The use of glucose‐lowering medications for the treatment of type 2 diabetes mellitus during pregnancy in the United States

Abstract Introduction Some guidelines allow for the use of either insulin or noninsulin antidiabetic agents for gestational diabetes, but only insulin is recommended for pregnant women with preexisting type 2 diabetes mellitus (T2DM). We aimed to document treatment patterns in routine care for women with preexisting T2DM. Methods We identified pregnancy cohorts within 2 US claims databases for publicly and privately insured individuals: the Medicaid Analytical eXtract (2000–2014) and OptumClinformatics (2004–2014). T2DM was classified with a validated algorithm using ICD‐9‐CM and CPT codes. We assessed medication usage over the years of the study, and changes in medication use before and after the beginning of pregnancy, using prescription fills as a proxy for the use of insulin, metformin, sulphonylureas and other noninsulin antidiabetic agents before pregnancy and during each trimester. Results Among 12,631 women with T2DM, insulin use in pregnancy was stable over the study years (55%–60% in the 2nd trimester), but 2nd trimester use of metformin increased from <5% to 20%. Over the study period, 41% of women filled a prescription for metformin before pregnancy, 37% in the 1st trimester and 17% in the 2nd trimester. By the 2nd trimester, few women used sulphonylureas (11%) or other noninsulin antidiabetic agents (1%). Of the women on metformin only before pregnancy, 36% switched to insulin only by 2nd trimester, 11% added insulin and 16% continued on metformin only. Of the women on metformin and insulin before pregnancy, 61% switched to insulin only by 2nd trimester, 22% continued with metformin and insulin and <5% used only metformin. Conclusion The use of insulin‐metformin combinations and other noninsulin antidiabetic drugs during pregnancy has increased. Safety studies for these medication regimens are needed.


| INTRODUC TI ON
Pre-existing type 2 diabetes mellitus (T2DM) complicates 1% of US pregnancies annually, with substantially higher prevalence in medically underserved populations. 1 Its prevalence is expected to increase in coming years given the obesity epidemic in the United States. 2 T2DM is associated with an increased risk of poor maternal, foetal and neonatal outcomes, including pregnancy loss and stillbirth, preeclampsia, congenital malformations, macrosomia and birth injury. 3,4 Glycemic control is an important intermediary for these risks, 4 highlighting the importance of preventing and treating hyperglycemia during pregnancy.
In nonpregnant women of reproductive age, T2DM is most often treated with lifestyle modifications; pharmacological treatment with antidiabetic agents is initiated if changes to diet and exercise are ineffective. Metformin is the preferred initial medication, and other agents, including insulin, can be added to achieve metabolic targets. 5 However, whilst some guidelines allow for the use of noninsulin antidiabetic agents for gestational diabetes, 6 the recommended treatment for preexisting T2DM during pregnancy is insulin. 6 For women with T2DM considering pregnancy, current American Diabetes Association (ADA) and American College of Obstetrics and Gynecology (ACOG) guidelines recommend initiating insulin therapy as soon as possible, and ideally before pregnancy, 6 although the use of metformin may be considered in rare cases. 7 Nonetheless, the use of noninsulin antidiabetic agents by women with preexisting T2DM does occur in pregnancy. [8][9][10] Recent studies have shown increases in the use of noninsulin antidiabetic medications during pregnancy, particularly metformin and glyburide. 8,9,11 However, these studies were limited in their ability to describe medication use specifically in T2DM; classification of preexisting diabetes depended mostly 9 or entirely 8,11 on filling prescriptions for antidiabetic medications before and/or during pregnancy and investigated populations that contained a mix of type 1, type 2 and gestational diabetes.
Furthermore, although previous studies examined the prevalence of antidiabetic medication use in the pregnancy period overall, they did not describe within-pregnancy longitudinal changes in treatment.
To address these limitations, we characterized prescription patterns and trends of antidiabetic medication use during pregnancy in two population-based cohorts of publicly and privately insured pregnant women with pregestational T2DM.

| Data source and study population
Publicly insured pregnancies were identified from the US Medicaid Analytical eXtract for the period of 2000-2014. Construction of the Medicaid mother-infant linked pregnancy cohort has been previously described in detail. 12 For the privately insured cohort, we used OptumClinformatics (Optum) files from 2004 to 2015. Motherinfant linkage in the Optum cohort was accomplished through a family identifier. Both Optum and Medicaid files contain diagnoses and procedure code emanating from inpatient stays and outpatient visits, as well as outpatient prescription fills. The date of the last menstrual period (LMP) was estimated based on the date of delivery and any codes for preterm birth, using a validated algorithm. 13 Both cohorts consisted of women aged 12-55 years at the time of delivery who had pregnancies ending in live births, and who had continuous insurance coverage from 180 days before the LMP date to 30 days after delivery to ensure completeness of their pregnancy healthcare claims. Data supporting this study are not publicly available due to privacy/ethical restrictions.

| Definition of pregestational type 2 diabetes mellitus
We previously developed and validated an algorithm based on

International Classification of Diseases, 9th revision, Clinical Modification
(ICD-9-CM) diagnosis codes, Current Procedural Terminology (CPT) procedure codes and prescription fills, to classify pregestational diabetes in pregnant women (Table S1), with a positive predictive value (PPV) of 91.7% for any pregestational and 87.0% for T2DM. 14 We modified the algorithm for this analysis to require ≥2 T2DM codes and 0 type 1 diabetes mellitus (T1DM) codes from 180 days before LMP through 90 days after LMP (PPV 100%; Table S2).

| Definitions of pregnancy periods and medication use
We assessed diabetes medication use before pregnancy and during each trimester. The prepregnancy baseline period was defined as 90 days before LMP date to the day before the LMP date (LMP-90 to LMP-1). The first trimester was defined as LMP to day 90 of pregnancy (LMP to LMP + 90), the second trimester as days 91-180 after LMP (LMP + 91 to LMP + 180) and the third trimester as the period between day 181 and the delivery date (LMP + 181 to delivery).
We categorized antidiabetic medications or classes as insulin, metformin, sulphonylureas and other noninsulin antidiabetic agents.
Women were considered to have used the medication of interest if they filled a prescription in the relevant pregnancy period, and those who used a combination product were considered users of both products (eg a prescription for glyburide/metformin HCl would be counted in both the metformin and sulphonylurea groups).
We also defined 8 mutually exclusive exposure groups based on the most frequently observed treatment strategies: no pharmacological treatment, metformin only, insulin only, metformin and insulin, sulphonylureas only, sulphonylureas and metformin, sulphonylureas and other noninsulin antidiabetic drugs, and "other" that included any pregnancies whose pharmacological treatment was not described by the previous definitions.

| Maternal characteristics
Maternal characteristics were coded using ICD-9-CM and CPT codes. Demographic characteristics were coded as the most recent value before the delivery date. Markers of diabetes severity, comorbid health conditions, laboratory test orders and concomitant medications were evaluated between LMP − 180 and LMP + 90.
Healthcare utilization and preventive services were evaluated from LMP − 180 to LMP − 1 (Table S4).

| Data analysis
We described secular trends in the use of specific antidiabetic medi-  No informed consent was required.

| Cohort characteristics
We identified 10,987 pregnancies from Medicaid and 1,644 from Optum ( Table 1). The cohorts differed in several ways, including maternal age, with 47% of the Optum cohort being age 35 or older, compared to 26% of the Medicaid cohort. In addition, the Optum cohort received more referrals for nutrition counselling (12% vs. 5% in Medicaid) and had a higher proportion diagnosed with hyperlipidemia (39% vs. 19%), hypothyroidism (21% vs. 6%) and polycystic ovarian syndrome (PCOS) (13% vs. 3%); they also had more laboratory tests ordered. By contrast, the Medicaid cohort had a greater number of inpatient and outpatient encounters and filled more prescriptions for medications to treat hypertension (32% vs. 24% in Optum) as well as pain and psychiatric illnesses. Despite these differences, the cohorts were similar in terms of prevalence of diabetic complications, diabetes-related comorbidities and other maternal health conditions.

| Time trends for the prevalence of prescriptions for specific antidiabetics
The percentages of women with T2DM filling a prescription for insulin, metformin, sulphonylureas, and other noninsulin antidiabetic  whilst the use of sulphonylureas had modest decreases (Figure 2A).
In the Medicaid cohort during the earlier years of observation, first trimester prescription fills for insulin were more than twice as common as metformin fills (11% filled a prescription for metformin vs. and during pregnancy (in the second trimester <2% in both Medicaid and Optum), which was stable over the study period. Figure S1 gives additional detail.

| Treatment strategies throughout pregnancy
Before pregnancy, the most common treatment strategy was metformin monotherapy (20% in Medicaid and 26% in Optum), followed by insulin monotherapy (9% in Medicaid and 5% in Optum). Insulin monotherapy was the most common treatment strategy by the second trimester (45% in Medicaid and 42% in Optum), with an increase in treatment strategies involving noninsulin medications over the study period ( Figures S2 and S3). Among women filling a prescription for an antidiabetic treatment before pregnancy, most (73%) were dispensed metformin alone or in combination.
Given that the first trimester seems to be a transitional period for antidiabetic treatment adjustment once a pregnancy is identified, we present trajectories of use from the prepregnancy period to the second trimester (Figures 3 and 4, with additional supporting information included in Tables S5 and S6). Five to seven percentage of T2DM women was on insulin monotherapy before pregnancy, and the majority continued on insulin monotherapy in the second trimester ( Figure 3A and B). Among women dispensed metformin alone before pregnancy, 36%-39% (in Medicaid and Optum respectively) switched to insulin monotherapy and an additional 12%-14% augmented the treatment with insulin in the second trimester ( Figure 3C and D, Figure 4). Among women on metformin combined with insulin before pregnancy, 65%-67% switched to insulin monotherapy in the second trimester and an additional 23%-27% continued on both metformin and insulin (Figure 4). Relative to patterns among women on metformin before pregnancy, fewer women with prepregnancy dispensations for sulphonylureas ( Figure 3E and F) or other noninsulin antidiabetic drugs ( Figure 3G and H) discontinued pharmacotherapy; instead, they switched or augmented treatment, most often with insulin. The supplemental material includes trajectories involving combinations of insulin with sulphonylureas ( Figure S4, Table S7).
Approximately 25% of women with T2DM were not prescribed a medication to treat diabetes before pregnancy and continued without pharmacotherapy through the second trimester. An additional 20%-22% filled no prescriptions for antidiabetic medications before pregnancy and initiated pharmacotherapy by the second trimester, mostly with insulin monotherapy (65%-71%; Figure 4).

| Sensitivity analyses
Stratified analyses before 2008 vs. 2008 or later ( Figures S5-S10) showed that, after 2008, fewer women switched to insulin monotherapy in the second trimester ( Figure S5), and more women switched to a treatment regime that included metformin ( Figure S6).
Sensitivity analyses examining utilization with a more sensitive definition of T2DM showed patterns of treatment that were consistent with the main analysis (Figures S11-S18).

| Principal findings
In a large cohort of publicly and privately insured US women with  Abbreviations: ACE, angiotensin-converting enzyme; LMP, last menstrual period; NOS, not otherwise specified; NSAID, nonsteroidal anti-inflammatory drugs; SD, standard deviation.

TA B L E 1 (Continued)
remained stable both throughout pregnancy and over time. The use of antidiabetic medications other than insulin, metformin and sulphonylureas before pregnancy was low, and most women switched to insulin early in pregnancy.
Our results are consistent with other recent studies describing increases in the use of noninsulin antidiabetic medications in pregnancy. [8][9][10][11] We observed that metformin was the most commonly prescribed antidiabetic medication before pregnancy, which differs from studies in which insulin was more prevalent 8,9 ; this discrepancy is likely due to differences in the definition of pregestational diabetes, used in other studies which allowed for inclusion of women with T1DM. to guideline-recommended treatment 6,7 are not occurring before pregnancy for many women in our study, which may result in first trimester foetal exposure to medications with unknown safety profiles and/or suboptimal glycemic control in early pregnancy.
Second, because women with T2DM often have insulin resistance that worsens during pregnancy, they may require high doses of insulin to achieve euglycemia. 6 The possibility that metformin in combination with insulin may help pregnant women with T2DM achieve glycemic control without high doses of insulin has been embraced by many clinicians and is the subject of two randomized

| Strengths and limitations
Our study has several important limitations. First, we used the date of a prescription fill to indicate medication use during a specific period, which may result in both overestimation and underestimation.
A woman may fill a prescription for a medication but may discontinue its use after learning she is pregnant; on the other hand, a woman may have filled a prescription in an earlier period and have a supply overlapping with a later period, such that she was truly exposed in both periods but would only be classified as exposed in the earlier period. Reliance on prescription fills could also produce the appearance of concomitant medication use, and fail to correctly identify treatment switching. Second, our study ends in 2014, which limits our ability to assess newer noninsulin antidiabetic treatments and recent changes in clinical practice. Third, misclassification of T2DM is possible and may differ within specific treatment patterns (eg metformin users may include more non-T2DM women incorrectly classified as T2DM because metformin is used for other indications than diabetes, whilst insulin users with T2DM may be misclassified as T1DM). However, because our validated algorithm required the presence of multiple specific diagnostic codes to classify individuals as T2DM, 14 we expect this type of misclassification to be infrequent.
Finally, we required women to have been continuously enrolled in their health insurance plan from 180 days before pregnancy start through 30 days post delivery, which may have selected a sample whose characteristics differ from the underlying population of pregnant women with T2DM in terms of wealth and/or disability.
These limitations are balanced by several strengths. We conducted our study in a large cohort that included both publicly and privately insured women in the United States and used a validated algorithm to identify preexisting T2DM based on ICD-9 codes and prescription fills before and early in pregnancy. The results were almost identical in the publicly and privately insured populations.
We also used data visualization methods to describe longitudinal trajectories of medication use to gain a better understanding of the transition of treatment after pregnancy onset. This study highlights important future research directions for the evaluation of pharmacological management of T2DM during pregnancy by identifying the most common treatment strategies.

| Conclusions
In a large population of US women with T2DM between 2000 and 2014, metformin was the most frequently used antidiabetic Given the increasing use of noninsulin antidiabetic medications in early pregnancy, safety studies are needed.