pregnancy effects of maternal drug exposure on both mothers and infants. 3 2 A well de signed observational study that applies a combined outcome can provide physicians with informat ion to identify drugs that can harm mother infan t pairs, help the FDA regulate t he drugs in the proper pregnancy categories, and ultimately improve the overall health of both mothers and infants. The first part of this study analyzes the utilization pattern of VPA prescriptions for pregnant women by usi ng Florida Medicaid data from 1999 to 2009. The analysis includes longitudinal trends in the utilization of VPA comparison of the use of VPA as mono and polytherapy in addition to other AEDs, and identification of significant predictors for VPA use in pr egnancy women This first part provides important insights into the evaluation of combined outcomes and set s the stage for the third part of the study The second part of this study detail s the process of developing an extended latent trait model to combi ne multiple outcomes. The p robability theory, statistical

PAGE 19

19 inference, and mathematical algorithms used to develop this model are discussed in this section. In the third part, the combined outcome is evaluated for validity and reliability using well estab lished association s between APOs and maternal exposure to VPA in the linked dataset of the Florida Medicaid and Florida Birth Vital Surveillance (BVS) from January 1 1999 to December 31 2009. Unidimensionality and local independence are evaluated using exp loratory factor analysis (EFA) and confirmatory factor analysis (CFA) 5 Construct validity is assessed by examining the convergent, discriminant, and discriminative validity. The length of hospital stays during infant delivery is used as a to examine whether and to what extent the combined outcome associates with the standard measure of health status for both mothers and infants. 33,34 The combined outcome will be considered invalid if it does not show convergent and discriminant validity b y failing to correlate with the relevant clinical endpoints and distinguish from the irrelevant events. A l ack of discriminative validity can make a combined outcome unable to distinguish the difference between extreme groups. This analysis ensures that th e combined outcome in drug safety studies not only improves statistical power, but also is valid and meaningful. Study Significance Previous pharmacoepidemiological studies o f APO s mostly investigate d the association between a drug and one specific outcom e, and were often limited by insufficient statistical power. Combining several rare outcomes into one combined outcome with an appropriate statistical model may help to power and efficiency.

PAGE 20

20 After the completion of this study, we expect to establish an alternative method with improved statistical power and adequate internal validity to detect a clinically significant risk ratio in the APO studies. However, the combined outcome also may obscure important differences for specific outcomes a nd could even mask AED differences by adding noise to the system This study will evaluate the strength and limitation of the combined outcome in birth defect study, and help health researchers choose more efficient and appropriate methods for outcome defi nition in the future. Furthermore, t he risk factors identified for prescribing VPA in the first part of this study provide information for clinical dec ision making in pregnant women. W hile randomized clinical trials are not feasible for APO outcome due to the ethical issues, the evidence from an observational APO study with adequate statistical power will provide informative clinical decision making and help the FDA to detect drug adverse effects, decide upon label changes, and develop treatment guidelin es to ultimately prevent adverse perinatal and pregnancy effects and improve public health. Research Questions and Hypotheses The research questions and specific aims in this dissertation are categorized in three parts, and apply to the studied mother infant pairs which are defined as Florida Medicaid enrollees who delivered a single live infant between April 01, 2000 and December 31, 2008. The detailed inclusion and exclusion criteria for the study populat ion are illustrated in C hapter 3. Spearman corr elation is applied for the discrete data, and Pearson correlation is calculated for the continuous variables that follows normal distribution. level of 0.05. H 0 refers to a null hypothesi s and H A to the alternative hypothesis. Part 1: Utilization of AEDs in Pregnant Women Included in the Study

PAGE 21

21 This study includes pregnant women who were enrolled in the Florida Medicaid fee for service program between the years 1999 200 9. Specific Aim 1 1: Assess the secular trend for the utilization of AEDs Hypothesis 1 1: H A : The slope of the secular trend for the utilization of AEDs is significantly less than 0 H 0 : The slope of the secular trend for utilization of AEDs is equal to 0 Specific Aim 1 2 : Assess the secular trend for the utilization of VPA. Hypothesis 1 2 : H A : The slope of the secular trend for the utilization of VPA is significantly less than 0 H 0 : The slope of the secular trend for utilization of VPA is equal to 0 Specific Aim 1 3: Com pare the utilization of AEDs from 1999 to 2009 in monotherapy and polytherapy in the pregnant women included in the study. Hypothesis 1 3 : H A : The secular trend for the utilization of AED in monotherapy is significantly different than that of polytherapy H 0 : The secular trend for the utilization of AED in monotherapy is not different than that of polytherapy. Specific Aim 1 4 : Compare the utilization of VPA from 1999 to 2009 in monotherapy and polytherapy in the pregnant women included in the study. Hypo thesis 1 4 : H A : The secular trend for the utilization of VPA in monotherapy is significantly different than that of polytherapy H 0 : The secular trend for the utilization of VPA in monotherapy is not different than that of polytherapy. Part 2: Apply a Late nt Variable Model to Combine the Four Birth Defect Outcomes

PAGE 22

22 In this section, a n LVM was developed to combin e the four components into one continues index of APO The combined outcome in this study is a latent variable obtained from the LVM, and the four co mponent outcomes refer to the observable manifest variables in the terminology of LVMs. Specific Aim 2: Establish a n LVM to combine BD ACBs, L BW, and PC /OC into one latent variable APO for the studied mother infant pairs. Hypothesis 2: H A : Estimati ons o f the parameters of a latent trait model using the modified Gauss Newton algorithm can converge t o 10 9 meeting the assumption that a latent trait model can be established to combine BD, ACNB, LBW, and PC/OC into one overall measure of APO for the stud ied mother infant pairs. H 0 : Estimating parameters of a latent trait model by the modified Gauss Newton algorithm is > 10 9 and cannot converge so that a latent trait model cannot be established to combine BD, ACNB, LBW, and PC/OC into an overall measure o f APO for the studied mother infant pairs. Part 3: Evaluate the Validity and Reliability of the C ombined Outcome The important assumption of LVM s local independence, was examined to ensure that the four components have no pairwise significant association after controlling for the combined outcome. Unidimensionality, reliability (referring to internal homogeneity ), and validity ( comprised of convergent validity, discriminant validity, and discriminative validity) of the combined outcome were evaluated in th is section Specific Aim 3 1 : Assess the unidimensionality of the combined outcome

PAGE 23

23 Hypothesis 3 1 : H A : Only one construct significantly correlates with all component outcomes with minimum factor loading 0.3 H 0 : More than one constructs significantly corr elate with all component o utcomes with minimum factor loading 0.3 Specific Aim 3 2: Examine the local independence of the four component outcomes Hypothesis 3 2 : H A : At least two component outcomes significantly correlate with each other after controlling for the combined outcome. H 0 : The four component outcomes are independent with each other after controlling for t he combined outcome Specific Aim 3 3 : Assess the internal homogeneity of the combined outcome Hypothesis 3 3 1: H A : The correlation between B D and ACNB is statistically significantly different from 0 in the studied mother infant pairs. H 0 : The correlation between BD and ACNB is equal to 0 in the studied mother infant pairs. Hypothesis 3 3 2: H A : The correlation between BD and L BW is significant ly different from 0 in the studied mother infant pairs. H 0 : The correlation between BD and L BW is equal to 0 in the studied mother infant pairs. Hypothesis 3 3 3: H A : The correlation between BD and PC/OC is significantly different from 0 in the studied mot her infant pairs. H 0 : The correlation between BD and PC/OC is equal to 0 in the studied mother infant pairs. Hypothesis 3 3 4: H A : The correlation between ACNB and L BW is significantly different from 0 in the studied mother infant pairs. H 0 : The correlatio n between ACNB and L BW is equal to 0 in the studied mother infant pairs.

PAGE 24

24 Hypothesis 3 3 5: H A : The correlation between ACNB and PC/OC is significantly different from 0 in the studied mother infant pairs. H 0 : The correlation between ACNB and PC/OC is equal to 0 in the studied mother infant pairs. Hypothesis 3 3 6: H A : The correlation between L BW and PC/OC is significantly different from 0 in the studied mother infant pairs. H 0 : The correlation between L BW and PC/OC is equal to 0 in the studied mother infant pairs. Specific Aim 3 4 : Evaluate the convergent validity of the combined outcome Hypothesis 3 4 : H A : The correlation between combined outcome and the length of hospital stay during delivery is significantly different from 0 in the studied mother infant pa irs. H 0 : The correlation between combined outcome and the length of hospital stay during delivery is equal to 0 in the studied mother infant pairs. Specific Aim 3 5 : Evaluate the discriminant validity of the combined outcome Hypothesis 3 5 : H A : The corre lation between the combined outcome and the status of breastfeeding is significantly different from 0 in the studied mother infant pairs. H 0 : The correlation between the combined outcome and the status of breastfeeding is equal to 0 in the studied mother i nfant pairs. Specific Aim 3 6 : Evaluate the discriminative validity of the combined outcome Hypothesis 3 6 : H A : The combined outcome is significantly different when comparing the group with BD and ELBW and the group with NBW only. H 0 : The

PAGE 25

25 combined outcome is not significantly different when comparing the group with BD and ELBW and the group with NBW only. Summary Evaluation of unidimensionality, local independence, construct validity and internal homogeneity address es the validity and reliability of the la tent variable and ensure the accuracy of the combined outcome. 5 The components that are locally dependent or fail to prove unidimensionality need to be excluded from the study 5 If construct validity and internal homogeneity are not established, the combin ed outcome is not valid and reliable, then the corresponding components need to be adjusted recombined, and reevaluated.

PAGE 26

26 T able 1 1. FDA R egulatory A ctions on VPA A ssociated with BDs 2 6 Drugs Year Month Event VPA 1978 February FDA approved for the trea tment of epilepsy 1982 November Researchers found 20 fold increased risk of spina bifida in children with in utero exposure to VPA. 1995 May FDA approv ed for the treatment of bipolar disorder. 2009 December FDA issued public health advisory on incre ased risk of neural tube defects and other major BDs in babies with in utero exposure to VPA and related products. 2011 June FDA issued a safety announcement regarding the increased risk of lower cognitive test scores for children with in utero exposure to VPA

PAGE 27

27 CHAPTER 2 LITERATURE REVIEW Definition and Epidemiology of Adverse Perinatal Outcomes (APOs) Birth Defects (BDs) APOs include maternal pregnancy complication s obstetric complication s and infant s BD s BDs involving MCMs and MAs, are also called congenital anomalies, congenital malformations, congenital defects, or congenital disorders. In general, a BD can be defined as a physical or physiological defect that is developed during pregnancy, presented at the time of birth, and may be in duced by genetic or environmental factors. 20 BDs occur in 3 5% of children born in the United States and account for 20% of all infant deaths. 20,3 5 ,3 6 It was reported that 2 3% of BDs are due to teratogen induced malformations, 20,21 which refer to malf ormations result ing from environmental or in utero exposure to teratogens. In the United States, about 3 million people currently live with teratogen induced malformations. 20 The pregnancy stage of maternal exposure to teratogens is very critical to the de velopment of malformations. The m ajority of malformations occur during weeks 3 to 9 (stage of organogenesis) of the pregnancy The f irst 3 5 weeks are the most importan t for the development of the c entral nervous system, whereas weeks 7 to 9 are more criti cal for urogital development. 20 Days 20 through 42 are critical for cardiopathic events. 3 7 Beyond week 8 is the fetal period, in which teratogenic effects mostly result in growth retardation or functional defects instead of major malformation. 3 7

PAGE 28

28 Major C ongenital Malformation ( MCM ) MCM is an abnormality of an essential anatomic structure that is present at birth and interferes significantly with function and/or requires major intervention 38 39 MCM include s heart malformations, urological def ects, oro facial defects, neural tube defects, and skeletal abnormalities, etc.. 38,40 ,4 1 Heart malformations involve atrial and ventricular septal defects (e.g., ostium secundum defect, tricuspid atersia, tetralogy of fallot), arterial system defects (e.g. patent ductus arteriosus, pulmonary stenosis, coarctation of the aorta), and venous system defects (e.g., double inferior vena cava, absence of the inferior vena cava). Urological defects include renal sias and agenesis, congenital polycystic kidney), abnormal location of the kidneys (e.g., pelvic kidney, accessory renal arteries), hypospadias (e.g. glandular hypospadias), and bladder defects (e.g., urachal fistula, exstrophy of the bladder or cloaca). C left lip with or without cleft palate is the major oro facial defect. Skeletal abnormalities include craniofacial defects and skeletal dysplasias (e.g., cranioschsis, craniosynostosis and d warfism), limb defects (e.g., ra dial ray defects), vertebral defect s (e.g., spina bifida). Malformation in central neural system are neural tube defects (e.g., spina bifida), hypophyseal defects (e.g., phalangeal hypoplasiac), cranial defects (e.g., holoprosencephaly, hydrocephalus), and congenital megacolon (hirschsprung disease). 4 2 Neural tube defects are caused by faulty neurulation or developmental abnormality of the neural tube. 38 Neural tube defects can be deteriorated by hydrocephaly or other midline defects. 4 3 T eratogen induced MCM s mostly occur between the third a nd eighth week of gestation. 4 4 Any impairment before three weeks is more likely to result in fatality. T he fetus becomes less sensitive to teratogenic effects a fter the eighth week when the

PAGE 29

29 organs have developed. Figure 2 1 delineates the time window of e xposure to teratogens and associated MCMs and MAs. 44 Minor Anomaly (MA ) Minor anomaly (MA), also called minor congenital malformation s is the abnormal morphologic feature that do es not cause serious medical or cosmetic consequences 45 Identification of MA can be difficult due to the definition and the easy variable occurrence area 4 6 Approximately 70% of MAs occur on the face or the hands. 46 The prevalence of MA is less than 4% in the g eneral population, and varies by race, ethnic ity and gender. 45,46 In h ealthy newborns, about 15% to 20% have one MA, 0.8% have two MAs, and 0.5% have three or more MAs. 46 Table 2 1 lists some commonly seen MAs categorized by the occurrence area 47 MA mostly occur s after the eighth week of gestation which is so called feta l period 4 4 The use of t eratogen s during this period may induce MAs by disturbing the growth of tissues or organs. 44 Abnormal Conditions of the Newborn (ACNB) ACNB was adapted into the birth certificates by all States and th e District of Columbia in 1992, but t he conditions included in this outcome vary from state to state Table 2 2 shows the conditions of ACNBs summarized and defined by a committee of Federal and State health officials from the Association of Vital Records and Health Statistics. 4 8 Low Bi rth Weight ( L BW) respectively as low birth weight (LBW), very low birth weight (VLBW), and extremely low birth weight (ELBW). Infants with low birth weight are likely to be born before 37 weeks

PAGE 30

30 of pregnancy. In 2009, 8.16% of live born infants showed low birth weight 50 The high rate of infant mortality and morbidity associated with low birth weight has been documented in previous studies. 51 Although this positive association has been a meliorated over time with improved perinatal technology and intens ive care, low birth weight and prematurity still have been identified as risk factors predisposing to cardiovascular dysfunction, lung disorder, hypertension, type 2 diabetes, renal diseases autism, and developmental delay. 5 2 5 6 Pregn ancy and Obstetric Complication (PC/OC) M aternal health during pregnancy has a profound effect on the health of the developing fetus and new born infants. Pregnancy complications are a wide range of adverse sym ptoms and health problems occurr ing in the mother during pregnancy. Previous studies have documented the association between maternal pregnancy complications and preterm birth, which leads to approximately 70 80% of infant mortality and morbidity. 5 5 Gestat ional hypertension, pre eclampsia, and eclampsia are major pregnancy complications associated with maternal exposure to AEDs. 5 8 ,5 9 Gestational hypertension, also called pregnancy induced hypertension, is d e fined as elevated blood pressure (>140/90 mmHg) a fter 20 weeks of pregnancy without proteinuria. 5 8 T hese risk s increase with an earlier occurrence and higher severity of hypertension. In general, about 25% of women with gestational hypertension develop pre eclampsia later and 50% have pre eclampsia if ge stational hypertension occurs prior to 30 weeks. 60 Pre eclampsia is diagnosed when gestational hypertension occurs in association with proteinuria in pregnant women after 20 weeks of pregnancy. 61 Pre eclampsia is a

PAGE 31

31 leading cause of preterm births, and may lead to other serious complications, including eclampsia and fetal death. 6 2 Eclampsia manifest s as tonic clonic seizures occurr ing in pregnant women who ha ve developed preeclampsia and ha ve not had a previous brain condition before pregnancy. 6 3 It is a ser ious medical condition that can cause fetal death. Preterm birth and caesarean section are two major obstetric outcomes in Women with Epilepsy ( WWE ) using AEDs. 5 8 Preterm birth is a birth at less than 37 weeks or at least three weeks before the due date. A ccording to the Centers for Disease Control and Prevention ( CDC ) 1 in every 8 births (more than a half million infants) in the United States are preterm each year. 6 4 Gestational hypertension, pre eclampsia, and eclampsia all can cause preterm birth. A c ae sarean section (or C section) is a surgical procedure used to deliver a uterus. The rate of Caesarean section s has increased worldwide, and was 31.8% in the United States in 2007 6 5 Postpartum hemorrhag e, the loss of greater than 500ml of blood after vaginal delivery or 1000ml of blood after cesarean section, is a major cause of maternal mortality in developed countries and worldwide. 6 6 A p revious study showed that AED is associated with the increased po stpartum hemorrhage after vaginal delivery (OR=1.29, 95%CI: 1.02 1.63). 6 7 Recent studies reported that WWE using AEDs had a 50% increase of gestational hypertension, 4 fold of pre eclampsia, 1.5 fold of caesarean section, 4.9 fold of preterm birth, 5.4 fo ld of early term term vaginal bleeding ( 28 weeks) after adjusting for other potential risk factors. 5 8 ,6 8 ,6 9

PAGE 32

32 Correlation between Component Outcomes Previous studies revealed that the aforementioned outcomes are highly relat ed to each other. 5 9 70 7 5 Significant morbidity, mortality, and childhood disability are associated with VLBW, ELBW, MCM MA, or serious pregnancy or obstetric complications. 5 8 70 7 5 Approximately 9 26% of neurosensory disabilit ies 6 42% of evolving cogn itive dysfunction, 1 15% of blindness, and 0 9% of deafness occurred in infants born with VLBW and ELBW. 71 A significant ly higher risk of developmental delay was found in infants born with MCM (prevalence ratio: 8.3, 95%CI: 7.6 9.0). 7 2 A 44% 86% mortali ty rate in infants with ELBW ( 500 750g ) was reported 7 3 Furthermore, infants with one, two, or three MAs w ere found to hav e a risk rate of associated MCMs of 3%, 10%, or 20%, respectively 46 Some common risk factors, involving social economic status, infan t gender, maternal age, race, BMI, smoking, alcohol consumption, nulliparity, comorbidity, and comedication during pregnancy, have been found associated with these outcomes. 7 5 7 8 Many p revious studies on the adverse perinatal and pregnancy effects of drug s are inconclusive. 8, 27, 28 It is difficult to distinguish between the real non inferior results and power deficiency owing to rare outcomes. Liu and Roth developed a n LVM to incorporate four important BD outcomes into a single measurement, the infant mor bidity index, to describe an BD 13 We will apply this method to combine all aforementioned adverse pregnancy and infant outcomes into a continuous inde x of overall APO in this drug safety study T he combined outcome will be eva luated in terms of validity and reliability to ensure the appropriate use of this new methodology.

PAGE 33

33 Important Teratogens Antiepileptic Drugs (AEDs) It has been noted that taking AEDs poses an increased risk of having child with congenital malformations in WWE. 7 9 The most common MCMs caused by in utero exposure to AEDs are orofacial clefts, cardiac abnormalities, neural tube defects, urologic defects, and skeletal abnormalities. 80 Previous studies reported a two to three fold increase in the malformation ra te among offspring with in utero exposure to AEDs. 21, 22, 81 8 2 The occurrence rates were reported 3.1% to 9.0 % for MCMs, 37% for one MA, and 11% for two MAs, in offspring with in utero exposure to AEDs. 21, 80 8 3 Several previous studies found that AED polyth erapy poses a higher risk of malformation than that of AED monotherapy. 8 3 8 5 The rate of MCMs in WWE wa s reported 4.5% for AED monotherapy, and 8.6% for AED polytherapy. 8 5 The mechanisms for the teratogenic effects caused by AEDs are still under investigat ion. Some researchers believed that unstable epoxides interfering with the regular development of the fetus could play a role. 3 7 AEDs such as VPA can interfere with folate metabolism and cause a deficiency of folic acid. 8 6 Phenytoin, barbiturates, and carb amazepine are folic acid antagonists that can cause folate malabsorption and lower folic acid levels in serum. 8 7 A prior study suggested that the metabolism of AEDs may produce cyto toxic free radicals. 8 8 r adical scavenging can enhance the teratogenic effects of these cytotoxins. 8 8 ,8 9 Furthermore, recent studies in an animal model found that in utero exposure to certain AEDs ( VPA clonazepam, diazepam, p henobarbital, phenytoin, or vigabatrin) can cause wides pread neuronal apoptosis of the offspring. 90 91 Th is effect was found to be dose

PAGE 34

34 dependent. 90 91 Further investigation is needed to examine whether similar phenomena occur in human beings. The current guideline s for AED use in WWE recommends monotherapy w ith the lowest effective dosage avoiding VPA, and supplementing with folate. 22 ,42 Despite the af ore mentioned concerns of BDs in AED use, current literature lacks comprehensive population based information on AED utilization trends o ver time in pregnant w omen Valproic Acid (VPA) Several studies have reported a greater risk (6.2% 20.3%) of MCM with in utero exposure to VPA 6,7,25, 9 2 9 6 MCMs caused by in utero exposure to VPA include neural tube and limb defect s (e.g. 1 5% of spina bifida), heart defects (hypoglycemia and cardiac malformations), hydrocephalus and micro cephaly, urogenital defects, and fetal valproate syndrome. 22,9 5 ,9 7 100 Fetal valproate syndrome, such as epicanthal folds, flat nasal bridge, small upper lips, and a downward turned mouth are also malformations caused by in utero exposure to VPA 9 9 Developmental delay is a major fetal valproate syndrome characterized by multiple anomalies. 9 5 101 It has been reported that in utero exposure to VPA was associated with elevated risk of impaire d cognitive function for children at 3 years of age, and reduced cognitive abilities in multiple domains for children at 6 years old compared to those without VPA exposure. 9 8 101 Overall, the risk of malformations for offspring with in utero exposure to V PA is 7.3 fold higher than that of no n expose d, and 4 fold higher than those exposed to all other AEDs. 7 In most of the studies, the teratogenic effect of VPA was dose dependent. 6,8 3 ,9 5 T wo critical dosage thresholds 1000 mg/day and 1400 mg/day have bee n reported for pregnant WWE. 7,24, 9 2 9 8 The risk of malformation increases as the VPA dosage exceeds 1000mg/day. T he risk of BDs is 34.5% f or VPA dosage s above 1400 mg/day and 5.5%

PAGE 35

35 as VPA dosages is below 1400 mg/day. T he teratogenic effects cannot be diff erentiated with the dose changes if the dosage falls below 1000mg/day 8 3 Theoretical Framework for Latent Variable Model ing Latent Variable Model (LVM) Latent variable s are unobserved constructs that can only be assessed indirectly b y observable manifest v ariables A n LVM is a statistical model that includes one or more unobserved latent variable s and a set of observable manifest variables. An important assumption for LVMs is l ocal Independence defined as that t he manifest variables are conditionally ind ependent upon a given latent variable and the relationship among the manifest variables is fully explained by the latent variable 10 2 10 3 Table 2 3 presents four different types of LVM s based on the scales of latent and manifest variables. 10 4 W hile the L VM became more popularly used in m od e rn psychometrics in 2008 Rabe Hesketh and Skrondal (2008) review ed and test ed all classical LVMs ( listed in T able 2 3 ) to demonstrate the extensive application of LVMs in medical research providing their standardized t erminology, no need of modification, and high efficiency. 12, 10 5 Latent Trait Analysis (LTA) Latent Trait Analysis (LTA), also called item response theory, has been widely used in psychometrics, such as in the Rasch model. LTA is conducted by deriving a co ntinuous latent variable from the observed categorical manifest variables. In LTA, the manifest variables are discrete, including dichotomous, nominal, or ordinal variables. The conditional distribution of manifest variables with a given latent variable ca n be

PAGE 36

36 binomial (if two levels) or multino mial (if multi level presented), whereas t he distribution of continuous latent variable can be assumed as normal, or log norma l 13,10 4 In Rabe Hesketh and Skrondal paper (2008) an IRT model with continuous latent variable s and categorical manifest variables has been described in detail. 12 Rabe Hesketh and Skrondal (2008) show s that the final inference i s based upon the marginal likelihood integrating out the latent variable This la tent variable indicate s the ability of a subject ( also called ability parameter in gen e ral or latent trait in IRT ) and is assumed as normally distributed. If the assumption of normal distribution practically for the latent variable a nonparametric maximum likelih ood estimation is suggested 12, 10 5 Latent Variable in Latent Trait Setting McCulloch (2008) advocated utilization of a latent variable joint model for multiple outcomes and suggested tha t it would be more scientific for BD stu dies to assess the overall ri sk or joint effects of teratogen on all the outcomes. 10 6 McCulloch (2008) further illustrated more benefits of using joint modeling fo r multiple outcomes, such as avoiding multiple testing and correlation among outcomes, achieving higher efficiency, and be tter handling of missing data especially data missing at random (MAR). 10 6 Liu and Roth ( 2008 ) developed a n LVM in a latent trait setting with a continuous latent variable and four categorical manifest variables : BD abnormal condition of new born ( ACNB ) LBW, and Developmental Delay or Disability (DDD) 13 This model combined four categorical BD outcomes: BD (dichotomous), ACNB (dichotomous), DDD (dichotomous), and LBW (4 level ordinal), into a single continuous latent variable labeled as infant morbidity index This latent variable is a probability of the morbidity BDs

PAGE 37

37 study (2008) infant morbidity index was assumed a continuous variable greater than and equal t o zero with a log normal marginal distribution due to the higher frequency of lower risk outcomes and an extended right tail for asymmetric distribution. 13 The density function for the log normal distributio n was described and plotted in F igure 2 2 As presented in F igure 2 2 the distribution of log normal is asymmetric and has a heavy and extended right tail. BD, ACNB and DDD a re categorized as a binary variable, and have Bernoulli distribution s 10 7 L BW was modeled as a multinomial variable since the four birth wei ght categories are mutually exclusive and each has its own probability The summation of the individual probabilities equals one. component outcomes are independent g iven the latent variable 10 2 ,10 3 Hence the overall probabilities of four component outcomes condit ional on the latent variable are equal to the products of conditional probability for each individual component outcome. 10 7 s the relati onship between joint distribution conditional distribution and marginal distribution. 10 7 (2008) delineated the general joint distribution of infant manifest outcomes as an integral of product of overall probability of four component outcomes conditional on the latent variable and the marginal distribution of the latent variable 13 (2008) 1 3 the conditional distribution for each categorical component outcome (CMM, ACNB DDD, or LBW) was model ed as a nonlinear function of the infant morbidity index In this nonlinear function, therefore, Liu and Roth (2008) assumed that the probability of an infant having an individual birth defect outcome

PAGE 38

38 would be zero if the infant morbidity index was zero, and every normal level (no birth defect or normal weight) was treated as a reference. 13 Although maximum likelihood estimation has been widely used for LVM s previous studies have reported that Maximum Likelihood estimation may converge to sult in multiple solutions due to the non concave of the log likelihood function. 1 0 9 1 11 The Gauss Newton algorithm is a method used to solve non linear least square function s by minimizing a sum of squared function and neglecting the second derivatives. 11 2 114 Several researchers suggested the modified Gauss Newton method, which incorporates the Fisher scoring algorithm with the Gauss Newton algorithm for Estimated Generalized Nonlinear Least Squares (EGNLS) estimation 15, 112 11 6 It was assumed that the s ample proportion for each outcome is consistent and an unbiased estimate of cell probability Therefore, based upon nonlinear least square theory, the quadratic form can be minimized to obtain EGNLS estimates using modified Gauss Newton method 1 5 ,114 (2008) 13 provided an advanced statistical technique for application in drug safety studies for detecting safety signal s from multiple rare adverse ev ents. In this study, we used this model to combine BD, ACNB LBW, and PC/OCs to quantify the associat ion of maternal drug exposure with overall APO for infants and mothers

PAGE 39

39 E valuation of Combined Outcome Assumptions for Composite Outcome A composite out come constitutes more than two component outcomes, one of which where the patients are considered to experience the composite outcome. 11 7 ,1 1 8 Composite outcome s ha ve been used extensively in clinical trials due to the increased event rates and improved sta tistical efficiency with the fixed small sample size. 11 7 ,1 1 9 ,1 20 Furthermore, using a composite outcome can help researchers select several outcomes 1 1 8 ,1 21 ,1 22 However, an argument remain s in the use of composite outcomes. M any researchers point ed out that composite outcome s c ould mislead and make the results difficult to interpret. 12 3 12 6 T hree important assumptions were proposed as a guidance of validation of composite outcome s : 1). s imilar impor tance to the patient; 2) s imilar event rates, and 3) s imilar treatment effects across the component outcomes. 12 5 12 7 Following these guidance, composite outcome s used in published clinical trials w ere systematically evaluated 12 6 12 8 Although a small gradi ent in importance might benefit the use of the composite outcome in clinical decision making 12 7 the large gradient in importance and treatment effects across the components could mislead results 12 6 ,1 2 8 Unid imensionality Given that only one latent variab le is inferred in the latent trait model in this study, unidimensionality is defined as the significant correlation between the combined latent variable and each component outcome EFA and CFA are two important method s for assess ing the unidimensionality o f a latent variable 5 ,12 9 Figure 2 2 (a) and (b) depict the general diagrammatic representations for EFA and CFA models. 5 In F igure 2 2 (a), Y 1 Y 5 are observable manifest variables that share common variances of two latent

PAGE 40

40 variables (S1 and S2). 12 is t he correlation between latent variable S 1 and S 2 ij are factor loadings, the correlation between ith observable manifest variable and jth latent variable. 1 5 are the residuals and represent the variance that are unique to each observable manife st variable and not explained by the latent variables. 5 As shown in F igure 2 3 (a), t here is no pre specified relationship between the latent variables and each observable manifest variable in EFA model. 5 EFA depends on the size of factor loadings ( the cor relation between the latent v a ria ble s and each observable manifest variable) for identifying unidimensionality. for EFA suggest s that the smallest correlation coefficient can be set at 0.3 to attribute the components to one factor 1 30 F igure 2 3 (b) delineates the pre assumption for CFA model. 5 I n CFA model it is assumed that latent variable S 1 only relates to the component outcome Y 1 Y 3 and S 2 only relates to Y 4 Y 5 5 Therefore, CFA model is pre specified by restricting the observ able manifest variables to a specific latent variable. Local Independence LVMs are based on the assumption of local i ndependence that assumes no significant association among the observable manifest variables after controlling for the latent variable. 13, 1 02 ,10 3 Previous studies have demonstrated that locally depen dent component variables in LVMs impacted the validity and reliability of the latent variable. 5 1 31 ,1 3 2 An EFA model can be employed to detect both unidimensionality and local dependence by examin ing the residual correlation matrix for each observable manifest variables. 5 12 9 ,1 31 O ne of the components that have significantly correlated residuals should be removed to eliminate the local dependence. 1 3 3

PAGE 41

41 Construct Validity The composite outcome in thi s study is not a regular composite outcome defined as the occurrence of any components. Instead, it is an unobservable latent variable inferred from observable component variables. Hence, the construct validity needs to be 18 Face validity and content validity address whether or to what exten t the components are adequate and representative of all the components depending upon the judgments of the experts and patients. 1 3 4 Construct validity consis ts of several types of validities that assess the construct from different aspects. Convergent validity measures how well the proposed measure similar health status or disea se process es 18 Discriminant validity examines the non correlation between the new construct and an irrelevant outcome. 1 3 5 Construct validi ty is established after verifying convergent and discriminant validity. Discrimina t ive validity examines whether the new construct can differentiate between the extreme groups, in which the group that is expected to perform worse performs worse 18 Reliability In classical psychometrics, reliability refers to the changes of measurement over time, including test retest re liability and internal consistency. 1 3 4 ,13 5 For the composite outcome obtained from a n LVM there is no difference between test and retest, thus, test retest reliability is guaranteed. An i n ternal consistency that tests the correlatio n between the componen ts can become each other or with the composite outcome. 1 3 4 ,13 5 Given that the composite o utcome is not a summed score, Therefore,

PAGE 42

42 correlations between each pa ir of components, and correlations between components and combined outcome, can be calculated to assess internal homogeneity.

PAGE 43

43 Figure 2 1 Types of Birth Defect and Time Window of Exposure to the Teratogens. Source: Persaud TVN, Chudley AE, Skalko RG, Basic concepts in teratology, Alan R. Liss, Inc. New York. 1985. 44

PAGE 44

44 Table 2 1. Commonly Seen Minor Anomalies (MAs). 47 Cranium and scalp Sinuses Skin Multiple hair whorls Branchial Shoulder dimples Absence of hair whorl Preauricular Sacrum di mples Patent metopic suture Ear lobe Dimples over other bones Metopic fontanel Helical Sole crease Sagittal fontanel Pilonidal Horizontal palmar crease (single) Parietal foramen Face and neck Bridged palmar crease Flat occiput Synophrys Single crease, finger V Prominent occiput Flat bridge of nose Skin tags Frontal bossing Prominent bridge of nose Hemangioma Flat brow Hypotelorism Nevi Ears Hypertelorism Pigmented spots Microtia Nostrils anteverted Hypopigmented spots Darwinian point Long nasal septum Trunk Darwinian tubercle Epicanthal fold Extra nipples Lack of helical folding Iris freckles Single umbilical artery Bridged concha Upward palpebral slant Umbilical hernia Ear lobe crease Downward palpe bral slant Diastasis rectus Ear lobe notched Short palpebral fissures Grandular hypospadias Ear lobe bifid Cleft uvula Shawl scrotum Lop ear Cleft lip microform Vaginal tag Cup shaped ear Cleft gum Limbs Retroverted ear Long philtrum C ubitus valgus Thickened helix Short philtrum Tapered fingers Helix excessively folded Smooth philtrum Overlapping fingers]toes Helix attached to scalp Microstomia Broad thumb, great toe Preauricular tags Macrostomia Clinodactyly Helical p its Maeroglossia Nails hypoplastic Preauricular pits Microglossia Nails hyperconvex Broad alveolar ridge Increased space, toes

PAGE 45

45 Table 2 1. Continued Face and neck Limbs Micrognathia Syndactyly, toes 2 3 Webbed neck Overlapping digits Redundant neck skin Heel prominent Ptosis Source: Stevenson RE, Hall JG, 1993. Terminology. Pp. 21 30 in Hall JG, Goodman RM (eds.) Human malformations and related anomalies, Vol. I, edited by Stevenson RE, Hall JG, and Goodman RM. Oxford Univ. Press, New York.47

PAGE 46

46 Table 2 2 Definitions and Symptoms of A bnormal C ondition s for the Newborns. 4 8 Conditions Definitions Symptoms Anemia less than 13.0 g/dI of Hemoglobin level, or less than 39% of Hematocrit Reduced activity, l imited cardiovascular reserve, p oor growth. Birth injury A structural or functional impairment of during birth Macrocephly, v ision disorders, h eadaches, i rritability or sleepiness, v omiting or nausea, seizures. Fetal alcohol syndrome A syndrome of altered prenatal development and growth in infants born of women with high levels of alcohol consumption during pregnancy. Poor growth, d ecreased muscle tone, p oor coordination, d elayed development in movement, speech, thinking, or social ski lls. Heart defects, f ace d efects. 4 9 Hyaline membrane disease/Neonatal respiratory distress syndrome (RDS) A defect due to prematurity, manifested by pulmonary hyaline membranes, incomplete expansion of the lungs, and respiratory distress during birth. Bl uish skin color and mucus membranes, apnea, decreased urine output, grunting, nasal flaring, rapid breathing, shallow breathing, shortness of breath and grunting sounds, drawing back of the chest muscles while breathing.

PAGE 47

47 Table 2 2 Continued Conditio ns Definitions Symptoms Meconium aspiration syndrome Aspiration of meconium and amniotic fluid into the lungs and impairing the lower respiratory system during birth. Bluish skin color, breathing disorders, limpness. Assisted ventilation A mechanical method of assisting respiration. Respiration difficulty Seizures Physical alters of behavior after an episode of sudden abnormal electrical activity in the brain. Isolated abnormal movements of a single limb, staring spells. Abnormal rhythmic jerking, st iffening, crying out.

PAGE 48

48 Table 2 3 Four Types of Latent Variable Models. 10 4 Unobservable Latent Variable Observable Manifest Variables Continuous Discrete Continuous Factor Analysis (FA) and Structure Equation Model (SEM) Latent Profile Analysis (LP A) Discrete Latent Trait Analysis (LTA) Latent Class Analysis (LCA) Source: Bartholomew, D.J., and Knott, M. (1999). Latent Variable Models and Factor Analysis. London: Arnold. 104

PAGE 49

49 Figure 2 2. Log 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 0 0.5 1 1.5 2 2.5 3 3.5 fx(x; ) x =0, =0.25 =0, =1

PAGE 50

50 Figure 2 3 Unidimentionality measurement M odels A ) Exploratory Factor Analysis (EFA) B ) Confirmatory Factor Analysis (CFA) Source: Kelly SW, Volurka RJ. The empirical assessment of c onstruct validity, Journal of Operations Management 16 (1998) 387 405. 5 11 Y 2 Y 5 Y 4 Y 3 S 2 S 1 21 12 13 23 14 24 15 25 22 1 2 3 4 5 Y 1 12 A

PAGE 51

51 Figure 2 3 continued 11 Y 2 Y 5 Y 4 Y 3 S 2 S 1 12 13 24 25 1 2 3 4 5 Y 1 12 B

PAGE 52

52 CHAPTER 3 METHODOLOGY The methodology part consists of five sections : 1. description of the data sources and linkage; 2. selection of study population and overall study design; 3. assessment of exposure outcome and covariates ; 4. descriptive analysis involving demographic chara cteristics of study population and evaluation of components for combined outcome ; 5. s tudy d esign and s tatistical a nalysis for t hree s pecific a ims Data Sources This study is based on four statewide, retrospective 10 year data sources: Florida Medicaid claims, Florida Birth Vital Statistics (BVS), Florida Birth Anomaly, and Florida H ospital Discharge Inpatient and Outpatient from January 1 1999 to December 31 2009. Florida BVS at Florida Department of Health (DoH) collects and manages a statewide birth certificate dataset. Florida Medicaid claims data and Hospital Discharge data are provided by Florida's Agency for Health Care Administration (AHCA). Florida Birth Anomaly data is generated and maintained by the Maternal and Child Health Education and Data Center (MCHERDC) at the University o f Florida, College of Medicine, Department o f Pediatrics. The multiple datasets were linked, de identified, and released by MCHERDC for the ana lysis in this study. This study was approved by the Institutional Review Board of the University of Florida ( project #: 426 2010) and the Florida Department of Health (DoH) (protocol #: H12012) Florida Medicaid Claims Dataset and National Drug Code (NDC) Florida Medicaid claims data is a clinical, patient level dataset provided by Florida's AHCA. 1 3 6 The data is comprised of eligibility, medical, and pharma cy claims for

PAGE 53

53 services from inpatient hospital s outpatient clinic s emergency room s, and pharmacies Brief demographics for enrolled members a re included in Medicaid claims data, such as age gender, race, residency, etc. Medicaid claims data do not incl ude claims for managed care or Medicare enrollees. We excluded patients with dual eligibility, and thus restricted the drug exposure cohort to pregnant women who were only in the fee for service or primary care case management program. Medicaid claims da ta has been used for many pharmacoepidemiology studies. In this study, maternal exposure to study drugs was determined using National Drug Codes (NDCs) from pharmacy claim records in Florida Medicaid claims data. An NDC is an 11 digit unique number to iden tify drug product and is provided for each dispensed prescription in pharmacy claims. NDCs for all AEDs were retrieved using active ingredients and American Hos pital Formulary Services drug therapeutic classes from Multum Lexicon database (Cerner Multum, Inc., Denver Colorado, 2008). 1 3 7 A high and satisfactory translation of NDCs with the Multum Lexicon database was reported in a previous study. 13 8 Florida Birth Vital Statistics (BVS) In the United States, 99% of births take place and are registered in ho spitals. 1 3 9 In Florida, birth certificates are collected in the hospital within 24 to 48 hours after the baby birth. The department of BVS in Florida DoH collects and manages birth certificate data for all infants born in Florida. Birth time, place, and da te for infants, and demographic characteristics for infant, mother, and father are recorded in the birth certificate s Although BDs were re corded, validity and reliability of identifying BDs in the birth certificates were not frequently reported in previou s studies 1 40 1 4 5

PAGE 54

54 A H C A Florida Hospital Discharge Data Florida AHCA is the chief health care administration and regulation agency for the state. The agency collects and manages several discharge database s (inpatient, outpatient surgery, and emergency) f rom almost all Florida hospitals, excepting s tate operated hospitals. Inpatient data is collected from acute care, short and long term psychiatric, and rehabilitation hospitals. The data are updated quarterly an d have approximately 2.5 million records per year. Each record corresponds to an individual inpatient hospital stay. This patient level data includes patient demographic characteristics, diagnosis code s (ICD 9CM), procedure code s (up to 31 CPT codes), sour ce of admission, physician type, discharge status, and total amount of charges. 1 4 6 HID data has been reported to be efficient supporting data for other data source s to assess some important clinical outcomes, such as cancer or BDs 1 4 6 1 4 7 Florida Hospital Outpatient Discharge Data (HOD), also called Florida AHCA Ambulatory data, includes the outpatient discharge information collected from hospitals, lithrotripsy centers, ambulatory surgery centers and cardiac catheterization laboratories. Similar to HID, s tate operated hospitals are not included in th is dataset Florida HOD is updated quarter ly and collects about 3.1 million records annually with each record representing an outpatient visit requiring surgery or an invasive diagnostic procedure. The major fields in HOD include: demographics of patients, diagnosis codes, procedure codes, physicians, discharge status, and charges by department. 1 4 6

PAGE 55

55 Florida Birth Anomaly The Florida Birth Anomaly data set consists of BD registry data prepared and maintained by MCHERDC based on information from four data sources: Florida BVS, A H C Medical Services Medical Data Systems Infant BDs including MCM s and MA s were identified 0 365 days after live b irth from the four data sources. The Florida Birth Anomaly dataset was generated by merging Florida BVS Data with each of the other three datasets A probabilistic merging strategy was used and the merging w as 14 8 Approximately 99.6% o f the matching rate is obtained for the current study by combin ing Florida BVS data with the other dataset. 1 4 8 A c omplete case ascertainment using medical record review s of Florida Birth Anomaly data has not been conducted yet However, the validity and case ascertainment ha ve been documented in previous studies for identifying BDs using specific dataset s such as birth certificate s hospital discharge data etc. 1 40 1 4 8 1 5 4 G ood quality and accuracy (93% of sensitivity and 95% of specificity ) have been s hown in t he combined dataset from these three sources. 1 5 5 Data Linkag e MCHERDC cleaned and linked the data from multiple sources. Each data source was cleaned first and then linked with other corresponding dataset s using a multi step linkage approach in which three methods of linkage are applied in sequence Deterministic, Fuzzy Matching, and Probabilistic. 1 5 6

PAGE 56

56 Records were first matched deterministically based on exact matches of unique combinations of personal identifiers includi ng Social Security Numb ers Date of Birth, and Names (used for the linkage of BVS to Medicaid only). Records that cannot be exactly matched due to missing or poor data quality were linked using Fuzzy M atching. 1 5 6 ,15 7 Fuzzy M atching allows at least one occurrence of Soci al Security Number digit transpositions, name misspelling, or day or month errors in birth date fields. 1 5 7 Remain ing unmatched records were linked using probabilistic techniques, based on statistical weighting of combinations of personal identifiers. Prob abilistic linkage involved a two step process 1 ) D eterministic matchin g from the first merging step empirically derive d weights to the non missing fields based on successful linkages. 2 ) After the unlinked data matched with several records by weights, the match es with the high est statistical probability (indicating by high weights) was chosen The record remained un matched when no high weights could be obtained Study Population Study Cohort Inclusion Criteria This study includes female Florida Medicaid enrollees who were older than 15 years of age delivered a live singleton infant between April 01, 2000 and December 31, 2008 and were enrolled in the Medicaid program as identified by pregnancy status The study cohort of maternal baby pair s was generat ed by linking the Florida Medicaid claim dataset and Florida Birth Anomaly data using strategies described above. Study Cohort Exclusion Criteria Many women joined the Medicaid program after becoming pregnant. We excluded the women who were enrolled in Medicaid program after a positive pregnant

PAGE 57

57 test More exclusion criteria for maternal infant pair include: mothers with less than 6 month s of eligibility for Medicaid before pregnancy; mothers who lost Medicaid eligibility during pregnancy; mothers with d ual eligibility for Medicare, HMO, or other private insurance; mothers giving multiple births; mothers with diabetes mellitus ( diagnosed with ICD 9 code: 249.x, 250.x 790.29, or used of any antidiabetics during baseline) hypertension (diagnosed with ICD 9 code: 401.x, 416.x, 796.2, 997.91, 459.3, or used of any antihypertensive drugs during baseline) or HIV pre pregnancy (diagnosed with ICD 9 code: 042, 079.53, V08, V01.79, 795.71, or used of any anti retroviral therapy) ; Infants who were t wins, triplets quadruplets or more; o utliers including infants with birth weight lower than 350 g or higher than 6000 g; mothers or infants with critical information or perinatal medical informat ion. Overall Study Design The s tudy design for each specific aim is delineated in the fol l owing section. Overall this is a retrospective cohort study. Figure 3 1 shows the study design and critical time points for defining exposure and outcomes. The stud y index date was the The d rug exposure window was defined as the subsequent 9 month pregnancy period after the first day of the last menstrual date. A six month baseline period before the first date of the last menstrual date was utili zed to determine the baseline demographic and clinical characteristics. BD o utcomes were detected 0 365 days after live birth The study period lasts from Janua ry 01 2000 to December 31 2009.

PAGE 58

58 Assessment of Exposure Drug Exposure In this study, two drug exp osure groups, VPA and AEDs (including VPA), were defined to imply two scenarios that have different association with four component outcomes. Drug exposure was determined from pharmacy claims using NDC codes derived by active ingredients and drug classes f rom the Multum Lexicon database 1 3 7 VPA use was defined as prescription s dispensed for VPA, sodium valproate, or a mixture of the two (valproate semisodium), including all available doses and release forms, and the various brand names (depakene, depakote, epilim, and stavzor). AEDs were classified as a group of drugs used in the treatment of epilepsy seizures, including: VPA, phenobarbital (solfoton), phenytoin (di phen, dilantin, phenytek), primidone (mysoline), ethosuximide (zarontin), carbamazepine (car batrol, epitol, equetro, tegr e tol), levetiracetam (keppa), felbamate (felbatrol), gabapentin (neurontin), lamotrigine (lamictal), topiramate (topamax), tiagabine (gabitril), oxcarbazepine (trileptal), pregabalin (lyrica), and zonisamide (zonegran). Pharmac y claims in Medicaid data has been approved as an accurate source for the assessment of drug exposure. 1 5 8 The rates of concordance with patient self reported medication use was high in previous studies. 1 5 8 ,1 5 9 Information of drug use from M edicaid pharmacy claims include d NDC codes for filled prescription drugs, and the number of days for which the drug was supplied. 1 60 As previously mentioned the birth anomalies are related to exposure time during pregnancy, MCM associates with teratogen exposure at the f irst trimester, and MA and LBW relate to the maternal drug exposure in the third trimester. 1 61 Maternal drug exposure during entire pregnancy period can affect the combined outcome T he exposure window thus, was established as the period of 14 days before the first day of

PAGE 59

59 The drug exposure was defined as any one dose of study drug dispensed during the exposure window, including which the drug was dispensed prior to the exposure window but its days of supply covers at least one day of the exposure window A dding 14 days prior to the pregnancy is to take into account of the conception period and the residual effects of AEDs. Sensitivity study was conducted to examine the effects of different drug exposure window s on the combined outcome. The m LMP was obtained using a multistep algorithm derived from linkages to the healthy start prenatal screen, which contains more accurate information because it is collected during the first trimes ter of the pregnancy 1 6 2 When healthy start information is not available, then the LMP date on the birth certificate was used. If these two dates are not available (about 13% of LMP in birth certificates are missing), then this information w as imputed from clinical estimat e 1 6 3 1 6 5 A p revious study suggested that LMP from birth certificates and clinical estimat es agreed within less than 2 weeks. 1 6 6 Assessment of Component Outcome s In this study, we identified four individual adverse pregnancy and infant ou tcomes: BD (involving MCM and MA) ACNB, LBW and PC /OC from multiple data sources. Major Congenital Malformation ( MCM ) and Minor Anomaly (MA) MCM s and MAs were collected from birth to the first 365 days of life using the 9 th edition of the International Classification of Diseases Clinical Modification (ICD 9 CM) code (740 759.9) from HID and HOD.

PAGE 60

60 Based upon the definition of MCM used by the CDC Metropolitan Atlanta Congenital Defects program, 42 CDC reportable congenital malformations were determined as MCM 16 7 These 42 specific MCM s were further grouped into eight major categories of congenital malformations: central nervous system, chromosomal, cardiac, gastrointestinal, genital and urinary, muscular and skeletal, oral clefts, and other defects. 1 6 7 1 6 8 The validation of the definition has been evaluated in previous studies. 1 6 9 The quality and relative contribution of different data sources for identifying MCM and MA cases were evaluated in previous studies. 1 6 5 ,1 70 Table 3 1 shows that the PPVs for over all or specific BDs identified using birth certificate and hospital inpatient claims can range from 68% to 94%. 1 71 It has been confirmed that Hospital Discharge data along with other CMS diagnostic information efficiently enhanced case ascertainment for BD cases from BVS data. 1 4 6 ,1 5 3 Abnormal Condition of New Born (ACNB) ACNB were identified from Florida Family Birth Anomaly data which was generated from birth certificate s and one year follow up s in infant hospital discharge data. Although birth certific ate are less reliable for reporting abnormal conditions, 1 40 previous validation studies showed that Positive Predictive Values (PPVs) ranged from 70.9% to 100% for identifying ACNB s from neonatal hospital discharge data. 16 9 Table 3 2 shows ICD 9 codes and CPT codes used for determining ACNB s from hospital discharge data. Low Birth Weight ( L BW) Infant birth weight i s accurately recorded in the birth certificate. 1 40 Only less than 1% of infants were missing birth weight records in the birth certificate, these will be excluded from the study. 1 7 4 The agreement between self reported birth weight and the

PAGE 61

61 birth weight recorded in the birth certificate is reported as fair to moderate. 1 7 5 In this study, birth weight was categorized in to four levels: Extremely Low Bi rth Weight (ELBW, 350 999 g), very Low Birth Weight (VLBW, 1000 1499 g), Low Birth Weight (LBW, 1500 2499 g) and Normal Birth Weight (NBW, 2500 5999 g). Pregnancy and Obstetric Complications ( PC/ OC s ) PC/OC s were identified either from BVS data or using IC D 9/CPT code s from Hospital Discharge data depending upon the extent of validity and reliability reported in the previous studies Appendix B 1 described the operational definition s of component outcomes Although BVS data showed poor validity and reliabil ity on PC /OC s other variables involving parental demographic characteristics, birth weight, delivery method, cesarean ind ications, and pregnancy history were accurately recorded in birth certificates 1 40 1 4 5 Fair to moderate PPV s of using ICD 9 cods in ho spital discharge data have been reported : 45.3% for mild preeclamsia 41.7% for eclampsia and 84.8% for severe preeclampsia 1 7 6 In our study, gestational hypertens ion, preeclamsia, and eclampsia were identified using ICD 9 c odes from HID and HOD. 17 7 ,17 8 P reterm birth was operationally defined as gestational age less than 37 weeks. 17 9 Gestational age was computed from the infant birth date and We have described the strategies of obtaining LMP from multiple resources. The evaluation of gestatio nal age from birth certificate data showed that the rate of false positive s and missed true cases for identifying preterm birth by using LMP is 15% and 20.5% specifically. 1 6 6 R educed error rate s in this study can be attributed to the use of the healthy sta rt prenatal screen data. 1 80 For OC, since birth certificates showed good validity on delivery method and cesarean indication, w e defined c esarean delivery and f orceps or vacuum extractor

PAGE 62

62 delivery from either birth certificates or ICD 9 codes in hospital di scharge data if it was missing in the birth certificates P ostpartum hemorrhage was identified solely using ICD 9 codes in hospital discharge data due to poor validity of birth certificate data on pregnancy complic ations and obstetric events. 1 40 Assessmen t of Controls and Covariates Control Group To assess treatment effect s on the combined outcome and each component outcome, a healthy control group defined as the infants with no maternal exposure to any AEDs during pregnancy, was selected for the compar ison. Compared with no AED user s, AEDs, as well as VPA ha ve shown significant increased risk of BD outcomes. 7 Previous prospective and retrospective cohort studies reported similar results in that the odds ratio of exposure to monotherapy VPA compared wi th no AED exposure was 7.3 with a 95% CI ( 4.4 12.2 ) 7 and the risk of VPA polytherapy on BDs is higher than that of VPA monotherapy 9 5 Covariates To assess treatment effects of the combined outcome and component outcomes the potential confounding facto rs were controlled using propensity score technique s The strategies of v ariable selection for propensity score ha ve been discussed by several investigators 1 81 1 83 It has been suggested that selecting variables associated with an outcome for propensity mo del may reduce bias. 1 81 Previous studies have documented that common risk factors for adverse maternal and infant outcomes include socioeconomic status, infant gender, maternal age, race, BMI, smo king, alcohol consumption, parity, and drug exposure during pregnancy. 7 5 7 8 Using the f ather residence postal code to assess socioeconomic status has been validated in previous

PAGE 63

63 studies 1 8 4 Though the FDA categorized drugs with teratogenic effects in the pregnancy categories D and X, prevalence of m aternal exposu re to category D and X drugs were reported as 4.8% and 4.6%, respectively 1 8 5 ACE inhibitors, antipsychotic, antidepressants, anxiolytics, antidiabetics, antihypertensive, hormones, isotretinoin, therapeutic radiation a ntineoplastic, coumarin based anticoa gulants, Iodides, antibiotics, glucocorticoids, quinine, NSAID, antiviral and other anticonvulsants are significant teratogens that were controlled during treatment effect assessment in this study 1 8 6 1 91 Furthermore, the medical indications, such as, epi lepsy, gestational diabetes, mental disorder, seizure, phenylketonuria, hypoxia, hyperthermia, myasthenia gravis, rheumatic disease, viri lizing tumors, and infection a nd parasitic diseases including v irus, r ubella, c ytomegalovirus, s yphilis, h erpes simplex virus, t oxopla s mosis, v aricella virus, and v enezu elan equine encephalitis virus were documented as teratogens in previous studies, and thus, controlled in this study 1 92 ,1 93 Demographic characteristics were identified f rom birth certificates, whereas co m orbidities or co medications during pregnancy were identified using ICD 9 and NCD codes from HOD or HID. The operational definition s for all covariates adjusted in data analysis were described in Appendix B 2 Propensity Score We collapsed all covariates i n to a propensity score and adjust ed in a multivariate model for evaluation of the treatment effect P ropensity score is a predicted probability of drug exposure and can be calculated using unconditional logistic regression with exposure as the dependent va riable ( Equation 3 7). T he exposure and nonexposure groups can be balanced with respect to the included covariate s to remove bias caused

PAGE 64

64 by confounding factors 1 9 4 The p ropensity score can be match ed stratifi ed or adjust ed depending upon study power and its distribution In this study, to avoid losing sample size due to unmatched pairs, we adjust ed the propensity score in a multivariate model. The distribution of the propensity score was examined and non overlapping area s w ere restricted due to exclusive clinical indications or contraindications for exposure and non exposure. 1 9 5 The p ropensity model is defined and described in Equation 3 1 for maternal exposure to VPA. Log [p/(1 2 X 2 3 X 3 4 X 4 5 X 5 6 X 6 7 X 7 8 X 8 m X m (3 1 ) w here m=4 0 and p: probability of dispensing VPA during pregnancy. Intercept, the value assumed all other covariates =0. 1 4 0 : Estimated parameters for the specific variables (x 1 x 4 0 ) X 1 : Maternal age at infant born X 2 : Race (White, Black, Native American, Asian, Hispanic, Unknown) X 3 : birth X 4 : X 5 : Fa X 6 : Mother previous adverse pregnancy experience X 7 : X 8 : Marital Status X 9 : X 10 : X 11 : X 12 :

PAGE 65

65 X 1 3 : X 1 4 : X 1 5 : X 1 6 : X 1 7 : ves and hypnotics) exposure during pregnancy X 1 8 : X 19 : Number of hospitalization for seizure during pregnancy X 2 0 : Number of physician visits with seizure diagnoses during pregnancy X 2 1 : use during pregnancy X 2 2 : X 2 3 : Rubella, Cytomegalovirus, HIV, Syphilis, Herpes simplex virus, Toxoplamosis, Varicella virus, V enezuelan equine encephalitis virus, Phenylketonuria, Hypoxia. X 2 4 : X 2 5 : X 2 6 : g baseline or pregnancy X 2 7 : X 2 8 : X 29 : X 3 0 : g tumors diagnoses during baseline or pregnancy X 3 1 : X 3 2 : X 3 3 : based anticoagulants exposure during pregnancy X 3 4 : re during pregnancy

PAGE 66

66 X 3 5 : X 3 6 : X 3 7 : X 3 8 : X 39 : uring pregnancy X 4 0 : Descriptive Analysis Descriptive analyses were performed to compare demographic and clinical characteristics between patients who used AEDs or VPA and non AED users Continued variables were compared using a student t test, and categorical variables were examined using a chi square test. Missing data w ere examined during descriptive analysis P atients missing important information, i.e., maternal age, LMP, or infant birth weight, were e xcluded from the study. For other demographic or clinical characteristics that were adjusted as covariates, the missing data were deleted during analysis if missing is less than 10%. For the covariates missing 10~ 35 %, we evaluated the variables and made th e judgment whether the data w ere missing at random. We conducted an imputation for the data missing at random, and excluded the variables from the study if the data was not missing at random. We excluded the covariates that were missing over 35 %. S election of component outcomes was evaluated based upon three assumptions underlying the composite outcome. This step of evaluation is more qualitative since the importance, frequen cy rate, and treatment effect are not exactly the same for all compon ent outcomes. It has been shown that the more important component is less

PAGE 67

67 frequen t and has a smaller treatment effect 1 9 6 ,19 7 However, the component needs to be reconsidered if its treatment effect or importance or f requency rate shows large differen ce with others. The importance of the component outcome was assessed by computing a correlation between individual outcomes and the clinically meaningful endpoints. The length of inpatient stay ha s been accepted widely as a clinically meaningful endpoint in either clinical t rials or epidemiology studies based upon administrative claims data 19 8 ,19 9 We assessed the length of inpatient stay for the mother infant pair during delivery. was calculated between the length of hospital stay during deli very and each component outcome. (3 2) w here x i is the rank of the true value of the component outcome for patient i, y i i s the rank of true value of length of inpatient stay or number of outpatient visit for patient i, and n is the sample size. Furthermore, the f requency of component outcomes for each component outcome was assessed from the study population. A l ogistic regression m odel was fitted for each component outcome to assess the adverse perinatal and pregnancy effects of maternal drug exposure. Propensity score was adjusted in the model as a covariate. Logit [P(Y i = 1 )] 1 X i1 (3 3 ) w here Y i denotes component outcome i, P(Y i =1) is the probability of the component outcome i, and X i1 denotes the propensity score. coefficient the frequency of component o utcome and t he estimates from Equation 3 4 between each component outcome were

PAGE 68

68 compared and justified. A s tudent t test with a Bonferroni correction was used for the pairwise comparison between each component outcome 200 (3 4) w here and are the estimat es, refer ring to above mentioned , or for component outcome s i and j. and are the standard deviation s for the estimates T he denominator is the standard error of the difference of two compared estimates Using a Bonferroni correction, the null hypothesis is rejected for P<0.008. The results of the student t test w ere used as a reference to finalize the component outcomes. For the component outcome that is significantly different with others, we first made adjustment on its definition to av oid the dominating on others and then justified it based upon the suggestion from Montori et al. (2005) that a s mall gradient in importance frequency rate or treatment effect across the components would be benefic ial for the composite outcome. 12 7 Therefore, the component outcome, which is different from the others, but is not do minant, was kept in the study. Study Design and Statistical Analysis for Three Specific Aims In concordance with the specific aims in this study the methodologies for study design and data analysis are presented in three parts: (1) T he u tilization of A EDs in pregnant women, (2) Combin ing the adverse pregnancy and infant outcomes using a n LVM in Latent Trait Setting and (3) Evaluation of combined outcome s with the established association of VPA with adverse pregnancy and infant outcome s SAS 9.3 (SAS In stitute, Cary, NC) was used for data analyses and modeling, and SigmaPlot 11.0 (Systat Software Inc, San Jose, CA) or Microsoft PowerPoint 2010 14.0 (Microsoft Corp., Redmond, WA) were used for graphs.

PAGE 69

69 Part 1 : Utilization of AEDs in Pregnant Women Descrip tive statistics are used to describe the secular trend of utilization of AEDs or VPA only between July 1999, the start of the drug exposure period, and the end of the drug exposure period in December 200 8 Continuous use of AEDs was defined as at least tw o consecutive prescriptions for AEDs or totaling more than 30 days of supply. Polytherapy was defined as one or more other AED drugs continuously used for at least 60 overlapping days. Annual prevalence was estimated and compared across different drug comb inations, including the first and second generation AEDs. AED use was further compared between different indications, which were operationally defined using diagnosis codes with AED prescriptions The compared indications for AED use include epilepsy (345. xx), depression (296.2 296.3, 300.4, 311.xx), bipolar (296.xx, 301.13), anxiety (300.0, 308.xx, 309.21, 309.81), migraine (346.xx), myoclonus (333.2) and neuropathic pain (338.xx). Part 2 : Application of Latent Variable Model F our component outcomes BD ACNB PC/OC, and L BW were identified using the operational definition described in A ppendix B 1 BD ACNB and PC/OC were categorized to a binary variable and followed a Bernoulli distribution, and f our levels of LBW were modeled as a multinomial variable The u nobserved index of APO was assumed log normally distributed Based on and Baye t he joint distribution for component outcomes can be expressed as a n integral of product of multinomial variable for conditional distribut ion of each component outcome and marginal distribution of latent variable 102,103,108 Marginal distribution of the latent

PAGE 70

70 variable is described as log normal. Given the observed outcomes, we can obtain the posterior distribution of the index of APO Fur thermore, we assumed that the conditional distribution of each categorical observed outcome ( BD ACNB PC/OC or LBW) was nonlinear function of the latent variable APO 13 The conditional distribution of observed outcome and the latent variable APO was link ed by two parameters in the non linear function. T he probability of any specific observed outcome is equal to 0 when APO =0 because APO accounts for all variation of the four observed outcomes and the relationship among these four outcomes. 13 In the non lin ear function the probability of an infant having an individual BD outcome was assumed zero if APO is zero, and every normal level (no BD or normal weight) was treated as a reference. The latent variable APO positively associates with the four observed out comes. T he larger the APO the higher the probability of the observed outcome 13 Latent Trait Model was performed using SAS Proc IML The p roportion of each outcome combination was calculated Then each parameter was estimated using the iter ation function for EGNLS starting from iteration 0 with initialized value until the stepping coefficient was less than 10 9 The final results are the estimate s of all parameter s The estimate of latent variable APO was obtained by e ntering the computed pa rameters into posterior function 13 Part 3 : Evaluation of the Combined Outcome This part of the evaluation targets validity and reliability of the combined outcome based on the psychometric properties of a latent variable

PAGE 71

71 A im 3 1 Assess the U nidim ensio nality of the C ombined O utcome s EFA was performed in SAS using Proc Factor and maximum likelihood method to assess the unidimensionality of the combined outcom e 5, 201 The Equation 3 5 shows the relationship of the latent variable and manifest variables. 202 (3 5) w here S denotes the latent variable Y is a matrix of manifest variables, is a matrix of factor loadings, and is a matrix of errors and represents the variance that is not explaine d by the latent variable Commun al ity computed by summing squares of factor loadings ( i 2 ) represents the variance of manifest variables accounted for by a common latent variable A large communality indicates the strong relationship between the latent v ariable and manifest variables 202 Aim 3 2 Examine the L ocal I ndependence of the F our C omponent O utcomes 3 statistic is a well recognized method to assess local independence in IRT. 1 31 13 3 203 Since this study only has 4 component outcomes and 1 la tent variable, we 3 statistic (shown in Equation 3 6). 1 32,203 To explicitly calculate the residual of one factor model for each component outcome, CFA was employed using SAS Proc Calis to calculate t he correlation between the residuals obtained from C FA model for component outcome s i and j (3 6) where are residuals of co mponent outcome i and j obtained from Equation 3 5 denotes the correlation of and denotes the Q 3 statistic for component

PAGE 72

72 outcome i and j. The components with the residual correlation coefficients greater than 0.2 were considered as possible local independence. 203 Aim 3 3 Assess the Internal Homogeneity of the Combined O utcome s rank correlation was calculated between each pair of component outcomes. Using Equation 3 7 x i denotes the rank of true value of one component outcome for patient i, y i is the rank of true value of another component outcome for patient i, n is the sample size outcomes Based on the rule of the thumb in psychometrics, the c orrelation between each pair of component outcome s m ) is expected between 0.3 and 0.7 so that the component outcome is moderately correlated but not so highly correlated as to cause redundancy. 18 (3 7) Aim 3 4 Evaluate the C onvergent V alidity of the C ombined O utcome the combined outcome and total length of hospital sta y for mother infant pair during delivery In Equation 3 8 x i denotes the rank of true value of the combined outcome for mother infant pair i; y i is the rank of true value of total length of hospital stay during delivery for mother infant pair i ; n is the sample size and are the mean s of the rank. (3 8) Aim 3 5 Evaluate the Discriminant Validity of the Combined O utcome The s tatus of breastfeeding (Yes/No) recorded in birth certificates is an event irrelevant to APO. To assess discriminant validity,

PAGE 73

73 calculated between the combined outcome and the status of breastfeeding Using Equation 3 1 9 x i denotes the rank of true value of the combined outcome for patient i, y i is the rank of true value of status of breastfeeding for mother infant pair i, n is the sample size and are the mean s of the rank. Aim 3 6 Evaluate the D iscriminative V alidity of the C ombined O utcome A s tudent t test is used to examine the difference of combined outcome between the group with major congenital malformation and extreme low birth weight and the group having normal birth weight only. (3 9) w here is the mean APO of the group having major congenital malforma tion and extreme low birth weight, and is the mean APO of the group having normal birth weight only. and are the standard deviation for and and the denominator is the standard error of the diffe rence of two compared means Sensitivity An a lysis Sensitivity analyses will be conducted to estimate the influence of possible biases in the study. The o perational definition of the exposure time window depends on many factors. We will explore changes in the final estimated association if defined exposure time window s are altered to account for drug s residual effects. We would define the exposure window as 10 days before LMP to infant birth 20 days before LMP to infant birth and 30 days before LMP to infant birth and examine the association of combined outcome APO with maternal exposure to the study drugs. As outlined previously, infant birth defects largely relate to the time window of maternal drug exposure. Drug induced MCM relates to the exposur e in the first

PAGE 74

74 trimester, whereas, minor birth defects and LBW relate to the exposure in the third trimester. Therefore, a subgroup sensitivity study will be performed to assess the association between maternal exposure to AEDs and four component outcomes in four pregnancy trimesters

PAGE 75

75 Figure 3 1. Schematic Diagram of Data Linkage. Florida Medicaid Pharmacy Claims: NDC Florida Birth Certificate: Demographics Delivery Events OC, ACNB, LBW AHCA Florida Hospital Inpatient Discharg e: PC/OC, Comorbidity Florida Birth Anomaly: BD AHCA Florida Hospital Outpatient Discharge: PC/OC, Comorbidity

PAGE 76

76 Figure 3 2. A Schematic Plot for the Study Design and Critical Time Points. Index date, Infants born, from April 2000 to December 2008 Study begins January 1999 Study ends December 2009 0 day to 1 year after birth date Pregnant period, 9 months, Exposure Time Window Outcome Last menstrual date, 0 day of pregnancy Exposure 6 months baseline

PAGE 77

77 Table 3 1. PPVs of overall and specific birth defects identified using birth certificate s 1985 2000. Defects Birth certificate, Positive predictive value Inpatient claims, Positive predictive value Either data source, Positive predictive value n % n % n % All defects 239 69.9 1316 69.9 1430 67.7 Cardiac defects 84 35.7 490 74.5 545 67.7 Gastrointestinal 17 94.1 182 73.6 185 73.5 Musculoskeletal 68 94.1 260 65.0 295 67.8 Genitourinary 20 65.0 244 68.4 254 67.3 Central nervous system 30 63.3 139 48.9 153 47.1 Ofofacial 31 93.6 60 93.3 66 90.9 Source: Cooper WO, Hernandez Diaz S, Gideon P, Dyer SM, Hall K, Dudley J, Cevasco M, Thompson AB, Ray WA, Positive predictive value of computerized records for major congenital malformations. Pharmacoepidemiology and Drug Safety 2008; 17: 455 460. 171

PAGE 78

78 Table 3 2. ICD 9 codes for Abnormal Condition s of the Newborn 1 72 Conditions ICD 9 Code Anemia 280.x 289.x Birth injury 767.x Fetal alcohol syndrome and Drug withdrawal syndrome 760.7x, 779.5 Hyaline membrane disease/Neonatal respiratory distress syndrome (RDS) 769.x Meconium aspiration syndrome 770.1x Assisted ventilation 93.9 Seizures* 779.0, 345.x *: About 98% of Neonatal seizure s was determined by IC D 9 code 779.0, and 2% by ICD code 345.1 to 345.9. 1 73

PAGE 79

79 CHAPTER 4 RESULTS Characteristics of Study Population A total of 753,377 infants were born in Florida between 2000 and 2009. After applying all inclusion and exclusion criteria, the final study c ohort consisted of 3,183 mother infant pairs in the AED exposure group, 226 mother infant pairs in the VPA exposure group, and 43,956 mother infant pairs in the healthy control group (Figure 4 1). Table 4 1 summarizes and compares the demographic character istics of the study cohort between the healthy controls and mother infant pairs with maternal exposure to AED or VPA. Because records of the diabetes, prenatal usage, and the status of infant breast fed were not col lected in Florida BVS data before 2004, large amounts of missing data (55 58%) were found in these four variables. We have to remove these four variables from multivariate analysis. S mall amounts of missing data were found in the rest of the variable s exce pt for age and The proportion of missingness ranged from 0.004 % in the m marital status to 24 % The pattern of missingness was similar across three comparison groups in most of the variables except slightly high lower in the m (45% and 32% vs 60%), the m prenatal use (42% and 31% vs 57%), and i nfant male gender (6% and 3% vs 15%) for AED and VPA exposure groups. Compared to the healthy controls, AED and VPA exposure groups include fathers that were older white males with lower than high school education, and mothers that were older, white, smokers, not married, and having ot her risk factors.

PAGE 80

80 Table 4 2 presents baseline clinical characteristics that were defined from Medicaid claim data using NDC, CPT, and ICD 9 codes. Indications that are significantly higher in AED and VPA exposure groups are: epilepsy (24% and 18% vs 0.2%), anxiety (6% and 7% vs 0.5%), bipolar disorder (23% and 14% vs 1.1%), depression(9% and 10.3% vs 1.7%), migraines (5% and 3% vs 0.4%), mental disorder s (37% and 35% vs 9%), and infection and parasitic diagnos e s (12% and 11% vs 7%). Compared to healthy cont rols, AED and VPA exposure groups have significantly higher rate of exposure to other medications: antipsychotic s (20% and 11% vs 0.6%), antidepressants (37% and 28% vs 3%), folic acid (61% and 62% vs 41%), anxiolytics (31% and 58% vs 3%), antibiotics (48% and 45% vs 32%), glucocorticoids (4% and 5% significantly higher rate of comorbidities (35% and 30% vs 4%), and co medication use (43% and 48% vs 8%).,Overall, mothers exposed t o AEDs during pregnancy have more risk factors than healthy controls, which may pose elevated risk of infant birth defects. Descriptive Analysis of Observed Outcomes All observed outcomes, including BD (major & minor CMs), ACBN, PCOC (PC, OC, and preterm born), and LBW, were assessed based on the definition proposed in C hapter 3. First of all, the four observed outcomes were evaluated specifically according to the assumptions for selection of components for composite outcome. We compared the frequency, imp ortance, and drug effects across th ese four observed outcomes. In F igure 4 2, we compared the frequencies of incidences among BD, ACNB, PCOC, and LBW in the study population. The incidence of PCOC (28.26%) was 2~3 times higher than that of BD (10.65%), ACN B (8.04%), and LBW (8.64%). BD, ACNB, and LBW seem have similar incidence rates in the study population.

PAGE 81

81 As proposed in C hapter 3, length of hospital stay during delivery was employed to assess the importance of four observed outcomes. Table 4 3 lists Spe arman correlation between the length of the hospital stay during delivery and four observed outcomes. Correlation coefficients (Rho) were all statistically significant and ranged from 0.10 (for BD and ACNB) to 0.21 (for PCOC) in the study population. A Stu dent t test was conducted to compare the calculated Spearman correlation coefficients between BD and PCOC, which has the lowest and highest correlation with the length of hospital stay during delivery. P value was of 0.9862 indicating similar importance ac ross four observed outcomes. Drug effects of AEDs and VPA on four observed outcomes were analyzed using Propensity Score adjustment in multivariate analysis. e stimate s in Table 4 4 indicate the magnitudes of the effects of AED or VPA use on observed outcomes compared with healthy controls. As shown in Table 4 4, AED significantly associated with all four observed outcomes: BD ( Estimate and SE: 0.340.16, P =0.0356) ACNB ( Estimate and SE: 0. 60 0.1 5 P =0.0001 ) PCOC ( Estimate and SE: 0. 70 0.1 1 P <.0001 ) and BW ( Estimate and SE: 0. 10 40. 02 P <.0001 ) Comparing the outcomes with the lowest and highest Estimate s (BW vs PCOC), the effects between BW and PCOC were not significantly different for AED and VPA (P=0.4434 for AED and P=0.4568 for VPA). The observed outcomes were compared between healthy controls and AED or VPA exposed group. Given the large healthy control group and relatively small AED or VPA exposures, the frequency of four observed outcomes and correlations between four observed outcomes and the length of hospital stay during delivery were dominated

PAGE 82

82 by large amounts of healthy controls and very similar in AED (AED exposure + healthy controls ) and VPA (VPA exposure + healthy controls) study groups. Figure 4 3 presents the incident rates of PC, OC, preterm birth, and overall PCOC in three study groups. Compared to healthy controls, AED exposed group has significantly higher rates on PC (12% vs 8%, P<.0001), OC (11% vs 5%, P<.0001), Preterm born (21% vs 19%, P=0.0063), and overall PCOC (36% vs 28%, P<.0001). However, the VPA exposed group has slightly lower PC, OC and preterm born than AED exposed group, and is not significantly higher than those of healthy controls on PC (11% vs 8%, P=0.0459), preterm(20% vs 19%, P=0.6569), and overall PCOC (34% vs 28%, P=0.0509). The VPA exposed group has significantly higher rate of OC (10% vs 5%, P=0.0006). In F igure 4 4, the incidence rates of major CM, mino r CM, BD, and ACNB were plotted and compared among three study groups. The VPA exposed group has highest rates of BD (20% vs 10.5%, P<.0001), including major CM (6.5% vs 2.6%, P<.0001) and minor CM (16.7% vs 9.1%, P<.0001), which are significantly higher t han those of healthy controls. ACNB in VPA exposed group is lower than that of the AED exposed group, and is not different with the healthy controls (10.2% vs 7.8%, P=0.1525). The AED exposed group has lower rate of BD than VPA group, whereas it has signif icantly higher rate than healthy controls on major CM (4.1% vs 2.6%, P<.0001), minor CM (10.7% vs 7.8%, P=0.0025), BD (12.8% vs 10.5%, P<.0001), and ACNB (12.1% vs 7.8%, P<.0001). Figure 4 5 delineates the distribution of four BW categories (Normal: >2500 g, LBW: 1500~2500g, VLBW: 1000~1500g, and ELBW: <1000g) in three study groups.

PAGE 83

83 than healthy controls (88.6% vs 91.6%, 10.6% vs 6.7%, 0.4% vs 0.7%, 0.4% vs 0.99%, P=0.0752) The rate of LBW is the highest in AED exposed group, and is significantly higher than that of healthy controls (88.1% vs 91.6%, 10.6% vs 6.7%, 0.9% vs 0.7%, 0.5% vs 0.99%, P<.0001). Therefore, AED users are significantly different with healthy controls o n all observed outcomes, whereas VPA users are mainly different with healthy controls on BD and OC. Two exposure groups have varied pattern of observed outcomes that were combined using LTM. The psychometric properties of combined outcome were evaluated an d compared among these three groups. Further analysis results are reported for each hypothesis under each specific aim. Part 1: Utilization of AEDs in Pregnant Women Included in the Study The secular trends of AED use in the study population from 2000 to 2009 w ere analyzed for each hypothesis under specific aims 1 1, 1 2, 1 3, and 1 4. Specific Aim 1 1: Assess the secular trend for the utilization of AEDs. The percentage of any AED use in the study population was calculated for each study year. Out of to tal 47,139 mother infant pairs in Florida Medicaid program, AED exposure in mothers during pregnancy ranged from 0.2% to 0.54% from year 2000 to 2009. The secular trend was analyzed and plotted in F igure 4 6. Despite the observed small fluctuation, AED use significantly from 2000 to 2009 ( =0.01, P=0.28). Specific Aim 1 2 : Assess the secular trend for the utilization of VPA The trend of VPA use in the study population (N=47,139) from 2000 to 2009 was an alyzed and plotted in F igure 4 7. VPA use significantly reduced from 0.05% in 2000 to

PAGE 84

84 0.02% in 2009 in the study population, about 3 per 1000 deliveries every year ( = 0.003, P=0.017). The reduction is statistically significant. Specific Aim 1 3 : Compare t he utilization of AEDs from 1999 to 2009 in monotherapy and polytherapy in the pregnant women included in the study. Out of a total of 3,183 pregnant women exposed to AEDs, approximately half were exposed to more t han two AEDs during pregnancy (F igure 4 8 reach the statistical significance level, the secular trends for AED use in polytherapy and monotherapy changed over time. A joint point was detected at year 2006 with JointPoint program. Before 2006, AED use in polytherapy increased from 52% in 2000 to 63% in 2006 about 19 per 1000 AED exposed pregnant women each year ( =1.90, P 1 =0.20), and then decreased from 48% in 2000 to 37% in 2006 in monotherapy. However, after 2006, AED polytherapy reduced from 63% in 2006 to 58% in 2009, about 26.1 per 1000 AED exposed pregnant women each year ( = 2.61, P 2 =0.52), and monotherapy increases from 37% in 2006 to 42% in 2009. Specific Aim 1 4 : Compare the utilization of VPA from 1999 to 2009 in monotherapy and polytherapy in the pregnant wome n included in the study. Figure 4 9 delineates the percentage of VPA use in polytherapy over the total of AED exposed mother infant pairs (N=3,183) between year 2000 and 2009. The secular trend of VPA use in polytherapy significantly decreased from 8% in 2000 to 3% in 2009 ( = 0.68, P=0.001), about 68 per 100 VPA exposed pregnant women every year from 2006 to 2009. I n monotherapy VPA use increased from 92% in 2000 to 97% in 2009, correpondingly.

PAGE 85

85 Part 2: Apply a Latent Variable Model to Combine the Four Birth Defect Outcomes Specific Aim 2: Establish a n LVM to combine BD ACNB, LBW, and PC /OC into one latent variable APO for the studied mother infant pairs. The first step in this part is the selections of initial estimates is to make iteration process converge fast. We utili zed six independent categories of observed outcomes: BD (Yes), AC (Yes), PCOC (Yes), and BW: 1500~2499g, 1000~1499g, 350~999g to calculate initial values of parameters. Table 4 5 lists the frequencies and percentages of each category of observed outcomes i n AED and VPA exposure. Using the percentage of each category of observed outcomes we deduced the initial values of parameters The modified Gauss Newton algorithm was run in SAS Proc IML, starting from the initialized value at iteration 0 until the dif ference of the last two estimates was less than 10 9 All final parameter s were estimated from the iteration process Table 4 6 lists the estimated parameters, which were used to further estimate the LV, APO. E xpected and observed frequencies and percentag es of each combination of four observed outcomes were enumerated in Table 4 7. Table 4 8 presents the estimated APO for 32 combina tions of four observed outcomes, each of which is associa ted with a unique score of APO ranging from 1 to 10 13 Part 3: Eval uate the Validity and Reliability of the Combined Outcome As proposed in C hapter 3, validity and reliability of the combined outcome, APO, were evaluated in this part using Factor analysis, Q statistics, and Spearman correlation. Specific Aim 3 1 : Assess the u nidimensionality of the c ombined o utcome The u nidimensionality of the combined outcome was assessed using EFA, which was conducted in SAS Proc Factors with the method of principal factor estimation The

PAGE 86

86 assumption of the basic factor analysis in SAS i s the linear relationship between the normally distributed manifest variables and latent variable. Our four component outcomes are three dichotomous variables (BD, ACNB, and PC/OC), and one ordered polytomous variable (BW). Thus, appropriate model selectio n and data input become important in this part of the study. We employed iterated Principal Factor Estimation with input of polychoric correlation matrix in our factor analysis to facilitate the factor extraction for dichotomous and ordinal variables. 203 T he polychoric correlations between each pair of four component outcomes were computed using the SAS macro 204 The polychoric correlation matrix was then submitted to SAS Proc Factor for factor ext raction with the method of Principle Factor Estimation. The factor analysis results were summarized in Table 4 9 (a) (d). The results in Table 4 9 (a) shows that one factor is retained, and its prel iminary eigenvalue and proportion are above 1 (eigenvalue and proportion for factor 1 is 1.7 and 1.1, respectively). The subsequent factors (factor 2 to 3) have small eigenvalues ( 0.20 to 0.12). Table 4 9 (b) displays the portion of variance explained by this one factor. The factor pattern in T able 4 9 (c) illu strates the correlation between each observed outcome and the single factor, which is ranged from 0.34 to 0.97. In Table 4 9 (d), the eigenvalue of the reduced correlation is 1.87, proportion of variance explained by factor 1 is 100%, and cumulative varian ce for factor 1 is 100% as well. Therefore, the EFA analysis results showed that a single factor is sufficient to explain the variance of four observed outcome. The unidimensionality of the combined outcome, thus, is approved.

PAGE 87

87 Specific Aim 3 2 : Examine t he local independence of the four component outcomes As proposed in C hapter 3, local indepe ndence was examined based upon Q Statistics. We conducted CFA with SAS Proc Calis to compute the residuals and the covariance of the residuals after controllin g for the single factor As shown in the last section, EFA results confirmed one factor structure for this data. In CFA analysis, we fitted the single factor model with an input of polychoric correlation matrix calculated previously. The model fit statisti cs, estimation of parameters and residuals, and the covariance of residuals, are included in Table 4 10 (a) (c). Table 4 10 (a) illustrates the fit statistics for CFA model. For this CFA model, chi square probability being 0 with large p value indicates t he non difference of the observed and expected matrices and acceptable model fit. Bentler's c omparative Fit Index (equals to 0.9999) and Bentler & Bonett's normed fix index (equals to 1) are large (0.90 is a cutoff) enough for acceptable model fit. Table 4 10 (b) lists the fitted model with standardized estimates for each component outcome with the single factor. We used standardized estimates because the observed outcomes have different scales. Table 4 10 (c) presents the estimates of variance from the fit ted model. Table 4 10 (d) shows the pairwise covariances between each pair of residuals. The corresponding correlations calculated from COV ij i j are listed in Table 4 10 (e). The consequences of correlation between each pair of residuals are ranged from 0.12 to 0.14, which is lower than 0.20, the cutoff point for local dependence. Thus, our CFA analyses confirm the local independence that th ere are no large correlations among four observed outcomes after controlling for the latent factor.

PAGE 88

88 Specific Aim 3 3: Assess the internal homogeneity of the combined outcome Internal homogeneity of the combined outcome was assessed by the calculation of correlation between each pair of the four observed outcomes. Table 4 11 lists the estimates of Spearman correlation and corresponding 95% confidence interval and p value for each pair of observed outcomes and observed outcome with combined outcome. Spearman correlations among observed outcomes are statistically significant and ranged from 0.09 to 0.36, indicating low internal homogeneity of the observed outcomes. However, the correlations between each observed outcome and combined outcome a re statistical significant and ranges from 0.36 to 0.86. For four observed outcomes, the lowest correlation is between PCOC and BD (Rho=0.09, 95%CI: 0.12~0.14), and the highest correlation is between PCOC and BW (Rho=0.36, 95%CI: 0.35~0.37). The combined o utcome APO has the highest correlation with PCOC (Rho =0.86, 95%CI: 0.86~0.86), and the lowest correlation with BD (Rho=0.36, 95% CI: 0.35~0.37). Specific Aim 3 4 : Evaluate the convergent validity of the combined outcome Spearman correlation between the c ombined outcome APO, and total length of the hospital stay was calculated to examine convergent validity of the combined outcome. T he length of hospital stay for mother infant pair during delivery was assessed from Medicaid inpatient claims admitted before the delivery and discharged after the infant born. Figure 4 10 delineates the relationship between APO and total length of the hospital stay. The correlation coefficience Rho equals to 0.27, which is statistically significant (P<.0001).

PAGE 89

89 Specific Aim 3 5: Evaluate the discriminant validity of the combined outcome The i nfant breast fed status was collected in infant birth certificate after 2004. In our study, we have 19,699 mother infant pairs have this data available, which provides enough statistical powe r for the calculation of Spearman correlation between the infant breast fed status and the APO score. Figure 4 11 delineate s the relationship between the infant breast fed status and the APO sore. The regression line is nearly fat. The analysis results sh ow the Spearman correlation coefficient equals to 0.07 and R square equals to 0.0049, indicating that only 0.49% of variance was shared by infant breast fed status and APO score. The statistical significance (P<.0001) for such a small correlation may be o wing to large sample size caused over powered analysis. Therefore, APO has no association with the infant bread fed status. The discriminant validity of combined outcome was approved. Specific Aim 3 6 : Evaluate the discriminative validity of the combine d outcome groups one was the mother infant pairs with BD and ELBW, and another was the mother infant pairs with NBW and without any BD, ACBN, and PCOC. These two groups were selected because they are supposed to have very different levels of birth defects, one is very serious, and another is normal and healthy. The co mparison results were shown in F igure 4 12 (a). The difference of APO is significantly different between these two grou ps (Mean SD: 8.47 0.01 vs 1.30 0.31, P<.0001). Thus, discriminative validity of the combined outcome was approved. The combined outcome APO can differentiate the extreme groups. The mother infant pairs with BD and ELBW have significantly higher APO t han those with NBW and without BD, ACNB, and PCOC.

PAGE 90

90 Thus, in F igure 4 13 (b), we further compared APO scores between AED, VPA, and healthy controls. The average APO score in AED group is significantly different with healthy controls (Mean SE: 2.04 0.02 vs 1.88 0.01, P<.0001), and in VPA group is not significantly different with health controls (Mean SE: 2.00 0.07 vs 1.88 0.01, P=0.1003).

PAGE 91

91 Figure 4 1. Flow Chart of Study Design. Total 3555 mother infant pairs exposed to AEDs during pregnancy in Florida Medicaid 2000 to 2009. Excluded 705,866 patients with HIV, chronic DM, HPN, multiple birth, and not enrolled in Florida Medica id 6 months prior to pregnancy. Excluded 372 patients with HIV, chronic DM, HPN, and multiple birth Total 3183 patients in AED ex posure Total 43,956 patients in healthy control Total 749,822 mother infant pairs not exposed to AEDs during pregnancy and enrolled in Florida Medicaid 6 months prior to pregnancy. Total 753,377 infants born in Florida from 2000 to 2009.

PAGE 92

92 Table 4 1. Demographic Charact eristics of Study Participants. Obtained from BVS. Characteristics All Patients N=47,139 VPA Users N=226 AED Users N=3,183 Controls N=43,956 P Value* P Value ** Maternal age at infant born, Mean SD 24.8 5.3 25.9 6.4 26.5 6.0 24.6 5.2 <.0001 0.0014 43.6 29.8 52.1 33.2 47.5 30.7 43.3 29.7 <.0001 <.0001 White Black Others Missing 22533(48) 14420 (31) 10102 (21) 84 (0.2) 177 (72) 27 (11) 41 (17) 0 (0) 2,200 (69) 429 (13) 552 (17) 2 (0.1) 20,333 (46) 13,991 (32) 9,550 (22) 82 (0.2) <.0001 <.0001 White Black Others Missing 17124 (36) 10060 (21) 8498 (18) 11457 (24) 95 (39) 31 (13) 32 (13) 87 (36) 1455 (46) 327 (10) 474 (15) 927 (29) 15,669 (36) 9,733 (22) 8,024 (18) 10,530 (24) <.0001 <.0001 Above High School Missing 14,984 (32) 10,111 (21) 66 (40) 78 (32) 765 (33) 890 (28) 14,219 (41) 9221 (21) <.0001 0.7107 experience, N( %) Missing, N(%) 330 (0.7) 27336 (58) 1 (0.7) 110 (45) 23 (1) 1030 (32) 307 (2) 26,306 (60) 0.0237 0.3482 Missing, N(%) 20762 (44) 26090 (55) 143 (99) 102 (42) 2172 (99) 986 (31) 18,590 (98) 25,104 (57) 0.5571 0.0246 Mothe Mean SD Missing, N(%) 8.9 15.8 2024 (4) 11.1 18.2 27 (1) 11.3 16.9 174 (5) 8.8 15.7 1850 (4) <.0001 0.0059 Missing, N(%) 19,178 (41) 2 (0) 86 (35) 0 (0) 1137 (36) 0 (0) 18,041 ( 41) 2 (0) <.0001 0.1679 1.7 3.8 1.1 1.3 1.3 5.1 1.7 3.7 <.0001 <.0001 Missing, N(%) 8069 (17) 538 (1) 74 (31) 6 (2) 936 (30) 96 (3) 7133 (16) 442 (1) <.0001 <.0001 n SD 1.8 9.3 4.8 15.9 4.0 13.5 1.7 9.0 <.0001 0.0028

PAGE 93

93 Missing, N(%) 928 (2) 2 (0) 25 (1) 903 (2) Missing, N(%) 213 (0.5) 37 (0.1) 3 (1) 1 (0.4) 38 (1) 6 (0.2) 175 (0.4) 31 (0.1) <.0001 0.0183 Abo ve High School Missing 17,945 (38) 868 (2) 86 (37) 12 (5) 1,173 (37) 46 (1.5) 16,772 (39) 822 (2) 0.0980 0.5377 Infant male gender, N(%) Missing, N(%) 20,297 (44) 6910 (15) 120 (49) 14 (6) 1,549 (49) 103 (3) 19,048 (43) 6807 (15) <.0001 0.0001 Infa nt breast fed, N(%) Missing, N(%) 13,034 (28) 27,441 (58) 60 (41) 112 (46) 1,150 (52) 1,043 (33) 11,884 (63) 26398 (60) <.0001 <.0001 Missing, N(%) 688 (3.5) 27,336 (58) 6 (4) 110 (45) 84 (4) 1030 (32) 604 (3) 26,306 (60) 0.1714 0.4170 N(%) Missing, N(%) 1378 (3) 3335 (7) 6 (3) 37 (15) 138 (5) 226 (7) 1240 (3) 3109 (7) <.0001 0.2559 Missing, N(%) 127 (0.3) 3335 (7) 0 (0) 37 (15 ) 10 (0.3) 226 (7) 117 (0.3) 3109 (7) 0.0089 0.1916 Missing, N(%) 992 (2.3) 3335 (7) 3 (1) 37 (15) 67 (2.3) 226 (7) 925 (2.3) 3109 (7) 0.0102 0.1882 Missing, N(%) 7668 (16) 3335 ( 7) 84 (40) 37 (15) 1029 (35) 226 (7) 6639 (16) 3109 (7) <.0001 <.0001 Missing, N(%) 3362 (7) 27336 (58) 17 (13) 110 (45) 306 (14) 1030 (32) 3056 (17) 26306 (60) 0.0005 0.1819 Note: *: By definition, AED users include the patients who used VPA. * : Compared between AED and healthy controls. ** : Compared between VPA and healthy controls.

PAGE 94

94 Table 4 2 Clinical Charact eristics of Study Participants. Obtained from Medicaid Claim Data. Characteristics All Patients N=47,139 VPA Users N=226 AED Users N=3,183 Controls N=43,956 P Value * P Value ** and pregnancy, N(%) 652 (1) 60 (24) 571 (18) 81 (0.2) <.0001 <.0001 p regnancy, N(%) 448 (1) 15 (6) 230 (7) 218 (0.5) <.0001 <.0001 and pregnancy, N(%) 54 (0.1) 0 (0) 27 (0.9) 27 (0.1) <.0001 >.999 pregnancy, N(%) 943 (2) 56 (23) 444 (14) 499 (1.1) <.0001 <.0001 and pregnancy, N(%) 1071 (2) 21 (9) 328 (10.3) 743 (1.7) <.0001 <.0001 and pregnancy, N(%) 269 (0.6) 12 (5) 96 (3) 173 (0.4) <.0001 <.0001 baseline and pregnancy, N(%) 4895 (10) 90 (37) 1118 (35) 3777 (9) <.0001 <.0001 pregnancy, N(%) 613 (1) 50 (20) 338 (11) 275 (0.6) <.0001 <.0001 ssants exposure during pregnancy,N(%) 2167 (5) 90 (37) 886 (28) 1281 (3) <.0001 <.0001 19913 (42) 150 (61) 1,965 (62) 17,948 (41) <.0001 <.0001 hypnotics) exposu re during pregnancy, N(%) 3284 (7) 76 (31) 1832 (58) 1452 (3) <.0001 <.0001 pregnancy, N(%) 23 (0.1) 1 (0.41) 10 (0.3) 13 (0.03) <.0001 0.0749 Number of hospitalization for seizure during pregnancy, Median (min, max) 0 (0, 6) 0 (0, 4) 0 (0, 6) 0 (0, 3) <.0001 <.0001 Number of physician visits with seizure diagnoses during pregnancy, Median (min, max) 0 (0, 5) 0 (0, 3) 0 (0, 5) 0 (0, 6) <.0001 0.0208 2558 (5) 19 (8) 171 (5) 2387 (5) 0.8888 0.1097

PAGE 95

95 Table 4 2. Continued. Characteristics All Patients N=47,139 VPA Users N=226 AED Users N=3,183 Controls N=43,956 P Value* P Value ** during baseline and pregnancy, including: Virus, Ru bella, Cytomegalovirus, HIV, Syphilis, Herpes simplex virus, Toxoplamosis, Varicella virus, Venezuelan equine encephalitis virus, Phenylketonuria, Hypoxia, N(%) 3314 (7) 30 (12) 355 (11) 2959 (7) <.0001 0.0006 pregn ancy, N(%) 30 (0.1) 2 (0.82) 12 (0.38) 18 (0.04) <.0001 0.0054 pregnancy, N(%) 184 (0.4) 3 (1.2) 14 (0.4) 170 (0.4) 0.6428 0.0720 N(%) 169 (0.4) 1 (0.4) 17 (0.5) 152 (0.4 ) 0.0861 0.5734 N(%) 15272 (32) 117 (48) 1418 (45) 13,854 (32) <.0001 <.0001 pregnancy, N(%) 767 (2) 11 (4) 160 (5) 607 (1.4) <.0001 <.0001 during pregnancy, N(%) 813 (2) 20 (8) 192 (6.0) 621 (1.4) <.0001 <.0001 N(%) 252 (0.5) 2 (0.8) 43 (1.4) 209 (0.5) <.0001 0.3267 Mothers with more than two disease diagnoses during pregnancy or baseline, N(%) 2,531 (5) 86 (35) 943 (30) 1588 (4) <.0001 <.0001 0 1 2 3 39,159 (83) 5,449 (12) 1,698 (4) 833 (2) 107 (44) 52 (21) 47 (19) 39 (16) 1,538 (48) 702 (22) 542 (17) 401 (13) 37,621 (86) 4,747 (11) 1,156 (3) 43 2 (1) <.0001 <.0001 Mothers used more than two medications during pregnancy, N(%) 4,850 (10) 106 (43) 1,538 (48) 3,312 (8) <.0001 <.0001

PAGE 96

96 Table 4 2. Continued. Characteristics All Patients N=47,139 VPA Users N=226 AED Users N=3,183 Controls N=43,956 P Val ue* P Value ** other medication, N(%) 0 1 2 3 27,705 (59) 14,584 (31) 3,476 (7) 1,374 (3) 62 (31) 77 (31) 40 (16) 66 (27) 619 (19) 1,026 (32) 846 (27) 692 (22) 27,086 (62) 13,558 (31) 2,630 (6) 682 (2) <.0001 <.0001 Note : *: By definition, AED users include the patients who used VPA. **: Compared between AED and healthy controls. ***: Compared between VPA and healthy controls.

PAGE 97

97 Figure 4 2. Frequencies of Four Observed Outcomes in the Study Population Table 4 3 Spearman Correlation of Four Observed Outcomes with Length of Hospital Stay During Delivery Variable With Variable N Correlation Estimate 95% CI P Value BD Length of Hospital Stay During Delivery 47139 0.10 0.07 0.10 <.0001 ACNB Length of Hospital Stay During Delivery 47139 0.10 0.08 0.10 <.0001 PCOC Length of Hospital Stay During Delivery 47139 0.21 0.20 0.22 <.0001 BW Length of Hospital Stay During Delivery 47139 0.16 0.15 0.17 <.0001 10.65 8.04 28.26 8.64 0 5 10 15 20 25 30 BD ACNB PCOC LBW Percent of Observed Outcomes, %

PAGE 98

98 Table 4 4 Effects of AEDs or VPA on Four Observed Outcomes AED users include the patients using VPA. Observed Outcome s Estimate Standard Error P Value AEDs BD 0.34 0.16 0.0356 ACNB 0.60 0.15 0.0001 PCOC 0.70 0.11 <.0001 BW 0.10 0.02 <.0001 VPA BD 0.96 0.41 0.0196 ACNB 0.67 0.43 0.1223 PCOC 0.99 0.32 0.0019 BW 0.10 0.06 0.0856

PAGE 99

99 Figure 4 3. PC, OC, Pret erm Born, and PCOC in AED, VPA users, and Healthy Controls. AED users include the patients using VPA. 0 5 10 15 20 25 30 35 40 45 Overall PCOC PC OC Preterm AED Users VPA Users Healthy Controls P<.0001 P=0.0459 P<.0001 P=0.0509 P=0.0063 P=0.0006 P<.0001 P=0.6569

PAGE 100

100 Figure 4 4. Birth Defects (Major Congenital Malformation and Minor Congenital Malformation) and Abnormal Condition in AED and VPA Users, and Healthy Controls AED users include the patients using VPA. 0 5 10 15 20 25 Major CM Minor CM BD (Major & Minor CM) ACNB AED Users VPA Users Heathy Controls P<.0001 P<.0001 P<.0001 P=0.1525 P=0.0025 P<.0001 P<.0001 P<.0001

PAGE 101

101 Figure 4 5. Distribution of Four BW Categories in AED and VPA Users, and Healthy Controls AED users include the patients using VPA. 82% 84% 86% 88% 90% 92% 94% 96% 98% 100% AED Users VPA Users Healthy Controls ELBW: 350~999g VLBW: 1000~1499g LBW: 1500~2499g Normal: 2500~5999g P<.0001 for AED Users vs Healthy Controls; P=0.0752 for VPA Users vs Healthy Controls.

PAGE 102

102 Figure 4 6. Percentage of AED Use in S P=0.28 AED users include the patients using VPA. 0 0.1 0.2 0.3 0.4 0.5 0.6 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Percent of AED use, % Years of Infant Born

PAGE 103

103 Figure 4 7. Percentage of VPA Use in Study Population from 2000 to 2009. = 0.003, P=0.017. 0 0.01 0.02 0.03 0.04 0.05 0.06 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Percent of VPA use, % Years of Infant Born

PAGE 104

104 Figure 4 8. Percentage of AED Use in Polytherapy in S tudy Population from 2000 to 2009. =1.90, P 1 =0.20; = 2.61, P 2 =0.52. AED users include the patients using VPA. 30 35 40 45 50 55 60 65 70 75 80 1999 2001 2003 2005 2007 2009 Percent of AED use in Polytherapy % Years of Infant Born

PAGE 105

105 Figure 4 9. Percentage of VPA Use in Polytheray in Study Population from 2000 to 2009. = 0.68, P=0.001. 0 1 2 3 4 5 6 7 8 9 10 1999 2001 2003 2005 2007 2009 Percent of VPA Use in Polytherapy % Years of Infant Born

PAGE 106

106 Table 4 5. Incident Rate in Each Category of Observed Outcomes. Variable Category Frequency Percentage AED Exposed Group + Healthy Controls BD Yes 5019 10.65 No 42120 89.35 ACNB Yes 3792 8.04 No 43347 91.96 PCOC Yes 13323 28.26 No 33816 71.74 BW 350~999g 451 0.96 1000~1499g 337 0.71 1500~2499g 3285 6.97 2500~5999g 43066 91.36 VPA Exposed Group+ Healthy Controls BD Yes 48 19.59 No 187 80.41 ACNB Yes 25 10.20 No 220 89.80 PCOC Yes 12276 27.77 No 31925 72.23 BW 350~999g 43 6 0.99 1000~1499g 311 0.70 1500~2499g 2975 6.73 2500~5999g 40479 91.58 Table 4 6. Estimates of Parameters in LTM. BD ACNB PCOC BW 11 11 2 1 2 1 3 1 3 1 4 1 4 1 42 42 43 43 AED Users 1.71 0.06 4.44 3.16 0.85 0.14 5.14 1.21 5.09 1.1 7 2.65 0.79

PAGE 107

107 Table 4 7 Observed Frequency (OB FREQ ), E xpected Frequency (EX FREQ ), O bserved P ercents (OB % ) and Expected Per cents (EX % ) by C ombinations of F our Observed O utcomes. BD ACNB PCOC BW OBFREQ EXFREQ OB% EX% Yes Yes Yes 350 999 105.00 47.94 0. 22 0.10 Yes Yes Yes 1000 1499 85.00 34.60 0.18 0.07 Yes Yes Yes 1500 2499 196.00 136.31 0.42 0.29 Yes Yes Yes 2500 5999 201.00 135.68 0.43 0.29 Yes Yes No 350 999 1.00 1.44 0.00 0.00 Yes Yes No 1000 1499 2.00 2.37 0.00 0.01 Yes Yes No 1500 2499 22.00 29.76 0.05 0.06 Yes Yes No 2500 5999 301.00 160.85 0.64 0.34 Yes No Yes 350 999 80.00 38.75 0.17 0.08 Yes No Yes 1000 1499 60.00 48.79 0.13 0.10 Yes No Yes 1500 2499 274.00 394.20 0.58 0.84 Yes No Yes 2500 5999 1006.00 1084.62 2.13 2.30 Yes No No 35 0 999 1.00 1.34 0.00 0.00 Yes No No 1000 1499 1.00 3.41 0.00 0.01 Yes No No 1500 2499 77.00 103.66 0.16 0.22 Yes No No 2500 5999 2608.00 2604.98 5.53 5.53 No Yes Yes 350 999 130.00 105.35 0.28 0.22 No Yes Yes 1000 1499 102.00 99.59 0.22 0.21 No Yes Y es 1500 2499 576.00 539.30 1.22 1.14 No Yes Yes 2500 5999 697.00 834.06 1.48 1.77 No Yes No 350 999 0.50 3.41 0.00 0.01 No Yes No 1000 1499 3.00 6.88 0.01 0.01 No Yes No 1500 2499 77.00 127.93 0.16 0.27 No Yes No 2500 5999 1295.00 1320.85 2.75 2.80 N o No Yes 350 999 134.00 89.72 0.28 0.19 No No Yes 1000 1499 83.00 141.25 0.18 0.30 No No Yes 1500 2499 1477.00 1642.46 3.13 3.48 No No Yes 2500 5999 8117.00 7863.94 17.22 16.68 No No No 350 999 1.00 3.27 0.00 0.01 No No No 1000 1499 1.00 9.95 0.00 0.0 2 No No No 1500 2499 586.00 473.60 1.24 1.00 No No No 2500 5999 28841.00 29049.74 61.18 61.62

PAGE 108

108 Table 4 8 Estimates of P osterior M ean of the L atent V ariable and the APO BD ACNB PCOC BW APO Yes Yes Yes 350 999 0.61 7.62 Yes Yes Yes 1000 14 99 0.44 5.50 Yes Yes Yes 1500 2499 0.31 3.87 Yes Yes Yes 2500 5999 0.19 2.39 Yes Yes No 350 999 0.56 6.95 Yes Yes No 1000 1499 0.44 5.45 Yes Yes No 1500 2499 0.28 3.52 Yes Yes No 2500 5999 0.14 1.76 Yes No Yes 350 999 0.57 7.13 Yes No Yes 1000 1499 0.44 5.47 Yes No Yes 1500 2499 0.29 3.65 Yes No Yes 2500 5999 0.16 2.01 Yes No No 350 999 0.54 6.71 Yes No No 1000 1499 0.43 5.41 Yes No No 1500 2499 0.26 3.29 Yes No No 2500 5999 0.10 1.28 No Yes Yes 350 999 0.59 7.36 No Yes Yes 1000 1499 0.44 5. 49 No Yes Yes 1500 2499 0.30 3.77 No Yes Yes 2500 5999 0.18 2.22 No Yes No 350 999 0.55 6.82 No Yes No 1000 1499 0.43 5.43 No Yes No 1500 2499 0.27 3.42 No Yes No 2500 5999 0.12 1.55 No No Yes 350 999 0.56 6.97 No No Yes 1000 1499 0.44 5.45 No No Yes 1500 2499 0.28 3.55 No No Yes 2500 5999 0.14 1.81 No No No 350 999 0.53 6.62 No No No 1000 1499 0.43 5.40 No No No 1500 2499 0.25 3.18 No No No 2500 5999 0.08 1.06

PAGE 109

109 Table 4 9 Results of Unidimentionality Assessment Using SAS Proc Facto rs with the Method of Maximum Likelihood. A) Preliminary Eigenvalues. B) Variance Explained by the Factor C ) Factor Pattern D ) Eigenvalues of the Reduced Correlation Matrix. A ) Preliminary Eigenvalues Factors Eigenvalue Difference Proportion Cumulative Factor 1 1.71670329 1.59095720 1.1236 1.1236 Factor 2 0.12574609 0.23732518 0.0823 1.2059 Factor 3 .11157909 0.09148729 0.0730 1.1329 Factor 4 .20306638 0.1329 1.0000 1 factor will be retained by the PROPORTION criterion. B ) Variance Explaine d by the Factor Factor 1 1.8727521 C ) Factor Pattern Observed Outcomes Factor 1 BD 0.33990 ACNB 0.58804 PCOC 0.68728 BW 0.96906 D ) Eigenvalues of the Reduced Correlation Matrix. Factors Eigenvalue Difference Proportion Cumulative Factor 1 1.8 7275211 1.72977919 1.0002 1.0002 Factor 2 0.14297292 0.17726334 0.0764 1.0766 Factor 3 .03429042 0.07481267 0.0183 1.0583 Factor 4 .10910310 0.0583 1.0000

PAGE 110

110 Table 4 10 CFA Results for Assessing Local Independence of the Observed Outcomes. A) F it Statistics B) Standardized Results for Linear Equations C) Standardized Results for Estimation of Variances. D) Standardized Results for Covariances a mong Residuals. E) Calculated Correlation Coefficients Among Residuals. A ) Fit Statistics Fi t Function 0.0000 Goodness of Fit Index (GFI) 1.0000 Chi Square 0.0000 Chi Square DF 4 Pr > Chi Square Bentler's Comparative Fit Index 0.999 9 Bentler & Bonett's (1980) Normal Fit Index 1.0000 B ) Standardized Results for Linear Equations BD = 0.3296 F1 + 1.0000 e1 Std Err 0.00104 B1 t Value 318.1 PCOC1 = 0.7172 F1 + 1.0000 e2 Std Err 0.000758 B2 t Value 946.5 ACNB = 0.6132 F1 + 1.0000 e3 Std Err 0.00115 B3 t Value 532.2 BW = 0.9329 F1 + 1.0000 e4 Std Err 0.00144 B4 t Value 645.9

PAGE 111

111 Table 4 10. Continued. C ) Standardized Results for Estimation of Variances. Variable Parameter Estimate Standard Error t Value F 1 1.00000 e1 varem1 0.89134 0.0006832 1305 e2 varem2 0.48569 0.00109 446.90124 e3 varem3 0.62399 0.00141 441.56098 e4 varem4 0.12962 0.00270 48.09027 D ) Standardized Results for Covariances a mong Residuals. Var1 Var2 Parameter Estimate Standard Er ror t Value e1 e2 cov12 0.04786 0.00387 12.37577 e1 e3 cov13 0.10481 0.00395 26.50695 e2 e3 cov23 0.01118 0.00360 3.10842 e1 e4 cov14 0.06373 0.00378 16.86067 e2 e4 cov24 0.02965 0.00310 9.54909 e3 e4 cov34 0.01994 0.00357 5.58203 E ) Calcu lated Correlation Coefficients Among Residuals. Var1 Var2 Formula Calculated Correlation e1 e2 cov12 1 2 0.07274 e1 e3 cov13 1 3 0.140537 e2 e3 cov23 2 3 0.03289 e1 e4 cov14 1 4 0.11576 e2 e4 cov24 2 4 0.118171 e3 e4 cov34 3 4 0.07011

PAGE 112

112 Table 4 11 Spearman Correlation Statistics for Four Observed Outcomes and One Combined Outcome. Variable With Variable Correlation Estimate 95% Confidence Limits p Value for H0:Rho=0 ACNB BD 0.13 0.12 0.14 <.0001 PCOC BD 0.09 0.08 0.10 <.0001 BW BD 0.12 0.11 0.13 <.0001 PCOC ACNB 0.18 0.17 0.19 <.0001 BW ACNB 0.27 0.27 0.28 <.0001 BW PCOC 0.36 0.35 0.37 <.0001 APO BD 0.36 0.35 0.37 <.0001 APO ACNB 0.39 0.39 0.40 <.0001 APO PCOC 0.86 0.86 0.86 <.0001 APO BW 0.56 0.55 0.56 <.0001

PAGE 113

113 Fig ure 4 10. Correlation between Combined Outcome APO and Total Length of Hospital Stay during Delivery. 0 1 2 3 4 5 6 7 8 0 20 40 60 80 100 APO Total Length of Hospital Stay During Delivery (Days) Rho = 0.27 P<.0001

PAGE 114

114 Figure 4 11. Correlation between Combined Outcome APO and Infant Breast Fed Status. R = 0.0044 0 1 2 3 4 5 6 7 8 9 APO Score Infant Breast Fed No Yes Rho= 0.07 P<.0001

PAGE 115

115 A Figure 4 12. Comparison of APO Scores between: A) Two Ext reme Groups. B) AED, VPA, and Healthy Controls AED users include the patients using VPA. B 0 1 2 3 4 5 6 7 8 9 10 Average APO Score Group with BD and ELBW, N=186 Group without any BD and NBW, N=28,841 P<.0001 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5 Average APO Score AED, N=3183 VPA, N=245 Heathy Controls, N=43956 P<.0001 P=0.1003

PAGE 116

116 CHAPTER 5 DISCUSSION This study was divided into four major sections. In the first section, we conducted descriptive analysis on baseline demographic and clini cal characteristics, and four component outcomes. All covariances were combined into a propensity score, which was employed in the multivariate analysis to assess the adjusted drug effects on each component outcome. The next three sections focused on thre e key research questions and hypotheses we proposed in C hapter 1. For the second section, we investigated the utilization trends of AEDs in pregnant women included in the study. Specifically we examined the secular trends of AED exposure during pregnancy a s a whole, in polytherapy, and monotherapy. The joint points of the changed trends were identified and explicitly presented in the figures. The subject of the third section was latent trait model. A combined outcome was generated as a latent variable from the l atent trait model described in C hapter 3. The last section was the evaluation of combined outcome based upon the latent variable theory and psychometric definition on construct validity and reliability. The unidimensionality, local independence, inte rnal homogeneity, and the validity of convergent, discriminant, and discriminative were assessed to confirm the validity and reliability of the combined outcome. Descriptive Analysis Previous studies revealed that the common risk factors of infant birth d efects involved lower socioeconomic status, infant gender, maternal age, race, smoking, alcohol consumption, and drug exposure during pregnancy 75 78 205 T he results noted in T able 4 1 and 4 2 delineated a profile of the study population and compared all clinical

PAGE 117

117 risk factors of birth defects among three groups. Epilepsy, anxiety, bipolar, depression, migraine, and other mental disorders are main indications that keep the mothers exposing to AEDs during pregnancy. In addition, mothers exposed to AEDs durin g pregnancy appeared older, more likely unmarried or married with older husband s more Caucasian, and had more exposure to tobacco, alcohol, and other teratogenic medications during pregnancy, and more prevalence of other indications that associated with birth defects than healthy controls in the study. These results demonstrated that the mother exposed to AEDs or VPA during pregnancy possessed systematically higher Medicaid enrollees who were poorer and sicker than the regular population, AED exposed women seem having lower socioeconomic and health status, and elevated risk of infant birth defects than Non AED users in Medicaid program. The factor that new born infan ts with maternal exposure to AEDs have more other risk factors of birth defects increases the difficulties of assessing adverse perinatal and pregnancy effects of AEDs. These demographical and clinical risk factors need to be addressed in the analyses of A ED effects on infant birth defects. Figure 4 2 and T ables 4 3 and 4 4 summarized the evaluation of four component outcomes following three critical assumptions for selecting components for a composite outcome. These three criteria required the selected co mponents to be similar on frequency, importance to patients, and treatment effects. 125,127 In our study, four component outcomes were found similar on the incidence rates, AED effects, and associations with the length of hospital stay during delivery, an e stablished health status measure. One exception might be PC/OC, which had the highest magnitude on three

PAGE 118

118 criteria and seemed a little different from other three component outcomes. However, neither the differences of AED effects on PC/OC nor the correlatio n between PC/OC and the length of the hospital stay during delivery was statistically significant between AED use and healthy controls. Thus, these four component outcomes exhibited a certain level of homogeneity and demonstrated the validity of component selection for AED safety study. The results in Table 4 4 showed that VPA use had no significant effects on ACNB and BW. Further analyses were conducted on the distribution of four component outcomes and corresponding components between healthy controls an d AED or VPA users. Figures 4 3, 4 4, and 4 5 provide d evidence for a pronounced difference between the mother infant pairs employed AED and VPA and brought up a concern for combining outcomes in VPA drug safety study. Compared to the same healthy controls AED use associated with significant difference on all four component outcomes, whereas, VPA use was mainly related to increased BD and PC/OC, and had no significant effect on ACNB and BW. Previous studies using birth registry data revealed the consistent results that VPA use associated with higher rate of BD than that of AED use. A prospective study based on UK Birth Registry reported 4.2% of MCM for AED and 6.2% of MCM for VPA. 85 However, the incidence of MA s in our study, 9.1% in healthy control and 10. 7% in AED users, was lower compared to what was reported in the literatures, 15% to 20% in the general population and 37% in AED exposed pregnant women 21,80 8 3 Considering the difficulties of identifying MAs, underreporting or misdiagnosis of MAs might ca use the discrepancy. Given that VPA is not consistently associated with four outcomes and violates the criteria for component selection for a

PAGE 119

119 composite outcome; a concern is raised in this part of study on the validity of combining these four outcomes in V PA safety study. Part 1: Utilization of AEDs in Pregnant Women F igure 4 5 showed that the prevalence of AED use in pregnant women in Florida Medicaid has not changed from 2000 to 2009. About 0.35% to 0.5% of pregnant women in our study were exposed to AED s during this time period, which is slightly higher than what was documented in the general population, 0.2% to 0.5%. 8,23 Of all prescribed AEDs, the rate of AED use in polytherapy increased from 2000 to 2006, and then decreased from 2006 to 2009 in our st udy. Our study identified a joint point of year 2006 for the change of trend. A previous study with respect to prescribing patter ns of AED s in the US from 2001 to 2007 concluded that increasing of AED use in pregnant women was driven by fivefold increase i n the use of the second generation AEDs. 20 5 Our study results showed the similar pattern that AED use was slightly increasing about 10 per 1000 deliveries from 2000 to 2009, and 20 per 1000 deliveries from 2001 to 2007 although it was not statistically sig nificant. It is noted in this study that VPA use decreased significantly from 2000 to 2009 in pregnant women in all treatments or polytherapy. The patterns are coincident with previously studies for AED exposure in Women with Epilepsy in child bearing age in Florida Medicaid and North American Epilepsy Registry. 206,207 Given that several studies during the time period of our study have reported an association of fetal valproate exposure with an increased risk of major congenital malformations and with impa irment pregnant women who used AEDs during pregnancy 41, 98,10 1

PAGE 120

120 Our study population is limited to the pregnant women enrolled in Medicaid program, which has lower socioec onomic status than the general population or private insurance holders. Our study spans over 10 years from 2000 to 2009 with more information about secular trends of AED prescribing in pregnant women. Analyses of the joint points were conducted to further disclose the details about the change of AED prescribing trends in the pregnant women. The change of the secular trend could be due to the rapid rise of add on therapy of newer generation AEDs, which explained the increased use of AED in polytherapy from 2 000 to 2007. New scientific findings or FDA black box warnings on the safety issues of AED influence the prescribing patterns of AED as well. In September 2006, the FDA issued a warning stating that taking lamortrigine during pregnancy may increase the ris k for cleft lip and palate malformation in infants. 20 8 This obviously affected the decision of prescribing newer AEDs to the pregnant women, and thus lower the utilization of AEDs in polytherapy in pregnant women after 2006. Given that several black warnin gs were issued by the FDA after 2009 for the use of newer AEDs in pregnant women, we expect to see more decline of AED use in pregnant women in recent years. Part 2 : Application of Latent Variable Model To the best of our knowledge combining outcomes wi th Latent Variable Model has not been utilized in any pharmacoepidemiological studies previously. This Latent Variable Model was developed and published in 200 8 for comb in ing four birth defect outcomes. We employed this model in this pharmacoepidemi o logica l study to assess the comprehensive effects of AED s on overall adverse perinatal and pregnancy outcomes for both mothers and infants. Stepwise model f ittings were presented through T able 4 5 to 4 8. Providing that the pregnant women included in this study are limited to

PAGE 121

121 Florida Medicaid enrollees, our results are similar but differ slightly with what was reported previously. 13 The distributions of four birth defects noted in T able 4 5 presented similar frequencies of four component outcomes between AED and VPA study population. Herein, the model was fitted in AED study population and an APO score was obtained for each subject based upon their initial combination of four outcomes. The final estimate of latent variable ranged from 0.08 for normal infant m other pairs to 0.61 for the mother infant pairs with BD, ACNB, PC/OC, and ELBW, is in the similar scope as documented previously. 13 Part 3: Evaluation of the Combined Outcome A comprehensive evaluation on the combined outcome was conducted based upon the properties of latent variable and psychometric theory. This study could be a pioneer for the evaluation of outcome combined by the latent variable model because few studies reported such a comprehensive evaluation for a combined outcome. Unidimensionality and Local Independence Unidimensionality and local independence are important properties and assumptions for the latent variable model employed in this study. As we proposed in 3 Statistics were utilized to assess unidimensi onality and local independence. 5 ,133,209 After fitted EFA model, one factor was retained for four component outcomes in the study population. As a consequence of the CFA model, the residual of one factor model was explicitly estimated for each component outcome, as well as the correlations between each pair of the residuals. Pairwise correlations between the residuals noted in T able 4 10 (d) are lower than the cu toff point of 0.2 mentioned in C hapter 3. This implies that there is no correlation between fo ur component outcomes after controlling

PAGE 122

122 for the latent variable, and the four component outcomes only correlate with each other through the latent variable APO. Therefore, local independence of the four component Q 3 Statistics. 133,20 9 Reliability As we outlined before, reliability in psychometric theory refers to internal consistency and rest retest reliability. 134,135 Due to the nature of the combined outcome from latent variable model, internal homogeneity was a ssessed to examine to what extent the combined outcome and four component outcomes measure the same construct. Spearman correlation statistics noted in Table 4 11 presented that all four component outcomes significantly correlate with each other and the c ombined outcome APO. According to the Rule of Thumb in psychometric theory, the correlations ranged from 0.3 to 0.7 indicate the idea relationship, related but no redundancy. 18 In this study, the estimates of Spearman correlation ranged from 0.09 (BD with PC/OC) to 0.36 (BW with PC/OC) for the correlation between components, and 0.36 (BD with APO) to 0.86 (PC/OC with APO) for the correlation between components and combined outcomes. This can be explained by the different mechanism and incidence rate across the selected components. PC/OCs defined in this study include g estational hypertension eclampsia, preeclampsia, preterm, caesarean section, and p ostpartum hemorrhage whi ch are all associated with AED exposure as reported in previous studies. 5 8 ,6 8,69 These PC/OCs are related to the increased risk of mortality for mothers and infants. 62 63 Considering that BD and fetal death are mutually exclusive, correlation between PC/O C and BD could be higher if adding fetal death to BD category. Due to the limitation of the

PAGE 123

123 dataset, feta death was not included in this study as a component outcome. Therefore, the correlation between PC/OC and BD is lower than other pairwise correlation s. Another concern brought up in Table 4 11 is the high correlation between PC/OC and APO (Rho=0.86), which is suspicious for the dominating effect of PC/OC on APO. Moderate correlation is found for the combined outcome APO with BD, ACNB, and BW (Rho=0.36 to 0.56). Validity Further evaluation assessed construct validity of the combined outcome. Based upon psychometric theory, construct validity can be established through convergent validity, discriminant validity, and discriminative validity. 18, 134 The f act that APO significantly correlated with the length of hospital stay during delivery (Figure 4 10: Rho=0.27, P<.0001), and has nearly no correlation with infant breast fed status (Figure 4 11, Rho= 0.07, P<.0001) indicates APO associated with a well esta blished health status measure, and not associated with an irrelevant variable. Therefore, APO is approved to be a measure of overall severity of adverse perinatal and pregnancy outcome for both mothers and infants. The higher the APO score, the longer the hospital stay for the mothers and infants during delivery. Discriminative validity was also established fo r the combined outcome APO. In F igure 4 with BD and ELBW, and another with NBD and no any BDs. Therefore, mothers and infants with different severity of these adverse outcomes can be differentiated by APO score. We further compared APO scores between AED, VPA, and healthy controls. Given the well established teratogenic eff ects of AED and VPA, 79 91 we hypothesize d

PAGE 124

124 that APO score is significantly higher in AED or VPA exposed group compared to non exposed mother infant pairs. Figure 4 12 (b) shows that mothers exposed to AEDs during pregnancy have a significantly higher APO sc ore than non exposed; whereas, mothers exposed to VPA during pregnancy have no different APO score versus non exposed. The non significant difference of the APO score between VPA and non exposed could be caused by the non significant effects of VPA on BW a nd ACNB. This brings up a caution for combining outcome in drug safety study that the study drug needs to have significant effects on all component outcomes. Combined outcome is an overall severity score summering all the drug effects with the consideratio n of the incidence rate in each component. Therefore, selection of components is very important that decides the validity of the combined outcome. Sensitivity Study Although the incidence of adverse perinatal and pregnant outcomes associates with the mate rnal drug exposure in different pregnancy trimesters, the time window of AED exposure cannot be defined in this study as f ew mothers enrolled to Medicaid program before pregnancy started. Therefore the originally proposed sensitivity study on different AED exposure windows cannot be conducted due to the limitation of data. We altered the definition of pregnancy period with calculated gestational age +10 day, 20 days, and 30 days to examine the changes of association of AED exposure during pregnancy and fou r component outcomes. However, there is no significant difference between these definitions.

PAGE 125

125 Generalizability The strength of using Medicaid data in our study is that a large cohort of pregnant women were enrolled in Medicaid from 2000 to 2009, and drug claims and diagnosis code were available and approved to be valid for the assessment of drug exposure and baseline medical history. However, generalizability could be an issue considering the lower socioeconomic status and worse health state of Medicaid en rollees than the general population. In this study, generalizability of the study findings needs to be considered for whether the results derived from Florida Medicaid sample are generalizable to the general population. Our study utilized linked data to birth certificate data. Table 5 1 briefly compared demographic characteristics of mother infant pairs between general population and Me dicaid enrollees. Medicaid enrollees reported medical history variables, including use , and infan t related information, involving i nfant breast fed status and infant gender. For the nomissing data, Medicaid enrollees have prenatal visits (8.9 15.8 vs 11.6 17.4), more likely to be married (41% vs 37%), higher instances of the mother having a previous preterm condition (2.3% vs 1.1%), higher tobacco use (17% vs 11%), and fewer instances of the infant being breast fed (28% vs 50%). etween Medicaid enrollee and general popu lation in Florida. As noted in F igure 5 1, Florida Medicaid enrollees are more while and black and less of other races, which could be due to the large

PAGE 126

126 population Hispanic immigrants that are not qualified for Medic aid program. Comparing to the general population, thus, Florida Medicaid enrollees are: 1. reluctant to complete birth certificates, therefore, have more missing information in BVS data; 2. more married mothers with fewer prenatal visits, and a higher rate of tobacco use and previous preterm, 3. more male infants with less infant breast feeding, 4. more often black and white and fewer of other races. Overall, Medicaid enrollees are more vulnerable to have the adverse pregnancy and perinatal outcomes. 75 78 As presented in F igure 5 2, component outcomes, BD, ACNB, and LBW, are slightly higher in Medicaid Enrollees compared to the general population. This is not surprising given that Medicaid enrollees have more risk factors of birth defects than the general p opulation as we outlined above. However, our study findings are pertaining to the validity of combined outcome from the latent variable model. Fair to poor generalizability of Medicaid data with the increased adverse perinatal and pregnancy outcomes is l ess likely to affect the validity of the combined outcome from the latent variable model. This validated method can be generalizable to the general population. We suggest applying this model to a drug safety study with larger study cohort from the general population and with valid component outcomes. Study Limitations Several limitations should be considered as a consequence of using linked claim data and the nature of the study design. First, this study is based on four linked data sources. Unmatched dat a due to poor quality or missing identifiers were not analyzed. The study population is limited to matched mother infant pairs with mothers enrolled in

PAGE 127

127 the Florida Medicaid fee for service program. Therefore, sample size and power are restricted due to rel atively rare outcomes and exposures. Second, two important outcomes, spontaneous abortion and fetal death, were not taken into consideration due to lack of correlation with other component outcomes. Future study is necessary to develop a more advanced LVM that can combine as many outcomes as possible. Third, cognitive (e.g., language delay) and behavioral outcomes (e.g., autism) were not assessed in this study due to the limitation of the claim data. Fourth, Medicaid claim data was used for reimburseme nt of health care providers. Pharmacy claims only show dispensed prescriptions, thus actual drug utilization in patients might differ due to drug incompliance and over the counter drug use. Fifth, some important risk factors for BDs are not measurable in administrative health data. Such variables are socioeconomic status, family history, genetic effects, chromosome abnormality, environmental effects( ie. lead exposure in house or work), clinical disease severity (e.g. DSM IV score), or substance abuse. S ixth, this study aims to improve analytical power and efficiency with increased outcome events. However, combining multiple outcomes also introduces more confounding factors, and thus the study results are more vulnerable to bias. The study results from LV M are compared with that of previous studies to estimate the extent of potential bias. Seventh, by combining MCM, MA, LBW, and PC/OC, the latent variable APO is an overall adverse outcome for mothers and infants. The association between

PAGE 128

128 teratogens and specific BD is unknown if latent variable APO is used as a dependent variable in the model. Eighth, BDs were identified from Florida Birth Anomaly, a Florida birth defect registry, which used multiple data sources to detect birth defects. Although the qua lity of Florida Birth Anomaly has been audited and reported in previous studies, 140,148 154 outcome ascertainment using electronic medical records has not been systematically conducted in this registry, as well as PPV and sensitivity. Misclassification du e to incorrect coding or unreliable diagnoses may exist in the dataset. 2 10 The underreport rate for MAs has been reported as 7% using hospital discharge data solely. 2 11 However, this study used the association between AEDs and BDs, which has been well esta blished, to examine the validity and reliability of the LVM. Any differentiation from the true association was monitored and adjusted accordingly. Ninth, AED exposure was defined as any AED use during the overall pregnant period. Given that large amount o f women enrolled to the Medicaid program after they were pregnant, AED exposure window and length cannot be identified. Finally, this LVM combines manifest outcomes based upon the probability of occurrence in the study population. The severity of each ou tcome is not mathematically weighted in the combining process. Future study is needed develop more advanced statistical models to combine more outcomes based upon not only the probability of occurrence, but also the severity of each outcome. Future Studie s This study combined four adverse perinatal and pregnancy outcomes together to form one severity index. More components can be added to better measure the mortality and morbidities for both mothers and infants during pregnancy and fully assess

PAGE 129

129 the adverse effects of maternal exposure to the study drugs. For example, we could add developmental delay as the fifth component for the study of teratogenic effects of VPA development dela y that can be identified 3~6 years after born. 98,101 Furthermore, the model can be upgraded with weighting severity for each health state. Severity weighting needs to be based on the clinical experience or scientific findings. Finally, future study also can be conducted on applying this model to a drug safety study to comprehensively evaluate the adverse perinatal and pregnancy effects of the study drug. Therefore, an in dex can be assigned to describe overall adverse effects of the study drug and severity of the adverse perinatal and pregnancy outcomes for both mothers and infants. Summary and Conclusions This study validated a statistical approach to combine the compone nt outcomes based on the incidences of the components in drug safety study. We recommend the use of this Latent Variable Model under the presumed conditions that the study drug poses the similar effects on all selected component outcomes. If the study drug like VPA, only significantly associates with part of selected component outcomes, its effects on the combined outcome may be diluted and presented non significance compared to the controls. This circumstance is detrimental to the drug safety study as the results can be driven towards the null and the true teratogenic effects of the drug can be covered. Hence, evaluation of selected components is essential before the Latent Variable Model is applied for the combined outcome.

PAGE 130

130 The evaluation of the combined outcome based on the psychometric theory and latent variable properties confirmed the validity and reliability of the model results. Our study suggests that the combined outcome can be a valid assessment of overall severity of adverse perinatal and pregna nt outcome, and can be employed to measure the overall adverse perinatal and pregnancy effects of interested drug for both mothers and infants. However, caution needs to be taken when combining outcomes since it could mask drug effects by adding component outcomes that are not associated with the drug exposure Our study also identified the change of utilization trends for AED polytherapy in 2007. The FDA black box warnings issued for the association between maternal exposure to l amotrigine and infant bi rth defects possibly prevented further increase of the second generation AEDs in add on therapy in pregnant women after 2006.

PAGE 131

131 Table 5 1. Demographic Charact eristics of Study Participants. Obtained from BVS. Characteristics All Mother Infant Pairs i n General Population N=753,377 All Mother Infant Pairs Included in This Study N=47,139 Maternal age at infant born, Mean SD 25.2 5.9 24.8 5.3 42.7 28.9 43.6 29.8 White Black Others Missing 2 56268(34) 170002 (23) 325428 (43) 1679 (0.2) 22533(48) 14420 (31) 10102 (21) 84 (0.2) White Black Others Missing 192534 (26) 119514 (16) 283360 (38) 157969 (21) 17124 (36) 10060 (21) 8498 (18) 11457 (24) (%) Above High School Missing 223,403 (30) 160,835 (21) 14,984 (32) 10,111 (21) Missing, N(%) 5,131 (0.7) 249,932 (33) 330 (0.7) 27336 (58) Missing, N(%) 505,120 (67) 239, 723 (32) 20762 (44) 26090 (55) Missing, N(%) 11.6 17.4 52,934 (7.0) 8.9 15.8 2024 (4) Missing, N(%) 277,408 (37) 65 (0) 19,178 (41) 2 (0) SD 1.2 5.2 1.7 3.8 Missing, N(%) 85,118 (11) 12,359 (2) 8069 (17) 538 (1) Missing, N(%) 1.9 10.2 10,936 (2) 1.8 9.3 928 (2) Missing, N(%) 2961 (0.4) 73 4 (0.1) 213 (0.5) 37 (0.1) Above High School Missing 281,285 (37) 11,936 (1.6) 17,945 (38) 868 (2) Infant male gender, N(%) Missing, N(%) 359,528 (48) 51,480 (7) 20,297 (44) 6910 (15) Infant breast fed, N(%) Missing, N(% ) 379,065 (50) 251,988 (33) 13,034 (28) 27,441 (58) Missing, N(%) 20,180 (3) 249,933 (33) 688 (3.5) 27,336 (58)

PAGE 132

132 Table 5 1. Continued Characteristics All Mother Infant Pairs in General Population N=753,377 All Mother Infant Pairs Included in This Study N=47,139 Missing, N(%) 31,588 (4) 64,930 (9) 1378 (3) 3335 (7) Missing, N(%) 1998 (0.3) 64,923 (9) 127 (0.3) 3335 (7) Moth Missing, N(%) 8404 (1.1) 64923 (9) 992 (2.3) 3335 (7) Missing, N(%) 111,782 (16) 1392 (0.2) 7668 (16) 3335 (7) Missing, N(%) 57614 (8) 249932 (33) 336 2 (7) 27336 (58) Note: *: Compared between AED and healthy controls. **: Compared between VPA and healthy controls.

PAGE 133

133 Figure 5 1 Frequencies of Component Outcomes in Medicaid Enrollees and General Population 10.65 8.04 8.64 11.03 9.94 9.14 7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12 BD ACNB LBW Percent of Component Outcomes, % Medicaid Enrollees General Population

PAGE 134

134 APPENDIX A ACTIVE INGREDIENT S AND DRUG CATEGORIES A 1. VPA active ingredients divalproex sodium fatty acid derivative anticonvulsants divalproex sodium (as valproic acid) fatty acid derivative anticonvulsants valproate sodium fatty acid derivative anticonvulsants valproic acid fa tty acid derivative anticonvulsants A 2. Other anticonvulsants active ingredients Active Ingredient Drug Category Carbamazepine dibenzazepine anticonvulsants Clonazepam benzodiazepine anticonvulsants, benzodiazepines clorazepate dipotassium benzodiaz epines Diazepam benzodiazepine anticonvulsants, benzodiazepines Ethosuximide succinimide anticonvulsants Ethotoin hydantoin anticonvulsants Felbamate carbamate anticonvulsants fosphenytoin (as phenytoin equivalent) hydantoin anticonvulsants Gabapenti n gamma aminobutyric acid analogs Lacosamide miscellaneous anticonvulsants Lamotrigine triazine anticonvulsants Levetiracetam pyrrolidine anticonvulsants Lorazepam benzodiazepine anticonvulsants, benzodiazepines Mephenytoin hydantoin anticonvulsants Mephobarbital barbiturate anticonvulsants, barbiturates Methsuximide succinimide anticonvulsants Oxcarbazepine dibenzazepine anticonvulsants Paramethadione oxazolidinedione anticonvulsants Phenacemide urea anticonvulsants Phenobarbital barbiturate ant iconvulsants, barbiturates phenobarbital sodium barbiturate anticonvulsants, barbiturates Phensuximide succinimide anticonvulsants Phenytoin group I antiarrhythmics, hydantoin anticonvulsants phenytoin sodium group I antiarrhythmics, hydantoin anticonv ulsants Pregabalin gamma aminobutyric acid analogs Primidone barbiturate anticonvulsants tiagabine hydrochloride gamma aminobutyric acid reuptake inhibitors Topiramate carbonic anhydrase inhibitor anticonvulsants

PAGE 135

135 Trimethadione oxazolidinedione anticon vulsants Zonisamide carbonic anhydrase inhibitor anticonvulsants A 3. SSRI antidepressants active ingredients Active Ingredient Drug Category paroxetine hydrochloride SSRI Antidepressants paroxetine mesylate SSRI Antidepressants citalopram (as citalo pram hydrobromide) SSRI Antidepressants desvenlafaxine (as succinate) SSRI Antidepressants duloxetine hydrochloride SSRI Antidepressants escitalopram oxalate SSRI Antidepressants fluoxetine hydrochloride SSRI Antidepressants fluvoxamine maleate SSRI A ntidepressants milnacipran hydrochloride SSRI Antidepressants sertraline hydrochloride SSRI Antidepressants venlafaxine hydrochloride SSRI Antidepressants

PAGE 136

136 A 4. Other antidepressants (Non SSRI) active ingredients Active Ingredient Drug Category amit riptyline hydrochloride tricyclic antidepressants amitriptyline hydrochloride, chlordiazepoxide psychotherapeutic combinations amitriptyline hydrochloride, perphenazine psychotherapeutic combinations amoxapine tricyclic antidepressants bupropion hydrob romide miscellaneous antidepressants, smoking cessation agents bupropion hydrochloride miscellaneous antidepressants, smoking cessation agents clomipramine hydrochloride tricyclic antidepressants desipramine hydrochloride tricyclic antidepressants doxe pin hydrochloride miscellaneous anxiolytics, sedatives and hypnotics, tricyclic antidepressants fluoxetine, olanzapine psychotherapeutic combinations imipramine hydrochloride tricyclic antidepressants imipramine pamoate tricyclic antidepressants maprot iline hydrochloride tetracyclic antidepressants mirtazapine tetracyclic antidepressants nefazodone hydrochloride phenylpiperazine antidepressants nortriptyline hydrochloride tricyclic antidepressants protriptyline hydrochloride tricyclic antidepressant s trazodone hydrochloride phenylpiperazine antidepressants trimipramine maleate tricyclic antidepressants A 5. Antipsychotic active ingredients Active Ingredient Drug Category amitriptyline hydrochloride, chlordiazepoxide psychotherapeutic combinatio ns amitriptyline hydrochloride, perphenazine psychotherapeutic combinations aripiprazole atypical antipsychotics asenapine atypical antipsychotics chlorpromazine phenothiazine antiemetics, phenothiazine antipsychotics chlorpromazine hydrochloride phen othiazine antiemetics, phenothiazine antipsychotics clozapine atypical antipsychotics fluoxetine, olanzapine psychotherapeutic combinations fluphenazine decanoate phenothiazine antipsychotics fluphenazine enanthate phenothiazine antipsychotics fluphen azine hydrochloride phenothiazine antipsychotics haloperidol miscellaneous antipsychotic agents haloperidol decanoate miscellaneous antipsychotic agents haloperidol lactate miscellaneous antipsychotic agents

PAGE 137

137 lamotrigine triazine anticonvulsants lithiu m carbonate miscellaneous antipsychotic agents lithium citrate miscellaneous antipsychotic agents loxapine hydrochloride miscellaneous antipsychotic agents loxapine succinate miscellaneous antipsychotic agents mesoridazine besylate phenothiazine antips ychotics methotrimeprazine phenothiazine antipsychotics molindone hydrochloride miscellaneous antipsychotic agents olanzapine atypical antipsychotics paliperidone atypical antipsychotics paliperidone palmitate atypical antipsychotics perphenazine phe nothiazine antiemetics, phenothiazine antipsychotics pimozide miscellaneous antipsychotic 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 thioridazi ne phenothiazine antipsychotics thioridazine hydrochloride phenothiazine antipsychotics thiothixene thioxanthenes thiothixene hydrochloride thioxanthenes trifluoperazine hydrochloride phenothiazine antipsychotics triflupromazine hydrochloride phenothi azine antiemetics, phenothiazine antipsychotics ziprasidone hydrochloride monohydrate atypical antipsychotics ziprasidone mesylate atypical antipsychotics A 6. Anxiolytic (including hypnotics and sedatives) active ingredients Active Ingredient Drug Cat egory alprazolam benzodiazepines amobarbital sodium barbiturates amobarbital sodium, secobarbital sodium barbiturates buspirone hydrochloride miscellaneous anxiolytics, sedatives and hypnotics butabarbital sodium barbiturates butalbital barbiturates chloral hydrate miscellaneous anxiolytics, sedatives and hypnotics

PAGE 138

138 chlordiazepoxide benzodiazepines chlordiazepoxide hydrochloride benzodiazepines clorazepate dipotassium benzodiazepines dexmedetomidine hydrochloride miscellaneous anxiolytics, sedativ es and hypnotics diazepam benzodiazepine anticonvulsants, benzodiazepines estazolam benzodiazepines eszopiclone miscellaneous anxiolytics, sedatives and hypnotics flurazepam hydrochloride benzodiazepines halazepam benzodiazepines hydroxyzine hydrochl oride antihistamines, miscellaneous anxiolytics, sedatives and hypnotics hydroxyzine pamoate antihistamines, miscellaneous anxiolytics, sedatives and hypnotics meprobamate miscellaneous anxiolytics, sedatives and hypnotics midazolam benzodiazepines mid azolam hydrochloride benzodiazepines midazolam hydrochloride, sodium chloride benzodiazepines oxazepam benzodiazepines pentobarbital barbiturates pentobarbital sodium barbiturates quazepam benzodiazepines ramelteon miscellaneous anxiolytics, sedative s and hypnotics secobarbital sodium barbiturates temazepam benzodiazepines triazolam benzodiazepines zaleplon miscellaneous anxiolytics, sedatives and hypnotics zolpidem tartrate miscellaneous anxiolytics, sedatives and hypnotics

PAGE 139

139 A 7. Folic Acid a ctive ingredients Active Ingredient Drug Category folic acid vitamin and mineral combinations folic acid vitamins A 8. Isotretinoin active ingredients Active Ingredient Drug Category isotretinoin miscellaneous antineoplastics isotretinoin miscellan eous uncategorized agents A 9. Antidiabetics active ingredients Active Ingredient Drug Category glimepiride antidiabetic combinations glipizide antidiabetic combinations glyburide antidiabetic combinations metformin hydrochloride antidiabetic combin ations pioglitazone hydrochloride antidiabetic combinations repaglinide antidiabetic combinations rosiglitazone maleate antidiabetic combinations sitagliptin antidiabetic combinations A 10. Antihypertensives active ingredients Active Ingredient Drug Category aliskiren antihypertensive combinations amiloride hydrochloride antihypertensive combinations amlodipine (as besylate) antihypertensive combinations amlodipine besylate antihypertensive combinations atenolol antihypertensive combinations at orvastatin antihypertensive combinations benazepril hydrochloride antihypertensive combinations bendroflumethiazide antihypertensive combinations bisoprolol fumarate antihypertensive combinations candesartan cilexitil antihypertensive combinations cap topril antihypertensive combinations chlorothiazide antihypertensive combinations chlorthalidone antihypertensive combinations clonidine antihypertensive combinations deserpidine antihypertensive combinations diltiazem hydrochloride antihypertensive c ombinations enalapril antihypertensive combinations enalapril maleate antihypertensive combinations eprosartan mesylate antihypertensive combinations felodipine antihypertensive combinations fosinopril sodium antihypertensive combinations guanethidin e monosulfate antihypertensive combinations hydralazine hydrochloride antihypertensive combinations

PAGE 140

140 hydrochlorothiazide antihypertensive combinations hydroflumethiazide antihypertensive combinations irbesartan antihypertensive combinations isosorbide dinitrate antihypertensive combinations lisinopril antihypertensive combinations losartan potassium antihypertensive combinations methyclothiazide antihypertensive combinations methyldopa antihypertensive combinations metoprolol tartrate antihypertens ive combinations moexipril hydrochloride antihypertensive combinations nadolol antihypertensive combinations olmesartan medoxomil antihypertensive combinations polythiazide antihypertensive combinations prazosin hydrochloride antihypertensive combinat ions propranolol hydrochloride antihypertensive combinations quinapril hydrochloride antihypertensive combinations rauwolfia serpentina antihypertensive combinations reserpine antihypertensive combinations spironolactone antihypertensive combinations telmisartan antihypertensive combinations timolol maleate antihypertensive combinations trandolapril antihypertensive combinations triamterene antihypertensive combinations trichlormethiazide antihypertensive combinations valsartan antihypertensive c ombinations verapamil hydrochloride antihypertensive combinations A 11. Hormone active ingredients Active Ingredient Drug Category desogestrel contraceptives drospirenone contraceptives estradiol cypionate contraceptives ethinyl estradiol contrace ptives ethynodiol diacetate contraceptives etonogestrel contraceptives levonorgestrel contraceptives medroxyprogesterone acetate contraceptives mestranol contraceptives norelgestromin contraceptives norethindrone contraceptives norethindrone aceta te contraceptives norgestimate contraceptives norgestrel contraceptives fulvestrant estrogen receptor antagonists chlorotrianisene estrogens conjugated estrogens estrogens

PAGE 141

141 dienestrol estrogens diethylstilbestrol estrogens diethylstilbestrol diphos phate estrogens esterified estrogens estrogens estradiol estrogens estradiol acetate estrogens estradiol benzoate estrogens estradiol cypionate estrogens estradiol valerate estrogens estradiol 17b estrogens estriol estrogens estrone estrogens estropipate estrogens ethinyl estradiol estrogens synthetic conjugated estrogens, A estrogens synthetic conjugated estrogens, B estrogens mifepristone progesterone receptor modulators etonogestrel progestins hydroxyprogesterone caproate progestins l evonorgestrel progestins medroxyprogesterone acetate progestins megestrol acetate progestins norethindrone progestins norethindrone acetate progestins norgestrel progestins progesterone progestins progesterone, micronized progestins levothyroxine s odium thyroid hormones liothyronine sodium thyroid hormones liotrix thyroid hormones thyroid thyroid hormones thyrotropin alpha thyroid hormones A 12. ACE active ingredients Active Ingredient Category Name benazepril hydrochloride angiotensin conv erting enzyme inhibitors captopril angiotensin converting enzyme inhibitors enalapril maleate angiotensin converting enzyme inhibitors enalaprilat angiotensin converting enzyme inhibitors fosinopril sodium angiotensin converting enzyme inhibitors lisi nopril angiotensin converting enzyme inhibitors moexipril hydrochloride angiotensin converting enzyme inhibitors perindopril erbumine angiotensin converting enzyme inhibitors quinapril hydrochloride angiotensin converting enzyme inhibitors ramipril ang iotensin converting enzyme inhibitors trandolapril angiotensin converting enzyme inhibitors

PAGE 142

142 A 13. Antineoplastic active ingredients Active Ingredient Drug Category bleomycin sulfate antineoplastic antibiotics dactinomycin antineoplastic antibiotics d aunorubicin citrate liposome antineoplastic antibiotics daunorubicin hydrochloride antineoplastic antibiotics doxorubicin hydrochloride antineoplastic antibiotics doxorubicin hydrochloride liposome antineoplastic antibiotics epirubicin hydrochloride an tineoplastic antibiotics idarubicin hydrochloride antineoplastic antibiotics mitomycin antineoplastic antibiotics mitoxantrone hydrochloride antineoplastic antibiotics pentostatin antineoplastic antibiotics plicamycin antineoplastic antibiotics valru bicin antineoplastic antibiotics amifostine antineoplastic detoxifying agents ifosfamide antineoplastic detoxifying agents mesna antineoplastic detoxifying agents abarelix antineoplastic hormones anastrozole antineoplastic hormones bicalutamide antin eoplastic hormones degarelix (as acetate) antineoplastic hormones diethylstilbestrol antineoplastic hormones diethylstilbestrol diphosphate antineoplastic hormones estramustine phosphate sodium antineoplastic hormones exemestane antineoplastic hormone s fluoxymesterone antineoplastic hormones flutamide antineoplastic hormones fulvestrant antineoplastic hormones goserelin acetate antineoplastic hormones histrelin acetate antineoplastic hormones letrozole antineoplastic hormones leuprolide (as leup rolide acetate) antineoplastic hormones leuprolide acetate antineoplastic hormones medroxyprogesterone acetate antineoplastic hormones megestrol acetate antineoplastic hormones nilutamide antineoplastic hormones raloxifene hydrochloride antineoplastic hormones tamoxifen (as tamoxifen citrate) antineoplastic hormones tamoxifen citrate antineoplastic hormones testolactone antineoplastic hormones toremifene citrate antineoplastic hormones triptorelin (as triptorelin pamoate) antineoplastic hormones interferon alfa 2a antineoplastic interferons interferon alfa 2b antineoplastic interferons

PAGE 143

143 aldesleukin miscellaneous antineoplastics altretamine miscellaneous antineoplastics arsenic trioxide miscellaneous antineoplastics asparaginase miscellaneous a ntineoplastics azacitidine miscellaneous antineoplastics bcg, tice strain miscellaneous antineoplastics bexarotene miscellaneous antineoplastics bortezomib miscellaneous antineoplastics denileukin diftitox miscellaneous antineoplastics irinotecan hyd rochloride miscellaneous antineoplastics isotretinoin miscellaneous antineoplastics lenalidomide miscellaneous antineoplastics levamisole hydrochloride miscellaneous antineoplastics mitotane miscellaneous antineoplastics pegaspargase miscellaneous ant ineoplastics porfimer sodium miscellaneous antineoplastics procarbazine hydrochloride miscellaneous antineoplastics thalidomide miscellaneous antineoplastics topotecan (as hydrochloride) miscellaneous antineoplastics topotecan hydrochloride miscellane ous antineoplastics tretinoin miscellaneous antineoplastics verteporfin miscellaneous antineoplastics A 14. Coumarin based anticoagulants Active Ingredients Active Ingredient Drug Category anisindione coumarins and indandiones dicumarol coumarins a nd indandiones warfarin sodium coumarins and indandiones aminocaproic acid miscellaneous coagulation modifiers antihemophilic factor, human miscellaneous coagulation modifiers antihemophilic factor, porcine miscellaneous coagulation modifiers antihemo philic factor, recombinant miscellaneous coagulation modifiers anti inhibitor coagulant complex miscellaneous coagulation modifiers antithrombin alfa miscellaneous coagulation modifiers antithrombin III (human) miscellaneous coagulation modifiers coagu lation factor viia miscellaneous coagulation modifiers complement C1 esterase inhibitor miscellaneous coagulation modifiers drotrecogin alfa (activated), lyophilized miscellaneous coagulation modifiers ecallantide miscellaneous coagulation modifiers fa ctor IX complex, human miscellaneous coagulation modifiers Factor IX, recombinant miscellaneous coagulation modifiers ferric subsulfate miscellaneous coagulation modifiers fibrinogen miscellaneous coagulation modifiers fibrinolysis inhibitor miscellane ous coagulation modifiers pentoxifylline miscellaneous coagulation modifiers

PAGE 144

144 protein C miscellaneous coagulation modifiers thrombin miscellaneous coagulation modifiers tranexamic acid miscellaneous coagulation modifiers von Willebrand factor miscellan eous coagulation modifiers A 15. Iodide active Ingredients Active Ingredient Drug Category sodium iodide I 131 antithyroid agents sodium iodide I 123 diagnostic radiopharmaceuticals iodides minerals and electrolytes sodium iodide minerals and electro lytes sodium iodide I 131 therapeutic radiopharmaceuticals A 16. Antibiotics active ingredients Active Ingredient Drug Category bleomycin sulfate antineoplastic antibiotics dactinomycin antineoplastic antibiotics daunorubicin citrate liposome antineo plastic antibiotics daunorubicin hydrochloride antineoplastic antibiotics doxorubicin hydrochloride antineoplastic antibiotics doxorubicin hydrochloride liposome antineoplastic antibiotics epirubicin hydrochloride antineoplastic antibiotics idarubicin hydrochloride antineoplastic antibiotics mitomycin antineoplastic antibiotics mitoxantrone hydrochloride antineoplastic antibiotics pentostatin antineoplastic antibiotics plicamycin antineoplastic antibiotics valrubicin antineoplastic antibiotics de xtrose glycopeptide antibiotics sodium chloride glycopeptide antibiotics telavancin glycopeptide antibiotics vancomycin hydrochloride glycopeptide antibiotics atovaquone miscellaneous antibiotics aztreonam miscellaneous antibiotics bacitracin miscell aneous antibiotics chloramphenicol miscellaneous antibiotics chloramphenicol palmitate miscellaneous antibiotics chloramphenicol sodium succinate miscellaneous antibiotics colistimethate sodium miscellaneous antibiotics dalfopristin miscellaneous anti biotics daptomycin miscellaneous antibiotics erythromycin ethylsuccinate miscellaneous antibiotics furazolidone miscellaneous antibiotics linezolid miscellaneous antibiotics metronidazole miscellaneous antibiotics metronidazole benzoate miscellaneous antibiotics

PAGE 145

145 metronidazole hydrochloride miscellaneous antibiotics novobiocin sodium miscellaneous antibiotics pentamidine isethionate miscellaneous antibiotics polymyxin B sulfate miscellaneous antibiotics quinupristin miscellaneous antibiotics rifa ximin miscellaneous antibiotics spectinomycin hydrochloride miscellaneous antibiotics sulfamethoxazole miscellaneous antibiotics sulfisoxazole miscellaneous antibiotics trimethoprim miscellaneous antibiotics trimetrexate glucuronate miscellaneous anti biotics bacitracin topical antibiotics bacitracin zinc topical antibiotics diperodon hydrochloride topical antibiotics erythromycin topical antibiotics gentamicin sulfate topical antibiotics lidocaine topical antibiotics mafenide acetate topical ant ibiotics mupirocin topical antibiotics mupirocin calcium topical antibiotics neomycin base (as neomycin sulfate) topical antibiotics neomycin sulfate topical antibiotics polymyxin B topical antibiotics polymyxin B sulfate topical antibiotics pramoxi ne hydrochloride topical antibiotics retapamulin topical antibiotics silver sulfadiazine topical antibiotics sulfacetamide sodium topical antibiotics sulfur topical antibiotics tetracycline hydrochloride topical antibiotics urea topical antibiotics acetic acid topical anti infectives aloe vera topical anti infectives benzyl alcohol topical anti infectives boric acid topical anti infectives brilliant green topical anti infectives chloroxine topical anti infectives crotamiton topical anti infecti ves gentian violet topical anti infectives hexylresorcinol topical anti infectives hydrogen peroxide topical anti infectives imiquimod topical anti infectives iodine (as cadexomer iodine) topical anti infectives iodoquinol topical anti infectives li ndane topical anti infectives malathion topical anti infectives

PAGE 146

146 n docosanol topical anti infectives permethrin topical anti infectives piperonyl butoxide topical anti infectives proflavine hemisulfate topical anti infectives pyrethrins topical anti i nfectives silver topical anti infectives sinecatechins topical anti infectives zinc acetate topical anti infectives benzoic acid urinary anti infectives cinoxacin urinary anti infectives fosfomycin tromethamine urinary anti infectives l hyoscyamine urinary anti infectives methenamine urinary anti infectives methenamine hippurate urinary anti infectives methenamine mandelate urinary anti infectives methylene blue urinary anti infectives nalidixic acid urinary anti infectives nitrofurantoin urina ry anti infectives nitrofurantoin monohydrate urinary anti infectives nitrofurantoin, macrocrystals urinary anti infectives oxytetracycline hydrochloride urinary anti infectives phenazopyridine urinary anti infectives sodium acid phosphate, monohydrat e urinary anti infectives sodium biphosphate urinary anti infectives sodium salicylate urinary anti infectives sulfamethizole urinary anti infectives trimethoprim urinary anti infectives butoconazole nitrate vaginal anti infectives clindamycin phosph ate vaginal anti infectives clotrimazole vaginal anti infectives gentian violet vaginal anti infectives hydroxyquinoline sulfate vaginal anti infectives metronidazole vaginal anti infectives miconazole vaginal anti infectives miconazole nitrate vagin al anti infectives povidone iodine vaginal anti infectives sodium sulfacetamide vaginal anti infectives sulfabenzamide vaginal anti infectives sulfanilamide vaginal anti infectives sulfathiazole vaginal anti infectives terconazole vaginal anti infect ives tioconazole vaginal anti infectives A 17. Glucocorticoids active ingredients Active Ingredient Drug Category betamethasone glucocorticoids

PAGE 147

147 betamethasone acetate glucocorticoids betamethasone dipropionate glucocorticoids betamethasone sodium pho sphate glucocorticoids betamethasone valerate glucocorticoids budesonide glucocorticoids cortisone acetate glucocorticoids dexamethasone glucocorticoids dexamethasone acetate glucocorticoids dexamethasone sodium phosphate glucocorticoids hydrocortis one glucocorticoids hydrocortisone acetate glucocorticoids hydrocortisone cypionate glucocorticoids hydrocortisone hemisuccinate glucocorticoids hydrocortisone sodium phosphate glucocorticoids hydrocortisone sodium succinate glucocorticoids lidocaine hydrochloride glucocorticoids methylprednisolone glucocorticoids methylprednisolone acetate glucocorticoids methylprednisolone sodium succinate glucocorticoids prednisolone glucocorticoids prednisolone (as sodium phosphate) glucocorticoids prednisol one acetate glucocorticoids prednisolone sodium phosphate glucocorticoids prednisolone tebutate glucocorticoids prednisone glucocorticoids triamcinolone glucocorticoids triamcinolone acetonide glucocorticoids triamcinolone diacetate glucocorticoids triamcinolone hexacetonide glucocorticoids A 18. Quinine active ingredients Active Ingredient Drug Category artemether antimalarial combinations atovaquone antimalarial combinations lumefantrine antimalarial combinations proguanil hydrochloride antim alarial combinations pyrimethamine antimalarial combinations sulfadoxine antimalarial combinations chloroquine hydrochloride antimalarial quinolines chloroquine phosphate antimalarial quinolines hydroxychloroquine sulfate antimalarial quinolines mefl oquine hydrochloride antimalarial quinolines primaquine phosphate antimalarial quinolines quinacrine hydrochloride dihydrate antimalarial quinolines quinine sulfate antimalarial quinolines doxycycline miscellaneous antimalarials

PAGE 148

148 doxycycline calcium mi scellaneous antimalarials doxycycline hyclate miscellaneous antimalarials doxycycline monohydrate miscellaneous antimalarials halofantrine hydrochloride miscellaneous antimalarials pyrimethamine miscellaneous antimalarials A 19. NSAID active ingredie nts Active Ingredient Drug Category analgesic balm nonsteroidal anti inflammatory agents bromfenac nonsteroidal anti inflammatory agents diclofenac nonsteroidal anti inflammatory agents diclofenac potassium nonsteroidal anti inflammatory agents diclof enac sodium nonsteroidal anti inflammatory agents etodolac nonsteroidal anti inflammatory agents fenoprofen calcium nonsteroidal anti inflammatory agents flurbiprofen nonsteroidal anti inflammatory agents ibuprofen nonsteroidal anti inflammatory agents ibuprofen (as l lysine) nonsteroidal anti inflammatory agents indomethacin nonsteroidal anti inflammatory agents indomethacin sodium trihydrate nonsteroidal anti inflammatory agents ketoprofen nonsteroidal anti inflammatory agents ketorolac trometham ine nonsteroidal anti inflammatory agents lansoprazole nonsteroidal anti inflammatory agents meclofenamate sodium nonsteroidal anti inflammatory agents mefenamic acid nonsteroidal anti inflammatory agents meloxicam nonsteroidal anti inflammatory agents misoprostol nonsteroidal anti inflammatory agents nabumetone nonsteroidal anti inflammatory agents naproxen nonsteroidal anti inflammatory agents naproxen sodium nonsteroidal anti inflammatory agents oxaprozin nonsteroidal anti inflammatory agents p iroxicam nonsteroidal anti inflammatory agents sulindac nonsteroidal anti inflammatory agents tolmetin sodium nonsteroidal anti inflammatory agents A 20. AntiViral active ingredients Active Ingredient Drug Category maraviroc antiviral chemokine recepto r antagonist abacavir sulfate antiviral combinations efavirenz antiviral combinations emtricitabine antiviral combinations interferon alfa 2b antiviral combinations lamivudine antiviral combinations lopinavir antiviral combinations nelfinavir antivi ral combinations ribavirin antiviral combinations ritonavir antiviral combinations

PAGE 149

149 tenofovir disoproxil fumarate antiviral combinations zidovudine antiviral combinations peginterferon alfa 2a antiviral interferons peginterferon alfa 2b antiviral inte rferons enfuvirtide miscellaneous antivirals fomivirsen sodium miscellaneous antivirals foscarnet sodium miscellaneous antivirals acyclovir topical antivirals penciclovir topical antivirals A 21. Antirectroviral active ingredients Active Ingredient Drug Category abacavir sulfate NRTIs adefovir dipivoxil NRTIs amprenavir protease inhibitors atazanavir sulfate protease inhibitors darunavir (as ethanolate) protease inhibitors delavirdine NNRTIs delavirdine mesylate NNRTIs didanosine NRTIs efa virenz NNRTIs emtricitabine NRTIs enfuvirtide miscellaneous antivirals entecavir NRTIs etravirine NNRTIs fosamprenavir calcium protease inhibitors indinavir protease inhibitors lamivudine NRTIs lopinavir protease inhibitors maraviroc antiviral che mokine receptor antagonist nelfinavir protease inhibitors nevirapine NNRTIs raltegravir integrase strand transfer inhibitor ritonavir protease inhibitors saquinavir protease inhibitors saquinavir (as mesylate) protease inhibitors saquinavir mesylate protease inhibitors stavudine NRTIs telbivudine NRTIs tenofovir disoproxil fumarate NRTIs tipranavir protease inhibitors zalcitabine NRTIs zidovudine NRTIs

PAGE 150

150 APPENDIX B OPERATIONAL DEFINITI ONS B 1. Component Outcomes Variable Operational Definit ion Birth Defects, including Congenital Major Malformations and Minor Anomalies at least one of the ICD 9 codes: 740.xx to 759.xx 710.xx to 739.xx, 520.xx to 629.xx, 320.xx to 389.xx, 140.xx to 239.xx, or 279.xx on hospital inpatient or outpatient di scharge claims any time from infant birth date to 365 days of follow up. Abnormal Condition of New Born checkboxes in BVS data: AC_VENT IMMED AC_VENT_30MIN AC_VENT_MORE_6HOUR AC_NICU AC_SURFACTANT AC_ANTIBIOTIC_SE PSIS AC_SEIZURE AC_OTHER AC_HYALINE_MEM AC_BIRTH_INJURY Or at least one of the ICD 9 codes 280.x 289.x, 767.x, 760.7x, 779.5, 769.x, 770.1x, 779.0, 345.x on hospital inpatient or outpatient discharge claims any time from infant birth date to 365 days of follow up. Birth Weight Used BIRTH_WEIGHT_GRAMS from BVS data and categorized to 4 levels: extremely Low BW ( 350 and <999 g), very Low BW ( 1000 and <1499 g), Low BW ( 1500 and <2499 g), normal BW ( 2500 and <5999 g).

PAGE 151

151 Pregnancy Complications Placenta previa, w/o bleeding, unspec. Placenta previa, v/bleeding, unspec. Abruptio placetae, unspec Hemorrhage in pr egnancy, unspec. Gestational hypertension Mild or unspecified pre eclampsia Severe pre eclampsia Eclampsia, unspec. Obstetrical Complications Cesarean delivery Forceps or vacuum extractor delivery Postpartum hemorrhage Preterm Birth Identified at l east one of the following ICD 9 or CPT codes: 641.0 641.1 641.2 641.9 642.3, 642.9 642.4 642.5 642.6 checkboxes from BVS data and at least one of the ICD 9 or CPT codes: MD_CES_LABOR_ATTEMPT or 669.7, 763.4, 59510, 59 514, 59515, 59612 MD_VAGINAL_FORCEPS, MD_VAGINAL_VACUUM, or 763.2, 763.3, 669.51 666.xx Gestational age <= 37 weeks

PAGE 152

152 B 2. Potential Confounders or Other Risk Factors for Birth Defects Variable Operational Definition infant born from BVS dataset Male, Female for infant from BVS dataset White, Hispanic, Black, Other from BVS dataset Calendar year Year of infant born from BVS dataset residence zip code from BVS dataset Mother previous adverse pregnancy experience Yes/No on mother previous poor pregnancy outcome from BVS dataset during this pregnancy from BVS dat aset Parity Continuous variable indicates the count of all live births from BVS dataset baseline At least one inpatient or two outpatient claims in Medicaid claims with the following ICD 9CM codes at any Diagnosis fi eld: 345.xx baseline baseline At least one inpatient or two outpatient claims in Medicaid claims with the following ICD 9CM codes at any Diagnosis field: 296.2 296.3, 300.4, 3 11.xx At least one inpatient or two outpatient claims in Medicaid claims with the following ICD 9CM codes at any Diagnosis field: 300.0, 308.xx, 309.21, 309.81 baseline At least one inpatient or two outpatient claims in Medicaid claims with the following ICD 9C M codes at any

PAGE 153

153 baseline baseline during baseline during pregnancy Diagnosis field: 296.xx, 301.13 At least one inpatient or two outpatient claims in Medicaid claims with the following ICD 9CM codes at any Diagnosis field: 346.xx At least one inpatient or two outpatient claims in Medicaid claims with the following ICD 9CM codes at any Diagnosis field: 333.2 At least one inpatient or two outpatient claims in Medicaid claims with the following ICD 9CM codes at any Diagnosis field: 338.xx At least one inpatient or two outpatient Medicaid claims during pregn ancy with the following ICD 9CM code at any Diagnosis field: 648.8 during baseline At least one inpatient or two outpatient Medicaid claims during baseline with the following ICD 9CM code at any Diagnosis field: 290 3 19 pregnancy At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A 5 pregnancy At least one pharmacy claim with NDC during pregna ncy for active ingredient listed in appendix A 3, A 4 At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A during pregnancy At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A 2 and hypnotics) exposure during pregnancy At least one pharmacy claim during pregnancy with NDC for active ingredient listed in appendix A 7

PAGE 154

154 baseline or pregnancy pregnancy At least one pharmacy claim with NDC during baseline or pregnancy for active ingredient listed in appendix A 8 At lea st one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A 9 during pregnancy At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A 10 Number of hospitalization for seizure during pregnancy Number of physician visits with seizure diagnoses during pregnancy during baseline, including: Virus, Rubella, Cytomega lovirus, Syphilis, Herpes simplex virus, Toxoplamosis, Varicella virus, Venezuelan equine encephalitis virus, Phenylketonuria, Hypoxia. during pregnancy Number of i npatient Medicaid claims during pregnancy and each with at least one inpatient diagnoses codes: 345.xx Number of outpatient Medicaid claims during pregnancy and each with at least one diagnoses codes: 345.xx At least one pharmacy claim with NDC during pr egnancy for active ingredient listed in appendix A 11 At least one inpatient or two outpatient Medicaid claims during baseline or pregnancy with the following ICD 9CM code at any Diagnosis field: 1.xx 139.xx, 270.1, 799.02, At least one inpatient or tw o outpatient Medicaid claims during pregnancy with the following ICD 9CM code at any Diagnosis field: 92.29, 92.41, 92.27, 92.3, 92.59, 93.59, V58 or CPT codes: 77261 77263, 77336, 77336 59, 77280 77295, 77370, 77305 77315, 77401 77418, 77321, 77432, 77326 77328, 77431 77432, 77331, 77470, 77332 77334, 77520 77525, 76370 76375, 76380, 76950, 76965, 36000, 36410, 61793, 76942, 77261 77263, 77305 77315, 77321, 77326 77328, 77427, 77950, 77370, 76000, C9722, G0242,

PAGE 155

155 during baselin e zinc) diagnoses during baseline baseline during baseline during baseline virillizing turmors diagnoses during baseline pregnancy pregnancy based anticoagulants exposure during pregnancy G0338, G0243, G0173, G0251, G0339, G0340, 992 4x, 76355, 77300, 0083T, 77781 59, 77782 59, C1717. At least one inpatient or two outpatient Medicaid claims during baseline or pregnancy with the following ICD 9CM code at any Diagnosis field: 281.2 At least one inpatient or two outpatient Medicaid cla ims during baseline or pregnancy with the following ICD 9CM code at any Diagnosis field: 269.3 At least one inpatient or two outpatient Medicaid claims during baseline or pregnancy with the following ICD 9CM code at any Diagnosis field: 991.6 At least on e inpatient or two outpatient Medicaid claims during baseline or pregnancy with the following ICD 9CM code at any Diagnosis field: 358.0x At least one inpatient or two outpatient Medicaid claims during baseline or pregnancy with the following ICD 9CM code at any Diagnosis field: 393 398 At least one inpatient or two outpatient Medicaid claims during pregnancy with the following ICD 9CM code at any Diagnosis field: 255.x At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A

PAGE 156

156 pr egnancy pregnancy pregnancy pregnancy pregnancy pregnancy At least one pharmacy claim with NDC during pre gnancy for active ingredient listed in appendix A At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A At least one pharmacy claim with NDC during pregnancy for active ingredient listed in appendix A At least one pharmacy claim with NDC during pregnancy for active ingr edient listed in appendix A

PAGE 157

157 LIST OF REFERENCES 1. The Centers for Disease Control and Prevention. Birth Defects. http://www.prevention news.com/general/cdc preve ntion.htm Accessed January 14, 2011. 2. Boivin A, Luo ZC, Audibert F, Msse B, Lefebvre F, Tessier R, Nuyt AM. Pregnancy complications among women born preterm. CMAJ 2012 Sep 24. 3. Bener A, Salameh KM, Yousafzai MT, Saleh NM. Pattern of Maternal Complicatio ns and Low Birth Weight: Associated Risk Factors among Highly Endogamous Women. ISRN Obstet Gynecol. 2012;2012:540495. Epub 2012 Sep 8. 4. Lund N, Pedersen LH, Henriksen TB, Selective Serotonin Reuptake Inhibitor Exposure in Utero and Pregnancy Outcomes. Arch Pediatr Adolesc Med. 2009; 163 (10): 949 954. 5. Kelly SW, Volurka RJ. The empirical assessment of construct validity, Journal of Operations Management 16 (1998) 387 405. 6. Meador KJ, Baker GA, Finnell RH, et al, In utero antiepileptic drug exposure: fetal death and malformations. Neurology 2006; 67; Pg407 12. 7. Wyszynski DF, Nambisan M, Surve T, Alsdorf RM, Smith CR, Holmes LB. Increased rate of major malformations in offspring exposed to valproate during pregnancy. Neurology 2005;64:961 965. 8. Mlgaard Nielsen D, Hviid A. Newer generation antiepileptic drugs and the risk of major birth defects. JAMA. 2011 May 18;305(19):1996 2002. 9. Bakker MK Kerstjens Frederikse WS Buys CH de Walle HE de Jong van den Berg LT First trimester use of paroxetine and congenital heart defects : a population based case control st udy. Birth Defects Res A Clin Mol Teratol. 2010 Feb;88(2):94 100. 10. Davis RL, Rubanowice D, McPhillips H, Raebel MA, Andrade SE, Smith D, Yood MU, Platt R; HMO Research Network Center for Education, Research in Therapeutics. Risks of congenital malformations and perinatal events among infants exposed to antidepressant medications during pregnancy. Pharmacoepidemiol Drug Saf. 2007 O ct;16(10):1086 94. 11. McCulloch C. Joint modelling of mixed outcome types using latent variables. Statistical Me thods in Medical Research 2008; 17:53 73 12. Rabe Hesketh S and Skrondal A. Classical latent variable models for medical research. Statistical Me thods i n Medical Research 2008; 17:5 32

PAGE 158

158 13. Liu X, Roth J. Development and validation of an infant morbidity index using latent variable models. Stat Med. 2008 Mar 30;27(7):971 89. 14. Sobel ME. Measurement, Causation and local independence in latent variable models. Late nt Variable Modeling and Applications to Causality, Lecture Notes i n Statistics, 1997, Volume 120, 11 28. 15. Hartley HO, The modified Gauss Newton method for the fitting of non linear regression functions by least squares. Techmetrics, 1961 3, 269 280. 16. Cordob a G, Schwartz L, Woloshin S, Bae H, Gotzsche C, Definition, reporting, and interpretation of composite outcomes in clinical trials: systematic review. BMJ 2010 341. 17. Freemantle N Calvert M Wood J Eastaugh J Griffin C Composite outcomes in randomized trials: greater precision but with greater uncertainty? JAMA. 2003 May 21;289(19):2554 9. 18. Streiner DL, Norman GR. Health Measurement Scales: A Practical Guide to their Development and Use. (2008) 4 th Edition. Oxford UK: Oxford University Press. 19. Saftlas AF, Lawson HW, Atrash HK, Pregnancy Related Morbidity. http://www.cdc.gov/reproductivehea lth/ProductsPubs/DatatoAction/pdf/rhow10.p df, Accessed October 16, 2012. 20. Finnell RH, Teratology: general considerations and principles. J allergy clin immunol, Feb 1999, vol 103, number 2, part 2. S337 42. 21. Kaneko S, Kondo T. Antiepileptic agents and birth defects: incidence, mechanisms and pre vention. CNS Drugs 1995; 3 (1): 41 55. 22. Tomson T, Which drug for the pregnant woman with ep ilepsy? N Engi J Med 2009; 360: 1667 1669. 23. Wide K, Winbladh B, Kallen B. Major malformations in infants exposed to antiepilepti c drugs in utero, with emphasis on carbamazepine and valproic acid: a nation wide, population based register study. Acta Paediatr. 2004;93: 174 24. Morrow J, Russell A, Guthrie E, et al. Malformation risks of antiepileptic drugs in pregnancy: a prospective stud y from the UK Epilepsy and Pregnancy Register. J Neurol Neurosung Psychiatry. 2006;77: 193 8. 25. Meador KJ, Pennell PB, Harden CL, Gordon JC, Tomson T, Kaplan PW, Holmes GL, French JA, Hauser WA, Wells PG, Cramer JA and for the HOPE work group, Pregnancy regis tries in epilepsy: A consensus statement on healt h outcomes, Neurology 2008; 71; Pg1109 1117.

PAGE 159

159 26. MedWatch: The FDA Safety Information and Adverse Event Reporting Program. http://www.fda.gov/Safety/MedWatch/default.htm. Accessed May 14 2012. 27. http://www.fda.gov/Drugs/DrugSafety/ucm245085.htm Accessed October 7th, 2012. 28. Cunnington MC, Weil JG, Messenheimer JA, Ferber S, Yerby M, Tennis P. Final results from 18 years of the International Lamotrigin e Pregnancy Registry. Neurology. 2011 May 24;76(21):1817 23. 29. Holmes LB, Mittendorf R, Shen A, Smith CR, Hernandez Diaz S. Fetal effects of anticonvulsant polytherapies: different risks from different drug combinations. Arch Neurol. 2011 Oct;68(10):1275 81. Epub 2011 Jun 13. 30. Dolk H, McElhatton P, Assessing epidemiological evidence for the teratogenic effects of anticonvulsant medications. J Med Genet 2002 39: 243 44. 31. Scheuerle A, Tilson H. Birth defect classification by organ system: a novel approach to hei ghten teratogenic signalling in a pregnancy registry. Pharmacoepidemiol Drug Saf. 2002 Sep;11(6):465 75. 32. Scheuerle A, Tilson H. Birth defect classification by organ system: a novel approach to heighten teratogenic signalling in a pregnancy registry. pharmacoepi demiology and drug safety 2002; 11:465 475 33. van Walraven C, Mamdani M, Fang J, et al: Continuity of care and patient outcomes after hospital discharge. Journal of Genera l Internal Medicine 19:624 631, 2004. 34. Horn SD, Sharkey PD, Buckle JM, et al: The relationship between severity of illness and hospital length of stay and mort ality. Medical Care 29:305 317, 1991. 35. CDC. National Center for Health Statistics. Annual summary of births, marriages, divorces, and deaths: United States, 1991. Hyattsville, Maryland: US Department of Health and Human Services, Public Health Service, CDC, 1992. (Monthly vital statistics report; vol.40, no13). 36. Mathews TJ, MacDorman MF. Infant mortality statistics from the 2005 period linked birth/infant death data set. Natl Vital Stat Rep 57(2):1 32. 37. Palmieri C, Canger R., Teratogenic potential of the newer antiepileptic drugs. CNS Drugs 2002; 16 (11):755 764. 38. Pennell PB, Antiepileptic drugs during pregnancy: what is known and which AEDs seem to be safest? Epilepsia, 49 (Suppl. 9):43 55, 2008.

PAGE 160

160 39. Pennell PB, Using current evidence in selecting antiepileptic drugs for use during pregnancy. Epilepsy currents, Vol. 5, No. 2 (March/April) 2005 pp. 45 51. 40. Finnell RH, Nau H, Yerby JA. General P rincipals: Teratogenenicity of antiepileptic drugs. In: Levy RH, Mattson RH, Meldrum BS, eds. Antiepileptic Drugs, 4th e d. New York: Raven Press,1995: 209 230. 41. Harden CL, Meador KJ, Pennell PB, et al. Practi ce Parameter update: Management issues for women with epilepsy -focus on pregnancy (an evidence based review): Teratogenesis and perinatal outcomes. Report of the Quality Standards Subcommittee and Therapeutics and Technology Subcommittee of the American A cademy of Neurology and American Epilepsy Society. Neurology 2009. 42. Wilkins, 2004. 43. Lindhout D, Meinardi H, Meijer J, Nau H. (1992) Antiepileptic drugs and teratogenesis in two co nsecutive cohorts: changes in prescription policy paralleled by changes in pattern of malforma tions. Neurology 42 (Suppl. 5): 94 110. 44. Persaud TVN, Chudley AE, Skalko RG, Basic concepts in teratology, Alan R. Liss, Inc. New York. 1985. 45. Hoyme HE, Minor anom alies: diagnostic clues to a aberrant human morphogenesis. Genetica 89: 307 315, 1993. 46. Margaret A, Hudgins L, The importance of minor anomalies in the evaluation of the newborn. NeoReviews 2003; 4;99. 47. Stevenson RE, Hall JG, 1993. Terminology. Pp. 21 30 in Hall JG, Goodman RM (eds.) Human malformations and related anomalies, Vol. I, edited by Stevenson RE, Hall JG, and Goodman RM. Oxford Univ. Press, New York. 48. Joyce A. Martin, M.P.H., and Fay Menacker, Expanded Health Data from the New Birth Certificate, 20 04, National Vital Statistics Reports. 2007. Vol 55, N12. http://www.cdc.gov/nchs/data/nvsr/nvsr55/nvsr55_12.pdf. Accessed May 15, 2012. 49. Carlo WA. Fetal alcohol syndrome. In: Kliegman RM, Behrman RE, Jenson HB, Stanton BF, eds. Nelson Textbook of Pediatri cs. 18th ed. Philadelphia, Pa: Saunders Elsevier; 2007:chap 100.2. 50. Kenneth D. Kochanek, Sharon E. Kirmeyer, Joyce A. Martin, Donna M. Strobino, Bernard Guyer. Annual Summary of Vital Statistics: 2009. Pediatrics 2012;129;338;

PAGE 161

161 51. McCormick MC. The contributio n of low birth weight to infant mortality and childhood morbidity. N Engl J Med. 1985 Jan 10;312(2):82 90. 52. Lorenz JM, Wooliever DE, Jetton JR, Paneth N. A quantitative review of mortality and developmental disability in extremely premature newborns. Arch P ediatr Adolesc Med. 1998 May;152(5):425 35. 53. Mercuro G, Bassareo PP, Flore G, Fanos V, Dentamaro I, Scicchitano P, Laforgia N, Ciccone MM. Prematurity and low weight at birth as new conditions predisposing to an increased cardiovascular risk. Eur J Prev Car diolog. 2012 Jan 19. 54. Greenough A. Long term pulmonary outcome in the preterm infant. Neonatology. 2008;93(4):324 7. Epub 2008 Jun 5. 55. Phillips DI. Birth weight and the future development of diabetes. A review of the evidence. Diabetes Care. 1998 Aug;21 Supp l 2:B150 5. 56. Luyckx VA, Brenner BM. Low birth weight, nephron number, and kidney disease. Kidney Int Suppl. 2005 Aug;(97):S68 77. 57. Bener A, Salameh KMK, Yousafzai MT, Saleh NM, Pattern of Maternal Complications and Low Birth Weight: Associated Risk Factors among Highly Endogamous Women ISRN Obstet Gynecol. 2012; 2012:540495. 58. Borthen I, Eide MG, Daltveit AK, Gilhus NE, Obstetric outcome in women with epilepsy: a hospital based, retrospective study. BJOG 2011; 118:956 965. 59. Veiby G Daltveit AK Engelsen BA Gilhus NE. Pregnancy, delivery, and outcome for the child in maternal epilepsy. Epilepsia. 2009 Sep;5 0(9):2130 9. 60. Gestational hypertension (pregnancy induced hypertension), http://www.babycenter.com/0_gestational hypertension pregnancy induced hypertension_1427402.bc. Accessed November 4, 2012. 61. Brown MA. Diagnosis and Classification of Preeclampsia and Other Hypertensive Disorders of Pregnancy. Hypertension in Pregnancy. Edited by Belfort MA, Thornton S, Saade GR. 2003. Marcel Dekker, Inc. Pg1 16. 62. Voto LS, Lapidus AM, Margulies M, Effects of preeclampsia on the mother, fetus and child. Gynaecology Forum Vol 4 No1, 1999. 63. Witlin AG. Eclampsia What is New? Hypertension in Pregnancy. Edited by Belfort MA, Thornton S, Saade GR. 2003. Marcel Dekker, Inc. Pg221 234. 64. Premature Birth, http://www.cdc.gov/features/prematurebirth, Accessed November 04, 2012

PAGE 162

162 65. Hamil ton BE, Martin JA, Ventura SJ. Births: Preliminary Data for 2007, National Vital Statistics Reports, Volume 57, Number 12, (3/18/2009). 66. Anderson JM, Etches D."Prevention and management of postpartum hemorrhage. American Family Physician, March 2007. 75 (6 ): 875 82. 67. Pilo C, Wide K, Winbladh B. Pregnancy, delivery, and neonatal complications after treatment with antiepileptic drugs. Acta Obstet ricia et Gynecologica 2006;85: 643 646. 68. I Borthen, MG Eide, G Veiby, AK Daltveit, NE Gilhus. Complications during pre gnancy in women with epilepsy: population based cohort study. BJOG 2009;116:1736 1742. 69. Borthen I, Eide MG, Daltveit AK, Gilhus NE. Delivery outcome of women with epilepsy: a population based cohort study. BJOG. 2010 Nov;117(12):1537 43. 70. Hansen M, Bower C, Milne E, de Klerk N, Kurinczuk JJ. Assisted reproductive technologies and the risk of birth defects a systematic review. Huma n Reproduction. Vol.20, No.2 pp. 328 338, 2005 71. Msall ME, Tremont MR. Measuring functional outcomes after prematurity: developmental impact of very low birth weight and extremely low birth weight status on childhood disability. Ment Retard Dev Disabil Res Rev. 2002;8(4):258 72. 72. Decoufl P, Boyle CA, Paulozzi LJ, Lary JM. Increased risk for developmental disabilities in children who hav e major birth defects: a population based study. Pediatrics. 2001 Sep;108(3):728 34 73. Agustines LA, Kub YG, Rumney PJ, Lu MC, Bonebrake R, Asrat T, Nageotte M. Outcomes of extremely low birth weight infants between 500 and 750 g. American Journal of Obstetr ics and Gynecology 2000; 182 :1113 1116 74. Schendel DE, Stockbauer JW, Hoffman HJ, Herman AA, Berg CJ, Schramm WF. Relation between very low birth weight and developmental delay among preschool children without disabilities. American Journal of Epidemiology 199 7; 146 :740 749. 75. Hogan DP, Park JM. Family factors and social support in the developmental outcomes of very low birth weight children. Clinical Perinatology 2000; 27 : 433 459. 76. Duong HT, Hoyt AT, Carmichael SL, Gilboa SM, Canfield MA, Case A, McNeese ML, Wall er DK; and the National Birth Defects Prevention Study. Is maternal parity an independent risk factor for birth defects? Birth Defects Res A Clin Mol Teratol. 2012 Feb 28.

PAGE 163

163 77. Baardman ME, Kerstjens Frederikse WS, Corpeleijn E, de Walle HE, Hofstra RM, Berger RM, Bakker MK. Combined adverse effects of maternal smoking and high body mass index on heart development in offspring: evidence for interaction? Heart. 2012 Mar;98(6):474 9. Epub 2012 Jan 30. 78. Siega Riz AM, Herring AH, Olshan AF, Smith J, Moore C; Nationa l Birth Defects Prevention Study. The joint effects of maternal prepregnancy body mass index and age on the risk of gastroschisis. Paediatr Perinat Epidemiol. 2009 Jan;23(1):51 7. 79. M. Shevell, S. Ashwal, D. Donley, J. Flint, M. Gingold, D. Hirtz, A. Majneme r, M. Noetzel, and R.D. Sheth, Practice parameter: Evaluation of the child with global developmental delay: Report of the Quality Standards Subcommittee of the American Academy of Neurology and the Practice. Committee of the Child Neurology Society. Neurol ogy 2003;60;367 380 80. Women with epilepsy: drug risks and safety during pregnancy. American academy of neurology. 2009, http://www.epilepsyfoundation.org/epilepsyusa/news/upload/AANWWEGuideline sPatientandClinicianSummariesEMBARGOED.pdf Accessed January 2, 2 011. 81. Samren EB, van Duijn CM, Christianens GC, et al. Antiepileptic drug regimens and major congenital abnormalities in the offspring. Ann Neurol 1999; 46:739 46. 82. Kaneko S, Battino D, Andermann E, et al. Congenital malformations due to antiepileptic drugs. Epilepsy Res 1999 ; 33:145 158. 83. Kaneko S, Otani K, Fukushima Y, et al. Teratogenicity of antiepileptic drugs: analysis of possible ri sk factors. Epilpesia 1988;29: 459 67. 84. Friis ML. Facial clefts and congenital heart defects in children of parents with epilep sy: genetic and environmental etiologic f actors. Acta Neurol scand 1989;79: 433 59. 85. Tomson T, Battino D, French J, Harden C, Holmes L, Morrow J, Robert Gnansia E, Scheuerle A, Vajda F, Wide K, Gordon J. Antiepileptic drug exposure and major congenital malf ormations: the role of pregnancy registries. Epilepsy Behav 2007 Nov;11(3):277 82. 86. Wegner C, Nau H. Alteration of embryonic folate metabolism by valproic acid during organogenesis: implications for mechanism of teratogenesi s. Neurology 1992; 42 Suppl. 5: 1 7 24. 87. Ogawa Y, Kaneko S, Otani K, et al. Serum folic acid in epileptic mothers and their relationship to congenital malform ations. Epilepsy Res 1991; (8): 75 8. 88. 1997; 21 (2):114 23

PAGE 164

164 89. Finnel RH, Buehler BA, Kerr BM, et al. Clinical and experimental studies linking oxidative metabolism to phenytoin induced teratogenesis. Neurology 1992; 42 Suppl. 5: 25 31. 90. Bittigau P, Sifringer M, Phol D, et al. Apoptotic neurodegeneration following t rauma is markedly enhanced in the immature brain. Ann Neurol 1999; 45: 724 35. 91. Bittigau P, Sifringer M, Ikonomidou C. Antiepileptic drugs and apoptosis in the developing br ain. Ann NY Acad Sci 2003; 993: 103 14. 92. Vajda FJ, O'Brien T J, Hitchcock A, et al. Critical relationship between sodium valproate dose and human teratogenicity: results of the Australian register of anti epileptic drugs in pregnancy. J Clin Neurosci 2004;11:854 858. 93. Artama M, Auvinen A, Raudaskoski T, Isojarvi I, Isojarvi J. Antiepilepti c drug use of women with epilepsy and congenital malformations in offspring. Neurology 2005;64:1874 1878. 94. Wide K, Winbladh B, Kallen B. Major malformations in infants exposed to antiepileptic drugs in utero, with emphasis on carbamazepine and valproic acid : a nation wide, population based register study. Acta Paediat r. 2004;93: 174 95. Morrow J, Russell A, Guthrie E, et al. Malformation risks of antiepileptic drugs in pregnancy: a prospective study from the UK Epilepsy and Pregnancy Register. J Neurol Neurosung Psychiatry. 2006; 77: 193 8. 96. Kaaja E, Kaaja R, Hiilesmaa V. Major malformations in offspring of women with epilepsy. Neurology. 2003; 60: 575 9. 97. Adab N, Kini U, Vinten J, et al. The longer term outcome of children born to mothers with epilepsy. J Neurol Neur osung Psychiatry. 2004; 75: 1575 83. 98. Kimford J. Meador, M.D., Gus A. Baker, Ph.D., Nancy Browning, Ph.D., Jill Clayton Smith, M.D., Deborah T. Combs Cantrell, M.D., Morris Cohen, Ed.D., Laura A. Kalayjian, M.D., Andres Kanner, M.D., Joyce D. Liporace, M.D., Page B. Pennell, M.D., Michael Privitera, M.D., and David W. Loring, Ph.D. for the NEAD Study Group Cognitive Function at 3 Years of Age after Fetal Exposure to Antiepileptic Drugs N Engl J Med 2009;360:1597 605 99. Gaily E, Kantola Sorsa E, Hiilesmaa V, et al. Normal intelligence in children with prenatal exposure to carbamazepine. Neurology. 2004;62:28 32. 100. Olafsson E, Hallgrimsson JT, Hauser WA, et al. Pregnancies of women with epilepsy: a population based study i n Iceland. Epilepsia. 1998; 39: 887 892. 101. Kimford J Meador, Gus A Baker, Nancy Browning, Morris J Cohen, Rebecca L Bromley, Jill Clayton Smith, Laura A Kalayjian, Andres Kanner, Joyce D

PAGE 165

165 Liporace, Page B Pennell, Michael Privitera, David W Loring, for the NEAD Study Group Fetal antiepileptic drug exposure and cognitive outcomes at age 6 years (NEAD study): a prospective observational study Lancet Neurol. 2013 Mar; 12(3): 244 52. 102. Sobel ME. Measurement, Causation and local independence in latent variable models. Latent Variable Modeling and Applic ations to Causality, Lecture Notes in Statistics, 1997, Volume 120, 11 28. 103. Bishop CM. Latent Variable Models, Published in Learning in Graphical Models, M. I. Jordan (Ed.), MIT Press (1999), 371{403}. 104. Bartholomew, D.J., and Knott, M. (1999). Latent Variab le Models and Factor Analysis. London: Arnold. 105. Rabe Hesketh S and Skrondal A. Multilevel and Latent Variable Modeling with Composite Links and Exploded Likelihoods Psychometrika 200 7 ; Vol.72, No.2, 123 140. 106. McCulloch C. Joint modelling of mixed outcome ty pes using latent variables. Statistical Method s in Medical Research 2008;17: 53 73 107. Evans, Merran; Hastings, Nicholas; Peacock, Brian (2000). Statistical Distributions. New York: Wiley. pp. 134 136. ISBN 0 471 37124 6. 3rd ed.. 108. Bock DR Estimating Item Param eters and Latent Ability When Responses Are Scored in Two or More Nominal Categories Psychometrika. Vol. 37 No. 1 March, 1972 109. Goodman LA, Exporatory Latent Structure Analysis Using Both Identifiable and Unidentifiable Models. Biometrika, 1974 61, 215 2 31. 110. Haberman SJ, Log Linear Models for Frequency Tables Derived by Indirect Observation: Maximum Likelihood Equations. Annals of Statistics, 1974 2, 911 924. 111. Haberman SJ, Product models for frequency tables involving indirect observation. The Annals of S tatistics 1977, Vol. 5, No. 6, 1124 1147. 112. Casella G, Berger RL. Statistical Inference, second edition, Duxbury Thomson Learning. 2002. Pg338 113. Jennrich, R. I., & Sampson, P. F. (1976). Newton Raphson and related algorithms for maximum likelihood variance co mponent estimation. Technometrics, 18, 11 17. 114. McHugh RB Efficient Estimation and Local Identification in Latent Class Analysis. Psychometrika 19 56 21 331 347

PAGE 166

166 115. Gallant RA. Nonlinear Statistical Models, 1987, New York: Wiley. 116. Haberman SJ, A Stablized Newt on Raphson Algorithm for Log Linear Models for Frequency Tables Drived by Indirect Observation Sociological Methodology, 1988 18, 193 211. 117. Cordoba G, Schwartz L, Woloshin S, Bae H, Gotzsche C, Definition, reporting, and interpretation of composite outcomes in clinical trials: systematic review. BMJ 2010 341. 118. Montori VM, Alonso Coello P, et al. Methodologic discussions for using and interpreting composite endpoints are limited, but still identify major concerns. J Clin Epidemiol 2007;60:651 7. 119. Moher D, Schu lz KF, Altman D, for the CONSORT Group. The CONSORT statement: revised recommendations for improving the quality of reports of parallel group randomized trials. JAMA. 2001; 285: 1987 1991. 120. The Heart Outcomes Prevention Evaluation Study Investigators. Effe cts of an anglotensin converting enzyme inhibitor, ramipril, on cardiovascular events in high risk patients. N Engl J Med. 2000; 342: 145 153. 121. Freemantle N, Calvert M, Wood J, Eastaugh J, Griffin C. Composite outcomes in randomized trials: greater precisi on but with greater uncertainty? JAMA 2003;289:2554 9. 122. FDA. Guidelines for industry: patient/reported outcome measures: use in medical product development to support labelling claims: draft guidance. BMC Health and Quality of Life Outcomes 2006;4:79. 123. Ross S. Composite outcomes in randomized clinical trials: arguments for and against. Am J Obstet gynecol 2007;196:119e1 6. 124. Tomlinson G, Detsky AS. Composite end points in randomized trials: there is no free lunch. JAMA 2010;303:267 8. 125. Ferreira Gonzalez I, Perma nyer Miralda G, Domingo Salvany A, Busse JW, Heels Ansdell D, Montori VM, et al. Problems with use of composite end points in cardiovascular trials: systematic review of randomised controlled trials. BMJ 2007;334:786. 126. Ferreira Gonzalez I, Permanyer Miralda G, Busse JW, Bryant DM, Montori VM, Alonso Coello P, et al. Methodologic discussions for using and interpreting composite endpoints are limited, but still identify major concerns. J Clin Epidemiol. 2007 Jul;60(7):651 7.

PAGE 167

167 127. Montori VM, Permanyer Miralda G, Ferreira Gonzalez I, Busse JW, Pacheco Huergo V, Bryant D, et al. Validity of composite end points in clinical trials. BMJ 2005;330:594 6. 128. Ferreira Gonzlez I, Alonso Coello P Sol I Pacheco Huergo V Domingo Salvany A Alonso J Montori V Permanyer Miralda G. Composite endpoints in clinical trials. Rev Esp Cardiol 2008 Mar;61(3):283 90. 129. Pedhazur, E.J., Schmelkin, L.P., 1991. Measurement, Design, and Analysis: An Integrated Approach. Lawrence Erlbaum Associates, Publishers, Hillsda le, NJ. 130. Hair, Jr., J.F., Anderson, R.E., Tatham, R.L., Black, W.C., 1992. Multivariate Data Analysis with Readings, 3rd edition. Macmillan, New York. 131. Wainer H, Thissen D. How is reliability related to the quality of test scores? What is the effect of local dependence on reliability? Educational Measure 1996;15:22 29. 132. Reeve BB, Hays RD, Bjorner BB, Cook KF, Crane PK, Teresi JA, Thissen D, Revicki DA, Weiss DJ, Hambleton RK, Liu H, Gershon R, Reise SP, Lai J, Cella D. Psychometric Evaluation and Calibration of Health Related Quality of Life Item Banks: Plans for the Patient Reported Outcomes Measurement Information System (PROMIS). Medical Care. May 2007, Vol 45, Issue 5, pp S22 S31. 133. Yen, W. M. (1993). Scaling performance assessments: Strategies for managing local item dependence. Journal of Educational Measurement, 30, 187 213. 134. Kimberlin C, Winterstein AG. Validity and reliability of measurement instruments use in research. Am J Health Syst Pharm. Vol 65 Dec 1, 2008, pg2276 84. 135. John, O.P., Benet Martinez, V. (2000). Measurement: Reliability, construct validation, and scale construction. In H. T. Reis & C. M. Judd (Eds.), Handbook of research methods in social psychology pp. 339 369. New York: Cambridge University Press. 136. AHCA Annual Report 2008 2009. http://w ww.fdhc.state.fl.us/docs/ahca_annual_report_online5.pdf. Accessed April 29, 2012. 137. Multum Home. 2012. http://www.multum.com. Accessed April, 2012. 138. Simonaitis L, McDonald CJ. Using National Drug Codes and drug knowledge bases to organize prescription record s from multiple sources. Am J Health Syst Pharm. 2009 Oct 1;66(19):1743 53. 139. National Vital Statistics Reports Vol. 56, No. 6, December 5, 2007.

PAGE 168

168 http://www.cdc.gov/nchs/CSS/data/nvsr/n vsr56/nvsr56_06.pdf, Accessed 010611. 140. Northam S Knapp TR The reliability and validity of birth certificates. J Obstet Gynecol Neonatal Nurs. 2006 Jan Feb;35(1):3 12. 141. Reichman NE Hade EM Validation of birth certificate data. A study of women in New Jersey's HealthStart p rogram. Ann Epidemiol. 2001 Apr;11(3):186 93. 142. Roohan PJ Josberger RE Acar J Dabir P Feder HM Gagliano PJ Validation of birth certificate data in New York State. J Commun ity Health. 2003 Oct;28(5):335 46. 143. Gore DC Chez RA Remmel RJ Hara han M Mock M Yelverton R Unre liable medical information on birth certificates. J Reprod Med. 2002 Apr;47(4):297 302. 144. DiGiuseppe DL Aron DC Ranbom L Harper DL Rosenthal GE Reliability of birth certificate data: a multi hospital comparison to medical records information. Matern Child Hea lth J. 2002 Sep;6(3):169 79. 145. Zollinger TW Przybylski MJ Gamache RE Reliability of Indiana birth certificate data compared to medical records. Ann Epidemiol. 2006 Jan;16(1):1 10. Epub 2005 Jul 21. 146. Salemi JL, Tanner JP, Block S, Bailey M, Correia JA, Watkins SM, Kirby RS. The relative contribution of data sources to a birth defects registry utilizing passive multisource ascertainment methods: does a smaller birth defects case ascertainment net lead to overall or disproportionate loss? J Registry Manag. 2011 Spring;38(1):30 8. 147. Penberthy L, McClish D, Pugh A, Smith W, Manning C, and Re tchin S. Using Hospital Discharge Files to Enhance Cancer Surveillance. Am J Epidemiol 2003;158: 27 34. 148. Tang Y, Ma C, Cui W, Chang V, Ariet M, Morse SB, Resnick MB, Roth J. The risk of birth defects in multiple births: a population based study. Maternal an d Child Health Journal, Vol. 10, No. 1, January 2006. 149. Kirby RS. The quality of data reported on birth certificates. Am J Public Health. 1997;87(2):301. 150. Wang Y, Druschel CM, Cross PK, Hwang SA, Gensburg LJ. Problems in using birth certificate files in the capture recapture model to estimate the completeness of case ascertainment in a population based birth defects registry in New York State. Birth Defects Res A Clin Mol Teratol. 2006;76(11):772 777.

PAGE 169

169 151. Calle EE, Khoury MJ. Completeness of the discharge diagnos es as a measure of birth defects recorded in the hospital birth record. Am J Epidemiol. 1991;134(1):69 77. 152. Callif Daley FA, Huether CA, Edmonds LD. Evaluating false positives in two hospital discharge data sets of the Birth Defects Monitoring Program. Publ ic Health Rep. 1995;110(2):154 160. 153. Wang Y, Sharpe Stimac M, Cross PK, Druschel CM, Hwang SA. Improving case ascertainment of a population based birth defects registry in New York State using hospital discharge data. Birth Defects Res A Clin Mol Teratol. 2 005;73(10):663 668. 154. Tanner JP, Salemi JL, Hauser KW, Correia JA, Watkins SM, Kirby RS. Birth defects surveillance in Florida: Infant death certificates as a case ascertainment source. Birth Defects Res A Clin Mol Teratol. 2010;88(12):1017 1022. 155. Carter RL, Grove D. Assessment of case ascertainment accuracy. Tallahassee, FL: Florida Department of Health, Bureau of Environmental Epidemiology. 1999. 156. Newcombe H. Handbook of Record Linkage: Methods for Health and Statistical Studies, Administration, and Business Oxford: Oxford University Press, 1988. 157. Blakely T, Salmond C. Probabilistic record linkage and a method to calculate the positive predictive value. Int J Epidemiol. 2002 Dec;31(6):1246 52. 158. Ray WA, Griffin MR. Use of Medicaid data for pharmacoepidemiolog y. Am J Epidemiol 1989;129:837 49. 159. West SL, Savitz DA, Koch G, Strom BL, Guess HA, Hartzema A. Recall accuracy for prescription medications: self report compared with database information. Am J Epidemiol 1995;142:1103 12. 160. Cooper WO, Hernandez Diaz S, Arbog ast PG, Dudley JA, Dyer S, Gideon PS, Hall K, Ray WA. Major congenital malformations after first trimester exposure to ACE inhibitors. N Engl J Med. 2006 Jun 8;354(23):2443 51. 161. Costei AM, Kozer E, Ho T, Ito S, Koren G. Perinatal outcome following third tri mester exposure to paroxetine. Arch Pediatr Adolesc Med. 2002 Nov;156(11):1129 32. 162. Simmons, M. Thompson D, Graham C. An evaluation of the healthy start prenatal screen.h ttp://www.doh.state.fl.us/rw_ webmaster/news/abstracts/prenatal_screen. pdf. Accessed December 13, 2012.

PAGE 170

170 163. Eworuke E Hampp C Saidi A Winterstein AG An algorithm to identify preterm infants in administrative claims data. Pharmacoepidemiol Drug Saf. 2012 Jun;21(6):640 50. Epub 2012 Apr 16.164. 164. Raebel MA, Ellis JL, Andrade SE. Evaluation of gestat ional age and admission date assumptions used to determine prenatal drug exposure from administrative data. Pharmacoepidemiol Drug Safety 2005;14(12):829 836. 165. Andrade SE, Gurwitz JH, Davis RL, et al. Prescription drug use in pregnancy. Am J Obstet Gynecol 2004;191:398 407. 166. Pearl M, Wier ML, Kharrazi M. Assessing the quality of last menstrual period date on California birth records. Paediatr Perinat Epidemiol. Sep 2007;21 Suppl 2:50 61. 167. Florida Birth Defects Consortium, Annual report on birth defects in Florida: 1996, http://www.doh.state.fl.us/environment/newsroom/brochures/pds/fbdr final.pdf, A ccessed January 13, 2011 168. Correa Villasenor A, Cragan J, Kucik J Metropolitan Atlanta Congenital Defects Program: 35 years of birth defects surveillance at the Centers for Disease Control and Prevention. Birth Defects Res A Clin Mol Teratol 2003;67:617 24. 169. Ford JB, Roberts CL, Alg ert CS, Bowen JR, Bajuk B, Henderson Smart DJ; NICUS group. Using hospital discharge data for determining neonatal morbidity and mortality: a validation study. BMC Health Serv Res. 2007 Nov 20;7:188. 170. Salemi JL, Hauser KW, Tanner JP, et al. Developing a dat abase management system to support birth defects surveillance in Florida. J Registry Manag. 2010; 37(1): 10 15. 171. Cooper WO, Hernandez Diaz S, Gideon P, Dyer SM, Hall K, Dudley J, Cevasco M, Thompson AB, Ray WA, Positive predictive value of computerized rec ords for major congenital malformations. Pharmacoepidemiology and Drug Safety 2008; 17: 455 460. 172. http://icd9cm.chrisendres.com Accessed May 6, 2012. 173. Blume HK, Garrison MM, C hristakis DA. Neonatal seizures: treatment and treatment variability in 31 United States pediatric hospitals. J Child Neurol. 2009 Feb;24(2):148 54. 174. Avchen RN, Scott KG, Mason CA. Birth weight and school age disabilities: a population based study. Am J Epi demiol. 2001 Nov 15;154(10):895 901.

PAGE 171

171 175. Hosler AS, Nayak SG, Radigan AM. Agreement between self report and birth certificate for gestational diabetes mellitus: New York State PRAMS. Matern Child Health J. 2010 Sep;14(5):786 9. 176. Geller SE Ahmed S, Brown ML, Co x SM, Rosenberg D, Kilpatrick SJ. International Classification of Diseases 9th revision coding for preeclampsia: how accurate is it? Am J Obstet Gynecol 2004 Jun;190(6):1629 33; discussion 1633 4. 177. Martel MJ, Rey E, Beauchesne MF, Perreault S, Lefebvre G, Forget A, Blais L. Use of inhaled corticosteroids during pregnancy and risk of pregnancy induced hypertension: nested case control study. BMJ. 2005 Jan 29;330(7485) :230. Epub 2005 Jan 19. 178. La Merrill M, Stein CR, Landrigan P Engel SM Savitz DA Prepregnancy body mass index, smoking during pregnancy, and infant birth weight. Ann Epidemiol. 2011 Jun;21(6):413 20. Epub 2011 Mar 21. 179. http://www.who.int/mediacentre/factsheets/fs363/en /index.html Accessed March 25, 2003. 180. Simmons M, Thompson D, Graham C. An Evaluation of the healthy start prenatal screen 1998 birth cohort. http://www.doh.state.fl.us/rw_webmaster/news/abstracts/prenatal_screen.pdf. Accessed December 27 2012. 181. Patrick AR, Schneeweiss S, Brookhart MA, Glynn RJ, Rothman KJ, Avorn J, Strmer T. The imp lications of propensity score variable selection strategies in pharmacoepidemiology: an empirical illustration. Pharmacoepidemiol Drug Saf 2011 Jun;20(6):551 9. Epub 2011 Mar 10. 182. Schneeweiss S, Rassen JA, Glynn RJ, Avorn J, Mogun H, Brookhart MA. High dimensional propensity score adjustment in studies of treatment effects using health care claims data. Epidemiology 2009 Jul;20(4):512 22. 183. Strmer T, Schneeweiss S, Rothman KJ, Avorn J, Glynn RJ. Performance of propensity score calibration -a simulation study. Am J Epidemiol 2007 May 15;165(10):1110 8. Epub 2007 Mar 28. 184. Deonandan R Campbell K, Ostbye T, Tummon I, Robertson J. A comparison of methods for measuring socio economic status by occupation or postal area. Chronic Dis Can 2000;21(3):114 8. 185. Andrade SE, Gurwitz JH, Prescription drug use in pregnan cy. Am J Obstet Gynecol 191(2): 39 8 407. 186. Andrade SE, Raebel, MA, Use of prescription medications with a potential for fetal harm among pregnant women. Pharm acoepidemiol Drug Saf 15 (8): 546 54.

PAGE 172

172 187. Koren G, Pastuszak A, Ito S, Drugs in Pregnancy, N Engl J Med 1998; 338:1128 1137. 188. Bracken MB, H olford TR (1981). Exposure to prescribed drugs in pregnancy and association with congenital malfor mations Obstetrics and gynecology 58 (3): 336 44. 189. Davis RL, Rubanowice D, McPhillips H, HMO Research Network Center for Education, Research in Therapeutics. R isk of congenital malformations and perinatal events among infants exposed to antidepressant medications during pregnancy. Pharm acoepidemiol Drug Saf 2007;16: 1086 1094. 190. Niebyl JR, Simpson JL, Teratology and Drugs in Pregnancy, Gynecology and Obstetrics. V olume 2, Chapter 6. http://www.glowm.com/resources/glowm/cd/index.html. Accessed December 26, 2012. 191. Chung W, Teratogens and their effects. http://www.columbia.edu/itc/hs/medical/humandev/2004/Chpt23 Teratogens.pdf. Accessed December 26 2012. 192. The Effects of Workplace Hazards on Female Reproductive Health. DHHS (NIOSH) Publication No. 99.104. Feb 1999. http://www.columbia.edu/itc/hs/medical/humandev/2004/Chpt23 Teratogens.pdf. Accessed December 26 2012. 193. Maternal Disease and Complications Summary Outline. http://prosono.ieasysite.com/obg notes_maternal_complications.pdf. Accessed December 26 2012. 194. Seeger JD, Williams PL, Walker AM, An application of propensity score ma tching using claims data. Pharma coepidemiol Drug Saf. 2005;14: 465 476. 195. Westreich D, Cole SR, Funk MJ, Brookhart MA, Strmer T. The role of the c statistic in variable selection for propensity score models. Pharmacoepidemiol Drug Saf. 2011 Mar;20(3):317 20 doi: 10.1002/pds.2074. Epub 2010 Dec 9. 196. Freemantle N, Calvert M, Wood J, Eastaugh J, Griffin C. Composite outcomes in randomized trials: greater precision but with greater uncertainty? JAMA 2003; 289: 2554 9. 197. Ferreira Gonzalez I, Permanyer Miralda G, Domi ngo Salvany A, Busse JW, Heels Ansdell D, Montori VM, et al. Problems with use of composite end points in cardiovascular trials: systematic review of randomised co ntrolled trials. BMJ 2007;334: 786.

PAGE 173

173 198. Kuzma J Randomized clinical trial to compare the length of hospital stay and morbidity for early feeding with opioid sparing analgesia versus traditional care after open appendectomy. Clin Nutr. 2008 Oct;27(5):694 9. Epub 2008 Sep. 199. Weintraub WS, Jones EL, Craver I, Guyton R, Cohen C. Determinants of prolonged length of hospital stay after coronary bypass surgery. Circulation 1989 Aug;80(2):276 84. 200. Abdi, H. In N.J. Salkind (ed.). Encyclopedia of Measurement and Statistics (2007) Thousand Oaks, CA: Sage http://www.utdallas.edu/~herve/Abdi Bonferroni2007 pretty.pdf Accessed December 19, 2012. 201. Muthen LK, Muthen BO. Los Angeles, CA: Muthen & Muthen; 1998 22. Jo¨reskog KG, So¨rbom D, Du 202. Suhr DD. Exploratory or Confirmatory Factor Analysis? Statistics and Data Analysis 31 http://www2.sas.com/proceedings/sugi31/200 31.pdf Accessed December 26, 2012. 203. Cohen J. & Cohen P. (1 983). Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences. Hillsdale, NJ, Lawrence Erlbaum Associates, Inc., Publishers. 204. http://www .sas.com/service/techsup/faq/stat_macro/polychor.html. Accessed 08/06/13 205. Bobo WV, Davis RL, Toh S, Li DK, Andrade SE, Cheetham TC, Pawloski P, Dublin S, Pinheiro S, Hammad T, Scott PE, Epstein RA Jr, Arbogast PG, Morrow JA, Dudley JA, Lawrence JM, Avalo s LA, Cooper WO. Trends in the use of antiepileptic drugs among pregnant women in the US, 2001 2007: a medication exposure in pregnancy risk evaluation program study. Paediatr Perinat Epidemiol 2012 Nov;26(6):578 88. 206. Xuerong Wen, Kimford J Meador, Almut G Winterstein, Abraham G Hartzema. Utilization of Antiepileptic Drugs in Florida Medicaid Women with Epilepsy of Childbearing Age. Pharmacoepidemiology and Drug Safety, 2012; 21: (Suppl. 3): Pg. 322. 207. Kimford J. Meador, Patricia Penovich, Gus A. Baker, Page B. Pennell, Edward Bromfield, Alison Pack,Joyce D. Liporace, Maria Sam, Laura A. Kalayjian, David J. Thurman, Eugene Moore, David W. Loring, for the NEAD Study Group. Antiepileptic drug use in women of childbearing age. Epilepsy & Behavior 15 (2009) 339 3 43 208. http://www.fda.gov/Safety/MedWatch/SafetyInformation/SafetyAlertsforHumanMe dicSafetyAler/ucm150637.htm. Accessed July 2013

PAGE 174

174 209. Yen WM. Effect of local item dependence on the fit and equating performance of the three parameter logistic model. Appl Psychol M easure 1984; 8:125 145. 210. Ray WA. Improving automated database studies. Epidemiology 2011 May; 22 (3): 302 4. 211. Wang Y Cross PK Druschel CM Hospital discharge data: can it serve as the sole source of case ascertainment for population based birth defects surveillance programs? J Public Health Manag Pract. 2010 May Jun;16(3):245 51.

PAGE 175

175 BIOGRAPHICAL SKETCH Xuerong Wen was born in Chengdu and raised in Changsha, China. I n 2008, s in s tatistics and public health from the University of Florida, she joined the D epartm e n t of Pharmaceutical Outcomes and Policy at the University of Florida where she received a doctorate in pharmaceutical sciences Xuerong has authored and coauthored several peer reviewed publications, and pres ented at national and international conferences. Her research interests focus on pharmacoepidemiology, pharmacogenetics, and statistical modeling.