Safety of GLP-1 Receptor Agonists and Other Second-Line Antidiabetics in Early Pregnancy

1Type 2 diabetes (T2D) is an increasingly common condition in female individuals of reproductive age,, which has resulted in increased use of antidiabetic medication (ADM) during pregnancy., In the general nonpregnant population, metformin is often the first-line pharmacological treatment for T2D, and insulin or other second-line noninsulin ADMs, including sulfonylureas, dipeptidyl peptidase 4 (DPP-4) inhibitors, glucagon-like peptide 1 (GLP-1) receptor agonists, and sodium-glucose cotransporter 2 (SGLT2) inhibitors, can be switched to or added to maintain glycemic control if needed. Notably, the use of second-line noninsulin ADM has increased in the last decade.,

2For patients with T2D who are planning pregnancy or who are already pregnant, the guideline-recommended treatment has traditionally been insulin, due to the limited data on the safety of noninsulin ADM for fetal development. However, the intentional use of metformin during pregnancy has become more common and unintended exposure to second-line noninsulin ADM medications during the first trimester, although still rare, has also increased over time., Unintentional pregnancy exposure arises because a proportion of pregnancies are unplanned,, and therefore the discontinuation of these medications often occurs during or after organogenesis. Hence, studies are urgently needed to be able to advise patients, clinicians, and regulatory bodies on the potential teratogenic risk of these medications.

3To generate evidence on the teratogenic risk of second-line noninsulin ADM, we combined data from 6 large population-based health care databases from 4 Nordic countries, the US, and Israel to identify a cohort of pregnant women with pharmacologically treated T2D around the time of conception. First, we described the time trends of second-line noninsulin ADM use in pregnancy over time. Next, we compared the risk of major congenital malformations (MCMs) overall, and cardiac MCMs specifically, in infants born to women with periconceptional use of second-line noninsulin ADM vs insulin.

4This study was conducted within the International Pregnancy Safety Study (InPreSS) Consortium, a collaboration among research groups in several countries, including the Nordic countries, the US, and Israel, all of whom have access to high-quality prospectively collected health care databases and registers.

5The Nordic cohort was derived from nationwide population registers and included all pregnancies resulting in singleton live-born infants in Finland, Iceland, Norway, and Sweden from 2009 to the end of available data in each country (2016-2020). The individual-level data from the 4 countries were pooled and harmonized using a common data model. Information on the Nordic population health registers is available in eMethods 1 in Supplement 1.

6The US cohort consisted of commercially insured pregnant women linked to their live-born (singleton and multiple) infants included in the MarketScan Research Database (2012-2021), one of the largest national health care administration databases. To ensure capture of all diagnosis codes and prescription fills during the study period, pregnant women were required to have continuous insurance coverage from at least 6 months before pregnancy to 1 month after delivery; infants were required to have coverage from birth until 90 days after birth, unless they died sooner.

7The Israeli cohort consisted of pregnancies resulting in a singleton live-born infant (2010-2020) from women continuously enrolled for at least 1 year preconception in the Maccabi Health Services (MHS) database; infants were required to have at least 1 year of complete follow-up postbirth, unless they died sooner. MHS is Israel’s second-largest health care organization serving as both insurer and health care service to approximately 25% of the Israeli population.

8Pregnancies with a diagnosis of a fetal chromosomal abnormality or with exposure to a known teratogenic medication (eTable 1 in Supplement 1) were excluded from each cohort (eFigure in Supplement 1).

9Use of the Nordic data was approved by applicable ethics review boards and/or data providing authorities (eMethods 1 in Supplement 1). Use of the MarketScan data was approved by the institutional review board at the Harvard T.H. Chan School of Public Health. Use of the Israeli MHS data was approved by the institutional review boards at MHS and Harvard T.H. Chan School of Public Health, which granted a waiver of informed consent.

10The study population included pregnancies in women with pregestational T2D linked to live-born infants. In the US data, T2D was identified using a validated algorithm based on diagnoses and medication prescription fills. In the validation study, the algorithm had a positive predictive value of 87% compared with electronic medical records; however, this estimate was likely conservative when applied to the present study where the exposures of interest, second-line noninsulin ADMs were indicated almost exclusively for T2D during the study period. The algorithm was adapted for use in the Nordic and MHS data, and optimized based on the best available information, including laboratory measurements in MHS. Algorithms and criteria are described in detail in eMethods 2 in Supplement 1. In this study, we refer to biologic sex and use the term pregnant woman to define pregnant human females of any gender identity. Gender identity was not recorded in the databases.

11Periconceptional exposure was defined based on the filling of 1 or more prescriptions of the respective drug class (eTable 2 in Supplement 1) from 90 days before the first day of the last menstrual period (LMP) to the end of the first trimester because drug supplies for chronic illness often cover 1 to 3 months. A secondary exposure definition for sensitivity analyses required the filling of 1 or more prescriptions from LMP to the end of the first trimester.

12Pregnancies were then classified into the following exposure groups: periconceptional use of no ADM, metformin only, insulin (with or without coprescriptions of metformin, but no other ADM), sulfonylureas, DPP-4 inhibitors, GLP-1 receptor agonists, or SGLT2 inhibitors. Women in the latter 4 exposure groups were allowed to have coprescriptions of any other ADM during the periconception period.

13The presence of any MCM overall and the subgroup of major cardiac malformations were identified using diagnosis and procedure codes as described in detail for each cohort in eTables 3 and 4 in Supplement 1. Briefly, MCMs were defined using infant diagnoses from the date of birth to 1 year after birth in the Nordic and Israeli cohorts, and using claims in the infant and the women’s records from date of birth to 90 days after birth in the US cohort.

14Key baseline characteristics were described for pregnant women with periconceptional use of the second-line ADM, including maternal age, comorbidities (ie, obesity, hypertension, cardiovascular disease, diabetic complications, polycystic ovary syndrome [PCOS]) and other prescription medication (ie, antihypertensive medication, lipid modifying agents; defined in eTable 5 in Supplement 1). Hemoglobin A1c (HbA1c) levels, which are a measure of glycemic control over the previous 3 months, were available for a subset of pregnancies in the US and Israeli cohorts (eTable 6 in Supplement 1). The mean (SD) and median (IQR) were calculated for each exposure group (HbA1c unit = %) with linked laboratory test results for HbA1c levels from between 90 days before LMP to the end of the first trimester.

15Analyses were conducted separately in the pooled Nordic cohort, the US cohort, and the Israeli cohort. We calculated the proportion of pregnancies with periconceptional exposure to second-line noninsulin ADM by birth year. The standardized weighted prevalence of any MCM and cardiac malformations were calculated for all exposure groups to account for regional differences in both utilization of ADM classes and baseline frequency of MCMs. Within each cohort, the crude and adjusted relative risks (RRs) with 95% CIs were estimated for each of the 4 second-line noninsulin exposure groups compared with insulin using a log-binomial model. In the adjusted model, maternal age, year of birth, obesity, and specific Nordic country (in the pooled Nordic cohort only) were added. The crude and adjusted estimates from the Nordic, US, and Israeli cohorts were then combined using fixed effect meta-analysis using R statistical software (version 4.3.1, R Foundation). The analysis was completed on June 23, 2023.

16In a total of 3 514 865 pregnancies from the 3 data sources combined, 51 826 (1.5%) were in women with pregestational T2D, of whom 15 148 (29.2%) were treated with ADM in the periconceptional period (Nordics, 9693; US, 4778; Israel, 677) (eFigure in Supplement 1). Among these pregnancies, 7440 (50%) used metformin only, 5078 (34%) insulin, 1352 (9.0%) sulfonylureas, 687 (4.5%) DPP-4 inhibitors, 938 (6.2%) GLP-1 receptor agonists, and 335 (2.2%) SGLT2 inhibitors. Periconceptional use of second-line noninsulin ADM increased over time, particularly in the US for GLP-1 receptor agonists, and except for sulfonylureas, which remained low in the Nordic countries, decreased in the US, and increased slightly in Israel (Figure 1). Use of other ADM classes such as glitazones, meglitinides, and α-glucosidase inhibitors remained very low.

17Table 1 shows the baseline maternal characteristics, by study cohort, for women with pregestational T2D in the 4 second-line noninsulin ADM and insulin exposure groups. Compared with those using insulin, women using second-line noninsulin ADMs were slightly younger in the US and Israel and slightly older in the Nordic countries. The prevalence of obesity and PCOS was highest in women using GLP-1 receptor agonists; chronic hypertension, cardiovascular disease, and use of antihypertensive and lipid-modifying agents were highest in SGLT2 inhibitor users. Diabetic complications were highest among women using insulin and SGLT2 inhibitors. Among second-line noninsulin ADM users, coprescription fills of insulin were common (37%-82%), as were coprescription fills of metformin (25%-90%).

18For the subsample of pregnant women in the US (n = 397) and Israel (n = 575) with available laboratory data, the median periconceptional HbA1c levels were highest among those using either insulin or the second-line noninsulin ADM, particularly DPP-4 inhibitors and SGLT2 inhibitors, relative to other pregnant women with T2D treated with metformin or not treated pharmacologically (eTable 6 in Supplement 1).

19Figure 2 (and eTable 7 in Supplement 1) shows the prevalence of any MCM and cardiac malformations for each exposure group. As a reference, there were 132 283 infants born with an MCM in the full pregnancy cohort (3.76%), and 2584 infants born with an MCM in the study population of women with T2D (5.28%). Within the study population, the prevalence of MCMs was lower among infants with periconceptional exposure to no ADM (4.77%) or metformin only (5.32%) than among those exposed to insulin (7.83%), sulfonylureas (9.71%), DPP-4 inhibitors (6.14%), GLP-1 receptor agonists (8.23%), or SGLT2 inhibitors (7.04%). For cardiac malformations, the prevalence was similarly elevated among infants born to women with T2D (2.25% vs 1.31% in the full pregnancy cohort) and lower among infants with periconceptional exposure to no ADM (2.30%) or metformin only (2.04%) than among those exposed to insulin (4.20%), sulfonylureas (4.85%), DPP-4 inhibitors (3.26%), GLP-1 receptor agonists (3.22%), and SGLT2 inhibitors (3.88%).

20Table 2 presents the meta-analyzed crude and adjusted RRs (aRRs). Compared with infants with periconceptional exposure to insulin, the aRRs were compatible with no substantial increased risk for MCMs among infants with periconceptional exposure to sulfonylureas (aRR, 1.18; 95% CI, 0.94-1.48), DPP-4 inhibitors (aRR, 0.83; 95% CI, 0.64-1.06), GLP-1 receptor agonists (aRR, 0.95; 95% CI, 0.72-1.26), or SGLT2 inhibitors (aRR, 0.98; 95% CI, 0.65-1.46). Similarly, results did not suggest an increased risk for cardiac malformations after periconceptional exposure to any of the second-line noninsulin ADMs studied compared with insulin. Individual cohort RRs are presented in eTable 8 in Supplement 1.

21Restricting the exposure definition in the sensitivity analysis to the filling of 1 or more prescriptions in the first trimester resulted in fewer exposed infants: 1070 to sulfonylureas, 400 to DPP-4 inhibitors, 461 to GLP-1 receptor agonists, and 181 to SGLT2 inhibitors. Consistent with the main analysis, the crude RR estimates did not suggest an increased risk for any MCM after exposure to any of the second-line noninsulin ADMs compared with insulin: sulfonylureas (RR, 1.02; 95% CI, 0.79-1.32), DPP-4 inhibitors (RR, 0.82; 95% CI, 0.55-1.23), GLP-1 receptor agonists (RR, 1.03; 95% CI, 0.73-1.47), or SGLT2 inhibitors (RR, 1.20; 95% CI, 0.69-2.11). Results were also similar to the main analysis for cardiac malformations (eTable 9 in Supplement 1).

22In this cohort study by the InPreSS consortium including more than 50 000 pregnancies in women with pregestational T2D from 6 countries, we observed no elevated risk of MCMs after periconceptional exposure to GLP-1 receptor agonists or any of the second-line noninsulin ADM classes evaluated compared with insulin, another second-line ADM and the traditional treatment for T2D in pregnancy. Further, we showed an increase in periconceptional use of second-line noninsulin ADMs, particularly GLP-1 receptor agonists in the US, highlighting that there has been a shift in how T2D in reproductive-aged women is treated. Although this study did not suggest that these medications have strong teratogenic effects, there is a need for further research to fully evaluate the safety of these medications in pregnancy.

23T2D is an increasingly common condition in female individuals of reproductive age, and consequently in pregnant patients., In-line with previous studies, we found an elevated prevalence (5.3%) of MCMs in infants born to women with pregestational T2D, compared with the general population (3.7%).,, The effect of T2D is believed to be at least partially mediated by hyperglycemia because poor glycemic control during pregnancy is associated with an increased risk of MCMs and other adverse pregnancy outcomes.,,, This supports the importance of glycemic control and having safe and effective medications available during pregnancy.

24Although insulin does not cross the placenta and is considered nonteratogenic, little to no data are available on what risks, if any, noninsulin ADMs may pose when used during the time of embryogenesis. However, use of metformin may be considered according to some guidelines. Because metformin use in pregnant women with T2D has increased over time, and is also used for treatment of infertility and PCOS, there is some information on the safety of metformin exposure during the first trimester., We observed that in all 3 study cohorts, use of DPP-4 inhibitors, GLP-1 receptor agonists, and SGLT2 inhibitors has also increased rapidly over the last decade, mimicking their use in the general population of patients with T2D., Notably, use of GLP-1 receptor agonists has increased substantially, particularly in the US, among individuals with T2D, likely due to their weight loss effects. With the recent approval for specific medications within this class to be used as anti-obesity treatment, the number of infants prenatally exposed to this class of medications will presumably continue to increase.

25To reduce confounding by indication, we restricted our analysis to women with T2D who had used second-line ADMs periconceptionally and compared noninsulin vs insulin treatments, much like a hypothetical clinical trial would have pregestational T2D as an inclusion criterion and use of insulin as an active comparator. We have previously shown the importance of comparing ADM strategies used for treating similar severity of T2D in pregnancy to achieve balance in markers of glycemic control between comparison groups (eg, HbA1c) and reduce confounding by the underlying diabetes progression.

26Although the study design optimized clinical equipoise between comparison groups, residual bias due to the channeling of T2D patients with specific characteristics to specific second-line ADM treatments is expected because these medications are recommended based on the presence of comorbidities such as obesity, cardiac, and kidney diseases. The clinical characteristics (eg, comorbidities and comedication use) of the exposure groups were in-line with these treatment recommendations. Confounding by obesity and cardiovascular conditions would preferentially affect GLP-1 receptor agonists and SGLT2 inhibitors, and bias the RR estimates for MCMs upward. Reassuringly, adjusting for obesity in this study did not substantially affect the RRs, and given the expected direction of confounding, adjustment for additional and maternal comorbidities would likely attenuate the estimates toward the null. Moreover, the HbA1c levels were slightly higher in some of the second-line noninsulin ADM groups compared with insulin, indicating that confounding by glycemic control could, if anything, bias the RR for MCMs upward. Despite the likely overestimation of the RRs due to residual confounding, they were most compatible with a null effect relative to insulin.

27Filled prescriptions around conception might not result in exposure during embryogenesis, particularly for those filled before LMP. Among second-line noninsulin ADM users, coprescription fills of insulin or metformin were common in the periconceptional period, indicating scenarios that are difficult to disentangle in our data: concomitant use or adherence to guideline recommendations for switching to metformin or insulin. If there are increased risks conferred by use of second-line noninsulin ADM throughout the first trimester, then early pregnancy switching among the exposed groups could lead to an underestimation of those risks. We conducted a sensitivity analysis including only pregnant women with prescription fills during the first trimester. Fewer pregnancies were included in this analysis, yet the conclusion that there was not a substantial increased risk of MCMs overall or cardiac malformations remained.

28The study population was restricted to pregnancies resulting in live births because information on MCM was not available or was only partially available for pregnancies that resulted in stillbirth, miscarriage, or termination. Conditioning on livebirth might introduce selection bias and potentially underestimate the RRs only if MCMs were more lethal or preferentially terminated in pregnancies exposed to specific ADMs relative to those exposed to insulin; however, to our knowledge, there are currently no studies that have investigated whether there is evidence toward this. Evaluating the potential effect of noninsulin ADM on fertility, miscarriages, or pregnancy termination is challenging and beyond the scope of this study.

29Despite including data from 6 countries, the number of infants exposed to specific second-line noninsulin ADM classes remained low during the study period and the estimates, thus, imprecise with the upper limits of the 95% CI including up to a 2-fold increased risk. Although this study is valuable because there is no available information on the teratogenicity of these medications in humans, and results are reassuring that these drugs are not major teratogens, confirmation from other studies is needed. Because the use of these medications is becoming more common for treatment of T2D and for other indications (ie, obesity), the number of infants with prenatal exposure will increase.

30In this study, infants born to women with pregestational T2D were associated with having a higher prevalence of MCMs, including cardiac malformations, compared with infants in the general population. However, in infants born to women with T2D treated with second-line ADM, we did not observe a greater risk of MCMs after periconceptional exposure to sulfonylureas, DPP-4 inhibitors, GLP-1 receptor agonists, or SGLT2 inhibitors compared with insulin. Although reassuring, confirmation from other studies is needed, and continuous monitoring will provide more precise risk estimates in the future as data accumulate.