Abstract
Background
GLP-1 receptor agonist (GLP-1RA) use has increased among women of reproductive age, but limited data exist on safety in pregnancy.
Objective
To estimate the risk of non-live birth, abnormal fetal growth, and major congenital malformations (MCM) with GLP-1RA dispensing in early pregnancy.
Design
In an observational cohort of pregnant women aged 16-55 with a GLP-1RA dispensation in the 90 days prior to last menstrual period (LMP), a target trial with two treatment strategies was emulated: continuation of dispensing into the first trimester (≥ 1 further dispensation) or non-continuation.
Setting
Merative MarketScan US insurance claims data (2011-2024).
Participants
3,572 pregnancies (41.1%, n=1,467 among women with type 2 diabetes).
Measurements
Risk of non-live birth was estimated using a weighted Kaplan-Meier estimator. Among live birth pregnancies linked to infants, the weighted prevalence of MCM, small for gestational age (SGA), and large for gestational age (LGA) was estimated.
Results
The weighted risk of non-live birth was 29.7% with continuation and 27.1% with non-continuation (adjusted risk ratio 1.09 95% CI 0.98-1.23). Among 2,529 live birth pregnancies, 1,443 (57.1%) received at least one GLP-1RA dispensation after LMP and 1,499 (829 continuers) were linked to an infant. Weighted prevalence ratios for continuation versus non-continuation were 1.29 (95% CI 0.82-2.06) for SGA, 1.08 (95% CI 0.84-1.40) for LGA, and 1.21 (95% CI 0.83-1.82) for MCM.
Limitations
Potential residual confounding by prior glycemic control.
Conclusion
Risks of non-live birth, SGA, LGA and MCM were not definitively higher with continuation of GLP-1RAs into early pregnancy. However, estimates for MCM and SGA were imprecise and were compatible with both no increased risk and clinically relevant differences in risk.
Primary Funding Source
National Institutes of Health
Introduction
Glucagon-like peptide-1 receptor agonists (GLP-1RAs) are effective medications for weight loss and the treatment of type 2 diabetes (1), with evidence from trials supporting improved glycemic control and a reduction in cardiorenal adverse events (2,3).
Despite increasing usage among women of reproductive age, safety of GLP-1RAs in pregnancy remains uncertain (4). Current recommendations are to discontinue use prior to pregnancy. However, as many pregnancies are unplanned (5), unintentional exposure arises in the first trimester, an important period for embryological and fetal development (6). In studies of pregnant animals, administration during pregnancy of several GLP-1RAs including semaglutide and liraglutide, led to increased pregnancy losses, a reduction in fetal weights, and increased congenital anomalies (7). The applicability of these findings to humans is unclear given differences in physiology between species, duration and dosage of exposure, and underlying cardiometabolic state. In humans with diabetes or obesity, potential adverse direct effects might plausibly be negated by improvements to cardiometabolic health. Preliminary evidence in humans from observational studies (8-12), did not indicate, from a clinical perspective, substantially increased risk of spontaneous abortions, small for gestational age infants, or congenital malformations with first trimester exposure relative to comparator groups. However, small sample sizes have limited precision, and interpretation is challenging given potential differences in glycemia between comparator groups (4). In addition to the potential direct effects of GLP-1RAs in pregnancy, treatment discontinuation may impact perinatal outcomes through potential effects on maternal cardiometabolic state (13,14). Discontinuation of GLP-1RAs outside of pregnancy has been linked in clinical trial data to rapid weight regain and reversion of glycemic levels (15-18).
Further evaluation of the risks associated with GLP-1RA exposure in pregnancy is needed with a larger study population. To provide evidence, we identified a cohort of pregnancies among women using GLP-1RAs prior to pregnancy in US MarketScan commercial insurance claims data (19). We compared using target trial emulation the risk of non-live birth, abnormal fetal growth, and major congenital malformation with continuation of GLP-1RA dispensing into pregnancy versus non-continuation.
Methods
Data sources
This study was conducted using data for years 2011-2024 from Merative MarketScan Commercial Claims and Encounters (MarketScan CCAE), a longitudinal US commercial insurance claims dataset. MarketScan CCAE, one of the largest US nationwide datasets of commercial health insurance beneficiaries, contains deidentified healthcare utilization data including demographic characteristics, inpatient and outpatient diagnoses and procedures, and dispensed outpatient prescription medications (20).
Study population
The study population comprised pregnant women aged 16 to 55 years (19). Pregnancies were identified from end-of-pregnancy codes (19). Gestational age was estimated using a validated algorithm that utilizes codes indicative of gestational week at birth, complemented with, where available, gestational week Z3A ICD-10 codes recorded in prenatal visits (21). Pregnancies with at least one dispensation of a GLP-1RA in the 90 days prior to last menstrual period (LMP) were included (21). LMP was estimated based on the pregnancy end date and gestational age at pregnancy end. We restricted to pregnancies with a last dispensation prior to LMP of either 28, 30, 84, or 90 days’ supply (covering ~95% of dispensations) and with last dispensation date plus days’ supply occurring on or after two weeks prior to LMP (i.e., current/recent users at LMP). Restriction on days’ supply facilitates adjustment for this variable in analyses. The GLP-1RAs under investigation were exenatide, liraglutide, semaglutide, tirzepatide and dulaglutide. The study cohort was restricted to pregnancies with insurance coverage from six months before estimated LMP until the latest of pregnancy end and 280 days after LMP. Pregnant women enrolled as the child of an employee were excluded given the grandchildren of employees are not typically covered by the grandparent’s insurance and are therefore not typically linkable. The eligibility criteria and broader design of this observational study were guided by specification and emulation of a hypothetical pragmatic target trial (Supplementary Methods and Supplement Table 1). Target trial emulation helps investigators specify a causal question and can facilitate the alignment of treatment assignment and start of follow-up, thus avoiding immortal time bias (22,23).
Treatment strategies
Treatment strategies under comparison were non-continuation of GLP-1RA dispensing after LMP, defined by no further GLP-1RA dispensations during pregnancy after LMP, versus continuation of dispensing into the first trimester, defined by at least one dispensation within a grace period from LMP until end of remaining weeks’ supply from pre-LMP dispensation plus six weeks. As these treatment strategies are not distinguishable at baseline (LMP), to avoid immortal time bias in analyses, cloning and censoring was used (24-26), in which each pregnancy is cloned (i.e., duplicated in the data), assigned to both treatment strategies, and censored when treatment deviates from the assigned strategy (Supplementary Methods, Supplement Figure 1, and Supplement Table 2).
Outcomes
The primary outcomes were non-live birth (a composite outcome of spontaneous abortion, elective termination, or stillbirth), small for gestational age (SGA) at birth, large for gestational age (LGA) at birth, and non-chromosomal major congenital malformation (MCM). Secondary outcomes were the two most common specific non-live birth outcomes (spontaneous abortion and elective termination). Outcomes were ascertained from coded clinical diagnoses and procedures in maternal and infant claims (details and related validation studies in Supplement Table 3) (27). SGA, LGA and MCM were ascertained among the subset of live birth pregnancies linked to infants with infant enrolment to at least 90 days following delivery (except in case of earlier infant death). Infants may be enrolled under the separate insurance of the father and therefore not linkable. Presence of a neonatal outcome was determined for each pregnancy by presence of a code in the mother’s or infant’s claims following delivery. In a multiple pregnancy, an outcome was considered present if one or more fetuses had a diagnosis (28). As a post-hoc analysis, we estimated among elective terminations the frequency of recording of codes for a known or suspected chromosomal or non-chromosomal fetal abnormality (ICD-9 655, ICD-10 O35) by continuation status. These codes are a marker of the relative frequency of termination for fetal abnormality but are likely to under-capture malformations among terminations in claims data.
Covariates
Potential confounders were identified and included as baseline covariates. These were ascertained in the 6 months prior to LMP, except for medications which were ascertained in the 3 months prior, and included calendar year, maternal age, tobacco use, overweight, obese, severely obese, hypothyroidism, pregestational type 2 diabetes, indicators of diabetes severity and other diabetes treatments (diabetes complications, hyperglycemia, hypoglycemia, metformin, insulin, thiazolidinediones, sulfonylureas, DPP4 inhibitors, SGLT2 inhibitors), cardiometabolic comorbidities and medications (chronic hypertension, hypercholesterolemia, antihypertensive or cholesterol-lowering medication prior to LMP), fertility characteristics (polycystic ovarian syndrome [PCOS], infertility, use of fertility medication), tests ordered (hemoglobin A1c [HbA1c], lipid panel, urine albumin-to-creatinine [uACR] ratio) and number of outpatient visits. Covariate definitions and their assessment windows are described in Supplement Table 4 and Supplement Figure 2. First trimester dispensing of other antidiabetic medications and potential teratogens was ascertained for descriptive comparison of continuers and non-continuers.
Statistical analysis
Risks were estimated for each treatment strategy and compared via risk ratios and risk differences. The risk of non-live birth for each treatment strategy was estimated using a weighted Kaplan-Meier (KM) estimator with gestational week as the underlying timescale (24,26). As treatment strategies were not distinguishable at baseline, pregnancies were cloned and then censored when non-adherent to the assigned strategy. Censoring weights were used to account for potential confounding arising from censoring after cloning due to non-adherence to the treatment strategy of interest and were estimated using a pooled logistic regression model for treatment including terms for week of pregnancy, weeks since last dispensation, prior weeks’ supply (i.e., 4 or 12 weeks), and baseline covariates. Continuous covariates of maternal age and gestational week were modelled with restricted cubic splines while year of LMP was modelled using linear and quadratic terms. For specific non-live birth outcomes, total effects for each treatment strategy were estimated in the presence of competing events of other non-live births using a weighted estimator of risk based on the subdistribution hazard (29).
For live birth outcomes (SGA, LGA, and MCM), both risk of live birth with outcome and prevalence of the outcome among live births were estimated. The prevalence among live births was estimated for each treatment strategy by a weighted proportion among pregnancies linked to infants with sufficient enrolment. The risk of live birth with outcome was estimated for each treatment strategy by multiplying one minus the weighted risk of non-live birth by the weighted prevalence of the outcome among live births.
Nonparametric bootstrapping with 1,000 resamples, using resampling at the person-level, was used to estimate 95% confidence intervals by the percentile method (30). Statistical analyses were conducted using R version 4.5.1.
As a secondary analysis, estimated risks and prevalences were stratified by prior type 2 diabetes in case of effect modification by diabetes status. To assess the sensitivity of results to exposure definition we 1) redefined the length of the grace period after weeks’ supply from six weeks to three weeks, 2) restricted to pregnancies with day’s supply of last prior dispensation ending on or after LMP (i.e., current users), 3) restricted to pregnancies with less than 6 weeks remaining supply at LMP to reduce potential exposure due to prior dispensation, and 4) redefined baseline when defining exposure and covariates as LMP plus 2 weeks rather than LMP.
As a sensitivity analysis, in case GLP-1RA continuation is associated with continued use of potential teratogens or recreational drugs, the cohort was restricted to pregnancies without a dispensation of known or suspected teratogens in the 90 days prior to LMP and without prior recorded tobacco use, alcohol abuse, or illicit drug abuse.
Ethics approval
Ethics approval was obtained from the institutional review board of Harvard T.H. Chan School of Public Health (IRB15-3895).
Role of the funding source
The funder had no role in the study design, data collection, analysis, interpretation, or approval of the manuscript for submission.
Results
Description of study population
Among 2,135,592 pregnancies meeting maternal age and enrolment eligibility criteria there were 3,572 pregnancies from 3,427 women with a GLP-1RA dispensation in the 90 days prior to LMP and with prior days’ supply meeting eligibility criteria (Supplement Figure 3). Numbers of pregnancies in MarketScan CCAE with GLP-1RA dispensations in the 90 days prior to LMP increased substantially in recent years from 2 per 1,000 in 2020 to 15 per 1,000 in 2024 (Figure 1). Among the 3,572 pregnancies with preconception GLP-1RA use, mean age at LMP was 34.0, 41.1% (n=1,467) were among women with type 2 diabetes, 23.1% (n=826) among women with chronic hypertension, and 25.6% (n=915) among women with hypercholesterolemia (Table 1). The last GLP-1RA dispensation before LMP was most commonly (Supplement Table 5) either liraglutide (n=903, 25.3%) or semaglutide (n=1,644, 46.0%).
Figure 1: Frequency of GLP-1RA dispensation in the 90 days prior to LMP by year.
Definitions: GLP-1RA, glucagon-like peptide-1 receptor agonist; LMP, last menstrual period.
Note: Counts are among 2,135,592 pregnancies meeting maternal age and enrolment criteria. Wilson score confidence intervals are presented.
Table 1:
Unweighted characteristics of all pregnancies and of live birth pregnancies by GLP-1RA continuation status
| Characteristic | All pregnancies, N=3,572 |
Live birth pregnancies, N=2,529 | ||
|---|---|---|---|---|
| Continuer, N=1,443 | Non-continuer, N=1,086 |
ASMD | ||
| Age at LMP – mean (SD) | 34.0 (5.0) | 33.7 (4.9) | 32.9 (4.7) | 0.16 |
| Year of LMP | 0.01 | |||
| 2011-2016 | 445 (12.5%) | 169 (11.7%) | 132 (12.2%) | |
| 2017-2024 | 3,127 (87.5%) | 1,274 (88.3%) | 954 (87.8%) | |
| Tobacco use | 102 (2.9%) | 43 (3.0%) | 28 (2.6%) | 0.02 |
| Overweight or obese | 1,972 (55.2%) | 820 (56.8%) | 583 (53.7%) | 0.06 |
| Severe obese | 680 (19.0%) | 268 (18.6%) | 203 (18.7%) | <0.01 |
| Type 2 diabetes | 1,467 (41.1%) | 616 (42.7%) | 377 (34.7%) | 0.16 |
| Diabetes complication | 189 (5.3%) | 70 (4.9%) | 56 (5.2%) | 0.01 |
| Hyperglycemia | 683 (19.1%) | 281 (19.5%) | 191 (17.6%) | 0.05 |
| Hypoglycemia | 61 (1.7%) | 30 (2.1%) | 19 (1.7%) | 0.02 |
| Hypercholesterolemia | 915 (25.6%) | 373 (25.8%) | 248 (22.8%) | 0.07 |
| Hypothyroidism | 512 (14.3%) | 218 (15.1%) | 155 (14.3%) | 0.02 |
| Chronic hypertension | 826 (23.1%) | 330 (22.9%) | 224 (20.6%) | 0.05 |
| Infertility | 313 (8.8%) | 120 (8.3%) | 107 (9.9%) | 0.05 |
| PCOS | 639 (17.9%) | 281 (19.5%) | 198 (18.2%) | 0.03 |
| Prior insulin | 349 (9.8%) | 146 (10.1%) | 94 (8.7%) | 0.05 |
| Prior metformin | 1,135 (31.8%) | 472 (32.7%) | 332 (30.6%) | 0.05 |
| Prior sulfonylurea | 125 (3.5%) | 48 (3.3%) | 28 (2.6%) | 0.04 |
| Prior thiazolidinedione | 49 (1.4%) | 16 (1.1%) | 16 (1.5%) | 0.03 |
| Prior DPP4 inhibitor | 65 (1.8%) | 23 (1.6%) | 16 (1.5%) | 0.01 |
| Prior SGLT2 inhibitor | 202 (5.7%) | 82 (5.7%) | 51 (4.7%) | 0.04 |
| Prior cholesterol-lowering medication | 371 (10.4%) | 163 (11.3%) | 82 (7.6%) | 0.13 |
| Prior any antihypertensive | 886 (24.8%) | 356 (24.7%) | 239 (22.0%) | 0.06 |
| Prior any fertility medication | 183 (5.1%) | 77 (5.3%) | 61 (5.6%) | 0.01 |
| Prior known or suspected teratogen | 485 (13.6%) | 195 (13.5%) | 153 (14.1%) | 0.02 |
| Number of outpatient visits – mean (SD) | 8.9 (8.1) | 8.7 (8.0) | 8.7 (7.9) | <0.01 |
| HbA1c test | 2,055 (57.5%) | 848 (58.8%) | 592 (54.5%) | 0.09 |
| uACR test | 460 (12.9%) | 196 (13.6%) | 129 (11.9%) | 0.05 |
| Lipid panel | 1,679 (47.0%) | 708 (49.1%) | 475 (43.7%) | 0.11 |
| Weeks remaining GLP-1RA supply – mean (SD) | 2.4 (3.0) | 1.7 (2.2) | 3.3 (3.6) | 0.52 |
Definitions: ASMD, absolute standardized mean difference; DPP4, dipeptidyl peptidase-4 inhibitor; GLP-1RA, glucagon-like peptide-1 receptor agonist; HbA1c, hemoglobin A1c; LMP, last menstrual period; PCOS, polycystic ovary syndrome; SD, standard deviation; SGLT2, Sodium-glucose cotransporter-2; uACR, urine albumin-creatinine ratio.
There were 996 (27.8%) non-live births (791 spontaneous abortions, 168 elective terminations, and 37 stillbirths) and 2,576 live births. Of the 2,576 live birth pregnancies, 47 were not adherent to either treatment strategy due to receipt of a first post-LMP GLP-1RA after the grace period and were censored for both treatment strategies in analyses. Among the 2,529 remaining live birth pregnancies, 1,443 (57.1%) received at least one GLP-1RA dispensation after LMP (Table 1). The majority of continuer live birth pregnancies had only one GLP-1RA dispensation after LMP (n=918, 63.6%) and had dispensations in the first trimester only (n=1,209, 83.8% - Table 2 and Supplement Table 6). Continuers relative to non-continuers were on average slightly older (mean age at LMP of 33.7 [Standard Deviation, SD 4.9] vs. 32.9 [SD 4.7] years) and more frequently had type 2 diabetes (42.7% vs. 34.7%), prior use of cholesterol-lowering medication (11.3% vs. 7.6%), and a HbA1c test ordered in the six months prior to LMP (58.8% vs. 54.5%). Continuers had on average fewer weeks remaining supply of medication at LMP (1.7 [SD 2.2] vs. 3.3 [SD 3.6]). After weighting, absolute standardized mean differences were less than 0.1 for all baseline covariates (Supplement Table 7) and less than or equal to 0.1 for first trimester other antidiabetic medications (ADM) (Supplement Table 8).
Table 2:
Unweighted periconceptional and first trimester medication dispensing among live birth pregnancies by continuation status
| Characteristic | Continuer, N=1,443 | Non-continuer, N=1,086 |
ASMD |
|---|---|---|---|
| Week of last GLP-1RA dispensation before pregnancy end relative to LMP – median (IQR) | 3.0 (1.0, 7.0) | −4.0 (−6.0, −2.0) | 1.59 |
| Last GLP-1RA dispensation before delivery | 1.94 | ||
| Prior to LMP | 0 (0.0%) | 1,086 (100.0%) | |
| First trimester | 1,209 (83.8%) | 0 (0.0%) | |
| Second trimester | 114 (7.9%) | 0 (0.0%) | |
| Third trimester | 120 (8.3%) | 0 (0.0%) | |
| Number of GLP-1RA dispensations during pregnancy | 0.40 | ||
| 0 | 0 (0.0%) | 1,086 (100.0%) | |
| 1 | 918 (63.6%) | 0 (0.0%) | |
| 2+ | 525 (36.4%) | 0 (0.0%) | |
| First trimester insulin | 365 (25.3%) | 273 (25.1%) | <0.01 |
| First trimester metformin | 440 (30.5%) | 274 (25.2%) | 0.12 |
| First trimester sulfonylurea | 57 (4.0%) | 26 (2.4%) | 0.09 |
| First trimester thiazolidinedione | 11 (0.8%) | 6 (0.6%) | 0.03 |
| First trimester DPP4 inhibitor | 17 (1.2%) | 7 (0.6%) | 0.06 |
| First trimester SGLT2 inhibitor | 59 (4.1%) | 20 (1.8%) | 0.13 |
| First trimester known or suspected teratogen | 109 (7.6%) | 81 (7.5%) | <0.01 |
Definitions: ASMD, absolute standardized mean difference; DPP4, dipeptidyl peptidase-4 inhibitor; GLP-1RA, glucagon-like peptide-1 receptor agonist; IQR, interquartile range; LMP, last menstrual period; SGLT2, Sodium-glucose cotransporter-2.
Risk of non-live birth
The weighted risk of non-live birth (Figure 2 and Supplement Figure 4) was 29.7% (95% CI 27.4%-31.9%) for continuation and 27.1% (24.8%-29.6%) for non-continuation with an adjusted risk ratio (aRR) of 1.09 (95% CI 0.98-1.23). Data were most compatible with either no difference or a small difference in risk of spontaneous abortion (aRR 0.99, 95% CI 0.88-1.13) and with an increased frequency of elective terminations with continuation (1.78, 95% CI 1.29-2.06). Crude estimates differed little from adjusted estimates with a crude RR of 1.09 (95% CI 1.00-1.18) for non-live birth (Supplement Figure 5).
Figure 2: Weighted risk for non-live birth, abnormal fetal growth, and major congenital malformations by GLP-1RA continuation status among 3,572 pregnancies.
* Some non-live birth events are counted in both treatment strategies because of cloning and censoring.
Definitions: CI; confidence interval; SGA, small for gestational age; LGA, large for gestational age; MCM, major congenital malformation.
Note: Prevalence of SGA, LGA, and MCM were ascertained only among the subset of uncensored live birth pregnancies with linkage to infants (n=1,499 [829 continuers]). Event numbers are among uncensored pregnancies.
Prevalence of MCM, SGA, and LGA
Of 2,529 uncensored live birth pregnancies, 1,980 (78.3%) could be linked to an infant and 1,499 (59.3%) additionally had adequate infant enrolment. Baseline characteristics of live birth pregnancies with and without linkage and adequate infant enrolment were broadly similar (Supplement Table 9). The adjusted prevalence ratios among live births for continuation relative to non-continuation were 1.29 (95% CI 0.82-2.06) for SGA, 1.08 (95% CI 0.84-1.40) for LGA, and 1.21 (95% CI 0.83-1.82) for MCM. Risk ratios for live birth with SGA, LGA, and MCM among all pregnancies were near identical to ratios of the prevalence of these outcomes among live births (Supplement Figure 6). Crude estimates did not differ considerably from adjusted estimates (Supplement Figure 5). Counts of specific malformations were small (Table 3). Seven of one hundred and thirty-two elective terminations occurring with continuation (5.3%) and 8 of 51 with non-continuation (15.7%) were accompanied by a code for a known or suspected fetal malformation.
Table 3:
Unweighted counts for different non-chromosomal malformations groups among uncensored linked live birth pregnancies by continuation status
| Malformation group | Continuer, N=829 | Non-continuer, N=670 |
|---|---|---|
| Any major | 68 (8.2) | 47 (7.0) |
| Cardiovascular | 20 (2.4) | 17 (2.5) |
| Oral | 4 (0.5) | 3 (0.4) |
| Central nervous system | 5 (0.6) | 6 (0.9) |
| Eye | 0 (0.0) | 1 (0.1) |
| Ear | 2 (0.2) | 1 (0.1) |
| Other vascular | 1 (0.1) | 0 (0.0) |
| Respiratory | 2 (0.2) | 2 (0.3) |
| Gastrointestinal | 8 (1.0) | 3 (0.4) |
| Genital | 13 (1.6) | 9 (1.3) |
| Urinary | 7 (0.8) | 5 (0.7) |
| Musculoskeletal | 5 (0.6) | 3 (0.4) |
| Limb | 5 (0.6) | 6 (0.9) |
| Abdominal wall | 1 (0.1) | 0 (0.0) |
| Other | 0 (0.0) | 1 (0.1) |
Note: Pregnancies may have infants with malformations in more than one specific grouping.
Risk by strata defined by type 2 diabetes
Estimates in strata defined by recorded type 2 diabetes were broadly consistent with primary analyses though with wide confidence intervals for termination, SGA, and MCM (Figure 3).
Figure 3: Weighted risks for non-live birth, abnormal fetal growth, and major congenital malformations by GLP-1RA continuation status among pregnancies in women A) without (n=2,105) or B) with (n=1,467) recorded type 2 diabetes.
* Some non-live birth events are counted in both treatment strategies because of cloning and censoring.
Definitions: CI; confidence interval; SGA, small for gestational age; LGA, large for gestational age; MCM, major congenital malformation.
Note: Prevalence of SGA, LGA, and MCM were ascertained only among the subset of uncensored live birth pregnancies with linkage to infants (without recorded diabetes, n=933 [492 continuers]; with diabetes, n=566 [337 continuers]). Event numbers are among uncensored pregnancies.
Sensitivity analyses
Sensitivity analyses were consistent with the primary analysis, though confidence intervals were wide, in particular for SGA, MCM, and elective termination (Supplement Figures 7-11).
Discussion
In this nationwide US cohort study of 3,572 pregnancies among women taking GLP-1RAs preconception, we quantified the risk of non-live birth, abnormal fetal growth, and MCMs with continuation of GLP-1RA dispensing into early pregnancy versus non-continuation of dispensing after LMP. Data were most compatible, as evident from confidence intervals, with no or a small difference in risk of non-live birth, spontaneous abortion, and LGA; with no or a small-to-moderate difference in prevalence of SGA and MCM; and with increased frequency of elective termination with continuation. For MCMs, the confidence interval ranged from 1.5% reduced to 4.7% increased prevalence among GLP-1RA continuers relative to non-continuers. The prevalence of LGA among both continuers and non-continuers was relatively high, as might be anticipated given the known effect of underlying type 2 diabetes and obesity on macrosomia.
The lack of evidence for a strong effect on congenital malformations, abnormal fetal growth, or spontaneous abortions is in concordance with previous cohort studies (8-10,12). A multi-country cohort study using data from MarketScan CCAE, Israel, and four Nordic countries, compared the prevalence of malformations with periconceptional dispensation in the three months before or after LMP of different ADMs among women with pregestational type 2 diabetes (8). The standardized prevalence of MCMs, in that study, among 938 GLP-1RA users (461 with a first trimester dispensation) was not higher than among insulin users (aRR 0.95, 95% CI 0.72-1.26). Relative to that previous study, we have added three more years of MarketScan data, included GLP-1RA users without type 2 diabetes, analyzed additional outcomes of non-live birth and abnormal fetal growth, and focused the analysis on dispensation after LMP rather than periconceptional use in the 90 days before or after LMP (8).
Smaller cohort studies did not find substantially increased frequency of SGA, LGA, pregnancy losses, or malformations with GLP-1RA usage relative to comparator groups but precision was majorly limited by sample size (9,12). A cohort study conducted using electronic healthcare records from the US, Lebanon and Saudi Arabia was large but was limited by an inability to adjust for covariates due to the use of aggregate data (10). In that study among pregnant women with type 2 diabetes (n=143,593) the crude prevalence of malformations in those with (n=35,083) and without GLP-1RA use at any time in pregnancy was similar (3.3% prevalence in both groups).
Elective terminations were more frequent with GLP-1RA continuation relative to non-continuation in our study. One likely explanation is that there may be more unintended or unexpected pregnancies among continuers. Unintended pregnancies are typically recognized later, have more continuation into pregnancy of medications with potential fetal harm, and higher rates of elective termination (5,31,32). Studies have suggested potential improvements to fertility among women taking GLP-1RAs, leading anecdotally to more unplanned pregnancies (33,34). In the previous cohort study of GLP-1RA exposed pregnancies reported to teratology information services, elective terminations for personal reasons were more frequent among GLP-1RA exposed pregnancies (9). Another potential contributor is that women unintentionally exposed to drugs in pregnancy have been shown to have a higher tendency to terminate due to perceived teratogenic risk (35,36). Differences in elective termination between treatment strategies could also arise due to differences in the frequency of congenital malformations (37). However, the frequency of recording of codes for known or suspected fetal abnormality was not higher among elective terminations with continuation relative to non-continuation.
Limitations of the study include the potential for residual confounding, particularly due to differences in prior glycemic levels. Test values, including HbA1c values, are not available for most enrollees in MarketScan CCAE. However, comparable glycemic levels were achievable among pregnant women in a study of US claims for different antidiabetic treatment strategies through adjustment for routinely measured covariates alone (38). Restricting to GLP-1RA users should reduce potential confounding by indication. There were minimal differences observed between crude and adjusted effect estimates in our study. Time-varying confounding relating to use of other ADMs might be present. However, we did not find major differences in ADM dispensing in the first trimester between treatment strategies. Moreover, in the context of pregnancy, the reason for non-continuation may be due more to earlier pregnancy recognition rather than glycemic control. A further limitation is that only recognized pregnancies among women that have presented to a clinician will be included in this study (39,40). For this reason, the data are not informative for any potential effect on very early losses among unrecognized pregnancies.
Although the data are informative about the safety of GLP-1RA exposure early in pregnancy, they provide little information on later pregnancy exposure. Most continuers only had one further dispensation, which is indicative of delayed discontinuation due to later pregnancy recognition. Also, there is a possibility some women obtained GLP-1RA dispensations out-of-pocket which would not be captured. However, restriction to a population with insurance-covered GLP-1RA dispensations would be expected to reduce the likelihood of this. Furthermore, dispensation of medication does not necessarily equate to usage. However, continuers might be expected to continue medication use until at the very least their last dispensation date. Finally, while in validation studies sensitivity for SGA was low, specificity was high (41), and given misclassification is not anticipated to be differential, bias in relative risk estimates should be minimal.
Findings are not indicative of substantially increased risk of non-live birth, abnormal fetal growth, or MCM with continuation of GLP-1RA dispensing into the first trimester providing some reassurance for pregnant women with unintentional first trimester exposure. However, while data are compatible with no increased risk, they are also, given substantive uncertainty in effect estimates, compatible with clinically relevant differences in risk, particularly for the less common outcomes of SGA and MCM for which the upper bound of the confidence interval for the risk ratio is close to 2. Given uncertainty over potential effects, current manufacturer recommendations to stop GLP-1RA use prior to pregnancy serve as a precaution (42). A larger study, which should be feasible as further data accumulate, would be valuable to calculate increasingly precise estimates of risk and to estimate risk with specific GLP-1RAs, of specific malformation subtypes, and in subgroups.
Conclusions
In this cohort study of pregnant women with preconception GLP-1RA use, there was no indication of a strong effect of GLP-1RA continuation into the first trimester on non-live birth, SGA, LGA, or MCM. However, there was uncertainty in estimates, particularly for the rarer outcomes of SGA and MCM, for which confidence intervals were compatible with both no effect and clinically relevant differences in risk. Accumulated evidence provides some reassurance regarding risk of malformations following inadvertent exposure in women who become pregnant while on GLP-1RAs.
Supplementary Material
Financial Support:
National Institute of Child Health and Development (NICHD R01 HD097778)
Competing interests:
The authors declare the following competing interests: LS is an investigator on grants to her institution from GSK, unrelated to this work; KFH is an investigator on grants to her institution from UCB and Takeda, unrelated to this work; SHD reports being an investigator on research grants to her institution from Takeda and consulting for Moderna, UCB and Jansen, all unrelated to the present study; JPB, EP, EWS, and BTB report no conflicts of interest.
Footnotes
Reproducible research statement: Study protocol and statistical code: Available through request to Dr. Brown (jeremybrown@hsph.harvard.edu). Data set: MarketScan data available from Merative under license.
References
- 1.Vilsbøll T, Christensen M, Junker AE, et al. Effects of glucagon-like peptide-1 receptor agonists on weight loss: systematic review and meta-analyses of randomised controlled trials. BMJ. 2012;344(jan10 2):d7771. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Kristensen SL, Rørth R, Jhund PS, et al. Cardiovascular, mortality, and kidney outcomes with GLP-1 receptor agonists in patients with type 2 diabetes: a systematic review and meta-analysis of cardiovascular outcome trials. Lancet Diabetes Endocrinol. 2019;7(10):776–85. [DOI] [PubMed] [Google Scholar]
- 3.Yao H, Zhang A, Li D, et al. Comparative effectiveness of GLP-1 receptor agonists on glycaemic control, body weight, and lipid profile for type 2 diabetes: systematic review and network meta-analysis. BMJ. 2024;384:e076410. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Drummond RF, Seif KE, Reece EA. Glucagon-like peptide-1 receptor agonist use in pregnancy: a review. Am J Obstet Gynecol. 2025;232(1):17–25. [DOI] [PubMed] [Google Scholar]
- 5.Finer LB, Zolna MR. Declines in Unintended Pregnancy in the United States, 2008–2011. N Engl J Med. 2016;374(9):843–52. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Cunningham FG, Leveno KJ, Dashe JS, et al. Williams Obstetrics, 26e. New York: McGraw Hill; 2022. [Google Scholar]
- 7.Muller DRP, Stenvers DJ, Malekzadeh A, et al. Effects of GLP-1 agonists and SGLT2 inhibitors during pregnancy and lactation on offspring outcomes: a systematic review of the evidence. Front Endocrinol. 2023;14:1215356. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Cesta CE, Rotem R, Bateman BT, et al. Safety of GLP-1 Receptor Agonists and Other Second-Line Antidiabetics in Early Pregnancy. JAMA Intern Med. 2024;184(2). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Dao K, Shechtman S, Weber-Schoendorfer C, et al. Use of GLP1 receptor agonists in early pregnancy and reproductive safety: a multicentre, observational, prospective cohort study based on the databases of six Teratology Information Services. BMJ Open. 2024;14(4):e083550. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Lewin Z, Snow S, Lee JA, et al. Glucagon-Like Peptide-1 Receptor Agonists among Pregnancies with Pregestational Diabetes and Its Relationship with Congenital Malformations. Am J Perinatol. 2025. [DOI] [PubMed] [Google Scholar]
- 11.Pondugula N, Culhane JF, Lundsberg LS, et al. Gestational Weight Gain and Hypertensive Disorders of Pregnancy With Prepregnancy and Early Pregnancy Glucagon-Like Peptide-1 Receptor Agonist Exposure. Obstet Gynecol. 2025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Chou Y, Weng S, Cheng F, et al. Unintentional periconceptional exposure to glucagon-like peptide-1 receptor agonists and adverse pregnancy outcomes: A nationwide cohort study in Taiwan. Diabetes, Obes Metab. 2026;28(2):1420–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Schultes B, Ernst B, Timper K, et al. Pharmacological interventions for weight loss before conception—putative effects on subsequent gestational weight gain should be considered. Int J Obes. 2023;47(5):335–7. [DOI] [PubMed] [Google Scholar]
- 14.Skov K, Mandic IN, Nyborg KM. Semaglutide and pregnancy. Int J Gynecol Obstet. 2023;163(2):699–700. [DOI] [PubMed] [Google Scholar]
- 15.Aronne LJ, Sattar N, Horn DB, et al. Continued Treatment With Tirzepatide for Maintenance of Weight Reduction in Adults With Obesity. JAMA. 2024;331(1):38–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Rubino D, Abrahamsson N, Davies M, et al. Effect of Continued Weekly Subcutaneous Semaglutide vs Placebo on Weight Loss Maintenance in Adults With Overweight or Obesity. JAMA. 2021;325(14):1414–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Wilding JPH, Batterham RL, Davies M, et al. Weight regain and cardiometabolic effects after withdrawal of semaglutide: The STEP 1 trial extension. Diabetes, Obes Metab. 2022;24(8):1553–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Foghsgaard S, Vedtofte L, Andersen ES, et al. Liraglutide treatment for the prevention of glucose tolerance deterioration in women with prior gestational diabetes mellitus: A 52-week randomized controlled clinical trial. Diabetes, Obes Metab. 2024;26(1):201–14. [DOI] [PubMed] [Google Scholar]
- 19.MacDonald SC, Cohen JM, Panchaud A, et al. Identifying pregnancies in insurance claims data: Methods and application to retinoid teratogenic surveillance. Pharmacoepidem Drug Safe. 2019;28(9):1211–21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Butler AM, Nickel KB, Overman RA, et al. IBM MarketScan Research Databases. In: Sturkenboom M, Schink T, editors. Databases for Pharmacoepidemiological Research. 2021. p. 243–51. [Google Scholar]
- 21.Chiu YH, Huybrechts KF, Zhu Y, et al. Internal validation of gestational age estimation algorithms in health-care databases using pregnancies conceived through fertility procedures. Am J Epidemiology. 2024;193(8):1168–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiology. 2016;183(8):758–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hernán MA, Sterne JAC, Higgins JPT, et al. A structural description of biases that generate immortal time. Epidemiology. 2025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Cain LE, Robins JM, Lanoy E, et al. When to Start Treatment? A Systematic Approach to the Comparison of Dynamic Regimes Using Observational Data. Int J Biostat. 2010;6(2). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Hernán MA. How to estimate the effect of treatment duration on survival outcomes using observational data. BMJ. 2017;360:k182. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Hernán MA, Robins JM. What if. Boca Raton: Chapman & Hall/CRC; 2020. [Google Scholar]
- 27.Zhu Y, Bateman BT, Hernandez-Diaz S, et al. Validation of claims-based algorithms to identify non-live birth outcomes. Pharmacoepidemiol Drug Saf. 2023;32(4):468–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Brown JP, Yland JJ, Williams PL, et al. Accounting for Twins and Other Multiple Births in Perinatal Studies of Live Births Conducted Using Healthcare Administration Data. Epidemiology. 2025;36(2):165–73. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Young JG, Stensrud MJ, Tchetgen EJT, et al. A causal framework for classical statistical estimands in failure-time settings with competing events. Statist Med. 2020;39(8):1199–236. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Efron B, Tibshirani RJ. An Introduction to the Bootstrap. Chapman and Hall/CRC; 1994. [Google Scholar]
- 31.Branum AM, Ahrens KA. Trends in Timing of Pregnancy Awareness Among US Women. Matern Child Heal J. 2017;21(4):715–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Cleary BJ, Butt H, Strawbridge JD, et al. Medication use in early pregnancy-prevalence and determinants of use in a prospective cohort of women. Pharmacoepidemiol Drug Saf. 2010;19(4):408–17. [DOI] [PubMed] [Google Scholar]
- 33.Zhou L, Qu H, Yang L, et al. Effects of GLP1RAs on pregnancy rate and menstrual cyclicity in women with polycystic ovary syndrome: a meta-analysis and systematic review. BMC Endocr Disord. 2023;23(1):245. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Nadarajah S. Ozempic babies: are weight loss drugs leading to unintended pregnancies? BMJ. 2025;388:q2440. [DOI] [PubMed] [Google Scholar]
- 35.Koren G, Bologa M, Long D, et al. Perception of teratogenic risk by pregnant women exposed to drugs and chemicals during the first trimester. Am J Obstet Gynecol. 1989;160(5):1190–4. [DOI] [PubMed] [Google Scholar]
- 36.Walfisch A, Sermer C, Matok I, et al. Perception of Teratogenic Risk and the Rated Likelihood of Pregnancy Termination: Association with Maternal Depression. Can J Psychiatry. 2011;56(12):761–7. [DOI] [PubMed] [Google Scholar]
- 37.Schechtman KB, Gray DL, Baty JD, et al. Decision-making for termination of pregnancies with fetal anomalies: analysis of 53,000 pregnancies. Obstet Gynecol. 2002;99(2):216–22. [DOI] [PubMed] [Google Scholar]
- 38.Cesta CE, Hernández-Díaz S, Huybrechts KF, et al. Achieving comparability in glycemic control between antidiabetic treatment strategies in pregnancy when using real world data. Pharmacoepidemiol Drug Saf. 2023;32(12):1350–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Wilcox AJ, Weinberg CR, O’Connor JF, et al. Incidence of Early Loss of Pregnancy. N Engl J Med. 1988;319(4):189–94. [DOI] [PubMed] [Google Scholar]
- 40.Macklon NS, Geraedts JPM, Fauser BCJM. Conception to ongoing pregnancy: the ‘black box’ of early pregnancy loss. Hum Reprod Updat. 2002;8(4):333–43. [DOI] [PubMed] [Google Scholar]
- 41.Phiri K, Hernandez-Diaz S, Tsen LC, et al. Accuracy of ICD-9-CM coding to identify small for gestational age newborns. Pharmacoepidemiol Drug Saf. 2015;24(4):381–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Nordisk Novo. Victoza. Prescribing Information. 2017. [Google Scholar]
- 43.Wanis KN, Sarvet AL, Wen L et al. Grace periods in comparative effectiveness studies of sustained treatments. Journal of the Royal Statistical Society Series A: Statistics in Society. 2024. Aug;187(3):796–810. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.