Citation
The Comparative Safety of Anti-Diabetic Treatment in Pregnant Women with Pre-Existing Diabetes

Material Information

Title:
The Comparative Safety of Anti-Diabetic Treatment in Pregnant Women with Pre-Existing Diabetes
Creator:
Knox, Caitlin A
Place of Publication:
[Gainesville, Fla.]
Florida
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (158 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Pharmaceutical Sciences
Pharmaceutical Outcomes and Policy
Committee Chair:
WINTERSTEIN,ALMUT GERTRUD
Committee Co-Chair:
SEGAL,RICHARD
Committee Members:
HAMPP,CHRISTIAN
XU,XIAOHUI
BRUMBACK,BABETTE A
Graduation Date:
8/9/2014

Subjects

Subjects / Keywords:
Diabetes ( jstor )
Diabetes complications ( jstor )
Infants ( jstor )
Insulin ( jstor )
Medicaid ( jstor )
Pregnancy ( jstor )
Pregnancy in diabetics ( jstor )
Type 1 diabetes mellitus ( jstor )
Type 2 diabetes mellitus ( jstor )
Women ( jstor )
Pharmaceutical Outcomes and Policy -- Dissertations, Academic -- UF
diabetes -- medicaid -- pharmacoepidemiology -- pregnancy -- safety -- treatment
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:
Research exploring the risk of anti-diabetic agents on obstetric outcomes in diabetic women is limited. To enhance the utility of Medicaid Analytic eXtract (MAX) data for epidemiologic research in pregnancy, we validated a Medicaid Case ID number algorithm to link mothers and infants, and identified factors that influenced linkage. We described longitudinal trends and determinants of initiation of anti-diabetic treatment in the first trimester, and estimated the risk of obstetric outcomes (preterm birth, cesarean section delivery, preeclampsia) in pre-existing diabetic women treated with anti-diabetic agents in the first trimester of pregnancy. We established retrospective cohorts of pregnant women, age 12 to 55 years with continuous Medicaid eligibility before and during pregnancy from 29 US states in MAX. Using linear regression, we estimated the prevalence of pre-existing diabetes, anti-diabetic use, and secular trends across the study period (2000-2006). We used logistic regression models to identify characteristics associated with initiation of anti-diabetic agents in the first trimester. Logistic regression and Cox proportional hazard models estimated the risk of obstetric outcomes. 3.6% of the 1,226,025 pregnancies in the pregnancy cohort had pre-existing diabetes. The utilization cohort included 658,485 deliveries with 2.4% prevalence of diabetes. The most commonly used anti-diabetics during pregnancy were insulin, metformin, sulfonylureas, and thiazolidinediones. Maternal characteristics associated with initiation of anti-diabetic agents in the first trimester were older age (women older than 40 years vs 20-29 years, odds ratio 1.60; 95% CI: 1.39, 1.84), and a diagnosis of hypertension (1.18; 95% CI: 1.05, 1.34). Compared to metformin, the adjusted odds ratio of preterm birth was 2.05 (95%CI: 1.09, 3.83) for sulfonylureas and 2.59 (95% CI: 1.61, 4.17) for insulin. Deliveries exposed to insulin in the first trimester had 42% (95% CI: 1.08,1.85) higher odds of cesarean section delivery compared first trimester metformin exposure. The adjusted hazard ratio for sulfonylureas was 0.61 (95% CI: 0.36, 1.04) as compared to metformin. This study emphasizes the increasing prevalence of pre-existing diabetes in pregnancy. It points to diverse risk profiles of anti-diabetic agents when considering different obstetric outcomes, highlighting the need for further research to optimize treatment of pre-existing diabetes during pregnancy. ( 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, 2014.
Local:
Adviser: WINTERSTEIN,ALMUT GERTRUD.
Local:
Co-adviser: SEGAL,RICHARD.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2015-08-31
Statement of Responsibility:
by Caitlin A Knox.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Embargo Date:
8/31/2015
Classification:
LD1780 2014 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

1 THE COMPARATIVE SAFETY OF ANTI DIABETIC TREATMENT IN PREGNANT WOMEN WITH PRE EXISTING DIABET ES By CAITLIN KNOX A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 4

PAGE 2

2 © 201 4 by Caitlin Knox

PAGE 3

3 To my grandmother, Betty Jean Root (1924 2014)

PAGE 4

4 ACKNOWLEDGMENTS I would like to express my gratitude to my supervisory committee: Richard Segal, Christian Hampp, Xiaohui Xu, and Babette Brumback, for their support and encouragement throughout my graduate career . I especially would like to thank my primary advisor , Almu t Winterstein , who has never allow ed me to rest on my laurels . Her consistent guidance, unwavering support and encouragement throughout my doctoral training, has been instrumental in my professional growth and development. Many thanks to Paul Kubilis for sharing his great statistical mind with me and always challenging me to improve my data analysis skills. To Linda Orr , Jill Hunt and Nicole Co r wine for sharing their administrative wisdom with me , a deserved and honored thank you. I am fore ver grateful to all the faculty and students in the department of Pharmaceutical Ou t c omes and Policy for their endless support and friendship that has been shown to me during my tenure at the University of Florida. For the provision of data, I thank the Office of Vital Statistics, Florida Department of Health and the Texas Department of State Health Services for the provision of birth certificates and the Centers for Medicare and Medicaid for provision of MAX data . My dissertation research was supported by the American F oundation for Pharmaceutical Education (AFPE) Pre doctoral fellowship, the Mary Kay Owen Fellowship, and the AHRQ Grants for Health Services Research Dissertation Program (1R36HS022384 01) . I want to thank my family for their unwavering support and encou ragement , especially my sister Tessa for providing her incredible proofreading skill s . Finally, I want to thank m y husband, Dan, for his endless love and support , thank you for always believ ing in m e , even when I have not.

PAGE 5

5 TABLE OF CONTENTS Page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ ........ 10 ABSTRACT ................................ ................................ ................................ ................... 12 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 14 Backgr ound ................................ ................................ ................................ ............. 14 Need for Study ................................ ................................ ................................ . 15 Purpose of Study ................................ ................................ .............................. 16 Research Que stions and Hypotheses ................................ ................................ ..... 17 Part I: Validation of Mother Infant Linkage Using the Medicaid Case ID Number Variable within the Medica id Analytic eXtract Database .................. 17 Part II: Epidemiology of Pre existing Diabetes in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ .................... 19 Part III: Utilization of Anti diabetic Agents in Pregnancy from 2000 2006 among Medicaid Patients in 29 US Sta tes ................................ .................... 19 Part IV: Comparative Safety of Anti diabetic aAgents in Pregnancy ................ 19 2 LITERATURE REVIEW ................................ ................................ .......................... 22 Diabetes Mellitus ................................ ................................ ................................ .... 22 Diagnosis and Classification of Diabetes Mellitus ................................ ............ 22 Diabetes Mellitus in Pregnancy ................................ ................................ ........ 24 Use of Anti diabetic Agents in Pregnancy ................................ ............................... 25 Metformin ................................ ................................ ................................ ......... 26 Sulfonylureas ................................ ................................ ................................ .... 28 Thiazolidinedi one (TZDs) ................................ ................................ ................. 30 Obstetric Outcomes ................................ ................................ ................................ 33 Preterm Birth ................................ ................................ ................................ .... 33 Cesarean Section Delivery ................................ ................................ ............... 34 Pr eeclampsia ................................ ................................ ................................ .... 35 3 METHODS ................................ ................................ ................................ .............. 38 Data Sources ................................ ................................ ................................ .......... 38 Medicaid Analytic eXtract (MAX) ................................ ................................ ...... 39 Internal Validity ................................ ................................ ................................ . 40 Vital Birth Certificate Records ................................ ................................ ........... 42 Source Population ................................ ................................ ................................ .. 42

PAGE 6

6 Source Population Measures ................................ ................................ ........... 43 Part I: Validation of Mother Infant Linkage Using the Medicaid Case ID Number Variable within the Medicaid Analytic eXtract Da tabase ................................ ...... 44 Sample Population ................................ ................................ ........................... 44 Data Analysis ................................ ................................ ................................ ... 44 Mother infant link ................................ ................................ ....................... 44 Validation of Medicaid Case ID number linkage ................................ ......... 46 Predicting linkage within a validated algorithm that identifies mother infant pairs within the Medicaid Analytic eXtract (MAX) database. ......... 47 Part II: Epidemiology of Pre existing Diabetes in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ........................... 49 Study Design ................................ ................................ ................................ .... 49 Sample Population ................................ ................................ ........................... 49 Data Analy sis ................................ ................................ ................................ ... 50 Part III: Utilization of Anti diabetic Agents in Pregnancy from 2000 2006 among Medicaid Patients in 5 US States ................................ ................................ ........ 51 Study Design ................................ ................................ ................................ .... 51 Sample population ................................ ................................ ..................... 52 Exposure d efinitions ................................ ................................ ................... 53 Data Analysis ................................ ................................ ................................ ... 54 Part IV: Comparative Safety of Oral Anti diabetic Agents in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ........ 55 Study Design ................................ ................................ ................................ .... 55 Sample population ................................ ................................ ..................... 56 Exposure definitions ................................ ................................ ................... 56 Outcome assessment ................................ ................................ ................ 57 Active comparison group ................................ ................................ ............ 57 Data Analysis ................................ ................................ ................................ ... 58 Intention to treat (ITT) analysis ................................ ................................ .. 58 Propensity scores ................................ ................................ ...................... 58 Statistical analysis ................................ ................................ ...................... 62 Explorat ory analysis ................................ ................................ ................... 63 4 RESULTS ................................ ................................ ................................ ............... 70 Part I: Validation of Mother Infant Linkage Using the Medicaid Case ID Number Variable within the Medicaid Analytic eXtract Database ................................ ...... 70 Sample Characteristics ................................ ................................ ..................... 70 Validation of Medicaid Case ID Number Linkage ................................ ............. 71 Predicting Linkage within a Validated Algorithm that Identifies Mother Infant Pairs within the Medicaid Analytic eXtract (MAX) Database. ........................ 72 Part II: Epidemiology of Pre existing Diabetes in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ........................... 73 Part III: Utilization of Anti diabetic Agents in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ................................ ...... 75 Part IV: Comparative Safety of Oral Anti diabetic Agents in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ........ 78

PAGE 7

7 Preterm Birth ................................ ................................ ................................ .... 80 Cesarean Section Delivery ................................ ................................ ............... 80 Preeclampsia ................................ ................................ ................................ .... 81 Exploratory Analysis ................................ ................................ ......................... 81 5 DISCUSSION ................................ ................................ ................................ ....... 116 Part I: Validation of Mother Infant Linkage Using the Medicaid Case ID Number Variable within the Medicaid Analytic eXtract Database ................................ .... 116 Part II: Epidemiology of Pre existing Diabetes in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ......................... 118 Part III: Utilization of Anti diabetic Agents in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ................................ .... 121 Part IV: Comparative Safety of Oral Anti diabetic Agents in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States ................................ ...... 124 Preterm Birth ................................ ................................ ................................ .. 124 Cesarean Section Delivery ................................ ................................ ............. 125 Pr eeclampsia ................................ ................................ ................................ .. 127 Exploratory Analysis ................................ ................................ ....................... 128 Strengths and Limitations ................................ ................................ ............... 129 Conclusions ................................ ................................ ................................ .......... 132 APPENDIX A OPERA TIONAL DEFINITIONS ................................ ................................ ............. 134 B SUPPLEMENTAL TABLES ................................ ................................ .................. 142 LIST OF REFERENCES ................................ ................................ ............................. 144 BIOGRAPH ICAL SKETCH ................................ ................................ .......................... 158

PAGE 8

8 LIST OF TABLES Table page 2 1 Safety and teratogenicity classification for use of selected oral anti diabetic agents during pregnancy and lactation ................................ ........................... 37 3 1 Calculation of linkage charact eristics. ................................ ................................ . 69 4 1 Successful Florida Birth Certificate Records to Medicaid linkages in percent by year (N=1,239,148 deliveries). ................................ ................................ ....... 95 4 2 Sensitivity of Medicaid Case ID Number algorithm varied by mother and ................................ ................................ ................. 96 4 3 Specificity of Medicaid Case ID Number algorithm varied by mother and ................................ ................................ ................. 97 4 4 Positive predictive value (PPV) of Medicaid Case ID Number algorithm varied ................................ ......................... 98 4 5 Determinants of accurate mother infant linkage by Medicaid Case ID Number. ................................ ................................ ................................ .............. 99 4 6 Baseline characteristics of the pregnancy cohort stratified by pre existing diabetes. ................................ ................................ ................................ ........... 102 4 7 Baseline characteristics of the pregnancy cohort stratified by type of diabetes. ................................ ................................ ................................ ........... 103 4 8 Annual prevalence of pre existing diabetes in pregnancy from 2000 2006 among Medicaid patients in 29 US states. ................................ ....................... 10 4 4 9 Baseline characteristics of th e delivery cohort. ................................ ................. 106 4 10 Baseline characteristics and pregnancy characteristics of the delivery cohort stratified by type of d iabetes. ................................ ................................ ............ 107 4 11 Predictors of initiation of anti diabetic agents in the first trimester of pregnancy among women with pre exist ing diabetes within the Medicaid Analytic eXtract database. ................................ ................................ ................ 108 4 12 Baseline characteristics of the safety cohort categorized by anti diabetic exposure. ................................ ................................ ................................ .......... 109 4 13 Risk of premature delivery among live deliveries exposed to anti diabetic agents during the first trimester of pregnancy. ................................ .................. 112

PAGE 9

9 4 14 Risk of cesarean section delivery identified among live births exposed to anti diabetic agents during the first trimester of pregnancy. ................................ .... 113 4 15 Risk of preeclampsia identified among live deliveries exposed to anti diabetic agents during the first trimester of pregnancy. ................................ .................. 114 4 16 Exploratory analyses: Assessment of the impact of requiring a live birth in the comparative safety analysis. ................................ ................................ ............. 115 A 1 ICD 9 CM codes for the identification of comorbid conditions during baseline. 135 A 2 Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of pregnancy. ................................ ................................ ............... 136 A 3 Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of abortion (spontaneous or induced)*. ................................ ........ 138 A 4 Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of a stillbirth*. ................................ ................................ ............... 139 A 5 Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of a delivery*. ................................ ................................ ............... 140 A 6 Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of a cesarean section delivery.* ................................ ................... 141

PAGE 10

10 LIST OF FIGURES Figure page 3 1 Medicaid Analytic eXtract (MAX) data included in pregnancy cohort from 29 US states. ................................ ................................ ................................ ........... 66 3 2 Health Insurance Coverage of Women of Reproductive Age (15 44), 2006. ...... 67 3 3 Medicaid Coverage of Birth. 2008. ................................ ................................ ..... 67 3 4 Pregnancy cohort study timeline, Medicaid Analytic eXtract data (MAX), 1999 2006. ................................ ................................ ................................ ......... 68 3 5 Anti diabetic drug utilization study timeline, Medicaid Analytic eXtract data (MAX), 2000 2006. ................................ ................................ ............................. 68 3 6 Comparative safety sensitivity analysis study timeline, Medicaid Analytic eXtract data (MAX), 2000 2006. 2 nd , 2 nd trimester. ................................ ............. 69 4 1 Assembly of the mother infant linked BCR cohort, Florida Birth Certificate Records (BCR) data, 1999 2004. DOB: date of birth, SSN: Social Security Number. ................................ ................................ ................................ .............. 83 4 2 Assembly of the mother infant linked Medicaid Case ID number cohort, Florida Birth Certificate Records (BCR) data, 1999 20 04. DOB: date of birth, SSN: Social Security Number. ................................ ................................ ............ 84 4 3 Prevalence of pre existing diabetes in pregnancy by state. ................................ 85 4 4 Prevalence of type 2 diabetes in pregnancy by state. ................................ ........ 86 4 5 Temporal trends of pre existing diabetes in pregnancy. ................................ ..... 87 4 6 Annual prevalence of anti diabetic utilization in women with pre existing diabetes before and during pregnancy. ................................ .............................. 88 4 7 Annual prevalence of anti diabetic utilization during pregnancy by drug class. .. 89 4 8 Pooled annual prevalence of anti diabetic drug utilization, by drug class and pregnancy period. ................................ ................................ ............................... 90 4 9 Assembly of the safety study cohort. ................................ ................................ .. 91 4 10 Propensit y score distributions of specific anti diabetic exposure for the premature delivery model. PS=propensity score, met=metformin, sulf=sulfonylureas, in=insulin, ms=metformin & sulfonylureas, mt=metformin & thiazolidinedione. ................................ ................................ ............................ 92

PAGE 11

11 4 11 Propensity score distributions of specific anti diabetic exposure for the cesarean section delivery model. PS=propensity score, met=metformin, sulf=sulfonylureas, in=insulin, ms=metformin & sulfonylureas, mt=metformin & thiazolidinedione. ................................ ................................ ............................ 93 4 12 Propensity score distribu tions of specific anti diabetic exposure for the preeclampsia model. PS=propensity score, met=metformin, sulf=sulfonylureas, in=insulin, ms=metformin & sulfonylureas, mt=metformin & thiazolidinedione. ................................ ................................ ............................ 94

PAGE 12

12 ABSTRACT Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THE COMPARATIVE SAFETY OF ANTI DIABETIC TREATMENT IN PREGNANT WOMEN WITH PRE EXISTING DIABETES By Caitlin Knox August 2014 Chair: Almut G. Winterstein Major: Pharmaceutical Sciences Research exploring the risk of anti diabetic agents o n obstetric outcomes in diabetic women is lim ited . T o enhance the utility of Medicaid Analytic eXtract (MAX) data for epidemiologic research in pregnancy, we validated a Medicaid Case ID number algorithm to link mothers and infants , and identified factors that influenc e d linkage . We describe d longitu dinal trends and determina nt s of initiation of anti diabetic treatment in the first trimester, and estimate d the risk of obstetric outcomes (preterm birth, cesarean section delivery, preeclampsia) in pre existing diabetic women treated with anti diabetic a gents in the first trimester of pregnancy . We establish ed retrospective cohort s of pregnant women, age 12 to 55 years with continuous Medicaid eligibility before and during pregnancy from 29 US states in MAX . Using linear regression, w e estimated the prevalence of pre existing diabetes, anti diabetic use, and secular trend s across the study period (2000 2006 ) . We used logistic regression models to identify characteristics associated with initiation of anti -

PAGE 13

13 diabetic agents in the first trimester . L o gistic regression and Cox proportional hazard models estimate d the risk of obstetric outcomes. 3.6% of the 1,226,025 pregnancies in the pregnancy cohort ha d pre existing diabetes . The utilization cohort included 658,485 deliveries with 2.4% prevalence of diabetes. The most common ly used anti diabetics during pregnancy were insulin, metformin, sulfonylureas, and thiazolidinediones . Maternal characteristics associated with initiation of anti diabetic agents in the first trimester were older age ( women older than 40 years vs 20 29 years , odds ratio 1 . 6 0 ; 95% CI: 1.39 , 1.84 ), and a diagnos is of hypertension (1. 18 ; 95% CI: 1. 05 , 1. 34 ). Compared to metformin, the adjusted odds ratio of preterm birth was 2.05 (95%CI: 1.09, 3.83) for sulfonylureas and 2.59 (95% C I: 1.61, 4.17) for insulin. Deliveries exposed to insulin in the first trimester had 4 2 % (95% CI: 1.08,1.85) higher odds of cesarean section delivery compared first trimester metformin exposure . The adjusted hazard ratio for sulfonylurea s was 0.61 (95% CI: 0.36, 1.04) as compared to metformin. This study emphasizes the increasing prevalence of pre existing diabetes in pregnancy . It points to diverse risk profiles of anti diabetic agents when considering different obstetric outcomes , highli ghting the need for further research to optimize treatment of pre existing diabetes during pregnancy .

PAGE 14

14 CHAPTER 1 INTRODUCTION Background In 2010, approximately 11% of women in the United States (US) aged 20 years or older had either diagnosed or undiagnosed diabetes. 1 Diabetes mellitus is one of the most common medical conditions to complicate pregnancy. 2 7 The general treatment of diabetes outside of pregnancy requires balancing complicated drug and dietary regimen s . Pregnancy further increases the complexity of manag ing this chronic disease, as the developing fetus both diabetic and non diabetic women. Pre existing diabetes in pregnancy is associated with perinatal and neonatal morbidity and mortality, as well as increased risk of adverse outcomes for the mother including pregna ncy induced hypertension, coronary artery disease, cesarean section delivery , and postpartum hemorrhage. 2 12 The m anagement of glycemic levels in pregnancy is largely dependent on diabetes type and severity . P oorly controlled diabetes is linked to major birth defects, spontaneous abortions, and stillbirths , attributed to both h yper glycemia and hypoglycemia . The p roper maintenance of glycemic levels is particularly critical during the early stages of pregnancy , because the first trimester is cruc ial for the development of fetal organs . 2,13 16 It is therefore generally accepted by prescribers that acquiring and sustaining glucose control in diabetic women prior to conception is essential , 17 because many women are unaware of their pregnancy unti l well into the first trimester. R esearch suggest s the risk of adverse pregnancy outcomes in women with pre existing type 2 d iabetes may actually be higher tha n in women with type 1 diabetes. 2,15 In the last 20 years , great strides have been made in improving pregnancy outcomes of

PAGE 15

15 type 1 diabetic women. However, with the emergence of the diabetes epidemic, type 2 diabetes has developed as a prominent chronic condition seen in pregnancy with le sser evidence and guidance on its management . 2,3,11,18 Need for S tudy Under the Code of Federal Regulations (CFR), pregnant women require additional protection in research (45CFR46.202). They are considered a vulnerable population in need of special co nsideration by the approving Institutional Review Board (IRB) to ensure the safety of the mother and fetus included in the study. For this reason, it is common practice to exclude pregnant women, fetuses, and neonates from r andomized cl inical trials . D rug effectiveness and safety studies for pregnant women are largely dependent on post marketing observational studies and prospective drug registries. However, f ew pregnancy registries have the ability to study rare outcomes of pregnancy due to small sample si ze or incomplete patient data (lack of medication or medical history). Furthermore, no pregnancy registry curentley follows women with pre existing diabetes (particularly focusing on type 2 diabetic women). When retrospectively studying medication outcomes in pregnancy, i t is generally required for researchers to employ multiple databases . The ideal way of linking multiple databases is to utilize unique patient identifiers, such as social security numbers (SSN). However, such unique identifiers have become increasingly difficult to employ because of stringent regulations regarding the use of identifiable information in research. In the Medicaid Analytic eXtract (MAX) database, the state assigned Medicaid Case ID number identifies family units. This variabl e allows for the possible unique linkage of mothers to infants, which enables researchers to follow an entire pregnancy , including pregnancy outcomes, within one database. Furthermore, established mother infant

PAGE 16

16 pairs can be then linked to other databases u sing either the mother or child and thus, enhancing linkage performance. Re search using the Medicaid Case ID number is however limited, and prior research has not demonstrated the reliability or validity of the Medicaid Case ID number variable in uniquely linking mothers to infants. Little evidence exists regarding the safety of oral agents during pregnancy. The research that has been conducted on glucose management during pregnancy focuses on gestational diabetes , 14,19 22 and congenital malformations . 7,23 I mp ortant clinical questions remain unanswered concerning the potential of adverse outcomes for both the mother and infant if glucose levels during pregnancy are uncontrolled . Presently , t here are no comparative safety studies available that address the s af ety of oral anti diabetic agent s use in pregnancy , with respect to obstetric outcomes such as cesarean section deliveries , preterm delivery or preeclampsia . Understanding the impact of oral anti diabetic agent use o n obstetric outcomes in light of the inc reasing prevalence of type 2 diabetes mellitus may help improve safety and effectiveness of treatment regimens . This will ultimately result in a healthier mothers and infants and more efficient use of health care resources. Moreover, because most studies d o not assess the potential in dependent effects of anti diabetic agents , it remains unclear to what extent poor obstetric outcomes are due to factors intrinsic to the disease state itself , to drugs used to treat the disease or to a combination of both. Purp ose of S tudy This research aim ed to explore the impact of anti diabetic agents on adverse obstetric outcomes. The management of diabetes in pregnancy is dependent on the type and severity of the disease. Unlike type 1 diabetes management, management of

PAGE 17

17 type 2 diabetes may be limited to life style modifications , utiliz ation of oral anti diabetic agents, and/or insulin. In order to investigate the safety of oral anti diabetic agents used during pregnancy, we first identif ied the utilization pattern and ex plore d the characteristics of women who use d and/or initiate d oral agents during pregnancy. To conduct this research we establish ed and validate d an algorithm within the MAX database that link ed mothers to infants, maximizing the utility of administrative claims data in population based pregnancy research. This linkage will make an important contribution to pregnancy research , as it enhance s the evaluation of pregnancy outcomes even after delivery by providing access to the health information of both mothe r and child . T his is to our knowledge, the first comparative safety study addressing the association of oral anti diabetic agents and adverse obstetric outcomes. E vidence suggests a strong correlation between diabetes and adverse obstetric outcomes , but th e extent to which these complications can be reduced through drug treatment interventions is unknown. 23 Research Q uestions and H ypotheses This dissertation is comprised of four parts. A statistical significance level of 0.05 is hypothesis are represented by H 0 and H A , respectively. Part I: Validation of M other I nfant L inkage U sing the Medicaid Case ID N umber V ariable within the M edicaid Analytic eXtract D atabase Research Question 1a: Is the Case ID number variable in the Medicaid Analytic eXtract database able to link mothers to infants?

PAGE 18

18 Hypothesis 1a 1: H A : The Case ID number variable has poor sensitiv ity (i.e. in identifying linked mother infant pairs. H 0 : The Case ID number variable is sensitive in identifying linked mother infant pairs. Hypothesis 1a 2: H A : The Case ID number variable has poor specificity (i.e. 9 0%) in identifying linked mother infant pairs. H 0 : The Case ID number variable is specific in identifying linked mother infant pairs. Research Question 1 b : Can varying the Medicaid eligibility window of the mother increase the sensitivity and specificity of linking by Case ID number variable ? Hypothesis 1 b 1: H A : Sensitivity of the linkage at 3, 6, 9, 12 or 15 continuous months before delivery for maternal Medicaid eligibility is significantly greater than sensitivity at the minimum requirement of one month eligibility at delivery. H 0 : There is no signific ant difference. Hypothesis 1 b 2: H A : Specificity of the linkage at 3, 6, 9,12, or 15 continuous months before delivery for maternal Medicaid eligibility is significantly greater than specificity at the minimum requirement of one month eligibility at delivery. H 0 : There is no significant difference . Research Question 1 c : Can varying the Medicaid eligibility window of the infant increase the sensitivity and specificity of linking by Case ID number variable ? Hypothesis 1 c 1: H A : Sensitivi ty of the linkage at birth, 1, 3, 6, 9, or 12 continuous months after birth for infant Medicaid eligibility is significantly higher than sensitivity at the minimum requirement of one month eligibility any month during the study period . H 0 : There is no sign ificant difference.

PAGE 19

19 H ypothesis 1 c 2: H A : Specificity of the linkage at birth, 1, 3, 6, 9, or 12 continuous months after birth for infant Medicaid eligibility is significantly higher than specificity at the minimum requirement of one month eligibility any month during the study period . H 0 : There is no significant difference Research Question 1 d : W hat are the determinants of a successful mother infant link? Part II: Epidemiology of P re existing D iabetes in P regnancy from 2000 20 06 among Medicaid P atients in 2 9 US S tates Research Question 2a: What is the overall prevalence of pre existing diabetes in pregnant women? Research Question 2b: What is the annual prevalence of pre existing diabetes in pregnant women from 2000 to 2006 ? Part III: Utilization of A nti d iabetic A gents in P regnancy from 2000 20 06 among Medicaid P atients in 29 US S tates Research Ques ti on 3a: What are the overall trends in the utilization of anti diabetic agents before and during pregnancy in women with pre existing diabetes? Research Questi on 3b: What are the annual trends in the utilization of anti diabetic agents before and during pregnancy in women with pre existing diabetes from 2000 to 2006 ? Research Question 3 c : What factors predict the initiation of anti diabetic agents during the fir st trimester of pregnancy in women with pre existing diabetes? Part IV: Comparative S afety of A nti diabetic a A gents in P regnancy Research Question 4: What is the comparative safety of selected anti diabetic agents used during the first trimester of pregnancy in terms of obstetric outcomes

PAGE 20

20 ( preterm birth, cesarean section delivery, or preeclampsia) in pregnant women with pre existing diabetes? Hypothesis 4 1 a : H A : Sulfonylureas are associated with an increased risk when compared to metformin concerning cesarean selection delivery . H 0 : Sulfonylureas are not associated with an increased risk when compared to metformin concerning cesarean selection delivery . Hypothesis 4 1b: H A : Sulfonylureas are associated with an increased risk when compared to metformin concerning preterm birth. H 0 : Sulfonylureas are not associated with an increased risk when compared to metformin concerning preterm birth . Hypothesis 4 1c: H A : Sulfonylureas are associated with an increased risk when compared to metformin concerning preeclampsia. H 0 : Sulfonylureas are not associated with an increased risk when compared to metformin concerning preeclampsia . Hypothesis 4 2a: H A : Thiazolidinediones are associated with an increased risk when compared to metformin concerning cesarean selection delivery. H 0 : Thiazolidinediones are not associated with an increased risk when compared to metformin concerning cesarean selection delivery . Hypothesis 4 2b: H A : Thiazolidine diones are associated with an increased risk when compared to metformin concerning preterm birth. H 0 : Thiazolidinediones not are associated with an increased risk when compared to metformin concerning preterm birth . Hypothesis 4 2c: H A : Thiazolidinediones are associated with an increased risk when compared to metformin concerning preeclampsia. H 0 : Thiazolidinediones are not

PAGE 21

21 associated with an increased risk when compared to metformin concerning preeclampsia . Hypothesis 4 3a: H A : The combination of t hiazolidinediones and metformin therapy are associated with an increased risk when compared to the combination sulfonylureas and metformin therapy concerning cesarean selection delivery. H 0 : The combination of t hiazolidinediones and metformin therapy are n ot associated with an increased risk when compared to the combination sulfonylureas and metformin therapy concerning cesarean selection delivery . Hypothesis 4 3b: H A : The combination of t hiazolidinediones and metformin therapy are associated with an increa sed risk when compared to the combination sulfonylureas and metformin therapy concerning preterm birth. H 0 : The combination of t hiazolidinediones and metformin therapy are not associated with an increased risk when compared to the combination sulfonylureas and metformin therapy concerning preterm birth . Hypothesis 4 3c: H A : The combination of t hiazolidinediones and metformin therapy are associated with an increased risk when compared to the combination sulfonylureas and metformin therapy concerning preeclam psia. H 0 : The combination of t hiazolidinediones and metformin therapy are not associated with an increased risk when compared to the combination sulfonylureas and metformin therapy concerning preeclampsia .

PAGE 22

22 CHAPTER 2 LITERATURE REVIEW Diabetes M ellitus Diabetes mellitus (DM) is a common metabolic condition that imposes a large burden of morbidity and mortality on the individuals afflicted by this chronic disease. It is characterized by the manifestation of hyperglycemia due to a shortage in the producti on or secretion of insulin, the reduced tissue response to insulin, or both. 24 26 Diabetes has reached epidemic levels in both developed and developing countries around the world . 27,28 Within the United States (US) , the prevalence of diabetes continue s to grow with th e number of people diagnosed with diabetes i n 2010 estimated at over 26 million Americans. 27 In 2012 , the total cost of diabetes was approximately $ 245 billion, with $ 1 76 billion in direct medical costs to treat diabetes, $ 105 billion in general medical costs, and $ 44 billion in prescription cost to treat diabetes related chronic complications. 29 C urrent estimates project that by 2025 one in three adults in the US will have diabetes. 1 Thus, diabetes directly or indirectly , a ffect s almost everyone in the US. P revention of diabetes and its associated health care utilization is important to minimiz e morbidity and mortality , as well as the enormous economic burden associated with this chronic condition. Diagnosis and C lassification of D iabetes M ellitus Until recently, the diagnosis of diabetes was based solely on plasma glucose criteria, including fasting plasma glucose levels (FPG) or the 75 g oral glucose tolerance test (OGTT). 24, 26,30 32 The standardization of glycated hemoglobin test ( Hb A1C ) assay testing via methods certified by the National Glycohemoglobin Standardization Program

PAGE 23

23 (NGSP) or the Diabetes Control and Complication Trial (DCCT) , le d to universal acceptance of A1C testing for diagnos tic purposes. 26 The criteria for the diagnosis of diabetes include 6.5% via certified methods, 2 a with a random plasma glucose measurement of at least 200 mg/dl. 26,30 The American Diabetes Association strongly suggests t he confirmation of a diabetes diagnosis with repeat testing, to minimize the potential for laboratory error. 26,30 There ar e four clinical classes of diabetes mellitus : (1) type 1 diabetes, (2) type 2 diabetes, (3) other specific types of diabetes, and (4) gestational diabetes mellitus (GDM). 24,26,30 It is important to note that some people cannot be clearly classified as type 1 or type 2 diabet ic and some people may actually have both type 1 and 2 diabetes. 30 Type 1 diabetes, previously known as juvenile onset diabetes or insulin dependent diabetes, occurs when there is autoimmune cell destruction of the pancreas that typically results in absolute insulin deficiency. 26,30 People with type 1 diabetes usually have autoantibodies , such as islet cell autoantibodies or autoantibodies to insulin, when initially diagnosed. 26 It accounts for 5 to 10% of individuals with diabetes. 1,26 Type 2 diabetes, previously known as adult onset or non insulin dependent diabetes, occurs when people have insulin resistance due to progressive insulin secretory defect s . 26,30 Type 2 diabetics , unlike type 1 diabetics , do not need insulin treatment to survive the earlier stages of the disease . The risk of developing type 2 diabetes is associated with obesity, age , and level of physical activity . Type 2 diabetes accounts for 90 % to 95% of the disease and is frequently undiagnosed for many years

PAGE 24

24 because the symptoms of hyperglycemia develop slow ly. 1,26 Other types of diabetes may result from infections, drugs, genetic defects, or endocrine disorders. 24 Gestational diabetes (GDM) occurs in pregnant women, without a previous diagnosis of diabetes. Typically, it is diagnosed between 24 and 28 we eks gestation, using a 75 g 2 h OGTT 24 Diabetes M ellitus i n P regnancy Diabetes mellitus is one of the most common medical conditions that complicate pregnancy. 2 7 U ntil the 1980s physicians usually advised against conception in women with pre existing diabetes because of the s evere adverse outcomes related to diabetic pregnancies . T oday, diabetic pregnancies have a better prognosis due to the improved understanding of the disease and available treatment options. Despite these advances, it is still associated with perinatal and neonatal morbidity and mortality, and adverse outcomes for the mother. 11,33 Studies have shown that p regnant diabetic women are at increased risk for : preterm labor, pregnancy induced hypertension, coronary artery disease, cesarean section delivery and postpartum hemorrhage. 2,7 9,34 W ith the emergence of the diabetes epidemic, type 2 diabetes has only recently developed as the most prominent chronic condition seen in pregnancy , with most of the current available information about the impact of diabetes on pregnancy outcomes stemming from studies on type 1 or gestational diabetes. 2,3,11,18 R esearch suggest s the risk of adverse pregnancy outcomes in women with pre existing type 2 diabetes may be higher than in women with type 1 diabetes. 2,15 Furthermore, while pregnancy outcomes of type 1 diabetic women have greatly improved , limited information about the safety and efficacy of the various treatment options for type 2 diabe tes is available .

PAGE 25

25 Use of A nti diabetic A gents in P regnancy Studies have linked p oor management of diabetes in the first trimester to major birth defects, spontaneous abortions, and stillbirths. 2,13 16 Therefore, it is generally accepted by physicians that acquiring and sustaining glucose control in diabetic women prior to conception is vital . 17 Despite this well established paradigm, one study found that more than 40% of women with pre existing diabetes have difficulty managing their glycemic control before pregnancy , and more tha n 60% of women hav e maintenance issues during pregnancy. 35 38 The management of glycemic control in pregnancy is largely dependent on the type of diabetes and severit y of the disease. While type 1 diabetes management requires insulin and thus leaves little choice during pregnancy, managing type 2 diabetes in its earlier stages may include life style modifications, single or combination treatment of oral anti diabetic agents, and/or insulin. In general, the use of insulin to treat type 2 diabetes mellitus during pregnancy is accepted and recommended as safe and effective in achieving normal blood glucose levels. 39,40 Currently, i nsulin i s the only drug approved for management of diabetes during pregnancy by the Food and Drug Administration (FDA). 41,42 However, i nsulin is difficult to administer because of the requirement for multiple daily injection s . Furthermore, hypoglycemia occurs in 71% of women who use insulin during their pregnancy. 43 Formerly, oral anti diabetic agents were not recommended to be given during pregnancy. 42,44 However, t he high risk of hypoglycemia associated with insulin use and t he growing prevalence of type 2 diabetes , which can be managed with oral anti diabetic agents i n its early stages, has changed drug utilization patterns . As a result, treatment regimen s of pregnant women with pre existing or gestational diabetes now

PAGE 26

26 incorporate the use of oral anti diabetic agents . This change is embraced by many women and prescribe rs, as it is believed that oral agents increase adherence to treatment because of the ease of the treatment regime n (e.g. no injections and no refrigeration needed). 45 These advantages notwithstanding , the increase in use of oral anti diabetic agents rais es various questions related to safety and effectiveness of these agents during pregnancy. Metformin Metformin was first approved in the US in 1995 and is the only biguanide a vailable in the US market today . 46 Metformin improves insulin sensitivity by reducing insulin concentrations and fasting plasma glucose. 43,46 49 Unless contraindicated (i.e. individuals with reduced kidney function) , guidelines recommend initiation of metformin and lifestyle interventions at the time of diagnosis of type 2 diabetes . 30 There are a variety of additional positive metabolic outcomes attributed to the use of metformin , including weight reduction, improvement of lipid profile, and positive vascular effects. 30,46,49 Furthermore, i mprovements in insulin sensitivity have been shown to restore ovulation in cases of polycystic ovary syndrome (PCOS), a co morbid condition commonly seen in type 2 diabetic women. 50 Metformin is used in combination therapy with other oral anti diabetic classes or insulin or as mono therapy and is classified under pregnancy category B ( Table 2 1 ) . 51 A major concern in the use of metformin during preg nancy is its ability to cross the placent a, thereby increasing the risk of neonatal hypoglycemia. It has been hypothesize d that , since it acts as an insulin sensitizer for peripheral tissue rather than an insulin analogue, metformin is less likely to affe ct fetal and neonate metabolism. 52 Additionally , several studies have shown that materna l glycemic control play s a large

PAGE 27

27 role in neonatal hypoglycemia after delivery. 53 58 These studies did not find an increased risk of congenital anomalies in infants born to mothers treated with metformin, compare d to those treated with insulin. 53 58 There are additional concerns in the use of metformin to maintain glycemic control. Outside of pregnancy, the risk of hypoglycemia in a diabetic patient treated with metformin ranges from 0 % to over 20% when compared to other oral anti diabetic agents. 59 Although mild, the adverse risks associate d with metformin are especially important for pregnant diabetic women because varying glycemic control is associated with an increased risk of congenital anomalies. 11,60 63 A systematic review of 216 controlled trials and cohort studies noted that metformin was typically associated with more gastrointestinal side effects (i.e. flatulence, diarrhea, nausea, abdominal cramping, and vomiting) when compared to other oral anti diabetic agents. 59 T he most serious potential side effect of metformin use is lactic acidosis , which can be fatal . 64 The rate of lactic acidosis in metformin user s is reported to be between 0 9 cases per 100,000 person years. 47,65 68 Although rare, the possible occurrence of lactic acidosis still influences the treatment strategies in patients with type 2 diabetes. Several studies that have evaluated the safety and efficacy of metformin during pregnancy are noteworthy . 19,53 55 T he MiG trial was the first clinical trial to investigate the comparative safet y of metformin in pregnancy. 19 Th is trial provid ed evidence that metformin is non infer ior to insulin in relation to neonatal safety (e.g. neonatal hypoglycemia, respiratory distress, birth trauma, or preterm birth) . 19 Furthermore , a

PAGE 28

28 meta analysis of eight studies found no evidence of increased risk of birth defects when metformin was taken in the first trimester of pregnancy. 6 9 However, several studies also indicate that there are potential drawbacks to the use of metformin in pregnant women. An a nalysis of a small consecutive sample of 25 type 2 diabetic women in Demark found that women treated with metformin during the first trimester of pregnancy had an increase d prevalence of preeclampsia when compared to women treated with insulin. 70 The increase risk of preeclampsia was also observed in another Danish study of 118 women with gestational diabetes or type 2 diabetes treated with oral anti diabetic agents for varying periods of their pregnancy. 21 However, there was n o difference in pregnancy induced hypertension, placental abruption or cesarean section rates in women treated with metformin compared to insulin users . 21 Overall, s everal metformin safety studies have been completed in women with polycystic ovarian syndrome or gestational diabetes investigat ing neonatal outcomes (including congenit al malformations ) . 19,53,56,69,71 However, f ew studies include women with pre existing diabetes, or distinguish between pre existing and gestation al diabetes. 14,19 22 T here is a gap in the literature regarding obstetric outcomes ( preterm birth , cesarean section delivery, or pre eclampsia ) as primary safety endpoints. Sulfonylureas Sulfonylureas have been successfully used to treat type 2 diabetes for almost 50 years. 46 They decrease blood glucose levels by stimulating insulin secretion in the pancreas and increasing insulin sensitivity in peripheral tissues . 43,46 Second generation sulfonylureas , including glyburid e, gliclazide, glipizide, and glimepiride, entered the market in the1970s and are still commonly used today . 41,46

PAGE 29

29 Sulfonylureas were the initial anti diabetic agent s prescribed for newly diagnosed type 2 diabeti cs , but lost market share after the introduction of metformin, in parts because w eight gain is a common undesirable side effect of sulfonylureas . 46 Therefore, t he se drugs are not recommended for overweight or obese patients. Sulfonylureas can be used with other classes of anti diabetic agents, except other insulin secretagogues. 46 Research suggests that adding a second anti diabetic drug class is typically more effective than increasing the sulfonylurea dose in patients with insufficient glycemic control. 72 U sing sulfonylureas in combination with insulin can decrease insulin requirements by 40 50%, and decrease fasting blood glucose and HbA1 c . 46,73,74 S ulfonylureas should always be introduced at a low dose and blood glucose levels should be carefully monitored due to the increased risk of hypoglycemia. 43 The pregnancy categorization of s ulfonylureas is not uniform ( Table 2 1 ) . First generation sulfonylureas (tolbutamide and chlorpropamide) are contraindicated in pregnancy as they are known to cross the placenta and can adversely affect f etal metabolism. 52 Second generation sulfonylureas (glyburide) have recei ved significant attention in recent years for the treatment of gestation al diabetes. Although a defin itive conclusion has yet to be made , second generation sulfonylureas are believed to cross the placenta to a much lesser extent tha n first generation sulfonylureas. 20,45,75 S everal studies have found no harmful effects and report acceptable glycemic control with the use of second generation sulfonylureas during pregnancy. 20,76 78 Among th is class of drugs, g lyburide has been determined to be a clinically effective alternative to in sulin in women with gestational diabetes in a randomized clinical trial . 20 T his trial did not find an increase in fetal complications in women who used glyburide in the seco nd

PAGE 30

30 and third trimester s of pregnancy compared with women who used insulin . 20 However, this clinical trial may have been underpowered to prove equivalence in perinatal outcome s, especially in women with pronounced hyperglycemia. 79 A retrospective study in a large manage d care organization compared the use of glyburide with insulin for the treatment of gestational diabetes in women who were unresponsive to diet therapy. 76 This study also found that glyburide w as at least as effective as insulin in obtaining glycemic control and normal birth weights , but resulted in an increased risk of preeclampsia. 76 The authors suggest ed that this finding deserves further investigation , as this was the first report of adverse pregnancy outcomes with the use of glyburide . 22,76 No clinical trials have been conducted on glyburide use in pregnant women with type 2 diabetes. 79 There is a noticeable gap in the evidence on the safety and efficacy of sulfonylureas use i n pregnancy . 20,52,80 Thiazolidinedione (TZDs) Thiazolidinedione s (TZDs) improve whole body insulin sensit ivity thr ough the stimulation of a nuclear receptor, peroxisome proliferator activated receptor ). 45,46 Like metformin, TZDs require sufficient availability of insulin to generate a significant loweri ng of blood glucose. 46 TZDs were first introduced to the US market in 1997 with t roglit azone, which was subsequently withdrawn from the marke t in 2000 due to fatal events of idiosyncratic hepato to xicity. 46 Pioglitazone and rosiglitazone were brought to the US market in 1999 , and ar e still used today utilization has dropped significantly after it was associated with major cardiovascular events . 41,46 TZDs are typically used as second line therapy when a patient fa ils to obtain glycemic control using metformin or sulfonylurea s . 30 Th is class of drugs can be used as

PAGE 31

31 mono therapy in non obese type 2 diabetes , or in combination with variety of other oral anti diabetic agents (e.g. metformin or glimepiride ) . Combination therapy of i nsulin with TZDs is a possible treatment option ; however, caution is suggested with this regimen . TZD s are also manufactured in c ombination preparations of TZD plus metformin or glimepiride ( i . e . Avandamet ®, Avandaryl ®, Actoplus Met ®, Actopl us Met XR ®, and Duetact ® ). In general, TZDs are well tolerated , but caution is suggested in people with active liver disease or heart failure . 4 6 The literature recommends prudence in diabe tic women with PCOS because TZDs have been found to cause ovulation in women with hyperandrogenism and chronic anovulation. 46,81 The treatment of anovulation a nd improvement of obstetric outcomes has typically defaulted to metformin. 56,57,82,83 Recently, studies have shown t hat TZDs stimulate ovulation, indicating a potential to serve as an alternative treatment choice when metformin is intolerable due to gastrointestinal side effects. 46,81,84 The use of TZDs in pregnancy is considered experimental, as no controlled data are available regarding its use during pregnancy. The only available study is a case series that investigated the safety of rosiglitazone, by observing eight women with P COS who were unable to tolerate metformin. 84 The women elected to continue the use of rosiglitazone during the first 12 weeks of pregnancy. All of the pregnancies ended in term deliveries, with no congenital malformations or pregnancy complications reported. The FDA categorizes TZDs as pregnancy categ ory C ( Table 2 1 ). T he reported use and safety of TZDs during pregnancy is limited to a small number of case reports and cases series that include a negligible sample of women exposed to TZDs during pregnancy. 84 86 In 2004, the first case of inadvertent exposure in the first trimester to

PAGE 32

32 rosiglitazone was reported . Upon the discovery of pregnancy , t he type 2 diabetic woman was switched to regular insulin after the 8 th week of gestation for the duration of her pregnancy. She gave birth to a healthy baby via cesarean section delivery at 36 weeks gestation. 85 The second case occurred in a woman only exposed to rosiglitazone for the first 3 weeks of her second trimester of pregnancy. She was switch ed to regular insulin in the 17 th week of gestation and at 37 weeks gestation, gave birth to a healthy boy by cesa rean section delivery. 86 Lastly, t he third case occurred in a newly diagnosed type 2 diabetic woman (6 diabetic month s history) , who was exposed to rosiglitazone in the first 8 weeks of pregnancy. 87 She also switch ed to regular insulin when her pregnancy was diagnosed and vaginally delivered a health boy at 40 weeks gestation. 87 In 2005, researchers identified 3 1 women who planning surgical termination of pregnancy between 8 and 12 weeks gestation. 88 Eac h woman received two doses of rosiglitazone before the procedure. Next, t he fetal tissue, coe l omic fluids, and amniotic fluids were tested for the concentration of rosiglitazone. The authors concluded that the risk of placental transfer of rosiglitazone is much higher in pregnancies at 10 or more gestation weeks . 88 Rosiglitazone was detect able in only two of the amniotic fluid samples , even though it was found in 19 samples of the fetal tissue, suggesting that fetuses may have the capability to metabolize rosiglitazone. 88 There is a scarce amount of research available on the safety of TZD use during pregnancy. No studies have looked at the utilization or safety of both rosiglitazone an d pioglitazone in pre existing diabetics during pregnancy.

PAGE 33

33 Obstetric Outcomes Preterm Birth In general, births occurring prior to 37 completed weeks of gestation are considered preterm births. Since 1990, the rate of preterm birth has risen by 21%, and in 2006, the preterm birth rate was estimated at 12.8% of all US births. 89 91 Preterm birth is associated with high rates of infant morbidity including bronchopulmonary dysplasia, congenital heart disease, anemia, jaundice, hypoglycemia, and infection. 91 About 7 5% of perinatal deaths occur in preterm infants, with the majority of deaths occurring in infants born at less than 32 weeks gestation. 92 T here are three general obstetric causes of preterm birth: maternal or fetal conditions , spontaneous preterm labor, or preterm premature rupture of the membranes (PPROM). 93 Black women have higher rates of preterm births when compared to white women (16 18% v. 5 9%). 93 Other m aternal demographic characteristics associated with preterm birth include single marital status, both low and high maternal age, and low education and socioeconomic status. 93 Obese women are more likely to have an infant with a birth defect and are at increased risk of preterm delivery. 93,94 Diabetes, hypertension, or obesity frequen tly persists within pregnancies that end in a preterm delivery. Pre existing diabetes is associated with an increased risk of preterm delivery . 42 A ret rospective study conducted in singleton pregnancies found that pre existing diabetes was associated with a 19.4% incidence of preterm birth with an adjust ed odds ratio (OR) of 2.4 (95% CI: 1.9 3.0). 95 The MiG trial suggested that women with gestational diabetes who were treated with metformin had a n unadju s ted relative risk (RR) of 1.60 (95% CI: 1.02 2.52) for preterm birth. 19 In contrast, a clinical trial conducted in women

PAGE 34

34 with gestational diabetes found no statistically significant difference in risk of preterm birth in the women treated with metformin compared to glyburide. 96 Furthermore, a retrospective study performed in a subset of women treated with glyburide or insulin for gestational di abetes found no difference in risk of preterm birth in mothers exposed to glyburide. 97 Importantly, m ost of the available studies investigated the risk of preterm birth in gestational diabetic women, and it was typically a secondary outcome. 19,76,95 97 Cesarean Section Delivery For years, cesarean delivery rates have created interest as a clinical measure of q uality of care and health outcomes. This interest stems from increased maternal morbidity (e.g. thromboembolic events) and mortality, large cost, and lack of data indicating better perinatal outcomes associated with this surgical procedure. 98 101 Cesarean section deliveries are the most common major surgical procedure for women in the US. In 2009, 33.5% of the hospital stays associated with childbirth , in women aged 15 44 years old, were due to cesarean sec tion deliveries. 99 Studies have shown that the pr imary cesarean section delivery rate s varied directly with socioeconomic status, institutional type, race/ethnicity, and type of practitioner. 98,99,102 S everal o ther maternal and birth characteristics were found to be independent risk factor s for cesarean section deliveries , includ ing maternal age 35 years old ) , non Hispanic African Americans race , pari ty, and medical (e.g. anemia, renal disease, cardiac disorder, chronic diabetes, gestational diabetes, lung disease, chronic hypertension, pregnancy induced hypertension, and genital herpes) and pregnancy related complications (e.g. eclampsia, preeclampsia , breech, placental abruption, placenta previa, and hydramnios). 99,101 103

PAGE 35

35 The prevalence of cesarean section delivery is consistently higher in pre existing diabetic pregnancies than in non diabetic pregnancies. 42,99 Cesarean delivery rates are reported to be 3 to 5 times hig her in women with pre existing diabetes than in non diabetic pregnancies. 23,104 Limited information is available on the independent risks as so ciated with cesarean section deliveries in diabetic women. It is believed that risk increases with the presence of diabetes complications; however, no study has explored the independent risk of oral anti diabetic agents as a primary safety outcome. Pree clampsia Preeclampsia is defined as the new onset of hypertension and proteinuria occurring after the 20 th week of gestation. 105 Preeclampsia can be classified into early on set and late onset. Early onset is associated with more severe maternal and neonatal outcomes (e.g. fetal growth restriction (FGR), eclampsia, or cardiovascular disease). 106 On the other hand, late onset is correlated with a milder disease state and a lower risk of adverse neonatal outcomes. Regardless of onset, preeclampsia is a serious complication of pregnancy estimated to affect between 2 5% of all pregnancies and endangers the health of both the mother and infant. 105,106 Furthermore, between 1990 and 1999 the rate of preeclampsia has increased over 40%. 107 S everal known risk factors for preeclampsia includ e a history of preeclampsia, nulliparity, new paternity, age, race, obesity, multiplet pregnancy, chronic hypertension, chronic kidney disease, lupus, rheumatoid arthritis, and chronic or gestational diabetes. 107 109 A population based study in Sweden found that preeclampsia and pregnancy induced hypertension occurred 3 4 times more frequently in women with pre existing diabetes than in non diabetic women. 110 A study conducted in the United

PAGE 36

36 Kingdom followed 182 pregnancies of women with type 2 diabetes and found that preeclampsia w as two times more common in diabetic than non diabetic wome n. 23 Two small observational studies found that women treated with metformin during the first trimester of pregnancy had an increased prevalence of preeclampsia when compared to women treated with insulin. 21,70 Similar findings were reported by a cohort study that compared glyburide to insulin (12% vs. 6%, p=0.02). 76 Finally, a clinical trial conducted in women with gestational diabetes found no difference in the incidence of preeclampsia in the women treated with glyburide compared to metformin (4% vs. 2.7%, p>0.5) . 96 Overall, the small number of published studies were conducted i n small samples with limited ability to explore the association between oral anti diabetic agents used in pre existing diabetics during pregnancy and the risk of preeclampsia.

PAGE 37

37 Table 2 1 . Safety and teratogenicity classificati on for use of selected oral anti diabetic agents during pregnancy and lactation . a Drug class Mechanism of action Medication (brand name) Pregnancy class b,c Lactation Fetal exposure Biguanides R educe hepatic glucose production Metformin (Glucophage XR®, Carbophage SR®, Riomet®, Fortamet®, Glumetza®, Obimet®, Gluformin®, Dianben®, Diabex®, and Diaformin®) B Unsafe Crosses placenta Sulfonylureas Increases insulin secretion Glipizide (Glucotrol®) C Unsafe Crosses placenta Glyburide ( DiaBeta®, Glynase®, Micronase®) B Unsafe Crosses placenta Glimepiride (Amaryl®) C Unsafe Crosses placenta Thiazolidinediones Enhances insulin sensitivity Rosiglitazone (Avandia®) C Safety unknown Crosses placenta Pioglitazone (Actos®) C Safety unknown Unknown a Data from the table are composed from the FDA drug Reference. b C ategory B: Animal studies have indicated no evidence of harm to the fetus, but there are no adequate and well controlled studies in pregnant women. Or a nimal studies have shown an adverse effect, but adequate and well controlled studies in pregnant women have failed to demonstrate a risk to the fetus. 51 c Category C: Animal studies have indicated an adverse effect and there are no adequate and well controlled studies in pregnant women. O r n o animal studies have been conducted and there are no adequate and well controlled studies in pregnant women. 51

PAGE 38

38 CHAPTER 3 METHODS There are four parts of our analysis. In part I, we aim ed to establish and validate an algorithm with the Medicaid database to lin k mothers to infants. In part II, we examined the prevalence of pre existing diabetes in pregnant women (overall and secular trends). In part III, we determined the anti diabetic utilization drug pattern before and during pregnancy in women with live birth s . In t he final part of this analysis (part IV), we evaluated the safety of the most commonly utilized drug classes of anti diabetic agents in women with pre existing diabetes in the first trimester of pregnancy in terms of risk of obstetric outcomes. Dat a Sources This study utilize d two data sources . The first include d Medicaid administrative claims data that was supplied by the Centers for Medicare and Medicaid Services (CMS), in the form of the Medicaid Analytic eXtract (MAX) database for 29 US states from 1999 to 20 06 ( Figure 3 1 ) . The second data source include d Vital Birth Certificate Records (BCR) f r o m Florida , which was one of the states for which MAX data was available . MAX data contains enrollment information and the final action claims for all Medicaid beneficiaries. This include d eligibility, inpatient claims, outpatient claims, pharmacy claims, long term care claims, and the date of death auxiliary file. Th e patient summary file contains patient demographic and eligibility information, collected for every beneficiary enrolled for at least one day during the year. MAX also supplie d SSN and Medicaid ID numbers, which allow ed linkage to the BCR data . The BCR pr ovide d clinical parameters concerning the birth, such as gestational age and a comprehensive medical and social history obtained through parent self report.

PAGE 39

39 Medicaid Analy tic eXtract (MAX) The Medicaid Analytic eXtract (MAX) data used in this study include d Medicaid fee for service (FFS) data from 29 US states for 1999 20 06 : Alabama, Arkansas , Florida, Georgia, Iowa, Idaho, Illino is, Indiana, Kansas, Louisiana, Massachusetts, Maryland, Minnesota, Missouri, Mississippi, North Carolina, Nebraska, New, Hampshi re, New Jersey, New York, Ohio, Pennsylvania, South Carolina, Tennessee, Texas, Virginia, Vermont, Wisconsin, and West Virginia. These s tates were included because of low managed care penetration and high Medicaid enrollment numbers, as reported by the C enters for Medicare and Medicaid Services (CMS). We restricted the analysis to beneficiaries who received care through for fee for service ( FFS ) or primary care case management programs . This restriction assures comprehensive access to all medical claims t hat were reimbursed by Medicaid. Furthermore, FFS data went quality and validity checks in MAX data, whereas encounter data (as submitted by providers under capitated arrangements) did not. 111 Critic ism surrounding the use of Medicaid data in epidemiologic research includes the limited generalizability to higher income individuals and the limited clinical and socio demographic information (as available in medical charts or through prospective data asc ertainment) , which can result in unmeasured confounding . However, Medicaid , as a health care utilization database , offers a number of advantages for studies of medications in pregnancy when compared to registries and prospective cohort studies. 112,113 First, Medicaid data facilitate d establishing a large population based cohort, allowing for the study of rare outcomes , which is particularly relevant for many maternal and neonatal safety or effe ctiveness studies . MAX data provide d detail of all pharmacy dispensed and Medicaid reimbursed medications , which

PAGE 40

40 minimize d exposure classification errors and resulting biases , such as recall bias through parental self report . In addition, MAX data are available at relatively low cost compared to the expense associated with prospective data collection. Medicaid covers 12% of US adult women of reproductive age ( Figure 3 2 ) . 114,115 It is the largest payer of maternity related health services, covering 4 in 10 births across the US, and in several states paying for more than half of total births ( Figure 3 3 ) . 116 118 Pregnancy related and neonatal hospitalizations make up almost 50% of all hospitalizations covered by Medicaid . 119 Depending on state, i ncome eligibility limits for pregnant women in Medicaid range from 133% of the federal poverty level (FPL) to 300% of FPL. 90,115,116 Forty two states and Washington DC provide continuous coverage during the course of pregnancy, including the first 60 days post partum. 116 Internal V alidity Medicaid prescription drug claims are consistent over time , complete , and have a high proportion of valid National Drug Codes (NDC) (used in the US to identify drug, strength and dose of dispens ed medications in claims data). 120 It has also been concluded that gross miscoding in MAX is rare , and occur s less often in in patient claims than in out patient claims. 120 Hennessy et al. (2003 ) examined several general categories for potential data errors including: the completeness of claims (inpatient, outpatient, prescription drug) for certain time periods, incomplete hospitalization data for beneficiaries over 65 years old, and diagnostic c odes. 121 The authors identified a varying degree of completeness for in patient and prescriptions claims . L imited gaps in out patient claims suggest ed satisfactory complet eness , however the authors caution that the suggestion of completeness in out patient claims requires further examination

PAGE 41

41 because there was no external reference for the data . 121 Furthermore, t he authors highlight ed the need for macro level descriptive analyses when using Medicaid data to assure data quality. 121 D iagnostic codes used in in and out patient claims may contain a number of errors that can influence code accuracy. 122 Identifying potential sources of cod ing bias and validating the codes can strengthen the interpretations and conclusions made by researchers. Oftentimes, researchers rely on multiple diagnos e s or procedur e codes to operationalize an outcome or exposure . A number of studies have validated diagnos e s claims based algorithms for several different disease states. Pertinent to this study, t he proposed algorithm to identify d iabetes mellitus was validated in several administrative databases, including Medicare and the HMO Research Network ( sensitivity: 50.1 74.4%, specificity: 99.3 99.4%, PPV: 91% 92.9% ) . 123 125 In addition, studies in administrative claims databases have validated some of the o utcomes employed in this study (i.e. preterm birth and preeclampsia ) . 126 128 In a validation study of hospital discharge codes, the PPV for specific forms of preeclampsia and eclampsia ranged from 41.7% to 84.8%. 128 Andrade et al. found that the overall PPV for preterm birth was 87% (ICD 9 CM codes: 644.21,765.0 765.28) . 126 Margulis et al. developed and validated algorithms to estimate the gestational age at birth based on ICD 9 CM codes identified in administrative claims. 129 We modified this algorithm, to use only maternal claims, and validated an algorithm to identify premature delivery (sensitivity: 4 6 . 0 %, specificity: 9 7 . 0 % and PPV 72. 8 % ) in the Florida and Texas Vital Birth Certificates and MAX databases (Appendi x : Preterm) . Our previous research also

PAGE 42

42 validated an algorithm to identify cesarean section deliveries (sensitivity: 54.3 %, specificity: 99.5 % and PPV 98.1 %) in the MAX database (Appendix: Cesarean Section). Vital Birth Certificate Records The Vital Birth Certificate Records (BCR) dataset used in this work represent s all births in the state of Florida from 1999 until 20 04 . State law requires the creation of BCRs for all births. Each state health department collects BCR s separately . Birth data is composed of all birth s occurring in the United States for both residents and nonresidents. BCR s include demographic characteristics of the infant, mother, and father (e.g. age, race, educ ation), as well as clinical characteristics of the mother and infant (e.g. APGAR score, weight gained during pregnancy, co morbid conditions). From 1989 to 2002, the BCR identified the clinical estimation of gestational age as the length of gestation esti mated by the attendant and not calculated from last menstrual period (LMP) . 130 In 2003, the obstetric esti mation of gestation was add to the all perinatal factors and assessments (e.g. ultrasound). The clinical or obstetric estimation of gestational age provides improved accuracy relative to the L MP calculation of gestational age. 131,132 Source P opulation Our analysis employ ed a retrospective cohort of pregnant women aged 12 55 years old. In part one , we identified all live births that occurred in the Florida MAX and Florida BCR database from 1999 20 04 . The unit of analysis was pregnancies, and the number of birth certificates and live birth claims identified within the cohort determined the count of pregnancies . In the second part of this analysis, w omen were included if they had a pregnancy claim within MAX ( Appendix Pregnancy ) and had 12 months

PAGE 43

43 continuous eligibility before their index date , which was defined as the c alculated start of pregnancy . Within the pregnancy cohort, we identified the pre existing diabetic pregnancy cohort as women with 1 in or 2 outpatient claims with ICD 9 CM code 250.XX before pregnancy . In part 3 of this analysis, we employ ed a fixed retrospective cohort of w omen with pre existing diabetes who maintain ed 12 month s of continuous eligibility before thei r index date ( calculated start of pregnancy ) until delivery . In p art 4 of this analysis, we again employed a fixed retrospective cohort of women with pre existing diabetes who maintained 6 months of continuous eligibility before their index date until delivery. We identified the delivery date from the MAX database ( Appendix Delivery ) in women with a live birth claim . Note that the prevalence estimates for the 1999 calendar year were not reported (in parts 2 and 3) because eligibly information preceding 1999 was not available. Source P opulation M easures We identified the delivery date from the first claim for live birth for each woman from MAX during the study period . We calculated p regnancy trimesters as 1 3 week segments starting from the index date (0 1 3 weeks, 1 4 2 7 weeks, and 2 8 weeks to birth ). B aseline characteristics of the cohort were determined by analyzing the 12 months (or 6 months for Part 4) preceding the index date , as was the pres ence of pre existing diabetes. The latter was defined as presence of 1 in or 2 out patient claims for diabetes ( ICD9 CM 250.xx) during the 12 month s baseline period ( pre conception period ) that were at least 1 day apart . 123 125 The National Drug Code (NDC) , which identifies unique medications in the MAX prescription drug claims , was cross walked to generic names using the Multum Drug Product Information Database ( Multum, Denver, Co ) .

PAGE 44

44 The following methods section addresses methodological detail for each of the four study parts. The institutional review and privacy boards (IRB) of the University of Florida, CMS, Florida Department of Health, and Texas Department o f Health approved this study . All the data analyses used SAS 9. 4 (SAS Institute, Cary, NC) and Microsoft Excel 2010 (Microsoft Corp., Redmond, WA). Part I: Validation of Mother Infant Linkage Using the Medicaid Case ID Number Variable within the Medicaid A nalytic eXtract Database Sample P opulation W e identified all live births that occurred in the Florida MAX and Florida BCR database. Births were included in this analysis study period. W e required a minimum of one month MAX eligibility at any point for infants. M others were required to have a minimum of one month MAX eligibility with at least one delivery claim . Women could contribute multiple pregnancies throughout the study period , but each delivery had to occ ur at least 24 weeks apart . If the multiple pregnancies were not 24 weeks apart, the second pregnancy was excluded from the analysis . We were only able to obtain infant s and mothers SSNs in BCR through re use of a pre existing data user agreement with th e Florida Department of Health . Therefore , this validation study was restricted to only Florida MAX and Florida BCR data from 1999 to 2004. Data A nalysis Mother infant link In Medicaid, it is possible to identify family units u sing the state assigned Medi caid Case ID number, which not only include s mothers and infants , but possibly other family members eligible under the same criteria. To establish unique mother infant

PAGE 45

45 linkages via Medicaid Case ID numbers, later called the Medicaid Case ID number algorithm , we required the mother infant pairs to have matching Medicaid Case ID to fall within the delivery date range. Since women could have multiple delivery claims for one delivery, and the delivery claim might not necessarily have one specific date associated with it, it was necessary to identify delivery ranges for inpatient and outpatient deliveries. The in patient delivery discharge date and the out service begin date and the service end date. The literature used a delivery range +/ 5 days of the maternal start and end dates of the delivery claim. 133 We explored the impact of modifying the range from +/ 5 days to 30 days of the delivery claim. We excluded all infants who were linked to more than one mother based on the Medicaid Case ID number algorithm . To establis h the validity of the mother infant link using the Medicaid Case ID number, we create d a mother infant linkage (gold standard) within the BCR, which was subsequently linked to MAX. To create the BCR linked cohort, we first linked all infants in MAX to thei r respective BCR via the ir SSN and DoB . We then identified the mothers by extracting the corresponding mothers from the link ed infant s . To identify mothers in MAX , we used t he mothers SSN s and DoB found in the BCR for linkage to the MAX data base . We estimated the annual proportion of live births detected in MAX claims by calculating the number of live births found in MAX, and dividing it by the total number of births identified in the BCR by delivery year. Then, we calculated the propor tion of

PAGE 46

46 mother infant pairs identified in the BCR that were linked to MAX via SSN and DoB . The calculation was repeated to identify the proportion of MAX births that could be linked to BCR via only the . The number of unique mothers who were linked to unique infants was determined after applying the Medicaid Case ID number algorithm (Medicaid Case ID number, DoB of the infant, and delivery date range as identified from the maternal delivery claim in MAX) . Validation of Medicaid Case ID number linkage For validity testing of th e mother infant linkage, we addressed two concepts separately: Is the Medicaid Case ID n umber able to correctly link mothers to infants in MAX? Which Medicaid eligibility window (m other and/or infant) increases the sensitivity and specify for accurate mother infant link s ? Using the BCR linked cohort, we determined the sensitivity of the Medicaid C ase ID number algorithm linkage as the proportion of correctly linked mother infant pairs (i.e. pairs linked by Medicaid Case ID algorithm and linked in BCR) over the total mother infant pairs identified in the BCR . S pecificity was calculated as the proportion of correctly non linked infants by the Medicaid Case ID n umber algorithm and BC R over the total mother infant pairs that should not be linked according to the BCR. P ositive predictive values (PPVs) were identified as the proportion of correctly linked mother infant pairs (i.e. pairs linked by Medicaid Case ID number algorithm and lin ked in BCR) divided by the total number of mother infant pairs linked via Medicaid Case ID number algorithm ( Table 3 1 ) .

PAGE 47

47 To assess the impact of MAX eligibility on t he sensitivity and specificity of linking mothers to infants by the Medicaid Case ID n umber algorithm , we varied the required continuous eligibility window in MAX of the mother to include only 3 months, 6 months, 9 months, 12 months , and 1 5 months immediately prior to the month of delivery. We then varied the continuous eligibility window in MAX for only the infant to include a minimum of 3 months, 6 months, 9 months, and 12 months subsequently following the birth month of the infant . Lastly , variou s combinations of continuous eligibility windows for both the mother and infant pair were tested . Predicting linkage within a validated algorithm that identifies mother infant pairs within the Medicaid Analytic eXtract (MAX) database . W e used a multivariat e logistic regression model to identify factors influencing true mother infant linkage in the MAX data when using the validated Medicaid Case ID number algorithm (equation 3 1) . Logit [P (Y=1)] = + 1 X 1 + 2 X 2 + 1 6 X 1 6 (3 1) Where: Y: Probability of accurate mother infant linkage via Medicaid Case ID Number X 1 : Calendar year of delivery (1999 2004) X 2 : Region (North, Northwest, Central, Southwest, Southeast) X 3 : Length of delivery hospitalization (days) X 4 : Type of delivery claim (in patient/out patie nt) X 5 : Maternal age (under 20, 20 29, 30 39, 40+) X 6 : Maternal race/ethnicity (White, Black, Asian, Hispanic, Unknown) X 7 : Maternal education (
PAGE 48

48 The model created (t 1) dummy variables (where t = the number of levels) for each categorical variable . We ch ose the predictor variables for this model based on prior experience using MAX data. For example, we assumed that the reporting of the Medicaid Case ID number would improve as the study years progressed. A lso , p reterm infants are quicker to receive their own Medicaid ID because they incur hospital costs sooner and to a larger extent than term infants do . Term infants are more likely to have their birth cost absorbed by their . Calendar year of delivery was defined from either the delivery date range, or if MAX patient summary file. Region ( Table B 1 ), type of delivery claim, maternal age, maternal race/ethnicity, maternal MAX eligibility, maternal eligibility for case assistance, maternal poverty status, infant gender, infant r ace/ethnicity, and infant MAX eligibility were all identified and defined from the MAX patient summary file . We ascertained the m aternal education level from the FL BCR. Sibling status was defined from the FL BCR live birth living variable and number of pr enatal visits was categorized from the FL BCR prenatal total visits variable. We created the preterm birth and infant birth weight variable s, >37 weeks gestation for preterm birth, <2500g birth weight, respectively, as identified from the FL BCR. Our model used forward stepwise selection with an alpha error e ntry criterion of 0. 2 0 and identified the predicted model with the best net fit using the Bayesian Information Criterion (BIC). We evaluate d the predicted

PAGE 49

49 between l inked and non linked mother infant pairs by us ing the area under the curve (AUC) of the receiver operator curves with 95% confidence intervals. 134,135 Part II: Epidemiology of Pre existing Diabetes in Pregnancy from 2000 2006 among Medicaid Patients in 29 US States Study D esign We conducted a cross sectional analysis of women aged 12 55 years to estimate the prevalence of pregnancy in conjunction with pre existing diabet es mellitus for seven consecutive years. We followed women fr om 12 months before their index date (calculated start of pregnancy ) until the end of their pregnancy ( e . g . spontaneous abortion, abortion, stillbirth , live birth , or censored ) . Note that the prevalence estimates for the 1999 calendar year are not repor ted because eligibly information preceding 1999 was not available. Sample P opulation Women were eligible for the pregnancy cohort if they ha d 2 ICD 9 CM codes for pregnancy or 1 ICD 9 CM codes for delivery in MAX ( Appendix Pregnancy ) . We required women to have at least 12 months continuous Medicaid eligibility before their index date ( beginning of pregnancy ) and continuous Medicaid eligibility until the end of pregnancy ( Figure 3 4 ) . Without the requirement of live birth, it was necessary to use a new algorithm to calculate the start of pregnancy. 129,136,137 Toh et al. validated a pregnancy indicat or algorithm, which uses the first claim indicating pregnancy as the conception date. 136,137 Wit hin our linked cohort of FL BCR data and FL MAX, we saw that the majority of pregnancies within MAX did not have a pregnancy code within the first trimester of pregnancy. Thus, it was inappropriate for us to use this algorithm to calculate the start

PAGE 50

50 of pr egnancy for our exploratory analysis, as we would misclassify the 1 st trimester. Thus, we u sed the delivery date algorithm to calculate the index date based on the end of the pregnancy claim. If a woman had a delivery claim, the index date was set at 27 0 d ays before the delivery claim. If there was a delivery claim and a claim for preterm birth , the index date was 245 days before delivery. For women identified as having a stillbirth or abortion, the end of pregnancy was calculated based on average length of pregnancies according to the literature (i.e. 1 96 days for stillbirth and 105 days for abortion). 138 142 If a woman did not have a termination of pregnancy claim ( e.g. no claim for delivery or other terminal pregnancy outcome ) ; we used the date of the first pregnancy claim to identify the index date. Pre existing diabetes was identified based on 1 in or 2 outpatient claims at least 1 day apart with ICD 9 CM 250.X X during the 12 month s pre index look back period . P revious research states that curre nt ICD 9CM coding practices do not provide sufficient accuracy to identify the type of diabetes from solely the fifth digit of the ICD 9 CM code . 122,143 Therefore, we conducted a stratified analysis to identify women with a high propensity to have type 2 diabe tes. 144 W e as sumed the presence of type 2 diabetes if 100% of diabetes claims in the baseline period indicated type 2 diabetes (ICD 9 CM 250.x0 or 250.x2). 144 Accordingly, t ype 1 diabetes was assumed if 100% of the diabetes claims in the baseline period indicated type1 diabetes, an d t mixed i.e. , missing the 5 th digit). Data A nalysis For each state, we estimated the annual prevalence and 95% confidence intervals (CIs) of pre existing diabetes in pregnancy by calculating the number of

PAGE 51

51 pregnancies aff ected by pre existing diabetes divided by the total number of pregnancies in a given calendar year. We describe d the demographic ( e.g . age, race/ethnicity , and state ) and clinical characteristics (i.e. delivery outc omes [ abortion (elective and spontaneous) or stillbirth ] and co morbid conditions ( Table A 1 ) [ e.g. polycystic ovary syndrome (P COS), hypertension, and obesity] ) in a ll pregnancies eligible for Medicaid in 29 states . We modeled the secular trends (P trends ) of pre existing diabetes in pregnancy prevalence using linear regression . We calculated the age standardized and the age and race/ethnicity standardized annual prevalence of pre existing diabetes using the direct standardization method. 145 We used the distribution of age and race/ethnicity of the entire pregnancy cohort as the standard population for the standardized annual prevalence estimates. The a nnual trend was modeled by fitting year as a continuous variable in a linear regression model using the empirical sa ndwich estimator , adjusting for race/ethnicity and/or age where appropriate. We considered p values <0.05 indicative of a significant trend. Part III: Utilization of A nti diabetic A gents in P regnancy from 2000 20 06 a mong Medicaid P atients in 5 US S tates Study D esign We conducted a cross sectional analysis of women aged 12 55 years old to estimate anti diabetic agent use throughout pregnancy in pre existing diabetic women for seven consecutive years. In addition, we examined variations in anti diabetic age nt use by patient characteristics and state. Women were required to have continuous Medicaid eligibility 12 months before their index date (calculated start of pregnancy ) until their MAX claim for live birth . W omen who contribute d multiple pregnancies were included , as long as they met continuous eligibility requirements (12 mo nths before their

PAGE 52

52 index date until delivery ) and their baseline period (12 months immediately preceding their index date ) did not overlap with a prior pregnancy (i.e. include a pregnancy or delivery claim) . If multiple pregnancies overlapped, (e.g. second pregnancy baseline period overlapped first pregnancy period), then the second pregnancy was excluded from the analysis. Note that the prevalence estimates for t he 1999 calendar year are not reported because eligib ilit y information preceding 1999 was not available. Sample population Women were eligible for inclusion into this cohort if they ha d a claim for live birth ( Appendix Delivery ), continuous Medicaid eligib ility from 12 months before pregnancy until delivery , and were identified has having pre existing diabetes ( Figure 3 5 ) . We calculated the index date (calculated start of pregnancy ) by subtracting 27 0 days from the identified delivery date. 129,137 If we identified a preterm delivery, we subtracted 245 days to identify the index date. 129,137 T he baseline period w as identified as 12 months immediately preceding the calculated start of pregnancy (index date) . Pre existing diabetes was defined as the presen ce of 1 in or 2 outpatient claims at least 1 day apart for diabetes ( ICD9 CM 250.xx) during the 12 month s baseline period. 123 125 We categorized diabetes type as previously described in part II of this analysis . 144 A delivery was assumed to be a ffected by type 2 diabet es if 100% of the diabetes claims , in the baseline period , were identified as type 2 diabe tes ( i.e. ICD 9 CM 250.x0 or 250.x2). 144 Type 1 diabetes was assumed if 100% of the diabetes claims in the baseline period indicated type1 diabetes, and t mixed included all women with mixed or unspecific claims ( i.e. , missing the 5 th digit).

PAGE 53

53 Exposure definitions We identified exposure windows ( pregnancy periods ) , as 3 months preceding the index date (or pre conception period), first trimester, second trimester, third trimester. The unit of analysis was delivery, i.e. w omen were allowed to cont ribute multiple pregnancies throughout the study period. For annual drug utilization prevalence estimates we considered each drug class, regardless of mono or combination therapy. Thus, deliveries could contribute to multiple exposures categories . We also classif ied overall deliveries into oral only treatment, insulin only treatment, mixed treatment (i.e. oral and insulin), and no treatment. Exposure drug class categories included metformin hydrochloride , sulfonylureas (i.e. glimepiride, glyburide, glipizide, and micronized glyburide) , thiazolidinedione s (i.e pioglitazone hydrochloride and rosiglitazone maleate) , single combination pill ( i . e . metformin hydrochloride pioglitazone hydrochloride, metformin hydrochloride ro siglitazone maleate, glimepiride pioglitazone hydrochloride, glimepiride rosiglitazone maleate, glipizide metformin hydrochloride, and glyburide metformin hydrochloride ), and insulin . We also examined if there was any anti diabetic therapy (e.g. exposure to any of the specified anti diabetic drug classes) , oral anti diabetic therapy only, and no therapy. We defined t he period of drug exposure starting with the date the anti diabetic agent prescription was dispensed (dispensing date) and using the total da the dispensing . For each pregnancy period, we then classified exposure based on a majority rule (i.e. total supply accounted for at least 45 days of the respective pregnancy period ), except for the insulin classification. The supply variable was unreliable with insulin, due to the varying dose requirements associated with its use.

PAGE 54

54 Therefore, we categorized a period as exposed to insulin if there was at least one claim that occurred in the identified period. I nitiation of an a nti diabetic agent was assumed if the pre conception period did not hav e pharmacy claims for any anti diabetic agent s and the first trimester had claims for an oral anti diabetic agent or insulin . Data A nalysis We first identified the number of deliveries affected by pre existing diabet es and allocated their follow up time in to the respective pregnancy period s . We then measured the prevalence and 95% CIs of anti diabetic agent utilization (overall and by the previously described categories) for each pregnancy period . We calculated prevalence of anti diabetic agent s use d as the proportion of deliveries with pre existing diabetes with a t least 45 days of drug coverage , as identified by supply , divided by the total number of deliveries with pre exi sting diabetes , for each pregnancy period. S ecular trends (P trends ) in drug utilization were estimated with using linear regression. We then evaluated patient characteristics stratified by diabetes type . We used univariable generalized estimating equations (GEE) models to identify characteristics associated with initiation of anti diabetic drug treatment compared to those who remain untreated in t he first trimester of pregnancy . We included the v ariables from the univariable models with a p value <0. 3 0 in the multivariate prediction model , and those variables with a p value <0.10 in the final adjusted model (equation 3 2) . We fit ted a model with GEE to account for clustering of multiple pregnancies within the same woman. Log it [P (Y=1)] = + 1 X 1 + 2 X 2 + 3 X 3 + 4 X 4 + 5 X 5 + 6 X 6 11 X 11 (3 2) Where:

PAGE 55

55 Y: Probabi lity of initiation of anti diabetic agent use in the first trimester of pregnancy . X 1 : Maternal age at baseline (under 20, 20 29, 30 39, 40+) X 2 : Maternal race/ethnicity (White, Black, Native American, Asian, Hispanic, Unknown) X 3 : Region (Northeast, Midwest, South, West) X 4 : Calendar year of index date (2000 20 06 ) X 5 : Use of in vitro fertilization during baseline (yes/no) X 6 : Alcohol use during baseline (yes/no) X 7 : Cigarette smoking during baseline (yes/no) X 8 : Illicit drug use during baseline (yes/no) X 9 : Obese during baseline (yes/no) X 10 : Chronic hypertension during baseline (yes/no) X 11 : Polycystic ovary syndrome (PCOS) during baseline (yes/no) The model created (t 1) dummy variables (where t = the number of levels) for each categorical variable . We choose the determinant variables for this model based on variables identified from the literature that may indicate the need for pharmaceutical interve ntion (e.g. additional comorbidities). Calendar year of index date was defined from the calculated start of pregnancy (index date) as identified from the MAX in or out patient claims . Region ( Table B 2 ), maternal age, and maternal race/ethnicity were all ascertained from the MAX patient summary file. We identified the comorbid conditions ( Table A 1 ) based or out patient claims during the 12 month baseline period . Part IV: Comparative S afety of O ral A nti diabetic A gents in P regnancy from 2000 20 06 among Medicaid P atients in 29 US S tates Study D esign We conducted a retrospective cohort study of women with pre existing diabet es aged 12 55 years old . Our objective was to estimate the risk of obstetric outcomes (i.e. cesarean section delivery , preterm birth , or preeclampsia ) in women exposed to various oral anti diabetic agent s in the first trimester of pregnancy . Women were required to have continuous Medicaid eligibility 6 months before their index date until delivery. Again, w omen were allowed to contribute multiple pregnancies, as long as the y retained

PAGE 56

56 continuous Medicaid eligibility with 6 months before their calculated start of pregnancy (index date) u ntil their claim for live birth and their baseline period ( 6 months immediately preceding their index date ) did not overlap with a prior pregn ancy (i.e. include a pregnancy or delivery claim). Sample population Women were eligible for this cohort if they ha d a claim for live birth ( Appendix Delivery ), continuous Medicaid eligibility as defined above , and were identified as having pre existing diabetes . W e required use of an oral anti diabetic agent, as either mono therapy or dual therapy, during the first trimester of pregnancy to be included in this analysis . We defined use of an oral anti diabetic agent as before, based on the estimated days as having drug supply for at least 45 days in the first trimester. To approximate similar disease severity among the diabetic women, we restricted study inclusion to women who did not have a pharmacy claim for insul in in the ir pre conception period. Exposure definitions Our analysis focused on the safety of the three most commonly utilized oral anti diabetic drug classes in women with pre existing diabetes: biguanides (metformin), sulfonylureas, and thiazolidinedion es. The exposure categories include d mono therapy only (sulfonylureas v. metformin or thiazolidinediones v. metformin) and dual therapy exposure (sulfonylureas and metformin v. thiazolidinediones and metformin). In an attempt to explore the non inferiority of metformin as compared to insulin, we created a third mono therapy group in which we compared women who were on only oral anti diabetic drugs in the baseline period and switch ed to insulin only in the first trimester . We defined the period of exposure to an anti diabetic agent for each woman by the date

PAGE 57

57 prescription. Dual therapy was defined as having 45 days or more of two target anti d iabetic agents overlap in the first trimester of pregnancy , or 45 days or more of a single combination product . The exposure window was restricted to the first trimester of pregnancy (0 13 weeks gestation). We focus ed on the first trimester of pregnancy for several reasons. Most importantly, r esearch has shown that i t is a critical period for fetal development and the determination of obstetric outcomes . 146 148 In addition, pregnancy related increasing insulin resistance typically does not present in the first trimester and therefore does not affect glycemic control. This allowed us to assume that changes in glycemic control in the first trimester did not affect the choice of the anti diabetic drug , t hus limiting the impact of time dependent confounding . Furthermore, it was important to set the exposure window before the diagnosis of the outcomes (cesarean sec tion, preterm birth, or preeclampsia), to avoid potential reverse causation. Outcome assessment The outcomes of interest were the following obstetric outcomes : cesarean section delivery , preterm birth , and preeclampsia . We used the MAX claims to identify cesarean section delivery ( Appendix Cesarean section delivery) and p reterm birth ( Appendix Preterm Birth) . Preeclampsia was defined by the presence of at least 1 in or out patient claim (Appendix Preeclampsia) occur ring between gestation week 2 1 and deliv ery. 105,128 Active comparison group We chose an active comparison group (metformin either as mono therapy or in combination) for several reasons . First, by choosing an active comparator, we

PAGE 58

58 minimize d confounding by indication, because all subjects ha d pre existing diabetes. Second , by restricting to comparable treatment regimens with an identical number of anti diabetic agents (e.g. , mono therapy to mono therapy) we reduced confounding by severity . Finally, metformin was consistently the most highly utilized oral anti diabetic agent during pregnancy in previous research , supporting its position as reference drug . 149,150 Data A nalysis I ntention to treat (ITT) analysis I ntention to treat (ITT) analysis was employed posure to anti diabetic agents during the first trimester to alleviate the issue of time dependent confounding , as diabetic treatment regimen according to changing glucose management requirements throughou t pregnancy . ITT maintains the original balance generated by the use of propensity scores. P ropensity scores A propensity score is a number between 0 and 1 that represents the conditional probability that a person is expose d to a certain group, given a select number of variables that have been identified as potential confounders. Propensity scores offer a method of matching on multiple confounding variables, without the limitations of covariate matching methods (i.e. restriction of the number of inclu ded confounders due to sample size limitations). P ropensity scores for each exposure class were separately calculated for both mono and dual therapy comparisons , us ed generalized estimating equations (GEE ) regression analysis to account for clustering o f multiple pregnancies within the same

PAGE 59

59 woman . We identified the dependent variable as a binary variable representing the two targeted exposure classes (e.g. mono therapy with metformin versus mono therapy with sulfonylureas ) . P redictor variables identified during the baseline period included maternal age, race/ethnicity, pregnancy year, and a number of co morbidities ( equation s 3 3 , 3 4, 3 5 ) . For each obstetric outcome model we modified the selection of covariates for the propensity score models to reflec t specific risk factors for the respective outcome (i.e. equation 3 3 was for premature delivery, 3 4 was for cesarean section delivery, and 3 5 was for preeclampsia). 91,99,101,104,105,151,152 Logit [P (Y=1)] = + 1 X 1 + 2 X 2 + 3 X 3 + 4 X 4 + 5 X 5 + 6 X 6 2 5 X 2 5 (3 3 ) Where: Y: Sulfonylureas only versus metformin only (Model 1), thiazolidinediones only versus metformin only (Model 2), insulin only versus metformin only (model 3), metformin and thiazolidinediones versus to metformin & sulfonylureas (Model 4 ) X 1 : Maternal age at baseline (<20, 20 29, 30 39, 40+) X 2 : Maternal race/ethnicity (White, Black, Native American, Asian, Hispanic, Unknown) X 3 : Region (Northeast, Midwest, South, West) X 4 : Calendar year of calculated LMP (2000 20 06 ) X 5 : Alcohol use during baseline X 6 : Cigarette smoking during baseline X 7 : Illicit drug use during baseline X 8 : Chronic hypertension during baseline X 9 : Obese during baseline X 10 : Polycystic ovary syndrome (PCOS) during baseline X 11 : VTE during baseline X 12 : Chronic kidney disease during baseline X 13 : Diabetic nephropathy during baseline X 14 : Claims indicating m ild uncontrolled diabetes during baseline X 15 : Claims indi cating s evere uncontrolled diabetes during baseline X 16 : Anemia during baseline X 17 : Depression during baseline X 18 : Anti depressant exposure during baseline X 19 : Statin exposure during baseline X 20 : Beta blocker exposure during baseline X 21 : Alpha agonist exposure during baseline

PAGE 60

60 X 22 : ACE inhibitor exposure during baseline X 23 : ARB exposure during baseline X 24 : Ca lcium channel blocker exposure during baseline X 25 : Diuretic exposure during baseline The model created (t 1) dummy variables (wher e t = the number of levels) for each categorical variable . Logit [P (Y=1)] = + 1 X 1 + 2 X 2 + 3 X 3 + 4 X 4 + 5 X 5 + 6 X 6 2 6 X 2 6 (3 4) Where: Y: Sulfonylureas only versus metformin only (Model 1), thiazolidinediones only versus metformin only (Model 2), insulin only versus metformin only (model 3), metformin and thiazolidinediones versus to metformin & sulfonylureas (Model 4) X 1 : Maternal age at baseline (<20, 20 29, 30 39, 40+) X 2 : Maternal race/ethnicity (White, Black, Native American, Asian, Hispanic, Unknown) X 3 : Region (Northeast, Midwest, South, West) X 4 : Calendar year of calculated LMP (2000 2006) X 5 : Alcohol use during baseline X 6 : Cigarette smoking during baseline X 7 : Illicit drug use during baseline X 8 : Chronic hyperte nsion during baseline X 9 : Obese during baseline X 10 : Polycystic ovary syndrome (PCOS) during baseline X 11 : VTE during baseline X 12 : Chronic kidney disease during baseline X 13 : Diabetic nephropathy during baseline X 14 : Claims indicating m ild uncontrolled diabetes during baseline X 15 : Claims indicating s evere uncontrolled diabetes during baseline X 16 : Anemia during baseline X 17 : Depression during baseline X 18 : Herpes simplex virus during baseline X 1 9 : Anti depressant exposure during baseline X 20 : Statin exp osure during baseline X 2 1 : Beta blocker exposure during baseline X 2 2 : Alpha agonist exposure during baseline X 2 3 : ACE inhibitor exposure during baseline X 2 4 : ARB exposure during baseline X 2 5 : Calcium channel blocker exposure during baseline X 2 6 : Diuretic exposure during baseline The model created (t 1) dummy variables (where t = the number of levels) for each categorical variable .

PAGE 61

61 Logit [P (Y=1)] = + 1 X 1 + 2 X 2 + 3 X 3 + 4 X 4 + 5 X 5 + 6 X 6 2 6 X 2 6 (3 5) Where: Y: Sulfonylureas only versus metformin only (Model 1), thiazolidinediones only versus metformin only (Model 2), insulin only versus metformin only (model 3), metformin and thiazolidinediones versus to metformin & sulfonylureas (Model 4) X 1 : Maternal age at baseline (<20, 20 29, 30 39, 40+) X 2 : Maternal race/ethnicity (White, Black, Native American, Asian, Hispanic, Unknown) X 3 : Region (Northeast, Midwest, South, West) X 4 : Calendar year of calculated LMP (2000 2006) X 5 : Alcohol use during baseline X 6 : Ciga rette smoking during baseline X 7 : Illicit drug use during baseline X 8 : Chronic hypertension during baseline X 9 : Obese during baseline X 10 : Polycystic ovary syndrome (PCOS) during baseline X 11 : Rheumatoid arthritis during baseline X 12 : Chronic kidney diseas e during baseline X 13 : Diabetic nephropathy during baseline X 14 : Claims indicating m ild uncontrolled diabetes during baseline X 15 : Claims indicating s evere uncontrolled diabetes during baseline X 16 : Anemia during baseline X 17 : Depression during baseline X 18 : Systemic lupus erythematosus (SLE) during baseline X 19 : Anti depressant exposure during baseline X 20 : Statin exposure during baseline X 21 : Beta blocker exposure during baseline X 22 : Alpha agonist exposure during baseline X 23 : ACE inhibitor exposure du ring baseline X 24 : ARB exposure during baseline X 25 : Calcium channel blocker exposure during baseline X 26 : Diuretic exposure during baseline The model created (t 1) dummy variables (where t = the number of levels) for each categorical variable . T he generalized estimating equations for binary data with logit link functions produced an empirically derived formula , which discriminated between the respective exposure groups , producing a predicted exposure score ( propensity score ) for a given individual . We plotted the propensity score distribution to evaluate its overlap between

PAGE 62

62 exposure groups, and truncated at the margins of the propensity score distributions that did not overlap to ensure exclusion of women who would be expected to always receive one t reatment over the other. The propensity scores were included in the regression models for each outcome as a covariate. Statistical analysis We calculate d both crude and adjusted odds ratios (OR) and their 95% CIs using conditional multivariate logistic r egression (equation 3 6 ) to assess the risk of cesarean section delivery and preterm birth in each exposure category . We calculated two adjusted models: 1) only adjusted for age as a categorical variable and 2) only adjusting for the propensity scores as a continuous variable. Risk factors that emerge during pregnancy were purposely ignored, because they might be direct consequence s of initial diabetes treatment choice. Logit [P (Y= 1 1 X 1 2 X 2 (3 6 ) Where: Y: Probability of cesarean section delivery compared to vaginal delivery (Model 1), Probability of preterm birth compared to term birth (Model 2) X 1 : Drug exposure ( i.e. sulfonylureas vs. metformin, thiazolidinediones vs. metformin, insulin vs. metformin, metformin & thiazolidinediones vs. metformin & sulfonylureas) X 2 : Exposure propensity score (i.e. continuous variable) W e employed time to event analys e s for the number of gestational days at the onset of p reeclampsia . The onset (in gestational days ) of preeclampsia is important, as early onset is associated with poor pregnancy outcomes and infant morbidity (e.g. cesarean section delivery or extreme prematurity) . The hazard function s [ h (t) ] applied to preeclampsia w ere the conditional probabilit ies that preeclampsia occur red in t he interval (t, t+dt) given that they ha d not occurred before (equation 3 7 ). Similar to the logistic regression models, within our time to event analyses, we calculated two

PAGE 63

63 adjusted models: 1) only adjusted for age as a categorical variable and 2) only ad justing for the propensity scores as a continuous variable. L og [ h (t , X ) ] = log [ h 0 (t) ] + 1 X 1 2 X 2 (3 7 ) Where: t denotes days until event preeclampsia (Model) , log [ h (t) ] represents the baseline hazard, and X =( X 1 , X 2 , X 3 ,..., X 10 ) signifies the collection of predictor variables X 1 : Drug exposure (i.e. sulfonylureas vs. metformin, thiazolidinediones vs. metformin, insulin vs. metformin, metformin & thiazolidinediones vs. metformin & sulfo nylureas) X 2 : Exposure propensity score (i.e. continuous variable) Under the assumption of proportional hazards, a single hazard ratio expresse s the probability of preeclampsia throughout pregnancy, with a 95% CI. The hazard ratio measure d the different risk associate d with anti diabetic agents (e.g. sulfonylureas v s . metformin) and represent ed the ratio of the hazard function of the two treatments, assuming that the ratio remain ed constant throughout pregnancy (proportional hazards). The outcomes models included the conditional probability of cesarean section delivery compared to vaginal delivery (Model 1), the conditional probability of preterm birth compared to term birth (Model 2), and the hazard rate of preeclampsia (Model 3). For each of the outcome models, we employ ed the recommendation for model building in epidemiologic studies, and retain ed covariates if they change d the point estimate for the given outcome by 10%. 153 E xploratory analys i s Preterm birth and ces arean section deliveries can occur due to a number of causes (e.g. maternal demographic characteristics, nutritional status, pregnancy history, psychological characteristics, infection, or hemorrhage), many that are also known to causes stillbirth or spont aneous abortions. 91,93,101,139,154 Thus i f the analyzed exposures

PAGE 64

64 affect the causes (e.g. increases the risk of placental hemorrhage which causes a spontaneo us abortion ) ; it will affect both the outcomes included in the comparative safety analysis, as well as the probability to have another pregnancy outcome (stillbirth or abortion). Since our analysis required that women have a live birth, we have the potential to introduce bias into the results of our comparative safety analysis, as we are assuming that the analyzed exposures did not have a differential effect on the probability of l ive birth. 155 In order to assess the impact of requiring a live birth in the c omparative safety analysis , we relaxed our inclusion criteria to allow women into the analysis if they had a live birth or other pregnancy outcomes claim s (i.e. abortion or stillbirth) (Appendices: Abortion, Stillbirth, and Delivery) ) . Without a valid alg orithm to calculate the start of pregnancy for both successful and other pregnanc y outcomes , we chose to use the delivery date algorithm and modif ied it for the other pregnanc y outcomes and preterm birth . 137,1 56 Furthermore, a pproximately 90% of our exploratory analysis cohort was classified as a live birth ; therefore, we employ ed high sensitivity and specificity for the majority of our cohort and minimize d the possible misclassification of the 1 st trimester. We used the same methods identified in part II, in which we used the variations of the delivery date algorithm to identify the start of pregnancy for live birth, preterm birth, abortion and stillbirth. However, s ince the delivery date algorith m has not been validated, we varied the assumptions of the average pregnancy length for the pregnancies resulting in an abortion or stillbirth to test the impact on the risk estimates. The varied lengths for the pregnancies that resulted in abortion or sti llbirth were calculated based on the literature to identify periods of low , medium , and high

PAGE 65

65 average lengths of pregnancies. 138 142 We iden ti was calculated the same as it was in part II (i.e. 105 days for abortions and 196 days for stillbirth ) days for abortions and 238 days for stillbirths. We require d 6 month s continuous eligibility from the index date in MAX until the end of the pregnancy ( Figure 3 6 ) . We use d the same methods used in the safety analysis to identify drug exposure categories , and used the trimmed propensity scores withi n th e outcome models . We used hierarchical modeling to fit two level models for the safety outcomes ( pre term birth, cesarean section delivery , and preeclampsia ) . For preterm birth and cesarean section delivery , we fit models with generalized estimating equ ations (GEE) to account for clustering of multiple pregnancies within the same woman. For preeclampsia, we again used a time to event analysis within hierarchical models, to examine the hazard of preeclampsia. We used this exploratory analysis to identify if there was a difference in the risk of target outcomes , (i.e. mono or dual oral anti diabetic therapy) given the risk of other pregnancy outcome s (stillbirth, spontaneous or elective abortions). This analysis explore s the magnitude of the bias due to our requirement of live birth in the comparative safety analysis.

PAGE 66

66 Figure 3 1 . Medicaid Analytic eXtract (MAX) data included in pregnancy cohort from 29 US states.

PAGE 67

67 Figure 3 2 . Health Insurance Coverage of Women of Reproductive Age (15 44), 2006 . 115 Figure 3 3 . Medicaid Coverage of Birth. 2008. 116

PAGE 68

68 Figure 3 4 . Pregnancy cohort study timeline, Medicaid Analytic eXtract data (MAX), 1999 2006. Figure 3 5 . Anti diabetic drug utilization study timeline, Medicaid Analytic eXtract data (MAX), 2000 2006.

PAGE 69

69 Figure 3 6 . Comparative safety sensitivity analysis study timeline, Medicaid Analytic eXtract data (MAX), 2000 20 06 . 2 nd , 2 nd trimester. Table 3 1 . Calculation of linkage characteristics . Florida Birth Certificate Record linked cohort (gold standard)* Yes No Medicaid linked cohort (Case ID number algorithm) Yes A True positive B False positive PPV: A/(A+B) No C False negative D True negative NPV: D/(C+D) Sensitivity: A/(A+C) Specificity: D/(B+D) *Florida Birth Certificate Record linked cohort consists of mother infant pairs as identified from the birth certificates (i.e. certificate number and year of birth match). Medicaid linked cohort consists of mother infant pairs as identified from the Medicaid Case ID number algorithm (i.e. matching Case ID number

PAGE 70

70 CHAPTER 4 RESULTS Part I: Validation of M other I nfant L inkage U sing the Medicaid Case ID Number V ariable within the Medicaid Analytic eXtract D atabase Sample Characteristics Analyzing the F L BC R database from 1999 to 2004, w e identified 1,239,148 distinct live births in Florida with valid SSNs ( Figure 4 1 ) . Of these, 678,209 infants and 4 88,807 mothers were matched to the FL MAX database on SSN and date of birth. Mother and infants were linked within the FL BCR based on matching birth certificate numbers and year of birth. The total number of mothers and infants pairs , in the FL BCR that c ould be linked to FL MAX , were 434,179 ( Figure 4 1 ) . We identified 485,828 deliveries from women aged 12 to 55 in the FL MAX database , from 1999 to 2004 using at least 1 ICD 9 CM or CPT code for delivery and requiring a valid SSN and date of birth, and eligibility for Medicaid during the month the delivery claim was issued ( Figure 4 2 ) . Of these deliveries , 267, 270 had a valid non missing Medicaid Case ID number. We identified 6 0 5 , 920 infants born in FL MAX between 1999 to 2004 with at least 1 month Medicaid eligibility during the study period, and 466 , 537 infants had a valid n on missing Medicaid Ca se ID number. We excluded 27,842 infants because their Medicaid Case ID number linked to more than one wom a n within the cohort. A total of 130,228 mothers and infants were matched in the FL MAX database by Medicaid Case ID number algorithm ( Figure 4 2 ) . Overall, FL MAX captured between 57.1% and 59.6% of all FL BCR deliveries th) in 1999 and 2004 , respectively ( Table 4 1 ) . We could link both mothers and infants to FL MAX in

PAGE 71

71 approximately 35% of all FL BCR deliveries, and 41% of these linked mothers and infants had a valid Medicaid Case ID number. Validation of Medicaid Case ID N umber L inkage After applying our exclusion criteria, w e identified a total of 7 32 , 837 distinct infant and mother records from the FL BCR database from 1999 to 2004. In the same time period , we identified 575 , 737 women with a delivery claim or infants with a date of birth in the MAX database. We saw much shorter delivery ranges for the outpatient claims , as approximately 70% of the ranges included a difference of 0 days from the start date to the end date. We modified the delivery range to explore the possibility of capturing additional mother date of birth with a r ange of +/ 5 days of the maternal start and end dates of the delivery claim. Extending t he delivery rang e (i.e. +/ 5 days to 30 days) slightly increased th e number of mother infant pairs, but minimally changed the sensitivity, or the specificity. Due to t he limited variability of the sensitivity or specificity, we identified our delivery range solely based on of the start and end dates associated with the delivery claims. When compared to mother infant pairs identified from the FL BCR (gold standard, the Medicaid Case ID number algorithm (matching of the mother to infant based on the Medicaid Case ID number, delivery date range and infant date of birth) , w e found a sensitivity ranging from 57.1% to 73. 7 % ( Table 4 2 ) with specificity of 98.0 to 99.9% ( Table 4 3 ) ; the positive predictive value (PPV) ranged from 86.6% to 88.2% ( Table 4 4 ) , w hen va rying the required MAX eligibility of both the mothers and/or infants. If you c hang e the required MAX eligibility period for mothers from 1 month continuous eligibility before delivery to 12 months continuous eligibility before delivery , we are able

PAGE 72

72 to inc rease the sensitivi ty from 57.1% to 71.0%. eligibility had minimal effect on the sensitivity of the algorithm. The specificity and PPV saw a marginal change when varying the eligibility of the mother or infant (+/ 1%) . Pred icting L inkage within a V alidated A lgorithm that I dentifies M other I nfant P airs within the Medicaid Analytic eXtract (MAX) D atabase. We explored the determinants of true mother infant linkage using the Medicaid Case ID algorithm. There were a total of 434 , 179 linked live births in FL BCR and FL MAX, and we assessed 179,103 lives births after excluding those that did not meet the eligibility requirements (e.g. non valid Medicaid Case ID number). We linked 72.7% of the identified live births via Medicaid Case ID number and delivery/birth dates, with 92.6% of the se linkages as true links . The final regression model exploring the determinants of true linkage, show ed good discriminative ability with an AUC of 0.8614. The c haracteristics associated with true link age are listed in Table 4 5 . Interestingly, characteristics associated with healthier infants (i.e. higher birth weight or term infants) indicated a higher likelihood for true linkage by the Medicaid Case ID number algorithm (e.g. term infants [ OR 2.09; 95% CI: 1.99, 2.21 ]). The Medicaid Case ID number is a variable that is supposed to identify family units within states. However, we saw that infants classifie d as not h aving a living sibling, as identified from their birth certificate, had increased odds of true linkage as compared to infants with 1 sibling (OR 2.96; 95% CI: 2.81, 3.12) . Furthermore, we observed that the odds of true linkage was higher in those with a 20 04 delivery year (OR 1.10; 95% CI: 1.02, 1.19) versus 1999 delivery year . There was a lower likelihood of linkage among o utpatient delivery claims (OR 0.54; 95% CI: 0.48, 0.61 ) .

PAGE 73

73 Part II: Epidemiology of P re existing D iabetes in P regnancy from 2000 2006 among Medicaid P atients in 29 US S tates After applying the inclusion and exclusion criteria, we identified 1, 226 , 025 pregnancies in the 29 US states from the MAX database from 2000 to 2006. W e found that the overall prevalence for pre existing diabete s within the pregnancy cohort was 3. 6 % and 2. 3 % for type 2 diabetes only . A total of 28 , 099 women had only ICD 9 CM diagnos is codes for type 2 diabetes , 2,208 with type 1 diabetes only, and 13 , 370 women had either ICD 9 CM diagnos is codes with undetermined diabetes type (i.e. missing fifth d igit) , both type 1 and type 2 diabetes, or mixture of both . The individual state prevalence for pre existing diabetes during pregnancy varied from 1. 8 % (Arkansas) to 7. 2 % (New Jersey) ( Figure 4 3 ) , and the prevalence of type 2 diabetes during pregnancy varied by state from 1.2% (Arkansas, Florida, Illinois) to 4.6 % (New Jersey) ( Figure 4 4 ). By region, as identified by the US Census, varied from 2.9 % (95% CI: 2.8 , 3. 0 ) to 3.8 % (95% CI: 3. 7 , 3.8 ) , w ith the Northeast region having the lowest prevalen ce and the South with the highest prevalence . The region al prevalence of t ype 2 diabetes exhibited a smaller range from 2. 0 % (95% CI: 1.9 , 2. 1 ) to 2. 5 % (95% CI: 2. 2 , 2.9 ) , with the Northeast having the lowest and the West region the highest prevalence of type 2 diabetes. demographic and clinical characteristics by diabetes status are presented in Table 4 6 . Pregnancies associate d with pre existing diabetes before the start of pregnancy included women who were on average older (mean age 3 2 . 4 vs. 24. 3 years, p<0.0001) , had longer continuous Medic aid eligibility (mean months 61.7 vs. 52.5 , p<0.0001) , and were Non Hispanic African American (4 3.9 % vs. 4 2.8 %, p <0.0001 ) when compared to pregnancies in non diabetic women. As expected,

PAGE 74

74 diabetic women had a higher prevalence of PCOS ( 3.0 % vs. 0. 8 % , p< 0.0001) , chronic hypertension ( 33.7 % vs 4. 8 %, p<0.0001) , obesity ( 1 6.2 % vs 3. 6 %, p<0.0001) , and infertility ( 0. 9 % vs 0. 5 %, p<0.0001) as identified in the 12 month s baseline period compared to non diabetic women . Additionally, d iabetic women also had a high er prevalence of stillbirths ( Table 4 6 ). Within the diabetes stratified analysis , type 2 diabetic and mixed diabetic women were o n average older ( Table 4 7 ) . Type 1 and mixed diabetic women had higher prevalence of abortion compared to the type 2 diabetics (5.5% vs. 4.0%, p<0.0001 ). As anticipate d , t he prevalence of PCOS ( 3.6 % vs. 1.1 %, p<0.0001) , hypertension (34.0% vs 17.4%, p <0.0001) and obesity ( 1 7 . 1 % vs. 7.8 %, p<0.0001) was higher in type 2 diabetic women compared to type 1 diabetic women ( Table 4 7 ) . Secular trends in overall pre existing diabetes and each stratified group of diabetes type are presented in Figure 4 5 . The annual prevalence of pre existing diabetes increased from 2.9 % (95% CI: 2.7 , 3. 1 ) in 2000 to 4. 8 % (95% CI: 4. 6 , 5.0 ) in 2006 (p trend = 0. 0 541 ) , indicating a rate change of 65.5 % from 2000 to 2006. From 2000 to 2006, there was a 92.8 % increase in the prevalence of pre existing type 2 diabetes in pregnancy, from 1. 7 % (95% CI: 1. 6 , 1.9 ) to 3.3 % (95% CI: 3.1 , 3. 5 ) (p trend =0. 0170 ). In 2003, we saw a modest drop (0.2%) in the prevalen ce of pre existing diabetes . This decrease was also seen in the stratified analysis . It is possible that the dip in prevalence could be an artifact of managed care penetration, thus a differential push could be occuring for women with chronic diseases into mangage care. The increase in prevalence of pre existing diabetes differed by maternal age, race/ethnicity, and pregnancy year. Owing to the observed secular increase in type 2

PAGE 75

75 diabetes, we noted a more pronounced increase of pre existing diabetes during pregnancy in older age groups ( Table 4 8 ). T he age specific prevalence of pre existing diabetes increased approximately 43.5 % from 2000 to 2006 in women aged 20 29 , a nd approximately 64.3% in women aged 30 39 years from 2000 to 2006. Women over the age of 40 years demonstrated only a 15.8 % increase in the prevalence of pre existing diabetes from 2000 to 2006 , but still had the highest age specific prevalence, almost do ubling the other three age categories combined ( Table 4 8 ). The annual age standardized prevalence of pre existing diabetes in Non Hispanic W hite women increased from 2.9 % in 2000 to 4.3 % in 2006. Asian/Pacific Islander women demonstrated a n increase from 3.1% to 4.5% over the study period. The age standardized prevalence of pre existing diabetes in Non Hispanic B lack women increased by 33.3 % from 2000 to 200 6 , and the p revalence more than doubled in Hispanic women from 2.6% to 5.5% . Part III: Utilization of A nti diabetic A gents in P regnancy from 2000 2006 among Medicaid P atients in 29 US S tates Over the course of the seven year study period from 2000 to 2006 , we identif ied 658,485 deliveries in the 29 US states. The overall prevalence of pre existing diabetes in the delivery cohort was 2. 4 %. A total of 9,546 women had ICD 9 CM codes consistent with type 2 diabetes only , 1,024 had type 1 diabetes claims only , and 5,04 7 had a mixture of claims that either had the fifth digit missing or that indicated presence of both type 1 and type 2 diabetes. Similar to the pregnancy cohort, the diabetic women in the delivery cohort w ere older, had a higher prevalence of additional co m orbid conditions, and had longer average baseline durations of Medicaid eligibility when compared to their non diabetic counter parts ( Table 4 9 ) . Diabetic women had a higher

PAGE 76

76 prevalence of cesarean section deliveries compared to non diabetic women, 39.8% vs . 22.8% respectively , as well as a higher prevalence of preterm birth s (15.1% vs . 10.9%) . There was also a higher prevalence of diabetics located in the Southern reg ion of the US compared to non diabetics ( Table 4 9 ). As with the pregnancy cohort, we stratified the women identified as pre existing diabetics into the probable dia betes types ( Table 4 10 ). Women identified as type 1 diabetics were on average younger than the type 2 and mixed diabetic women. T he mixed diabetic women were on avera ge older, had longer Medicaid eligibility and had a higher prevalence of co morbid conditions ( Table 4 10 ). When we explor ed the anti diabetic utilization during pregnancy by diabetes type, as expected , the probable type 1 diabetics utilized only oral agents at a prevalence of 2.0% and type 2 dia betics utilized only insulin at a prevalence of 4.8%. Surprisingly, the type 1 diabetics had a 43.1% prevalence of no anti diabetic agent utilization during pregnancy. The fact that almost half of all identified type 1 diabetics did not use insulin is a fu rther indication of a limitation in the granularity of ICD 9 CM codes in order to distinguish between diabetes types using only the fifth digit of administrative claims. The prevalence of anti diabetic treatment (insulin or oral anti diabetic drugs) among women with pre existing diabetes before pregnancy (i.e. during the pre conception period) increased from 27.5 % (95% CI: 23.4%, 31.7% ) to 35.9 % (95%CI: 33.8%, 38.0% ) from 2000 to 2006 (p= 0.520 ) ( Figure 4 6 ) . During pregnancy, there was a n increase in anti diabetic treatment from 47.0 % (95%CI: 42.5%, 51.5% ) in 2000 to 53.9 % (95%CI: 51.7%, 56.1% ) in 2006 (p= 0.3727 ). In general, we saw a comparable

PAGE 77

77 increase in the utili zation of overall oral anti diabetic treatment both before pregnancy and during pregnancy over the course of our study period ( Figure 4 6 ). Next, w e examined the secular trend of specific utilization of anti diabetic treatment classes during pregnancy throughout the study period. The prevalence of both metformin and t hiazolidinedione use during pregnancy increased approximately 66.0% from 2000 to 200 6. Oral anti diabetic combination products entered the market in 2000, thus it was not surprising that we did not see utilization during pregnancy until 2001. We saw a steady increase in the prevalence of use in combination product use to 1.4% (95% CI: 0.9 %, 1.9%) in 2006. The prevalence of insulin utilization during pregnancy increase d marginally by 9.5% from 2000 to 2006 ( Figure 4 7 ). Sulfony lurea use showed a similar annual prevalence to metformin, with an increase until 2002, followed by a steady decrease until 2005 to a prevalence of 4.6%, only to rebound to 6.2% in 2006. The decrease in the use of sulfonylureas was similar to what Hampp et al. found in their analysi s exploring the market trends for anti diabetic drugs. 157 Each of the oral anti diabetic drug classes increased in pooled prevalence from the pre conception period to the first trimester of pregnancy but then steadily decreased in prevalence in the second and third trimesters of pregnancy ( Figure 4 8 ) . On the other hand, i nsulin use steadily increased from the pre conception period until the third trimester ( Figure 4 8 ) . We saw that 35% of all the pre existing diabetic deliveries included in our analysis , initiat ed anti diabetic treatment during the first trimester. O nly 25% of pre existing deliveries that were identified as no t having received anti diabetic treatment in the pre conception period remained without anti diabetic treatment in the first trime ster

PAGE 78

78 of pregnancy . Of the deliveries with newly initiated treatment in the first trimester, 59.6% were identified as insulin only users, 26% were exposed to more than one anti diabetic drug class, 6.7% were metformin only users, 5.5% were sulfonylureas on ly users, and 2.3% were thiazolidinedione only users. The results from the prediction model to identify characteristics associate d with the initiation of anti diabetic agents compared to women who remain untreated in the first trimester of pregnancy are sh own in Table 4 11 . Previously untreated diabetic w omen who initiate d anti diabetic treatment compared to those who remain ed untreated in the first trimester of pregnancy were more likely to be either younger than 20 years [adjusted odds ratio (OR)=1.37, 95% CI: 1.16,1.62) or older than 40 years compared to 20 29 y ears of age [ ( OR)=2.69, 95% CI: 2.42, 2.98]. Women who were also diagnos ed with chronic hypertension during the 12 months before pregnancy had increased odds of initiating anti diabetic treatment in the first trimester [OR=1.66, 95% CI: 1.54, 1.80]. Interestingly, women diagnosed with PCOS during the 12 months before pregnancy h ad decreased odds of initiating anti diabetic treatment in the first trimester [OR= 0.55 , 95% CI: 0.39, 0.77 ]. Part IV: Comparative S afety of O ral A nti diabetic A gents in P regnancy from 2000 20 06 among Medicaid P atients in 29 US S tates The study cohort had 1,980 deliveries with mono or dual therapy exposure in the first trimester of pregnancy ( Figure 4 9 ). Mono therapy of metformin in the first trimester of pregnancy was the most common exposure cate g ory (26.0%), followed by the initiation of mono therapy of with insulin in the first trimeste r of pregnancy (25.1%), mono therapy with sulfonylureas (21.4%), dual therapy with metformin and sulfonylureas (11.5%), mono therapy with thiazolidinedione (8.9%), and dual therapy

PAGE 79

79 with metformin and thiazolidinedione (7.2%). Although we did not restrict t he cohort to the presence of type 2 diabetes, the majority of deliveries included in the analysis (i.e.>75%) were identified as deliveries to women with only type 2 diabetes. The cohort was predominantly White, Non Hispanic, located in the Southern region of the US, and had a mean age of 37.1 years . We saw similar prevalence rates of PCOS, chronic hypertension and obesity during the six month baseline period in our safety cohort (Part IV) as in the utilization cohort (Part III). When stratified by exposure category, we saw differences in several characteristics ( Table 4 12 ) . Most notably, women on sulfonylureas only , thiazolidinediones only or dual therapy (e.g. metformin and sulfonylureas only) had higher rates of chronic hypertension than those on metformin only or insulin only in the first trimester of pregnancy. Interestingly, women on insulin only or dual therapy had uncontrolled diabet es (i.e. ICD 9 CM codes associated with un controlled diabetes but no hospitalization in the baseline period) tha n women on mono oral anti diabetic therapy in the first trimester of pregnancy. Differences were also apparent in the utilization of anti hyper tensive drugs , statin s , and anti depressants during the baseline period , suggesting greater need for complex therapy in women on dual therapy or thiazolidinediones ( Table 4 12 ). We included characteristics identified in Table 4 12 in the propensity score models for each exposure comparison and outcome, as noted in the part IV method s section. Within each propensity score model, we saw m inimal sample size loss when trimming the exposure categories. For example, the largest loss was seen in the propensity score model for exposure to sulfonylureas vs. metformin in preterm birth, as

PAGE 80

80 the number of metformin exposed deliveries went from 514 to 477 and sulfonylureas from 424 to 414. Due to the limited number of outcomes, we restricted the adjusted models to include maternal age only or the trimmed propensity scores that summarized all basel ine covariates, including maternal age. Preterm B irth Across all agents, we observed 163 (8.2%) preterm b irths with in the comparative safety cohort. In th e unadjusted comparisons with metformin (or metformin & sulfonylureas for dual therapy), insulin exposure had more than double the odds for preterm birth , whereas the other mono therapy comparators showed a null effect (sulfonylureas) or a 38% reduction in odds (thiazolidinedione s ) ( Table 4 13 ). Deliveries exposed to metformin and thiazolidinedione s had 64% increased odds of preterm birth when compared to metformin and sulfonylureas ( Table 4 13 ) . When we adjusted for maternal age within the regression models, sulfonylureas demonstrated a 56% increase in the odds of preterm birth, although not statistically significant. Finally, adjustment for the exposure specific propensity scores ( Figure 4 10 ) resulted in a significant increase in the odds of preterm birth for deliveries exposed t o sulfonylureas during the first trimester of pregnancy when compared to metformin [OR: 2.05 (95% CI: 1.09, 3.83)]. Results for comparison of thiazolidinedione s versus metformin, as well as for the comparison of the t wo dual therapy groups were inc onclusiv e due to wide confidence intervals, but point estimates suggested similar risk. Cesarean S ection D elivery Within the safety cohort, 29.6% of live births occurred via c esarean section delivery . The adjust ment for age or the propensity scores ( Figure 4 11 ) minimally

PAGE 81

81 changed the odds of cesarean section delivery for specific anti diabetic drug classes ( Table 4 14 ). Unadjusted and after propensity score adjustment, oral agents showed similar risk for cesarean section delivery (sulfonylureas versus metformin OR 0.94 (95% CI: 0.68, 1.30), and thiazolidin ediones versus metformin ( OR 0.98 , 95% CI: 0.65, 1.47)). In contrast, insulin showed increased odds, which was not altered after adjustment OR 1.41 (95% CI: 1.05, 1.89). The comparison of dual therapies showed similar risk but the confidence intervals were wide ( Table 4 14 ). Preeclampsia We observed an overall incidence of 6.5% for the diagnosis of preeclampsia during pregnancy within the safety study cohort. During pregnancy, the unadjusted hazard ratio (HR) for preeclampsia in women exposed to sulfonylureas was 0.71 (95% CI: 0.46, 1.10), 0.65 (95% CI: 0.38, 1.09) for thiazolidinedione, and 0.88 (95% CI: 0.60, 1.29) for insulin as compared to mono therapy met formin use. In comparison to mono therapy with metformin, both sulfonylureas and thiazolidinediones showed a lower but insignificant risk for preeclampsia ( adjusted HR ( Figure 4 12 ) for sulfonylurea s use was 0.61 (95% CI: 0.36, 1.04) and 0.58 (95% CI: 0.32, 1.04) for thiazolidinedione s ( Table 4 15 ) ) . Resembling the unadjusted hazard ratios , dual therapy of metformin and thiazolidinedione had an increased adjusted HR of 1.74 (95% CI: 0.78, 3.89), although not statistically significant. Exploratory Analys i s We identified over 1.6 mi llion pregnancies in women with 6 months continuous Medicaid eligibility before their calculated start of pregnancy . In the pregnancies a ffected by pre existing diabetes, the overall prevalence of other pregnancy outcomes (i.e. stillbirth, spontaneous or e lective abortion) was 5.4% . When we restricted our

PAGE 82

82 cohort to pregnancie s resulting in a confirmed failed pregnancy (ICD 9 CM or CPT 4) or confirmed delivery, we maintained 21,901 pregnancies with an overall prevalence of other pregnancy outcomes of 9.4%. When we considered the additional risks of other pregnancy outcomes , we found estimates that were different for the primary comparative safety analysis ( Table 4 16 ). W e found that sulfonylureas no longer increased the risk for preterm delivery, the protective effect of thiazolidinedione was increased , and the increased odds of preterm birth were lessened in women exposed to insulin. In cesarean section deliveries, we sa w a modest increase in odds associated with thiazolidinedione s . As for preeclampsia , consideration of other pregnancy outcomes (i.e. stillbirth, spontaneous or elective abortion) resulted in essentially eliminating the effect seen within the primary compar ative safety analysis, as all agents saw no difference in risk when compared to metformin ( Table 4 16 ) . Varying the pregnancy length for each type of other pregnancy outcome and thus the definition of on and offset of the first trimester did not result in marked differences across the estimates of the exploratory analysis.

PAGE 83

83 Figure 4 1 . Assembly of the mother infant linked BCR cohort , Florida Birth Certificate Records (BCR) data, 1999 2004 . DOB: date of birth, SSN: Social Security Number.

PAGE 84

84 Figure 4 2 . Assembly of the mother infant linked Medicaid Case ID number cohort , Florida Birth Certificate Records (B CR) data, 1999 2004 . DOB: date of birth, SSN: Social Security Number.

PAGE 85

85 Figure 4 3 . Prevalence of pre existing diabetes in pregnancy by state.

PAGE 86

86 Figure 4 4 . Prevalence of type 2 diabetes in pregna ncy by state.

PAGE 87

87 Figure 4 5 . Temporal trends of pre existing diabetes in pregnancy. 0.0% 1.0% 2.0% 3.0% 4.0% 5.0% 6.0% 2000 2001 2002 2003 2004 2005 2006 Prevalence of Diabetes Year Total DM T2DM Only Mixed DM T1DM Only

PAGE 88

88 Figure 4 6 . Annual prevalence of anti diabetic utilization in women with pre existing diabetes before and during pregnancy. 0 10 20 30 40 50 60 70 80 90 100 2000 2001 2002 2003 2004 2005 2006 Prevalence of utilization (%) Year Oral anti-diabetic utilization during pregnancy Anti-diabetic utilization during pregnancy Pre-conception oral antidiabetic utilization Pre-conception antidiabetic utilization

PAGE 89

89 Figure 4 7 . Annual prevalence of anti diabetic utilization during pregnancy by drug class. 0 10 20 30 40 50 60 70 80 90 100 2000 2001 2002 2003 2004 2005 2006 Prevalence of Utilization (%) Year Metformin Sulfonyulreas Thiazolidinedione Combination Insulin

PAGE 90

90 Figure 4 8 . Pooled annual prevalence of anti diabetic drug utilization, by drug class and pregnancy period . 0 5 10 15 20 25 30 35 40 45 50 Prevalence of Utilization (%) Metformin Sulfonylureas Thiazolidinedione Combination Insulin

PAGE 91

91 Figure 4 9 . Assembly of the safety study cohort.

PAGE 92

92 Figure 4 10 . P ropensity score distributions of specific anti diabetic exposure for the premature delivery model. PS=propensity score, m et=metformin, s ulf=sulfonylureas, i n=insulin, ms =metformin & sulfonylureas, mt =metformin & thiazolidinedione .

PAGE 93

93 Figure 4 11 . P ropensity score distributions of specific anti diabetic exposure for the cesarean section delivery model. PS=propensity score, met=metformin, sulf=sulfonylureas, in=insulin, ms=metformin & sulfonylureas, mt=metformin & thiazolidinedione.

PAGE 94

94 Figure 4 12 . P ropensity score distributions of specific anti diabetic exposure for the preeclampsia model. PS=propensity score, met=metformin, sulf=sulfonylureas, in=insulin, ms=metformin & sulfonylureas, mt=metformin & thiazolidinedione.

PAGE 95

95 Table 4 1 . Successful Florida Birth Certificate Records to Medicaid linkages in percent by year (N=1,239,148 deliveries) . * Year F L BCR deliveries linked to MAX FL BCR deliveries with both mother and infant linked to MAX FL BCR deliveries with only infant linked to MAX FL BCR deliveries with only mothers linked to MAX 1999 57.09% 34.45% 18.73% 3.90% 2000 58.17% 34.51% 19.62% 4.05% 2001 59.41% 34.36% 20.82% 4.24% 2002 60.32% 35.72% 20.67% 3.92% 2003 60.16% 35.66% 19.80% 4.69% 2004 59.55% 35.46% 18.54% 5.55% * FL BCR was linked to MAX via social security number and date of birth. FL BCR = Florida birth certificate records database; MAX=Medicaid Analytic eXtract database.

PAGE 96

96 Table 4 2 . Sensitivity of Medicaid Case ID Number algorithm varied by mother and infants M edicaid eligibility. Mother MAX eligibility At delivery N=729,420 At delivery + 1 month before N=713,038 At delivery + 3 month before N=685,067 At delivery + 6 month before N=632,851 At delivery + 9 month before N=551,516 At delivery + 12 month before N=546,303 At delivery + 15 month before N=543,700 Infant MAX eligibility 57.1% 58.8% 61.7% 66.7% 70.2% 71.0% 70.4% 59.6% 61.5% 64.5% 69.9% 73.6% 73.7% 73.1% 59.5% 61.4% 64.4% 69.8% 73.5% 73.6% 73.0% 58.6% 60.5% 63.4% 68.7% 72.6% 72.5% 72.0% 58.2% 59.8% 62.6% 67.6% 72.6% 72.5% 72.1% 57.7% 59.2% 62.0% 66.9% 72.5% 72.5% 71.9% 57.2% 58.7% 61.4% 66.3% 71.8% 71.8% 71.1% * Sensitivity was calculated as the number of mother infant linked by Case ID number algorithm and FL BCR divided by the total number of mother infant pairs linked within the FL BCR. MAX=Medicaid Analytic eXtract database , FL BCR= Florida Birth Certificate Records .

PAGE 97

97 Table 4 3 . Specificity of Medicaid Case ID Number algorithm varied by mother and Mother MAX eligibility At delivery N=729,420 At delivery + 1 month before N=713,038 At delivery + 3 month before N=685,067 At delivery + 6 month before N=632,851 At delivery + 9 month before N=551,516 At delivery + 12 month before N=546,303 At delivery + 15 month before N=543,700 Infant MAX eligibility 98.8% 98.9% 99.0% 99.2% 99.9% 99.9% 99.9% 98.0% 98.2% 98.4% 98.9% 99.8% 99.9% 99.9% 98.0% 98.2% 98.4% 98.8% 99.8% 99.9% 99.9% 98.5% 98.6% 98.7% 99.0% 99.8% 99.9% 99.9% 99.2% 99.2% 99.2% 99.3% 99.8% 99.9% 99.9% 99.3% 99.3% 99.3% 99.4% 99.8% 99.9% 99.9% 99.4% 99.4% 99.4% 99.4% 99.8% 99.9% 99.9% * Specificity was calculated as the number of mother infant not linked by Case ID number algorithm and FL BCR divided by the total number of mother infant pairs not linked within the FL BCR. MAX=Medicaid Analytic eXtract database , FL BCR= Florida Birth Certificate Records

PAGE 98

98 Table 4 4 . Positive predictive value (PPV) of Medicaid Case ID Number algorithm eligibility. Mother MAX eligibility At delivery N=729,420 At delivery + 1 month before N=713,038 At delivery + 3 month before N=685,067 At delivery + 6 month before N=632,851 At delivery + 9 month before N=551,516 At delivery + 12 month before N=546,303 At delivery + 15 month before N=543,700 Infant MAX eligibility 86.6% 86.6% 86.8% 87.1% 87.1% 86.9% 87.0% 86.8% 87.0% 87.1% 87.5% 86.7% 86.4% 86.5% 86.8% 86.9% 87.1% 87.5% 86.7% 86.4% 86.5% 86.8% 87.0% 87.1% 87.4% 86.8% 86.1% 86.2% 87.5% 87.6% 87.7% 88.0% 87.1% 86.7% 86.9% 87.6% 87.8% 87.9% 88.2% 87.4% 86.9% 87.0% 87.7% 87.8% 87.9% 88.2% 87.5% 87.1% 87.5% * PPV was calculated as the number of mother infant linked by Case ID number algorithm and FL BCR divided by the total number of mother infant pairs linked with the Medicaid Case ID number algorithm. MAX=Medicaid Analytic eXtract database , FL BCR= Florida Birth Certificate Re cords

PAGE 99

99 Table 4 5 . Determinants of accurate mother infant linkage by Medicaid Case ID Number. Variable Odds Ratio 95% Confidence Interval Delivery year 1999 1.00 1.00 2000 0.79 0.7 4 , 0.8 5 2001 0.8 5 0.7 9 , 0.9 1 2002 0.85 0.79, 0.9 2 2003 1.0 4 0.9 7 , 1.1 2 2004 1. 12 1.0 3 , 1. 21 Length of delivery hospitalization (days) 0.7 6 0.74, 0.77 Delivery Claim IP 1.00 1.00 OT 0.54 0.48, 0.61 Mother's continuous MAX eligibility before delivery 1 month 1.00 1.00 3 months 0.69 0.6 6 , 0.73 6 months 0.9 5 0. 90 , 1.00 9 months 1. 90 1. 68 , 2. 14 12 months 1. 51 1. 31 , 1. 74 15 months 1.03 0.9 3 , 1.15 Age of mother Under 20 years 0.80 0.75, 0.86 20 29 years 1.00 1.00 30 39 years 1.22 1.17 , 1.2 8 40+ years 2.1 5 1.87, 2.47 Race/ethnicity of mother White, Non Hispanic 1.00 1.00 Black, Non Hispanic 1. 40 1. 34 , 1. 45 Asian 1.03 0. 87 , 1.21 Hispanic 1. 26 1. 20 , 1. 33 Unknown 1. 53 1.26 , 1. 86 Education of mother Less than high school 1. 49 1. 43 , 1. 55 High school graduate 1.00 1.00 Some college 1.08 1.0 4 , 1.13 College graduate 0.9 4 0.8 5 , 1. 03 Graduate school 0. 63 0.54, 0.7 2 Unknown 0. 35 0.2 8 , 0.4 3

PAGE 100

100 Table 4 5 . Continued. Variable Odds Ratio 95% Confidence Interval Region North 0. 93 0. 88 , 0. 99 Northwest 1.46 1.37, 1.56 Central 1.0 6 1.02, 1.11 Southwest 0. 89 0.8 4 , 0.95 Southeast 1.00 1.00 Mother qualifies for cash assistance Yes 0. 94 0. 90 , 0.9 8 No 1.00 1.00 Mother identified as poverty eligible in MAX Yes 0.93 0.87, 1.00 No 1.00 1.00 Sibling (live births) None 2.9 7 2.8 2 , 3.1 3 1 1.00 1.00 2 to 4 0.4 8 0.46, 0. 50 5+ 0.22 0.2 1 , 0.24 Number of prenatal visits 1 to 9 1.3 9 1.3 3 , 1.4 5 10 to 14 1.00 1.00 15 to 20 0.40 0.3 9 , 0.4 2 21 to 30 0.29 0.25, 0.33 31+ 0.3 7 0.27, 0.49 None/Unknown 1.61 1.4 6 , 1.7 9 Preterm delivery Yes 1.00 1.00 No 2.0 9 1.98, 2. 20 Infant eligible for MAX at birth Yes 0.6 6 0.5 6 , 0.7 7 No 1.00 1.00

PAGE 101

101 Table 4 5 . Continued. Variable Odds Ratio 95% Confidence Interval Infant's continuous eligibility after birth 1 month 1.00 1.00 3 months 1.1 3 1.0 8 , 1.1 8 6 months 1. 10 1.0 1 , 1.19 9 months 1.0 8 0.97, 1.1 9 12 months 0.87 0.81, 0.9 3 Gender of infant Male 0.8 9 0.86, 0.9 2 Female 1.00 1.00 Birth weight of infant Under 2500 grams 0.11 0.10, 0.1 2 Over 2500 grams 1.00 1.00 * MAX=Medicaid Analytic eXtract database.

PAGE 102

102 Table 4 6 . Baseline characteristics of the p regnancy cohort stratified by pre existing diabetes . Variable Category DM* N=44,037 Not DM* N=1,181,988 Age: Years (SD) 32.4 (9.9) 24.3 (7.4) Eligibility: Months (SD) 61.7 (27.4) 52.5 (24.9) Race/ethnicity White, Non Hispanic 43.7% 45.8% Black, Non Hispanic 43.9% 42.8% American Indian or Alaskan Native 1.0% 0.7% Asian or Pacific Islander or Native Hawaiian 0.8% 0.7% Hispanic or Latino 8.4% 8.7% Unknown 2.2% 1.2% End of Pregnancy Abortion 4. 6 % 4. 4 % Stillbirth 0.7% 0.5% Preterm Birth 1 1 . 5 % 8 . 8 % Comorbid Conditions PCOS 3.0% 0.8% Hypertension 33.8% 4.8% Obesity 16.2% 3.6% Infertility 0.9% 0.5% Smok ing 9.1% 5.8% Alcohol abuse 3.7% 2.0% Drug abuse 4.8% 2.7% Region Northeast 9. 1 % 11.2% Midwest 28.6% 30.0% South 61.6% 58.1% West 0.8% 0.8% * DM = pre existing diabetes mellitus .

PAGE 103

103 Table 4 7 . Baseline characteristics of the pregnancy cohort stratified by type of diabetes . Variable Category T1DM* N=2,208 T2DM * N=28,099 Mixed DM * N=13,730 Age: Years (SD) Eligibility: Months (SD) Race/ethnicity White, Non Hispanic Black, Non Hispanic American Indian or Alaskan Native Asian or Pacific Islander or Native Hawaiian Hispanic or Latino Unknown End of Pregnancy Abortion Stillbirth Preterm Birth Comorbid Conditions PCOS Hypertension Obesity Infertility Smok ing Alcohol abuse Drug abuse Region Northeast Midwest South West * T1DM = type 1 diabetes mellitus, T2DM= type 2 diabetes mellitus, DM= diabetes mellitus .

PAGE 104

104 Table 4 8 . Annual prevalence of pre existing diabetes in pregnancy from 2000 2006 among Medicaid patients in 29 US states . 2000 2001 2002 2003 2004 2005 2006 P trend # Pregnancies 106,760 127,693 146,752 161,084 207,649 238,625 237,462 Mean age (SE) 23.9 (7.5) 24.3 (7.8) 24.5 (7.8) 24.3 (7.4) 24.4 (7.3) 24.6 (7.5) 25.1 (8.1) # with Diabetes 3,120 4,290 4,955 5,138 6,803 8,400 11,331 Age <20 1.2 (0.25) 1.3 (0.23) 1.1 (0.22) 1.1 (0.21) 1.1 (0.19) 1.1 (0.17) 1.6 (0.18) 0.1860 20 29 2.3 (0.23) 2.4 (0.21) 2.4 (0.19) 2.4 (0.18) 2.5 (0.16) 2.6 (0.15) 3.3 (0.16) <0.0001 30 39 5.6 (0.41) 6.3 (0.36) 6.5 (0.34) 6.5 (0.33) 6.9 (0.28) 7.5 (0.26) 9.2 (0.25) <0.0001 40+ 16.4 (0.74) 17.5 (0.61) 17.2 (0.56) 16.2 (0.58) 15.6 (0.52) 16.9 (0.46) 19.0 (0.39) 0.0016 Race/ethnicity White, Non Hispanic crude 2.9 (0.24) 3.2 (0.22) 3.3 (0.20) 3.0 (0.19) 3.0 (0.16) 3.3 (0.15) 4.6 (0.15) <0.0001 age standardized 2.9 (0.08) 3.1 (0.07) 3.2 (0.07) 3.0 (0.06) 3.1 (0.05) 3.3 (0.05) 4.3 (0.06) <0.0001 Black, Non Hispanic crude 3.0 (0.23) 3.4 (0.21) 3.4 (0.20) 3.4 (0.19) 3.4 (0.17) 3.6 (0.16) 4.8 (0.15) <0.0001 age standardized 3.3 (0.08) 3.6 (0.08) 3.6 (0.07) 3.5 (0.07) 3.6 (0.06) 3.7 (0.6) 4.4 (0.06) <0.0001 Asian crude 3.6 (1.1) 4.0 (1.1) 4.4 (1.0) 4.4 (1.0) 4.2 (0.94) 4.4 (0.89) 5.6 (0.90) 0.0316 age standardized 3.1 (0.37) 3.2 (0.35) 3.7 (0.36) 3.4 (0.33) 3.5 (0.33) 3.7 (0.32) 4.5 (0.36) 0.0067

PAGE 105

105 Table 4 8 . Continued . 2000 2001 2002 2003 2004 2005 2006 P trend Hispanic crude 2.4 (0.43) 2.9 (0.40) 2.7 (0.40) 3.1 (0.40) 3.7 (0.40) 3.8 (0.38) 5.7 (0.43) <0.0001 age standardized 2.6 (0.14) 3.1 (0.14) 2.9 (0.14) 3.4 (0.15) 4.1 (0.16) 4.3 (0.16) 5.5 (0.19) <0.0001 All Women Crude 2.9 (0.15) 3.3 (0.14) 3.3 (0.13) 3.2 (0.13) 3.3 (0.11) 3.5 (0.10) 4.7 (0.10) <0.0001 Age standardized 3.1 (0.05) 3.4 (0.05) 3.3 (0.05) 3.3 (0.04) 3.4 (0.04) 3.5 (0.04) 4.4 (0.04) < 0.0001 Race and age standardized 3.1 (0.05) 3.3 (0.05) 3.3 (0.05) 3.3 (0.04) 3.4 (0.04) 3.6 (0.04) 4.4 (0.04) <0.0001 Data are prevalence per 100 (SEM). * P values are derived from linear regression model using pre existing diabetes as the outcome variable and year as a continuous variable in the model after adjustment for other variables specified for the row. £ Pre existing diabetes. Standardized estimates are standardized to the distribu tion of age and race/ethnicity categories in the overall pregnancy sample.

PAGE 106

106 Table 4 9 . Baseline characteristics of the delivery cohort. Variable Category DM N=15,637 Not DM N=642,848 Age: Years (SD) 30.5 (10.2) 23.0 (6.5) Eligibility: Months (SD) 63.9 (26.2) 53.6 (24.3) Race/ethnicity White, Non Hispanic 42.3% 44.0% Black, Non Hispanic 43.0% 42.8% American Indian or Alaskan Native 1.3% 0.8% Asian or Pacific Islander or Native Hawaiian 1.0% 0.8% Hispanic or Latino 10.0% 10.5% Unknown 2.5% 1.1% Delivery Route Vaginal 60.2% 77.2% Cesarean Section 39.8% 22.8% Preterm Birth 15.1% 10.9% Comorbid Conditions PCOS 2.2% 0.4% Hypertension 20.6% 2.7% Obesity 11.4% 2.4% Infertility 1.0% 0.4% Smok ing 6.7% 4.1% Alcohol abuse 2.9% 1.5% Drug abuse 2.8% 1.8% Region Northeast 10.8% 13.3% Midwest 22.7% 30.3% South 65.9% 55.7% West 0.6% 0.7% * DM = pre existing diabetes mellitus .

PAGE 107

10 7 Table 4 10 . Baseline characteristics and pregnancy characteristics of the delivery cohort stratified by type of diabetes. Variable Category T1DM N=1,02 4 T2DM N=9,5 46 Mixed DM N=5,0 47 Age: Years (SD) Eligibility: Months (SD) Race/ethnicity White, Non Hispanic Black, Non Hispanic American Indian or Alaskan Native Asian or Pacific Islander or Native Hawaiian Hispanic or Latino Unknown Region Northeast Midwest South West Comorbid Conditions PCOS Hypertension Obesity Infertility Smok ing Alcohol abuse Drug abuse Drug Exposure during pregnancy Oral Only Insulin Only Mixed None Delivery Route Vaginal Cesarean Section Preterm Birth * T1DM = type 1 diabetes mellitus, T2DM= type 2 diabetes mellitus, DM= diabetes mellitus Combination of use of insulin, oral, or none throughout pregnancy.

PAGE 108

108 Table 4 11 . Predictors of initiation of anti diabetic agents in the first trimester of pregnancy among women with pre existing diabetes within the Medicaid Analytic eXtract database. Univari ate models Final adjusted model Covariates Odds Ratio 95% Confidence limits p value Odds Ratio 95% Confidence limits p value Race / ethnicity Black vs. white 0.78 0.70 0.86 <0.0001 0.76 0.68 0.84 <0.0001 American Indian vs. white 0.40 0.28 0.59 <0.0001 0.38 0.26 0.55 <0.0001 Asian vs. white 1.29 0.73 2.29 0.3661 1.21 0.68 2.17 0.5177 Hispanic vs. white 0.73 0.62 0.86 0.0001 0.73 0.62 0.86 0.0002 Unknown vs. white 1.75 1.22 2.52 0.0023 1.47 1.02 2.12 0.0381 Maternal Age Under 20 vs. 20 29 1.35 1.15 1.60 0.0003 1.37 1.16 1.62 0.0002 30 39 vs. 20 29 1.07 0.95 1.20 0.2703 1.04 0.93 1.18 0.4164 40+ vs 20 29 1.74 1.52 1.98 <0.0001 1.60 1.39 1.84 <0.0001 Region Northeast vs. south 1.06 0.90 1.23 0.4877 Midwest vs. south 0.86 0.76 0.96 0.0103 West vs. south 0.85 0.48 1.50 0.5744 Delivery Year 2001 vs. 2000 1.19 0.87 1.63 0.2672 2002 vs. 2000 1.25 0.92 1.70 0.1499 2003 vs. 2000 1.19 0.88 1.61 0.2591 2004 vs. 2000 1.16 0.85 1.58 0.3426 2005 vs. 2000 1.01 0.74 1.38 0.9553 2006 vs. 2000 1.19 0.87 1.63 0.2762 Comorbid Conditions PCOS 0.53 0.39 0.73 <0.0001 0.55 0.39 0.77 0.0005 Hypertension 1.29 1.16 1.45 <0.0001 1.18 1.05 1.34 0.0067 Obesity 0.92 0.78 1.09 0.3525 Infertility 0.67 0.40 1.11 0.1199 Smoking 1.04 0.83 1.29 0.7560 Alcohol abuse 1.09 0.78 1.54 0.6096 Drug abuse 0.95 0.67 1.33 0.7586

PAGE 109

109 Table 4 12 . Baseline characteristics of the safety cohort categorized by anti diabetic exposure. Variable Category Only metformin N=514 Only sulfonylureas N=424 Only thiazolidinediones N=176 Only insulin N=497 Metformin & sulfonylureas N=227 Metformin & thiazolidinediones N=142 Age: Years (SD) 35.5 (10.6) 40.9 (9.8) 39.8 (9.7) 32.6 (7.3) 41.6 (9.6) 36.9 (10.2) Eligibility: Months (SD) 62.1 (28.5) 65.7 (28.5) 67.6 (27.8) 56.4 (26.4) 67.1 (30.2) 63.9 (28.8) Race/ ethnicity White, Non Hispanic 47.3% 44.8% 48.9% 37.6% 35.2% 45.1% Black, Non Hispanic 35.8% 42.2% 41.5% 41.9% 47.6% 40.8% Asian or American Indian or Alaskan Native 3.3% 2.1% 1.1% 2.6% 2.6% 4.2% Hispanic or Latino 10.5% 7.1% 5.1% 16.3% 9.7% 7.0% Unknown 3.1% 3.8% 3.4% 1.6% 4.8% 2.8% Region Northeast 11.7% 7.5% 4.5% 10.5% 11.0% 9.2% Midwest 16.9% 8.3% 5.7% 19.9% 11.0% 14.8% South 69.8% 84.2% 89.2% 61.0% 77.5% 72.5% West 1.6% 0.0% 0.6% 0.4% 0.4% 3.5% Uncontrolled Diabetes Mild 24.3% 23.8% 24.4% 34.4% 32.6% 33.8% Severe 0.2% 0.2% 0.0% 0.4% 0.0% 0.0%

PAGE 110

110 Table 4 1 2 . Continued. Variable Category Only metformin N=514 Only sulfonylureas N=424 Only thiazolidinediones N=176 Only insulin N=497 Metformin & sulfonylureas N=227 Metformin & thiazolidinediones N=142 Comorbid Conditions during baseline PCOS 7.0% 0.2% 2.8% 2.2% 0.4% 2.1% Hypertension 25.5% 34.2% 39.2% 25.8% 35.2% 35.9% Obesity 11.1% 7.1% 8.5% 11.7% 9.7% 15.5% VTE 1.8% 1.9% 2.3% 1.0% 0.9% 0.7% Diabetic Nephropathy 0.8% 0.7% 2.3% 1.2% 1.8% 0.7% Chronic Kidney Disease 0.0% 0.2% 0.0% 0.0% 0.0% 0.0% Anemia 4.9% 7.8% 11.4% 6.4% 9.7% 7.7% Depression 29.2% 26.9% 26.1% 21.9% 22.9% 28.2% Herpes Simplex Virus 1.0% 0.9% 0.6% 1.2% 0.9% 0.7% Rheumatoid arthritis 1.6% 3.1% 2.3% 1.0% 0.9% 2.1% Lupus 0.4% 0.7% 0.6% 0.4% 0.4% 0.0% Multiple sclerosis 0.0% 0.5% 0.6% 0.6% 0.0% 0.7% IBD 0.0% 0.0% 0.6% 0.2% 0.0% 0.7% Smoke 3.1% 5.2% 4.5% 3.0% 5.7% 2.8% Alcohol 2.1% 2.8% 1.1% 1.4% 2.6% 2.8% Drug abuse 1.2% 1.9% 1.1% 1.2% 0.9% 3.5%

PAGE 111

111 Table 4 1 2 . Continued. Variable Category Only metformin N=514 Only sulfonylureas N=424 Only thiazolidinediones N=176 Only insulin N=497 Metformin & sulfonylureas N=227 Metformin & thiazolidinediones N=142 Anti hypertensive prescription filled during baseline Alpha agonist 1.0% 6.4% 2.3% 2.6% 3.1% 1.4% ACE inhibitor 29.0% 39.6% 36.9% 33.4% 43.6% 43.7% ARB 5.3% 6.6% 11.4% 5.4% 8.4% 12.7% Beta blocker 13.0% 16.0% 15.9% 6.8% 20.7% 8.5% Ca channel blocker 8.4% 11.1% 16.5% 7.2% 6.6% 10.6% Diuretic 24.9% 34.4% 40.9% 19.3% 29.5% 28.2% Statin prescription filled during baseline 15.8% 21.9% 23.9% 15.1% 24.2% 31.0% Anti depressant prescription filled during baseline 40.9% 40.3% 43.8% 33.6% 36.6% 44.4%

PAGE 112

112 Table 4 13 . Risk of premature delivery among live deliveries exposed to anti diabetic agents during the first trimester of pregnancy. Odds Ratio (95% CI) Adjusted Event Risk Unadjusted P value Age P value Propensity Score P value Mono therapy Metformin 5.87 % 1.00 1.00 1.00 Sulfonylureas 6. 52 % 0.99 (0.59, 1.70) 0.9900 1.56 (0.89, 2.75) 0.1223 2.05 (1.09, 3.83) 0.0252 Thiazolidinedione s 4.05 % 0.62 (0.27, 1.42) 0.2552 0.80 (0.33, 1.91) 0.6170 0.85 (0.36, 2.01) 0.7129 Insulin 14.6 0 % 2.44 (1.59, 3.76) <0.0001 2.16 (1.36, 3.42) 0.0010 2.59 (1.61, 4.17) <0.0001 Dual therapy Metformin & Sulfonylureas 4.15 % 1.00 1.00 1.00 Metformin & Thiazolidinedione s 5.04 % 1.64 (0.63, 4.23) 0.3072 1.32 (0.49, 3.55) 0.5845 1.05 (0.37, 3.01) 0.9219

PAGE 113

113 Table 4 14 . Risk of cesarean section delivery identified among live births exposed to anti diabetic agents during the first trimester of pregnancy. Odds Ratio (95% CI) Adjusted Event Risk Unadjusted P value Age P value Propensity Score P value Mono therapy Metformin 27.04% 1.00 1.00 1.00 Sulfonylureas 27. 78 % 1.04 (0.78, 1.39) 0.7912 1.03 (0.77, 1.38) 0.8579 0.94 (0.68, 1.30) 0.7131 Thiazolidinedione s 3 0 . 06 % 1.20 (0.82, 1.75) 0.3345 1.19 (0.81, 1.74) 0.3712 0.98 (0.65, 1.47) 0.9225 Insulin 34. 08 % 1.42 (1.08, 1.85) 0.0117 1.43 (1.09, 1.88) 0.0097 1.41 (1.05, 1.89) 0.0207 Dual therapy Metformin & Sulfonylureas 27 . 19 % 1.00 1.00 Metformin & Thiazolidinedione s 25 . 21 % 1.01 (0.62, 1.63) 0.9712 0.98 (0.60, 1.61) 0.9403 0.89 (0.49, 1.61) 0.6994

PAGE 114

114 Table 4 15 . Risk of preecl a mpsia identified among live deliveries exposed to anti diabetic agents during the first trimester of pregnancy. Hazard Ratio (95% CI) Adjusted Unadjusted P value Age P value P r opensity Score P value Mono therapy Metformin 1.00 1.00 1.00 Sulfonylureas 0.71 (0.46, 1.10) 0.1238 0.66 (0.33, 1.31) 0.0931 0.61 (0.36, 1.04) 0.0683 Thiazolidinedione s 0.65 (0.38, 1.09) 0.1026 0.48 (0.26, 0.87) 0.0157 0.58 (0.32, 1.04) 0.066 0 Insulin 0.88 (0.60, 1.29) 0.4982 0.77 (0. 44 , 1. 35 ) 0.2199 0. 86 (0.5 6 , 1. 32 ) 0. 4851 Dual therapy Metformin & Sulfonylureas 1.00 1.00 1.00 Metformin & Thiazolidinedione s 1.57 (0.82, 3.00) 0.171 0 9.97 ( 1.49 , 49.41) 0.0 176 1.74 (0.78, 3.89) 0.1742

PAGE 115

115 Table 4 16 . Exploratory analyses: Assessment of the impact of requiring a live birth in the comparative safety analysis . Sensitivity Analysis Outcome Exposure Safety Study Low Medium High Preterm birth (OR) Sulfonylureas 2.05 (1.09, 3.83) 0.98 (0.55, 1.72) 0.98 (0.50, 1.93) 0.92 (0.48, 1.77) Thiazolidinedione s 0.85 (0.36, 2.01) 0.71 (0.34, 1.47) 0.69 (0.30, 1.61) 0.62 (0.27, 1.42) Insulin 2.59 (1.61, 4.17) 0 . 7 5 (0.36, 1.56) 1.46 (0.75, 2.85) 1.34 (0.68, 2.62) Metformin & Thiazolidinedione s 1.05 (0.37, 3.01) 2.15 (0.89, 5.20) 0.99 (0.36, 2.74) 0.95 (0.33, 2.63) Cesarean Section (OR) Sulfonylureas 0.94 (0.68, 1.30) 1.40 (0.96, 2.04) 1.16 (0.81, 1.66) 1.13 (0.79, 1.63) Thiazolidinedione s 0.98 (0.65, 1.47) 2.12 (1.35, 3.32) 1.56 (1.02, 2.36) 1.49 (0.98, 2.27) Insulin 1.41 (1.05, 1.89) 1.23 (0.77, 1.99) 1.13 (0.71, 1.80) 1.15 (0.72, 1.83) Metformin & Thiazolidinedione s 0.89 (0.49, 1.61) 1.00 (0.99, 1.01) 0.94 (0.52, 1.69) 0.88 (0.47, 1.63) Preeclampsia (HR) Sulfonylureas 0.61 (0.36, 1.04) 0.89 (0.60, 1.34) 1.12 (0.69, 1.82) 1.12 (0.68, 1.83) Thiazolidinedione s 0.58 (0.32, 1.04) 0.78 (0.51, 1.19) 0.86 (0.48, 1.52) 0.69 (0.36, 1.33) Insulin 0.91 (0.58, 1.44) 0.78 (0.47, 1.30) 1.18 (0.66, 2.12) 1.20 (0.66, 2.17) Metformin & Thiazolidinedione s 1.74 (0.78, 3.89) 1.58 (0.85, 2.95) 1.07 (0.60, 1.92) 1.34 (0.65, 2.75) * V ariations of the delivery date algorithm identif ied the start of pregnancy for live birth, preterm birth, abortion and stillbirth. T he average pregnancy lengths for the pregnancies resulting in an abortion or stillbirth were var ied to test the impact on the risk estimates. L ow = length of pregnancy as 91 days for abortions and 168 days for stillbirths, medium = 105 days for abortions and 196 days for stillbirth, and high = 168 days for abortions and 238 days for stillbirths .

PAGE 116

116 CHAPTER 5 DISCUSSION Part I: Validation of M other I nfant L inkage U sing the Medicaid Case ID Number V ariable within the Medicaid Analytic eXtract D atabase The increased use of administrative databases for pregnancy research requires the development of methods to link multiple databases together, or if possible , link ing mothers to infants with in one database, in order to obtain comprehensive health information . Ideally, the use of unique identifiers would allow researchers to link maternal record s to birth certificates; however, with the implementation of strict privacy laws, this is often an impossible feat. Our analysis suggests that using the Medicaid Case ID variable and date of birth/delivery range s results in a successful linkage of m others to infants in MAX. Other studies have utilized the Medicaid Case ID variable to link mothers to infants within the MAX database, but none ha ve validated th e linkage using birth certificates records . 133,158 160 In this study, we validated the Medicaid Case ID variable in the FL MAX database by using the F L BCR data base . We identified 59% of the possible deliveries in FL BCR in the FL MAX database . O nly 55% of deliveries included in the Medicaid cohort had a Medicaid Case ID variable associated with it, but 77% of infants included a valid Medicaid Case ID variable. We wer e able to link mothers to infants by Case ID number algorithm for 17.8% of the identified deliveries in the FL BCR that linked to FL MAX . To improve the sensitivity and specificity of the linkage , we varied the requirement of continuous Medicaid eligibili ty f or both the mother and infant. We found that requiring the mother to have continuous Medicaid eligibility from 12 months before delivery and infants to be eligible at birth produce d the highest sensitivity (73.7%, 95%

PAGE 117

117 CI : 73.5%, 73.8% ) and specificity ( 99.9 %, 95% CI: 99.8%, 99.9% ) for the mother infant match. While this eligibility requirement will result in a loss of sample size, it produces the most useful scenario for pregnancy research, because the full exposure history of the mother during pregnan cy can be examined and immediate birth outcomes concerning the infant are ascertainable. We also identified factors that influenced the Medicaid Case ID number algorithm performance using limited Medicaid eligibility requirements ( mothers only during the month of delivery and infants were required to have one month Medicaid eligibility at any time throughout the study period ) . Term infants, infants without living siblings at birth, and women older than 40 years old were more likely to result in a true linkage. This analysis indicates that attention is warranted when using the Medicaid Case ID variable to link mother and infant, as there is the potential to under represent certain populations. Our study was subject to additional limitations. Since we lin ked to infant records, the validation cohort is limited to pregnancies that result ed in a live birth, thus excluding all other pregnanc y outcomes (i.e. spontaneous abortion, elective abortion, and stillbirth). We were only able to link approximately 2 2 % of the MAX identified deliveries via Medicaid Case ID number , indicating a drastic decrease in the usable sample size and generalizability of study results, if the linkage was to be used in an analysis. Nevertheless, using the Medicaid Case ID number algorit hm to create the mother infant pairs may reduce the occurrence of false positive deliveries instead of identifying deliveries from the mother only. Using the linkage can also improve the validity of the

PAGE 118

118 last menstrual period date assignment, and allow for the investigation of infant outcomes. Our evaluation was limited to only mothers and infants with valid SSN in both FL BCR and FL MAX. We also limit ed the analysis to mothers and infants who had fee for service benefits in MAX. The restriction to fee for service only was done to assure comprehensive access to all claims, but result ed in excluding an increasing number of women and children d ue to Medicaid managed care penetration. Furthermore, t he definition and availability of the Medicaid Case ID variable varies across each US state. Currently, we were limited to using only FL birth certificates. As reported by Palmsten et al., the prevalence of deliveries linked via Medicaid Case ID variable varied greatly across each of the individual US states. 133 Additional validation studies will be needed to ensure the generalizability of our findings to other states. Part II: Epidemiology of P re existing D iabetes in P regnancy from 2000 2006 among Medicaid P atients in 29 US S tates We identified all pregnancies in the MAX database for 29 US states from 2000 to 2006. T he overall prevalence of pre existing diabetes was 3.6% and the state specific prevalence of pre existing diabetes varied from 1.8% to 7.2%. The age and race/ethnicity adjusted prevalence of pre existing diabetes during pregnancy increased from 3.1% in 2000 to 4.4% in 2006 (p<0.0001). The age specific prevalence increased 64.2% from 2000 to 2006 for women in categories aged between 30 to 39 years and 43.4% for women aged between 20 to 29 years. Similar to the literature, we found a small er (i.e. 15.8%) but significant increase in the prevalence of pre existing diabetes in 3 The significant increase in the prevalence of pre existing

PAGE 119

119 diabetes in pregnancies of younger women (i.e. 20 39 years) is evidence of an increasing onset of type 2 diabetes in earlier age (i.e. during reproductive years) . Unlike previous studies that explore d the temporal trends of diabetes in pregnancy, we did not restrict to our sample t o only pregnancies resulting in live births. 3 5 Furthermore, o ur prevalence estimate focused on the prevalence of pre existing diabetes, rather than estimating the combined prevalence of gestat ional and pre existing diabetes . 3 5 Finally , we were able to estimate the prevalence in a large and geographically diverse population by utilizing 29 Medicaid states. To our knowledge, Albrecht et al conducted the only other study that has documented national trends in pre existing diabetes . The authors estimated an overall prevalence of pre existing diabetes of 1.3% in their study cohort, and when they appl ied their approach to other published estimates, they calculated a 0.4% and 0.5% prevalence of pre existing diabetes in two studies . 3 5 These estimates of pre existing diabetes are a third of our estimate (3.6%), but the authors utilized hospital discharge data, which required a live birth and only identified pre existing dia betes from the delivery codes , potentially underestimating the true prevalence . 142,161,162 We conducted this study using a large subset of the MAX administrative database. MAX allowed us to investigate the prevalence of pre existing diabetes in pregnancy in a large and geographically diverse cohort of m ore than 1.2 million pregnant women , from 29 US states, with medical claims. Our study period spanned seven years, allowing us to examine the temporal trends of pre existing diabetes a ffecting pregnancy. Utilizing the longitudinal data for all included women allowed us to distinguish pre existing from gestational diabetes and to consider other pregnanc y

PAGE 120

120 outcomes (i.e. pregnancies ending in spontaneous abortion, abortion, or stillbirth) that are a ffected by pre existing diabetes. Our study is not without limitations. The results of this study are only representative of pregnant women in the Medicaid fee for service population . Medicaid is the largest payer of maternity related health services, covering 4 in 10 births, a nd in several states paying for more than half of the total births . 116 118 However, our focus on the fee for service population (to assure comprehensive access to all claims) excludes an increasing number of women especially in more recent years due to an increasing Medicaid managed care penetration . The small drop in the prevalence of pre existing diabetes in 20 03, could be a feature of the differential shift of women with chronic diseases into Medicaid managed care , which would artificially reduce the prevalence of pre existing diabetes over time . Th e analysis of this study was based on automated healthcare administration claims that are intended for reimbursement purposes, not electronic medical records , there by limiting the validity of our outcome (i.e. pre existing diabetes) identified by ICD 9 CM codes . Nevertheless, we util ized a validated algorithm to identify diabetes with a sensitivity of more than 75% and a positive predictive value of greater than 90 %. 123 125 Similar to previous research, we presented the prevalence of type 1 and 2 diabetes as a stratified analysis because of the limited granularity of ICD 9 CM codes, and because previous studies have been largely unsuccessful in validating algorithms to distinguish the typing of diabetes based solely on ICD 9 CM codes. 122,143,144,150,161,163

PAGE 121

121 Part III: Utilization of A nti diabetic A gents in P regnancy from 2000 2006 among Medicaid P atients in 29 US S tates We conducted a descriptive analysis o f anti diabetic drug utilization before and during pregnancy in women with pre existing diabetes in 29 state s of the US Medicaid program between 2000 and 2006. Based on our algorithm to identify pre existing diabetes, we observed an overall diabetes prevalence of 2.4% in live deliveries . Diabetic women were on average older, had higher prevalence of additional co morbid conditions, higher prevalence of preterm birth and cesarean section deliveries compared to non diabetic women with a live birth. We stratified our analysis by diabetes type to obtain characteristics of anti diabetic utilization during pregnancy by disease type . As expected, women categorized with type 1 diabetes only had a low prevalence of oral anti diabetic exposure (2.0%). However, they also show a high prevalence of no anti diabetic treatment (43%) , which is highly unlikely because of insulin de pendence . This find ing further supports previous literature in the conclusion that distinguishing diabetes type solely based on the fifth digit of the ICD 9 CM code is not a valid method. 122,143,144,1 63 Women assumed to be diagnosed with type 2 diabet es exhibited anticipated anti diabetic drug utilization patterns . Pharmacological interventions are currently recommended for the initial therapy of diabetes. 164 This is reflected by the 2007 2009 National Health Interview Survey, which reported that only 16% of all adults with a diagnosis of diabetes do not take either insulin or oral medication. 165 T he discovery that nearly 40% of diabetic women, with medical insurance before pregnancy, did not fill any anti diabetic prescriptions during pregnancy is might be co ncerning. However, it was not until 2007 that the American

PAGE 122

122 Diabetes Association first published the recommendation for early intervention with metformin in combination with lifestyle changes as the primary intervention at the time of diagnosis. 166 Thus, the push of pharmacological interventions at the time of diagnosis did not occur until after our study period, and may explain the high prevalence of no treatment with an anti diabetic agent, both before and during pregnancy. S ince we were utilizing prescript ion claims data for our analysis, we cannot be certain how many women were prescribed anti diabetic drugs but never filled their prescriptions . Nevertheless, a similar study that was conducted in commercial health plan data from 1999 to 2009 found a simila r prevalence of non treatment during the pre conception period. 150 The use of anti diabetic drugs in the preconception period increased from 27% in 2000 to 36% in 2006, whereas the use of these drugs during pregnancy increased from 47% in 2000 to 54% in 2006 . From Figure 4 8 , it is evident that although all of the drug classes s aw an increase in prevalence from the pre conception period to the first trimester of pregnancy, insuli n observed the largest increase at 9%. Furthermore, of those who initiated treatment in the first trimester approximately 60% initiated treatment of ins ulin only . Although the reasons for initiation are unknown, it is possibly an indication of physicians following guidelines in prescribing insulin to manage glycemic control in pregnant women, as insulin is currently the only FDA approve d drug to treat dia betes in pregnancy. 30 Previously published literature suggests that in women with pre existing diabetes, glucose control does not deteriorate during the first trimester but does in the later trimesters of pregnancy. 42,167 W e e xplored the characteristics of women who i nitiated

PAGE 123

123 anti diabetic treatment compared to those who remained untreated in the first trimester of pregnancy . Our results indicate that additional factors, other than pre existing diabetes, that might contribute to a high risk pregnancy also are indicators for the initiation of anti diabetic treatment in the first trimester of pregnancy. Interestingly, we observed that those with a diagnosis of PCOS were less likely to initiate anti diabetic therapy in the first trimester of pregnancy, even though the current literature hypothesizes that exposure to anti diabetic agents throughout pregnancy in women with PCOS will improve the outcomes of pregnancy (e.g. decrease risk of spontaneous abortion or preterm birth). 50,69,82 Our findings with PCOS could be an artifact of physicians preference to not introduce additional medications in the critical period of fetal development when a woman is well controlled with life style modifications. However, anti diabetics , specifically metformin, have been found to have beneficial effect s on metabolic outcomes in women with hypertension . 30,46,49 This would be particularly important because several anti hypertensive agents are contraindicated in Our study had a number of limitations. We report results for a Medicaid po pulation that was eligible for fee for service benefits, which may not be representative of pre existing diabetes treatment pattern during pregnancy in the US. We did not have information on gestational length and last menstrual period (LMP) in the MAX dat a , which introduces the possibility of misclassification of the trimesters of pregnancy. However, we utilized validated algorithms to identify the calculated LMP, and the employment of a majority rule to determine exposure for each window likely mean s the misclassification is minor. 124,137,168 Finally, the inability to measure disease severity in

PAGE 124

124 our analysis of the initiation of anti diabetic treatment introduces a significant unmeasured confounder. Addi tional studies are needed to investigate the role glycemic control plays in the initiation of anti diabetic treatment in the first trimester of pregnancy. Part IV: Comparative S afety of O ral A nti diabetic A gents in P regnancy from 2000 20 06 among Medicaid P atients in 29 US S tates We assessed the comparative safety of anti diabetic drugs during the first trimester of pregnancy in women aged 12 to 55 years eligible for in Medicaid fee for service benefi ts . Unlike previous studies that explored the safety of oral anti diabetic agents, we restrict ed our study to women with pre existing diabetes and focused on the obstetric outcomes as primary analyses. 19,23,76,96,97 We found that t here were varying degrees of risk in the targeted obstetric outcomes with the commonly prescribed anti diabetic drugs. Preterm B irth We know from previous literature that women with pre existing diabet es are at higher risk of preterm delivery . 42 However, the comparative safety of various anti diabetic agents in this regard have not been assess ed . We found that premature delivery was substatially more likely in women exposed to sulfonyulrea s or insulin , as compared to metformin as a mono therapy . Our results further support the finding identified in PCOS research that metformin exposure in the first tr imester of pregnancy may reduce the risk of preterm birth. 56,82 Likewise , thiazolidinediones have also been found to im prove pregnancy outcomes in women with PCOS, mirroring the trend for a decreased risk in our study . 84 Because confidence intervals for thiazolidinedione or dual therapy of metformin and thiazolidinedione were wide, additional stud ies of their comparative risk are needed .

PAGE 125

125 Our results differed from those identified by the MiG and the glyburide randomized control t rials (RCTs) with regards to the risk of preterm birth , in that we found an increased risk of preterm birth in insulin users compared to metformin, as well as an increased risk in sulfonylureas compared to metformin , while the RCTs did not find a differenc e in risk . 19,76,96 Unlike our study, RCTs observed the risk of preterm birth in women with gestational diabetes, thus , anti diabetic exposure was limited to the second and/or third trime ster of pregnancy. 19,76,96 Furthermore , within our study we had a larger proportion of women with hypertension and different baseline risks for both preeclampsia and preterm birth compared to the RCTs , which could account for the higher risk of preterm birth we found in those exposed to sulfonylureas or insulin as compare to the literature . 76,96 The glyburide RCTs anal yzed the risk associate d with the exposure to glyburide only, whereas we observed the risk associated with the therapeutic drug class of sulfonylureas , not just deliveries exposed to glyburide . Thus , the fact that the glyburide RCT s did not find a differe nce in risk of preterm birth and we did may be attributed to several factors including the disease etiology, exposure period, or type of drug (i.e. glyburide vs. glipizide). Cesarean S ection D elivery Cesarean section deliveries are performed for a number of reasons including maternal preference and medical need, which indicate a necess ity for ensuring optimal maternal and neonatal outcomes. 169 171 There is agreement among researchers that the cesarean section delivery rate is higher in women with diabetes , and that diabetes mellitus is an independent risk for cesarean section delivery . 15,23,42,99,172 174 However , there is limited information on the comparative safety of oral anti diabetic treatment and the associated risk of cesarean section delivery . 42,99

PAGE 126

126 Within our safety study, we found no marked difference in the odds of cesarean section delivery between sulfonylureas or thiazolidinedione when compared to metformin . Whereas , w omen exposed to insulin in the first trimester had 41% higher od ds of cesarean section delivery compared to metformin. These results reflect the incidence rates identified by Hellm uth et al., who found no difference in the incidence of cesarean section delivery in metformin or sulfonylurea users (30% vs 28%), while ins ulin users had a slightly increased rate of 38%. 21 The literature provides substantial evidence that women with pre existing diabetes are at increased risk of cesarean section delivery, however the direct cause of this association is still unknown. We know that physicians are inclined to be cautious with w omen who are considered high risk in pregnancy (i.e. women with pre existing diabetes) . Specifically, p hysicians are concerned about the increased risk of stillbirth, child birth in women with pre existing diabetes . Thus , the lack of difference in risk among the oral anti diabetic agents could be due to physician preference to conduct a cesarean section delivery regardless of exposure status . Recently, a study found that wom en with diabetes have impaired uterine contractility , increasing the need for cesarean sections . 175 This case control study looked at women with type 1 and gestational diabetes compared to match ed non diabetic s to ex plore the contractility of the myometrium. 175 The majority of diabetic women in the study were treated with insulin ; however , the authors conducted an analysis to see if there was a difference in the contractility amo ng women treated with insulin (n=31) compared to diet (n=9) . 175 The fact that we did not see a difference in risk

PAGE 127

127 among the oral anti diabetic agents, but did in the insulin users could be due to the effect the exposu res (i.e. anti diabetic agents) have on contractility (i.e. insulin decreases the contratility, whereas oral anti agent have minimal effects). Further studies are needed to explore the effect that anti diabetic agents have on the impaired uterine contracti lity. Preeclampsia Characteristics of preeclampsia are h igh blood pressure and signs of damage to other organ systems, often the kidneys. The exact cause of preeclampsia is unknown, however, it is accepted that preeclampsia is an endothelial cell disorder (i.e. narrowing of the placental blood vessels). Key factors for the control of endothelial cell function include insulin effectiveness and angiogenesis (i.e. new blood vessel formation) . T hus , in women w ith pre existing diabetes the decreased insulin sensitivity enhances the defects in new blood vessel formation therefore increasing the risk of preeclampsia. 176 178 Dunne et al. reported that preeclampsia was two times more common in type 2 diabetic women than in non diabetic women. 23 In their analysis, the authors obtained laboratory values , which al lowed categoriz ation of women as having normal, average, or poor glycemic control. In the women identified as having preeclampsia, 94% were categori zed as having normal or average control. This indicates that factors other than glycemic control may increas e the risk of preeclampsia in women with pre existing diabetes. We found that both sulfonylurea and thiazolidinedione exposure in the first trimester of pregnancy resulted in a decreased risk of preeclampsia when compared to metformin only. Our results are consistent with other research that suggests an

PAGE 128

128 increased risk of preeclampsia in women treated with metformin in the first trimester. 19,21,70,179 This result may be perplexing because metformin tends to improve blood pressure in men and non pregnant women with diabetes, as well as the fact that preeclampsia is a major indication for preterm birth, an d metformin has been found to 56,82,179 However, when we consider the fact that preeclampsia is an endothelial cell disorder, which is a ffected by insulin sensitivity, it is understandable how different anti diabetic drug expo sure s may affect the risk of preeclampsia. Specially, metformin acts mainly by decreasing hepatic glucose production, thereby reducing glucose levels without the stimulation of insulin secretion. However, sulfonylureas stimulate insulin secretion by the pa cell and thiazolidinedione s improve insulin sensitivity through the stimulation of a nuclear receptor, PPAR . Thus, because metformin does not stimulate insulin secretion , it is plausible that the different mechanisms of actions for each anti diabetic drug class differentially affect insulin effectiveness and thus angiogenesis , creating a difference in the risk of preeclampsia. Further studies are needed to explore the effect of anti diabetic agents have on the risk of preeclampsia. Exploratory Analysis Previous literature has reported the increase d risk of stillbirth and sponanteous (or elective) abortion in women with pre exisiting diabetes ; however , limited information is available on anti diabetic drug exposure in th e first trimester of pregnancy. 7,23,42 Our exploratory analysis allowed us to investigate a fuller safety profile of the anti diabetic agents, as well as to quantify the effect of exposures on the probability of live birth . 155,180 We found differences in the comparative safety estimates when we considered the additional risks of other pregnancy outcomes , however only thiazolidinedione s in

PAGE 129

129 c esarean section deliveries produced a statistically significant risk . Nevertheless, caution is recommended with i nter preting the results of our exploratory analyses. We were only able to capture abortions and stillbirth s that resulted in a medical claim wi th respective ICD 9 CM coding . Since approximately 25% of pregnancies are lost before clinical recognition, it is possible that we are masking the true risk of exposure and even introduce misclassification bias. Also, we were unable to differentiate betwee n induced abortion and spontaneous abortions. This is particularly important because induced abortions should not be considered when exploring the risk of spontaneous abortions . P regnancies destined for induced abortions can contribute to spontaneous abortions but not births, therefore distorting the risk of spontaneous abortions. 155 Furthermore, by identifying the start of pregnancy from average pregnancy lengths , we may have introduced misclassification of the exposure period. However, w e estimated the effect of misclassification of the first trimester on our outcome estimates by conducting several sensitivity analyses, which showed little influence on the outcome estimates. 181 Finally, there is a possibility of reverse causation when considering the risk of the other pregnancy outcomes ( stillbirth, spontaneous or elective abortion ) in the compara tive safety estimates. For example, metformin has been found to reduce the risk of spontaneous abortions, thus women with an increased risk of spontaneous abortions could be using metformin for its effective ness in glucose control, eas e of administration , assumed safe ty in pregnancy, and lower risk of other pregnancy outcome in high risk pregnancies . Strengths and L imitations Our study had several strengths. T his is to our knowledge, the first comparative safety study addressing the association of oral anti diabetic agents and adverse

PAGE 130

130 obstetric outcomes . Our study included comprehensive administrative data from seven years, in 29 US states , providing a geograp hically diverse population. The use of Medicaid data permitted us to adress a critical population of lower income and minority pregnant women. Furthermore, it allowed for observation of prescribing practices and healthcare utilization, because access to ca re (e.g. cost) is not a major limitation due to minimal co pays associated with our the Medicaid benefit plans . Several study design features aided in the establish ment of balanced comparison groups , including the requirement of pre existing diabetes diag nosis, restriction of drug exposure to mono or dual therapy with a high propensity for treatment within each exposure group and the broad selection of variables for the propensity score adjustment. In order to minize the potential for misclassification o f our outcomes, we utilized a validate d algorithm with moderate to high specificity for identifying preeclampsia, as well as validat ed algorithms for both preterm birth and cesarean section delivery in MAX data. We found that the codes for cesarean sectio n delivery and preeclampsia produce high specificity, thus we expect a minimally biased risk ratio. The moderate specificity of preterm birth potentially introduces non differential misclassification of the outcome that could effect the risk estimates. All administrative data have limitations and this study is no exception. W e utilized the MAX database because it is the largest payer in maternity related health services and covers approximat ely 50% of all births in the US . N evertheless, due to our strict in clusion criteria for drug exposure, our study struggled with small samples and event rates.

PAGE 131

131 We cannot exclude the possibility of non adherence, because even if a prescription is filled , it does not ensure the utilization of the drug, which introduces the possibility of misclassification. We did explore the average day supply fo r each exposure category and found that with the exception of insulin, the majority of the oral anti diabetic drugs maintained a 30 day supply. This resulted in a regular patt ern of prescription refills allowing for frequent update s of exposure windows . Since laboratory values are not available in claims data, it is unclear whether the women included in our analysis were able to achieve or maintain normoglycemia before or duri ng pregnancy. Our solution to this issue was to restrict comparisons according to therapy regimens, under the assumption that a woman on only mono therapy should be comparable to another woman on only mono therapy as it concerns glycemic control. We furthe r adjusted for additional potential markers of disease severity within the propensity score, which allowed us to balance our exposure groups based on known and adequately measured confounders. Finally, the inability to control for disease severity, because laboratory values are not captured in administrative claims data, introduces a significant unmeasured confounder. We attempted to control for disease severity with strict inclusion criteria, as well as utilizing propensity scores to bala n ces the probabili ty of exposure between the different exposure categories. Even if we were able to measure disease severity, assessments would be challenging, as anti diabetic treatment would likely be adjusted to reflect varying glucose levels throughout pregnancy, result ing in time dependent confounding. We used an ITT analysis to alleviate the issue of time dependent confounding , as diabetic treatment

PAGE 132

132 regimen according to changing glucose management requirements througho ut pregnancy . Unfortunately, because of sample size con s tra i nts we could not conduct a sensitivity analysis on women who continuously utilized the same anti diabetic agent throughout pregnancy until delivery occurred. Conclusion s In our validation analysi s, we were successful in identifying and validating a mother infant linkage algorithm in MAX. The ability to link mothers to infants using the Medicaid Case ID number maximizes the utility of administrative claims data in population based pregnancy researc h and enhances the evaluation of pregnancy outcomes by providing access to the health information of both mother and child. We recommend caution when using this mother infant linkage algorithm, as there is the potential to under represent certain populations (e.g. preterm infants). Our analysis investigating the comparative safety of anti diabetic treatment in pregnant women with pre existing diabetes uncovered important findings. Over the course of the study period , the prevalence of pre existing diabetes in pregnancy increase d approximately 42%. Because women with diabetes might be increasingly shifted to managed care plans, we may be underestimating secular increases of pre existing diabetes in pregnancy. Following the secular increase o f the prevalence of pre existing diabetes in pregnancy, w e saw a rise in the utilization of overall oral anti diabetic drugs both before and during pregnancy . In our comparative safety analysis, the odds of preterm birth were elevated among women exposed to sulfonylureas in the first trimester of pregnancy. In contrast, women exposed to only insulin during the first trimester had an increase d odds of cesarean section delivery. Lastly, we confirmed earlier findings that metformin exposure

PAGE 133

133 in the first trimester may be associated with increased risk of preeclampsia. O ur exploratory analyses highlight the critical need for additional research on the safety o f anti diabetic treatment in relation to other pregnancy outcomes (s pontaneous abortion, stillbirth , or elective abortions) . Understanding the role anti pregnancy outcomes in light of the increasing prevalence pre existing diabetes will permit for more successful treatment regimens for these women, ult imately resulting in healthier mothers and infants and efficient use of health care resources.

PAGE 134

134 APPENDIX A OPERATIONAL DEFINITIONS Pre existing D iabetes M ellitus Pre existing d iabetes mellitus was defined as presence of two or more in or out patient claims with primary or secondary diagnosis of diabetes ( ICD 9 CM: 250. XX ) during the baseline study period . Uncontrolled Diabetes We categorized claims indicating mild uncontrolled diabetes as those with (ICD 9 CM: 250.02, 250.42, 250.52, 250.62, and 250.72 ). Claims indicating severe uncontrolled diabetes included 250.12, 250.13, 250.22, 250.23, 250.32, and 250.33 .

PAGE 135

135 Co morbid Conditions Table A 1 . ICD 9 C M codes for the identification of comorbid conditions during baseline . Condition ICD 9 CM Polycystic ovary syndrome (PCOS) 256.4X, 628.0, 704.1 Hypertension 401.XX, V81.1 Obesity 278.XX, V77.8 Venous Thromboembolism (VTE) 671.3X, 671.4X, 671.5X, 671.9X, 673.2X, 673.8, 451.XX, 452.XX, 453.XX, 415.XX Chronic Kidney Disease (CKD) 585.1 585.9X Diabetic Nephropathy 250.4X Anemia 280.XX 285.XX Depression 296.XX, 300.XX, 309.XX, 311.XX Rheumatoid Arthritis 714.XX, 696.XX Systemic Lupus Erythematosus (SLE) 710.XX Inflammatory Bowel Disease (IBD) 555.XX, 556.XX Multiple Sclerosis 340.XX Alcohol abuse 303.XX, 305.0X, V11.3, V79.1 Cigarette smoking V15.82, 305.1X Illicit drug abuse 305.2X 305.4X, 305.6X 305.9X

PAGE 136

136 Pregnancy We identified pregnancy with 2 in or out patient claims for pregnancy or 1in or out patient claim for delivery, stillbirth, or abortion. Table A 2 . Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of pregnancy . Coding system Cod e s Code definitions Current procedure terminology (CPT 4) 81025 Diagnosis of pregnancy 84163 Pregnancy associated p lasma protein A (PAPP A) 59000 Amniocentesis; diagnostic 59012 Cordocentesis (intrauterine), any method 59015 Chorionic villus sampling, any method 59025 Fetal non stress test 59400 Routine obstetric care including antepartum care 59610 Routine obstetric care including antepartum care after previous cesarean delivery 59618 Routine obstetric care including antepartum care following attempted vaginal delivery after previous cesarean delivery ICD 9 CM (procedures) V22 Supervision of normal pregnancy V23 Supervision of high risk pregnancy V72.42 Pregnancy examination or test, positive result ICD 9 CM (diagnoses) 634 639 Other pregnancy with abortive outcome 640 649 Complications mainly related to pregnancy 678 679 Other maternal and fetal complications

PAGE 137

137 Trimesters Pregnancy is measured in trimesters , f ro menstrual period (referred to as conception) to the birth of the infant (typically 40 weeks gestation). First trimester: Includes the first day of the calculated last menstrual period to the end of the 1 3 th week gestation (90 days) . Second trimester: Includes the start of the 1 4 th week of gestation to the end of the 27 th week of gestation (90 days) . Third trimester: Includes the start of the 28 th week of gestation until birth (65 or 90 days) .

PAGE 138

138 Abortion Table A 3 . Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of abortion (spontaneous or induced)*. Coding system Codes Code definitions Current procedure terminology (CPT 4) 59812 Treatment of incomplete abortion, any trimester, completed surgically 59820 Treatment of missed abortion, completed surgically; first trimester 59821 Treatment of missed abortion, completed surgically; second trimester 59830 Treatment of septic abortion, completed surgically 59840 Induced abortion, by dilation and curettage 59841 Induced abortion, by dilation and evacuation 59850 59852 Induced abortion, by one or more intra amniotic injections (amniocentesis injections ) 59855 59857 Abortion, induction ICD 9 CM (diagnoses) 63 2 638 Abortion (includes miscarriage, spontaneous abortion) 639 Complications following abortion and ectopic molar pregnancies 779.6 Termination of pregnancy (fetus) *Presence of at least one in or outpatient claim with a CPT or ICD 9 CM code in women who had an abortion (spontaneous or induced).

PAGE 139

139 Stillbirth Table A 4 . Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of a stillbirth *. Coding system Codes Code definitions ICD 9 CM (diagnoses) V27. 1 Single stillborn V27.4 Twins, both stillborn V27.3 Twins one live born and one stillborn V27.7 Other multiple birth, all stillborn V32 Twin, mate stillborn V 3 5 Other multiple, mates all stillborn V36 Other multiple, mates live and stillborn * Presence of at least one in or out patient claim with an ICD 9 CM code after 20 weeks of pregnancy, in women who delivered a stillborn infant.

PAGE 140

140 Delivery ( L ive B irth) Table A 5 . Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of a delivery*. Coding system Codes Code definitions Current procedure terminology (CPT 4) 01960 01969 Anesthesia for delivery 5940 9 59410 Vaginal delivery codes 59510 59515 Cesarean delivery codes 59610 59614 Vaginal Birth After Cesarean Section (VBAC) 59618 59622 Cesarean delivery after attempted VBAC 59525 Cesarean delivery with hysterectomy ICD 9 CM (procedures) 72.0 7 5 .4 Procedures assisting delivery 763 763.6 Fetus or newborn affected by other complications of labor and delivery ICD 9 CM (diagnoses) V27 V27.9 Outcome of delivery V30.0 V37.1, V39.0 V39.1 Live born infants according to type of birth 640 648.92 Complications mainly related to pregnancy 650 659 Normal delivery, and other indication for care in pregnancy, labor, and delivery 660 669 Complications occurring mainly in the course of labor and delivery 670 677 Complications of the pue r perium *Presence of at least one in or out patient claim with a CPT or ICD 9 CM code

PAGE 141

141 Pre term B irth Presence of at least one in or out patient claim with an ICD 9 CM code: 644.0X, 644.42, 765.0X, 765.1X, 765.20, 765.21, 765.22, 765.23, 765.24, 765.25, 765.27, 765.28. Cesarean S ection D elivery Table A 6 . Current procedure terminology (CPT 4) and ICD 9 CM codes for the identification of a cesarean section delivery.* Coding system Codes Code definitions Current procedure terminology (CPT 4) 01961, 01963, 01968 9 Anesthesia for delivery 59510 59515 Cesarean delivery codes 59618 59622 Cesarean delivery after attempted VBAC ICD 9 CM (procedures) 74 .XX Procedures assisting delivery ICD 9 CM (diagnoses) 763.4 Cesarean delivery 669.7 Cesarean delivery, without mention of indication * Presence of at least one in or out patient claim with a CPT or ICD 9 CM code Preeclampsia Preeclampsia is defined by the presence of hypertension and proteinuria occur r ing after the 20 th week of gestation. 105 We operationalized preeclampsia as presence of at leas t 1 in or out patient claim with any of the following diagnosis codes (ICD 9 CM: 642.4x, 642.5x, 642.6x, or 642.7x) between gestation week 2 1 (20 weeks 1 day) to delivery .

PAGE 142

142 APPENDIX B SUPPLEMENTAL TABLES Table B 1 . List of counties in Florida County Region County Region ALACHUA N LAKE C BAKER N LEE SW BAY NW LEON NW BRADFORD N LEVY N BREVARD C LIBERTY NW BROWARD SE MADISON N CALHOUN NW MANATEE SW CHARLOTTE SW MARION C CITRUS C MARTIN SE CLAY N MONROE SE COLLIER SW NASSAU N COLUMBIA N OKALOOSA NW DADE SE OKEECHOBEE SW DESOTO SW ORANGE C DIXIE N OSCEOLA C DUVAL N PALM BEACH SE ESCAMBIA NW PASCO C FLAGLER C PINELLAS C FRANKLIN NW POLK SW GADSDEN NW PUTNAM N GILCHRIST N SANTA ROSA NW GLADES SW SARASOTA SW GULF NW SEMINOLE C HAMILTON N ST. JOHNS N HARDEE SW ST. LUCIE SE HENDRY SW SUMTER C HERNANDO C SUWANNEE N HIGHLANDS SW TAYLOR N HILLSBOROUGH C UNION N HOLMES NW VOLUSIA C INDIAN RIVER SE WAKULLA NW JACKSON NW WALTON NW JEFFERSON LAFAYETTE NW N WASHINGTON NW

PAGE 143

143 Table B 2 . List of Medicaid Analytic eXtract (MAX) States and Region in analysis State State Abbreviation Region Alabama AL South Arkansas AR South Florida FL South Georgia GA South Iowa IA Midwest Idaho ID West Illinois IL Midwest Indiana IN Midwest Kansas KS Midwest Louisiana LA South Massachusetts MA Northeast Maryland MD South Minnesota MN Midwest Missouri MO Midwest Mississippi MS South North Carolina NC South Nebraska NE Midwest New Hampshire NH Northeast New Jersey NJ Northeast New York NY Northeast Ohio OH Midwest Pennsylvania PA Northeast South Carolina SC South Tennessee TN South Texas TX South Virginia VA South Vermont VT Northeast Wisconsin WI Midwest West Virginia WV South

PAGE 144

144 LIST OF REFERENCES 1. Diagnosed and undiagnosed diabetes in the United States, all ages, 2010. 2011; http://www.cdc.gov/diabetes/pubs/estimates11.htm#1 , 2012. 2. Weindling AM. The confidential enquiry into maternal and child health (CEMACH). Arch Dis Child. Dec 2003;88(12):1034 1037. 3. Lawrence JM, Contreras R, Chen W, Sacks DA. Trends in the prevalence of preexisting diabetes and gestational diabetes mellitus amo ng a racially/ethnically diverse population of pregnant women, 1999 2005. Diabetes Care. May 2008;31(5):899 904. 4. Ferrara A, Kahn HS, Quesenberry CP, Riley C, Hedderson MM. An increase in the incidence of gestational diabetes mellitus: Northern Californi a, 1991 2000. Obstet Gynecol. Mar 2004;103(3):526 533. 5. Dabelea D, Snell Bergeon JK, Hartsfield CL, et al. Increasing prevalence of gestational diabetes mellitus (GDM) over time and by birth cohort: Kaiser Permanente of Colorado GDM Screening Program. Di abetes Care. Mar 2005;28(3):579 584. 6. From the Centers for Disease Control. Perinatal mortality and congenital malformations in infants born to women with insulin dependent diabetes mellitus United States, Canada, and Europe, 1940 1988. JAMA. Jul 1990;2 64(4):437, 441. 7. Dunne F. Type 2 diabetes and pregnancy. Semin Fetal Neonatal Med. Aug 2005;10(4):333 339. 8. Casson IF, Clarke CA, Howard CV, et al. Outcomes of pregnancy in insulin dependent diabetic women: results of a five year population cohort stud y. BMJ. Aug 1997;315(7103):275 278. 9. Ray JG, O'Brien TE, Chan WS. Preconception care and the risk of congenital anomalies in the offspring of women with diabetes mellitus: a meta analysis. QJM. Aug 2001;94(8):435 444. 10. Evers IM, de Valk HW, Visser GH. Risk of complications of pregnancy in women with type 1 diabetes: nationwide prospective study in the Netherlands. BMJ. Apr 2004;328(7445):915. 11. Macintosh MC, Fleming KM, Bailey JA, et al. Perinatal mortality and congenital anomalies in babies of women with type 1 or type 2 diabetes in England, Wales, and Northern Ireland: population based study. BMJ. Jul 2006;333(7560):177.

PAGE 145

145 12. Simmons RK, Unwin N, Griffin SJ. International Diabetes Federation: An update of the evidence concerning the prevention of typ e 2 diabetes. Diabetes Res Clin Pract. Feb 2010;87(2):143 149. 13. Association AD. Preconception care of women with diabetes. Diabetes Care. Jan 2000;23 Suppl 1:S65 68. 14. Kitzmiller JL, Buchanan TA, Kjos S, Combs CA, Ratner RE. Pre conception care of dia betes, congenital malformations, and spontaneous abortions. Diabetes Care. May 1996;19(5):514 541. 15. Jensen DM, Damm P, Moelsted Pedersen L, et al. Outcomes in type 1 diabetic pregnancies: a nationwide, population based study. Diabetes Care. Dec 2004;27( 12):2819 2823. 16. Suhonen L, Hiilesmaa V, Teramo K. Glycaemic control during early pregnancy and fetal malformations in women with type I diabetes mellitus. Diabetologia. Jan 2000;43(1):79 82. 17. Andrade SE, Raebel MA, Morse AN, et al. Use of prescriptio n medications with a potential for fetal harm among pregnant women. Pharmacoepidemiol Drug Saf. Aug 2006;15(8):546 554. 18. Williams A, Modder J. Management of pregnancy complicated by diabetes -maternal glycaemic control during pregnancy and neonatal mana gement. Early Hum Dev. May 2010;86(5):269 273. 19. Rowan JA, Hague WM, Gao W, Battin MR, Moore MP, Investigators MT. Metformin versus insulin for the treatment of gestational diabetes. N Engl J Med. May 2008;358(19):2003 2015. 20. Langer O, Conway DL, Berk us MD, Xenakis EM, Gonzales O. A comparison of glyburide and insulin in women with gestational diabetes mellitus. N Engl J Med. Oct 2000;343(16):1134 1138. 21. Hellmuth E, Damm P, Mølsted Pedersen L. Oral hypoglycaemic agents in 118 diabetic pregnancies. D iabet Med. Jul 2000;17(7):507 511. 22. Homko CJ, Reece EA. Insulins and oral hypoglycemic agents in pregnancy. J Matern Fetal Neonatal Med. Nov 2006;19(11):679 686. 23. Dunne F, Brydon P, Smith K, Gee H. Pregnancy in women with Type 2 diabetes: 12 years outcome data 1990 2002. Diabet Med. Sep 2003;20(9):734 738. 24. Alberti KG, Zimmet PZ. Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus provisional report of a WHO consultation. Diabet Med. Jul 1998;15(7):539 553.

PAGE 146

146 25. Report of the Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. Diabetes Care. Jul 1997;20 (7):1183 1197. 26. Association AD. Diagnosis and classification of diabetes mellitus. Diabetes Care. Jan 2013;36 Suppl 1:S67 74. 27. Shaw JE, Sicree RA, Zimmet PZ. Global estimates of the prevalence of diabetes for 2010 and 2030. Diabetes Res Clin Pract. J an 2010;87(1):4 14. 28. King H, Aubert RE, Herman WH. Global burden of diabetes, 1995 2025: prevalence, numerical estimates, and projections. Diabetes Care. Sep 1998;21(9):1414 1431. 29. Association AD. Economic costs of diabetes in the U.S. in 2012. Diabe tes Care. Apr 2013;36(4):1033 1046. 30. Association AD. Standards of medical care in diabetes -2012. Diabetes Care. Jan 2012;35 Suppl 1:S11 63. 31. Genuth S, Alberti KG, Bennett P, et al. Follow up report on the diagnosis of diabetes mellitus. Diabetes Car e. Nov 2003;26(11):3160 3167. 32. Committee IE. International Expert Committee report on the role of the A1C assay in the diagnosis of diabetes. Diabetes Care. Jul 2009;32(7):1327 1334. 33. Higgins M, Galvin D, McAuliffe F, et al. Pregnancy in women with T ype 1 and Type 2 diabetes in Dublin. Ir J Med Sci. Jun 2011;180(2):469 473. 34. Bald M, Ratjen F, Nikolaizik W, Wingen AM. Ciprofloxacin induced acute renal failure in a patient with cystic fibrosis. Pediatric Infectious Disease Journal. Mar 2001;20(3):320 321. 35. Holing EV, Beyer CS, Brown ZA, Connell FA. Why don't women with diabetes plan their pregnancies? Diabetes Care. Jun 1998;21(6):889 895. 36. Coustan DR. Pre conception planning. The relationship's the thing. Diabetes Care. Jun 1998;21(6):887 888. 37. Janz NK, Herman WH, Becker MP, et al. Diabetes and pregnancy. Factors associated with seeking pre conception care. Diabetes Care. Feb 1995;18(2):157 165. 38. Casele HL, Laifer SA. Factors influencing preconception control of glycemia in diabetic women. Arch Intern Med. Jun 1998;158(12):1321 1324. 39. Association AD. Standards of medical care in diabetes -2009. Diabetes Care. Jan 2009;32 Suppl 1:S13 61.

PAGE 147

147 40. Bulletins ACoP. ACOG Practice Bulletin. Clinical Management Guidelines for Obstetrician Gynecologi sts. Number 60, March 2005. Pregestational diabetes mellitus. Obstet Gynecol. Mar 2005;105(3):675 685. 41. Drugs@FDA. http://www.accessdata.fda.gov/scripts/cder/drugsatfda/ . 42. Leahy J. Diabetes and Carbohydrate Metabolism. In: Barbour L, Friedman J, eds. Management of Diabetes in Pregnancy 2010: http://www.endotext.org/diabetes/index.htm . 43. Refuerzo JS. Oral hypoglycemi c agents in pregnancy. Obstet Gynecol Clin North Am. Jun 2011;38(2):227 234, ix. 44. Gutzin SJ, Kozer E, Magee LA, Feig DS, Koren G. The safety of oral hypoglycemic agents in the first trimester of pregnancy: a meta analysis. Can J Clin Pharmacol. 2003;10( 4):179 183. 45. Feig DS, Briggs GG, Koren G. Oral antidiabetic agents in pregnancy and lactation: a paradigm shift? Ann Pharmacother. Jul 2007;41(7):1174 1180. 46. Krentz AJ, Bailey CJ. Oral antidiabetic agents: current role in type 2 diabetes mellitus. Dr ugs. 2005;65(3):385 411. 47. Bailey CJ, Turner RC. Metformin. N Engl J Med. Feb 1996;334(9):574 579. 48. Kirpichnikov D, McFarlane SI, Sowers JR. Metformin: an update. Ann Intern Med. Jul 2002;137(1):25 33. 49. Howlett HC. Metformin in the management of ty pe 2 diabetes. Diabet Med. Mar 1989;6(2):179 180. 50. Fauser BC, Tarlatzis BC, Rebar RW, et al. Consensus on women's health aspects of polycystic ovary syndrome (PCOS): the Amsterdam ESHRE/ASRM Sponsored 3rd PCOS Consensus Workshop Group. Fertil Steril. Jan 2012;97(1):28 38.e25. 51. Law R, Bozzo P, Koren G, Einarson A. FDA pregnancy risk categories and the CPS: do they help or are they a hindrance? Can Fam Physician. Mar 2010;56(3):239 241. 52. Nicholson W, Baptiste Roberts K. Oral hypoglycaemic agents du ring pregnancy: The evidence for effectiveness and safety. Best Pract Res Clin Obstet Gynaecol. Feb 2011;25(1):51 63. 53. Coetzee EJ, Jackson WP. Pregnancy in established non insulin dependent diabetics. A five and a half year study at Groote Schuur Hospit al. S Afr Med J. Nov 1980;58(20):795 802.

PAGE 148

148 54. Coetzee EJ, Jackson WP. Oral hypoglycaemics in the first trimester and fetal outcome. S Afr Med J. Apr 1984;65(16):635 637. 55. Coetzee EJ, Jackson WP. The management of non insulin dependent diabetes during pr egnancy. Diabetes Res Clin Pract. 1985 1986 Feb 1985;1(5):281 287. 56. Glueck CJ, Wang P, Goldenberg N, Sieve Smith L. Pregnancy outcomes among women with polycystic ovary syndrome treated with metformin. Hum Reprod. Nov 2002;17(11):2858 2864. 57. Glueck C J, Wang P, Kobayashi S, Phillips H, Sieve Smith L. Metformin therapy throughout pregnancy reduces the development of gestational diabetes in women with polycystic ovary syndrome. Fertil Steril. Mar 2002;77(3):520 525. 58. Glueck CJ, Goldenberg N, Streicher P, Wang P. Metformin and gestational diabetes. Curr Diab Rep. Aug 2003;3(4):303 312. 59. Bolen S, Feldman L, Vassy J, et al. Systematic review: comparative effectiveness and safety of oral medications for type 2 diabetes mellitus. Ann Intern Med. Sep 2007 ;147(6):386 399. 60. Damm P, Mølsted Pedersen L. Significant decrease in congenital malformations in newborn infants of an unselected population of diabetic women. Am J Obstet Gynecol. Nov 1989;161(5):1163 1167. 61. Fuhrmann K, Reiher H, Semmler K, Glöckner E. The effect of intensified conventional insulin therapy before and during pregnancy on the malformation rate in offspring of diabetic mothers. Exp Clin Endocrinol. Apr 1984;83(2):173 177. 62. Pregnancy outcomes i n the Diabetes Control and Complications Trial. Am J Obstet Gynecol. Apr 1996;174(4):1343 1353. 63. Reece EA, Wiznitzer A, Homko CJ, Hagay Z, Wu YK. Synchronization of the factors critical for diabetic teratogenesis: an in vitro model. Am J Obstet Gynecol. Apr 1996;174(4):1284 1288. 64. Lalau JD. Lactic acidosis induced by metformin: incidence, management and prevention. Drug Saf. Sep 2010;33(9):727 740. 65. Stang M, Wysowski DK, Butler Jones D. Incidence of lactic acidosis in metformin users. Diabetes Care . Jun 1999;22(6):925 927. 66. Bailey CJ. Biguanides and NIDDM. Diabetes Care. Jun 1992;15(6):755 772. 67. Krentz AJ, Ferner RE, Bailey CJ. Comparative tolerability profiles of oral antidiabetic agents. Drug Saf. Oct 1994;11(4):223 241.

PAGE 149

149 68. Misbin RI, Green L, Stadel BV, Gueriguian JL, Gubbi A, Fleming GA. Lactic acidosis in patients with diabetes treated with metformin. N Engl J Med. Jan 1998;338(4):265 266. 69. Gilbert C, Valois M, Koren G. Pregnancy outcome after first trimester exposure to metformin: a m eta analysis. Fertil Steril. Sep 2006;86(3):658 663. 70. Hellmuth E, Damm P, Mølsted Pedersen L. Congenital malformations in offspring of diabetic women treated with oral hypoglycaemic agents during embryogenesis. Diabet Med. Jun 1994;11(5):471 474. 71. Co etzee EJ, Jackson WP. Metformin in management of pregnant insulin independent diabetics. Diabetologia. Apr 1979;16(4):241 245. 72. Rodbard HW, Jellinger PS, Davidson JA, et al. Statement by an American Association of Clinical Endocrinologists/American Coll ege of Endocrinology consensus panel on type 2 diabetes mellitus: an algorithm for glycemic control. Endocr Pract. 2009 Sep Oct 2009;15(6):540 559. 73. Lebovitz HE, Pasmantier R. Combination insulin sulfonylurea therapy. Diabetes Care. Jun 1990;13(6):667 6 75. 74. Quatraro A, Consoli G, Ceriello A, Giugliano D. Combined insulin and sulfonylurea therapy in non insulin dependent diabetics with secondary failure to oral drugs: a one year follow up. Diabete Metab. Dec 1986;12(6):315 318. 75. Elliott BD, Schenker S, Langer O, Johnson R, Prihoda T. Comparative placental transport of oral hypoglycemic agents in humans: a model of human placental drug transfer. Am J Obstet Gynecol. Sep 1994;171(3):653 660. 76. Jacobson GF, Ramos GA, Ching JY, Kirby RS, Ferrara A, Fie ld DR. Comparison of glyburide and insulin for the management of gestational diabetes in a large managed care organization. Am J Obstet Gynecol. Jul 2005;193(1):118 124. 77. Koren G. Glyburide and fetal safety; transplacental pharmacokinetic considerations . Reprod Toxicol. 2001 May Jun 2001;15(3):227 229. 78. Carpenter MW, Coustan DR. Criteria for screening tests for gestational diabetes. Am J Obstet Gynecol. Dec 1982;144(7):768 773. 79. Kitzmiller JL, Block JM, Brown FM, et al. Managing preexisting diabetes for pregnancy: summary of evidence and consensus recommendations for care. Diabetes Care. May 2008;31(5):1060 1079. 80. Lee Parritz A. New technologies for the management of pregestati onal diabetes mellitus. Obstet Gynecol Surv. Mar 2012;67(3):167 175.

PAGE 150

150 81. Ghazeeri G, Kutteh WH, Bryer Ash M, Haas D, Ke RW. Effect of rosiglitazone on spontaneous and clomiphene citrate induced ovulation in women with polycystic ovary syndrome. Fertil Ster il. Mar 2003;79(3):562 566. 82. Glueck CJ, Phillips H, Cameron D, Sieve Smith L, Wang P. Continuing metformin throughout pregnancy in women with polycystic ovary syndrome appears to safely reduce first trimester spontaneous abortion: a pilot study. Fertil Steril. Jan 2001;75(1):46 52. 83. Glueck CJ, Bornovali S, Pranikoff J, Goldenberg N, Dharashivkar S, Wang P. Metformin, pre eclampsia, and pregnancy outcomes in women with polycystic ovary syndrome. Diabet Med. Aug 2004;21(8):829 836. 84. Haddad GF, Jodicke C, Thomas MA, Williams DB, Aubuchon M. Case series of rosiglitazone used during the first trimester of pregnancy. Reprod Toxicol. Oct 2008;26(2):183 184. 85. Yaris F, Yaris E, Kadioglu M, Ulku C, Kesim M, Kalyoncu NI. Normal pregnancy outcome foll owing inadvertent exposure to rosiglitazone, gliclazide, and atorvastatin in a diabetic and hypertensive woman. Reprod Toxicol. Jun 2004;18(4):619 621. 86. Kalyoncu NI, Yaris F, Ulku C, et al. A case of rosiglitazone exposure in the second trimester of pre gnancy. Reprod Toxicol. 2005 Mar Apr 2005;19(4):563 564. 87. Choi JS, Han JY, Ahn HK, et al. Exposure to rosiglitazone and fluoxetine in the first trimester of pregnancy. Diabetes Care. Sep 2006;29(9):2176. 88. Chan LY, Yeung JH, Lau TK. Placental transfer of rosiglitazone in the first trimester of human pregnancy. Fertil Steril. Apr 2005;83(4):955 958. 89. Martin JA, Kung HC, Mathews TJ, et al. Annual summary of vital statistics: 2006. Pediatrics. Apr 2008;121(4):788 801. 90. State Health Facts. 2011; http://www.statehealthfacts.org/ . Accessed August 12, 2011. 91. Warner M, National Center for Health Statistics (U.S.). Born a bit too early recent trends in late preterm births. [Hyattsville, MD: U.S. Dept. of Health and Human Services, Centers for Disease Control and Prevention, National Center for Health Statistics; 2009: http://purl.access.gpo.gov/GPO/LPS125594 . 92. McCormick MC. The contribution of low birth weight to infant mortality and childhood morbidity. N Engl J Med. Jan 1985;312(2):82 90. 93. Goldenberg RL, Culhane JF, Iams JD, Romero R. Epidemiology and causes of preterm birth. Lancet. Jan 2008;371(9606):75 84.

PAGE 151

151 94. Goldenberg RL, Tamura T. Prepr egnancy weight and pregnancy outcome. JAMA. Apr 1996;275(14):1127 1128. 95. Yanit KE, Snowden JM, Cheng YW, Caughey AB. The impact of chronic hypertension and pregestational diabetes on pregnancy outcomes. Am J Obstet Gynecol. Oct 2012;207(4):333.e331 336. 96. Moore LE, Clokey D, Rappaport VJ, Curet LB. Metformin compared with glyburide in gestational diabetes: a randomized controlled trial. Obstet Gynecol. Jan 2010;115(1):55 59. 97. Ramos GA, Jacobson GF, Kirby RS, Ching JY, Field DR. Comparison of glyburi de and insulin for the management of gestational diabetics with markedly elevated oral glucose challenge test and fasting hyperglycemia. J Perinatol. May 2007;27(5):262 267. 98. Gregory KD, Ramicone E, Chan L, Kahn KL. Cesarean deliveries for medicaid pati ents: a comparison in public and private hospitals in Los Angeles county. Am J Obstet Gynecol. May 1999;180(5):1177 1184. 99. Stranges E, Wier L, Elixhauser A. Complicating Conditions of Vaginal Deliveries and Cesarean Sections, 2009: Statistical Brief #131. Rockville, MD: Agency for Health Care Policy and Research; 2012. 100. MacDorman MF, Declercq E, Menacker F, Malloy MH. Infant and neonatal mortality for primary cesarean and vaginal births to women with "no indicated risk," United States, 1998 2001 b irth cohorts. Birth. Sep 2006;33(3):175 182. 101. Menacker F, Declercq E, Macdorman MF. Cesarean delivery: background, trends, and epidemiology. Semin Perinatol. Oct 2006;30(5):235 241. 102. Lipkind HS, Duzyj C, Rosenberg TJ, Funai EF, Chavkin W, Chiasson MA. Disparities in cesarean delivery rates and associated adverse neonatal outcomes in New York City hospitals. Obstet Gynecol. Jun 2009;113(6):1239 1247. 103. Bailit JL, Garrett JM, Miller WC, McMahon MJ, Cefalo RC. Hospital primary cesarean delivery rate s and the risk of poor neonatal outcomes. Am J Obstet Gynecol. Sep 2002;187(3):721 727. 104. Schuit E, Kwee A, Westerhuis ME, et al. A clinical prediction model to assess the risk of operative delivery. BJOG. Jul 2012;119(8):915 923. 105. Palmsten K, Setog uchi S, Margulis AV, Patrick AR, Hernández Díaz S. Elevated risk of preeclampsia in pregnant women with depression: depression or antidepressants? Am J Epidemiol. May 2012;175(10):988 997. 106. Bezerra Maia E Holanda Moura S, Marques Lopes L, Murthi P, da Silva Costa F. Prevention of preeclampsia. J Pregnancy. 2012;2012:435090.

PAGE 152

152 107. Roberts JM, Pearson GD, Cutler JA, Lindheimer MD, Institute NHLaB. Summary of the NHLBI Working Group on Research o n Hypertension During Pregnancy. Hypertens Pregnancy. 2003;22(2):109 127. 108. Roberts JM, Lain KY. Recent Insights into the pathogenesis of pre eclampsia. Placenta. May 2002;23(5):359 372. 109. Hernández Díaz S, Toh S, Cnattingius S. Risk of pre eclampsia in first and subsequent pregnancies: prospective cohort study. BMJ. 2009;338:b2255. 110. Hanson U, Persson B. Outcome of pregnancies complicated by type 1 insulin dependent diabetes in Sweden: acute pregnancy complications, neonatal mortality and morbidit y. Am J Perinatol. Jul 1993;10(4):330 333. 111. Byrd V, Hedley Dodd A. Assessing the Usability of Encounter Data for Enrollees in Comprehensive Managed Care Across MAX 2007 2009. Vol 15. Washington, DC: Mathematica Policy Research; 2012:11. 112. Schneeweis s S, Avorn J. A review of uses of health care utilization databases for epidemiologic research on therapeutics. J Clin Epidemiol. Apr 2005;58(4):323 337. 113. Strom BL, Kimmel SE, Hennessy S, eds. Pharmacoepidemiology. 5 ed. West Sussex: Wiley Blackwell; 2 012. 114. Health Insurance Coverage of Adults 19 64, states (2009 2010), U.S. (2010). 2010; http://www.statehealthfacts.org/comparemaptable.js p?ind=130&cat=3&sub=39&y r=252&typ=2&o=a&sortc=3 . Accessed February, 2012. 115. Medicaid's Role in Family Planning. The Kaiser Family Foundation; 2007. 116. Ranji U, Salganicoff A, Stewart A, Cox M, Doamekpor L. State Medicaid Coverage of Perinatal Services : Summary of State Survey Findings. The Kaiser Family Foundation; 2009. 117. Garcia G. Maternal and Child Health (MCH) Update: States Increase Eligibility for Children's Health in 2007. 2008. http://www.nga.org/files/live/sites/NGA/files/pdf/0811MCHUPDATE.PDF . Accessed 10.10.11. 118. Medicaid's Role for Women Across the Lifespan: Current Issues and the Impact of the Affordable Care Act. The Kaiser Family Foundation; 2012. 11 9. Stranges E Fau Ryan K, Ryan K Fau Elixhauser A, Elixhauser A. Medicaid Hospitalizations, 2008: Statistical Brief #104 BTI Healthcare Cost and Utilization Project (HCUP) Statistical Briefs.

PAGE 153

153 120. Hennessy S, Leonard CE, Palumbo CM, Newcomb C, Bilker WB. Quality of Medicaid and Medicare data obtained through Centers for Medicare and Medicaid Services (CMS). Med Care. Dec 2007;45(12):1216 1220. 121. Hennessy S, Bilker WB, Weber A, Strom BL. Descriptive analyses of the integrity of a US Medicaid claims database. Pharmacoepidemiol Drug Saf. Mar 2003;12(2):103 111. 122. O'Malley KJ, Cook KF, Price MD, Wildes KR, Hurdle JF, Ashton CM. Measuring diagnoses: ICD code accuracy. Health Serv Res. Oct 2005;40(5 Pt 2):1620 1639. 123. Chen G, Khan N, Walker R, Quan H. Validating ICD coding algorithms for diabetes mellitus from administrative data. Diabetes Res Clin Pract. Aug 2010;89(2):189 195. 124. Andrade SE, Mo ore Simas TA, Boudreau D, et al. Validation of algorithms to ascertain clinical conditions and medical procedures used during pregnancy. Pharmacoepidemiol Drug Saf. Nov 2011;20(11):1168 1176. 125. Hebert PL, Geiss LS, Tierney EF, Engelgau MM, Yawn BP, McBe an AM. Identifying persons with diabetes using Medicare claims data. Am J Med Qual. 1999 Nov Dec 1999;14(6):270 277. 126. Andrade SE, Scott PE, Davis RL, et al. Validity of health plan and birth certificate data for pregnancy research. Pharmacoepidemiol Dr ug Saf. Jan 2013;22(1):7 15. 127. White RH, Brickner LA, Scannell KA. ICD 9 CM codes poorly indentified venous thromboembolism during pregnancy. J Clin Epidemiol. Sep 2004;57(9):985 988. 128. Geller SE, Ahmed S, Brown ML, Cox SM, Rosenberg D, Kilpatrick SJ . International Classification of Diseases 9th revision coding for preeclampsia: how accurate is it? Am J Obstet Gynecol. Jun 2004;190(6):1629 1633; discussion 1633 1624. 129. Margulis AV, Setoguchi S, Mittleman MA, Glynn RJ, Dormuth CR, Hernández Díaz S. Algorithms to estimate the beginning of pregnancy in administrative databases. Pharmacoepidemiol Drug Saf. May 2012. 130. Wier ML, Pearl M, Kharrazi M. Gestational age estimation on United States livebirth certificates: a historical overview. Paediatr Peri nat Epidemiol. Sep 2007;21 Suppl 2:4 12. 131. Qin C, Hsia J, Berg CJ. Variation between last menstrual period and clinical estimates of gestational age in vital records. Am J Epidemiol. Mar 2008;167(6):646 652.

PAGE 154

154 132. Ambrose CS, Rizzo C, Caspard H. Methods to estimate gestational age can significantly affect study results. JAMA Pediatr. Apr 2014;168(4):388. 133. Palmsten K, Huybrechts KF, Mogun H, et al. Harnessing the Medicaid Analytic eXtract (MAX) to Evaluate Medications in Pregnancy: Design Consideration s. PLoS One. 2013;8(6):e67405. 134. Swets JA. Measuring the accuracy of diagnostic systems. Science. Jun 1988;240(4857):1285 1293. 135. Gönen M. Receiver Operating Characteristic (ROC) Curves.210 231. 136. Hardy JR, Leaderer BP, Holford TR, Hall GC, Bracken MB. Safety of medications prescribed before and during early pregnancy in a cohort of 81,975 mothers from the UK General Practice Research Database. Pharmacoepidemiol Drug Saf. Aug 2006;15(8):555 564. 137 . Toh S, Mitchell AA, Werler MM, Hernández Díaz S. Sensitivity and specificity of computerized algorithms to classify gestational periods in the absence of information on date of conception. Am J Epidemiol. Mar 2008;167(6):633 640. 138. Pazol K, Creanga AA , Burley KD, Hayes B, Jamieson DJ, (CDC) CfDCaP. Abortion surveillance United States, 2010. MMWR Surveill Summ. Nov 2013;62(8):1 44. 139. Silver RM, Varner MW, Reddy U, et al. Work up of stillbirth: a review of the evidence. Am J Obstet Gynecol. May 2007 ;196(5):433 444. 140. Wyatt PR, Owolabi T, Meier C, Huang T. Age specific risk of fetal loss observed in a second trimester serum screening population. Am J Obstet Gynecol. Jan 2005;192(1):240 246. 141. Likis FE, Sathe NA, Carnahan R, McPheeters ML. A syst ematic review of validated methods to capture stillbirth and spontaneous abortion using administrative or claims data. Vaccine. Dec 2013;31 Suppl 10:K74 82. 142. Yasmeen S, Romano PS, Schembri ME, Keyzer JM, Gilbert WM. Accuracy of obstetric diagnoses and procedures in hospital discharge data. Am J Obstet Gynecol. Apr 2006;194(4):992 1001. 143. Wei WQ, Leibson CL, Ransom JE, et al. Impact of data fragmentation across healthcare centers on the accuracy of a high throughput clinical phenotyping algorithm for specifying subjects with type 2 diabetes mellitus. J Am Med Inform Assoc. 2012 Mar Apr 2012 ;19(2):219 224. 144. Zgibor JC, Orchard TJ, Saul M, et al. Developing and validating a diabetes database in a large health system. Diabetes Res Clin Pract. Mar 2007;75(3):313 319.

PAGE 155

155 145. Fleiss J, Levin B, Paik M. Statistical Methods for Rates and Proportion s. Vol 3rd. New York: Wiley Interscience; 2003. 146. Miller E, Hare JW, Cloherty JP, et al. Elevated maternal hemoglobin A1c in early pregnancy and major congenital anomalies in infants of diabetic mothers. N Engl J Med. May 1981;304(22):1331 1334. 147. Ki nsley B. Achieving better outcomes in pregnancies complicated by type 1 and type 2 diabetes mellitus. Clin Ther. 2007;29 Suppl D:S153 160. 148. Becerra JE, Khoury MJ, Cordero JF, Erickson JD. Diabetes mellitus during pregnancy and the risks for specific bi rth defects: a population based case control study. Pediatrics. Jan 1990;85(1):1 9. 149. Lawrence JM, Andrade SE, Avalos LA, et al. Prevalence, trends, and patterns of use of antidiabetic medications among pregnant women, 2001 2007. Obstet Gynecol. Jan 201 3;121(1):106 114. 150. Knox CA, Delaney JA, Winterstein AG. Anti diabetic drug utilization of pregnant diabetic women in us managed care. BMC Pregnancy Childbirth. 2014;14:28. 151. Palmsten K, Hernández Díaz S, Huybrechts KF, et al. Use of antidepressants near delivery and risk of postpartum hemorrhage: cohort study of low income women in the United States. BMJ. 2013;347:f4877. 152. Palmsten K, Hernández Díaz S, Kuriya B, Solomon DH, Setoguchi S. Use of disease modifying antirheumatic drugs during pregnancy and risk of preeclampsia. Arthritis Care Res (Hoboken). Nov 2012;64(11):1730 1738. 153. Maldonado G, Greenland S. Simulation study of confounder selection strategies. Am J Epidemiol. Dec 1993;138(11):923 936. 154. Griebel CP, Halvorsen J, Golemon TB, Day AA. Management of spontaneous abortion. Am Fam Physician. Oct 2005;72(7):1243 1250. 155. Rothman KJ, Greenland S, Lash TL. Modern epidemiology. 3rd ed. Philadelphia: Wolters Kluwer Health/Lippincott Williams & Wilkins; 2008. 156. Andrade SE, Raebel MA, Mor se AN, et al. Use of prescription medications with a potential for fetal harm among pregnant women. Pharmacoepidemiol Drug Saf. Aug 2006;15(8):546 554. 157. Hampp C, Borders Hemphill V, Moeny DG, Wysowski DK. Use of antidiabetic drugs in the U.S., 2003 201 2. Diabetes Care. May 2014;37(5):1367 1374. 158. Palmsten K, Huybrechts KF, Michels KB, et al. Antidepressant use and risk for preeclampsia. Epidemiology. Sep 2013;24(5):682 691.

PAGE 156

156 159. Phiri K, Fischer MA, Mogun H, et al. Trends in antiretroviral drug use d uring pregnancy among HIV infected women on medicaid: 2000 2007. AIDS Patient Care STDS. Feb 2014;28(2):56 65. 160. Huybrechts KF, Palmsten K, Mogun H, et al. National trends in antidepressant medication treatment among publicly insured pregnant women. Gen Hosp Psychiatry. 2013 May Jun 2013;35(3):265 271. 161. Albrecht SS, Kuklina EV, Bansil P, et al. Diabetes trends among delivery hospitalizations in the U.S., 1994 2004. Diabetes Care. Apr 2010;33(4):768 773. 162. Devlin HM, Desai J, Walaszek A. Reviewing performance of birth certificate and hospital discharge data to identify births complicated by maternal diabetes. Matern Child Health J. Sep 2009;13(5):660 666. 163. Kho AN, Hayes MG, Rasmussen Torvik L, et al. Use of diverse electronic medical record systems to identify genetic risk for type 2 diabetes within a genome wide association study. J Am Med Inform Assoc. 2012 Mar Apr 2012;19(2):212 218. 164. Update of ADA's major position statement, "Standards of Medical Care in Diabetes". Introduction. Diabetes Care. Jan 2011;34 Suppl 1:S1 2. 165. 2003 National Diabetes Fact Sheet. 2003; http://www.cdc.gov/diabetes/pubs/general.htm#treatment , 2012. 166. Association AD. Standards of medical care in diabetes -2007. Diabetes Care. Jan 2007;30 Suppl 1:S4 S41. 167. McCance DR. Pregnancy and diabetes. Best Pract Res Clin Endocrinol Metab. De c 2011;25(6):945 958. 168. Margulis AV, Setoguchi S, Mittleman MA, Glynn RJ, Dormuth CR, Hernández Díaz S. Algorithms to estimate the beginning of pregnancy in administrative databases. Pharmacoepidemiol Drug Saf. Jan 2013;22(1):16 24. 169. Lydon Rochelle MT, Cárdenas V, Nelson JC, Holt VL, Gardella C, Easterling TR. Induction of labor in the absence of standard medical indications: incidence and correlates. Med Care. Jun 2007;45(6):505 512. 170. Menacker F, Hamilton BE. Recent trends in cesarean delivery i n the United States. NCHS Data Brief. Mar 2010(35):1 8. 171. Bailit JL, Downs SM, Thorp JM. Reducing the caesarean delivery risk in elective inductions of labour: a decision analysis. Paediatr Perinat Epidemiol. Jan 2002;16(1):90 96.

PAGE 157

157 172. Ehrenberg HM, Dur nwald CP, Catalano P, Mercer BM. The influence of obesity and diabetes on the risk of cesarean delivery. Am J Obstet Gynecol. Sep 2004;191(3):969 974. 173. Kjos SL, Berkowitz K, Xiang A. Independent predictors of cesarean delivery in women with diabetes. J Matern Fetal Neonatal Med. Jan 2004;15(1):61 67. 174. Bo S, Menato G, Signorile A, et al. Obesity or diabetes: what is worse for the mother and for the baby? Diabetes Metab. Apr 2003;29(2 Pt 1):175 178. 175. Al Qahtani S, Heath A, Quenby S, et al. Diabete s is associated with impairment of uterine contractility and high Caesarean section rate. Diabetologia. Feb 2012;55(2):489 498. 176. Chou E, Suzuma I, Way KJ, et al. Decreased cardiac expression of vascular endothelial growth factor and its receptors in in sulin resistant and diabetic States: a possible explanation for impaired collateral formation in cardiac tissue. Circulation. Jan 2002;105(3):373 379. 177. Montagnani M, Ravichandran LV, Chen H, Esposito DL, Quon MJ. Insulin receptor substrate 1 and phosph oinositide dependent kinase 1 are required for insulin stimulated production of nitric oxide in endothelial cells. Mol Endocrinol. Aug 2002;16(8):1931 1942. 178. Vicent D, Ilany J, Kondo T, et al. The role of endothelial insulin signaling in the regulation of vascular tone and insulin resistance. J Clin Invest. May 2003;111(9):1373 1380. 179. Vanky E, Stridsklev S, Heimstad R, et al. Metformin versus placebo from first trimester to delivery in polycystic ovary syndrome: a randomized, controlled multicenter study. J Clin Endocrinol Metab. Dec 2010;95(12):E448 455. 180. Hernán MA, Hernández Díaz S, Robins JM. A structural approach to selection bias. Epidemiology. Sep 2004;15(5):615 625. 181. Gerlach R, Stamey J. Bayesian model selection for logistic regression with misclassified outcomes. Statistical Modelling. 2007;7(3).

PAGE 158

158 BIOGRAPHICAL SKETCH Caitlin Knox was born in Chicago, Illinois. She received her B achelor of S cience degree in f ood s cience and h uman n utrition with a minor in Spanish Language from the University of Florida in 2007; she continued her education and received her Master of Public Health with a concentration in e pidemiology from the University of Florida in 2009. Later that year, s he joined D epartment of Pharmaceutical Outcomes and Policy. In 2011, Caitlin founded and was president of the Florida International Society of Pharmacoepidemiology (ISPE) Student Chapter, she was appointed as chair of the Student Council i n ISPE, and was elected as the g raduate s tudent c ouncil r and Policy. In 2012, she was elected s tudent p resident of the College of Pharmacy. Caitlin received the Linton E. Grinter Fellowship in 2009 from the University of Florida. She was appointed as a Pre doctoral fellow by the American Foundation for Pharmaceutical Education in 2011 and 2012. Later in 2012, she was also the recipient of the Mary Kay Owen Fellowship from the College of Pharmacy at the University of Florida. Caitlin was awarded an R36 Dissertation grant (1R36HS022384 01) from the AHRQ Grants for Health Services Research Dissertation Program in August 2013. Caitlin has authored and co authored several peer reviewed publications and presented her research at national and international conferences. Her research interests include drug utilization, safety , and effectiveness. Clinical areas of pharmacoepidemiologic studie s include maternal and child health, medicines in pregnancy, diabetes mellitus, and infectious disease.