Women who kept taking a GLP-1 drug into the first trimester of pregnancy were not clearly more likely to lose the pregnancy or to have a baby with a birth defect. They were also not clearly in the clear. That is the honest shape of a study published June 9 ↗ in Annals of Internal Medicine, one of the first sizable human cohorts to ask what happens when these drugs are not stopped before conception.
The question is not hypothetical. GLP-1 drugs like semaglutide (sold as Ozempic and Wegovy) and tirzepatide (Mounjaro and Zepbound) are taken by a fast-growing number of women of childbearing age, and the labels tell them to stop before trying to conceive. Animal studies found fetal harm at high doses, and human data has been thin. Weight loss can also restore fertility in women who were not ovulating, so some conceive without planning to, while still on the drug. The guidance to stop assumes a clean handoff that does not always happen.
The researchers used a method called target trial emulation, which takes insurance claims and analyzes them as if patients had been randomly assigned. From Merative MarketScan US claims spanning 2011 to 2024 they found 3,572 pregnancies in women aged 16 to 55 who had filled a GLP-1 prescription in the 90 days before their last period. About 41 percent (1,467) had type 2 diabetes. They then compared two strategies: keep filling the drug into the first trimester, or stop.
The headline outcome was reassuring on its face. Nonlive birth, a category that covers miscarriage and stillbirth, occurred in 29.7 percent of pregnancies where the drug continued and 27.1 percent where it stopped. After adjustment that came to a risk ratio of 1.09, with a confidence interval (0.98 to 1.23) that brushes up against no difference. The rate is high in both arms because these are higher-risk pregnancies to begin with, not because of the drug.
Reassuring, then imprecise
The birth outcomes among live babies are where the caution lives. Compared with stopping, continuing the drug carried a prevalence ratio of 1.29 for a baby being small for gestational age, 1.08 for large for gestational age, and 1.21 for a major congenital malformation, meaning a structural birth defect. None of those reached statistical significance. But read the spread, not just the middle. The interval for small-for-gestational-age ran from 0.82 to 2.06, and for malformations from 0.83 to 1.82. Those ranges are wide enough to include both "no effect" and "a real, clinically meaningful increase in risk." The study cannot tell the two apart.
That is the difference between no evidence of harm and evidence of no harm, and this is the first one. The authors say as much. Risks were not definitively higher, but the estimates for malformations and growth restriction were imprecise and compatible with both no increased risk and a difference that would matter to a parent. Residual confounding by prior blood-sugar control is the limitation they flag, since diabetes itself raises some of these risks.
For the millions of people on these drugs, the practical read is narrow. A woman who finds out she is pregnant while taking semaglutide or tirzepatide now has a real human dataset rather than only animal studies, and the central estimates do not show alarm. The dataset is also too small at the tails to promise safety. Both of those are true at once, which is the part that does not fit on a reassuring chart.
The two drugs that anchor this question sit on peptidemodel as semaglutide ↗ and tirzepatide ↗, both acting on the GLP-1 receptor ↗. Pregnancy is one of the last large blanks in their safety record, and it will take a prospective registry, not a claims database, to fill it.