GLP-1 Exposure Around Conception: Why “No Detectable Increase” Is Not Proof of Pregnancy Safety

The best current pooled evidence did not detect a clear increase in several adverse pregnancy outcomes after GLP-1 receptor-agonist exposure around conception. That is limited reassurance about reported exposures—not proof that GLP-1 medicines are safe in pregnancy. The distinction matters because the headline population exceeded 2.1 million pregnancies, while only 8,325 were reported as exposed, and some outcome analyses were far smaller.

A large overall study population can narrow quickly once researchers isolate exposed pregnancies and individual outcomes. A null pooled estimate means no increase was clearly detected in the available data; it does not establish zero risk.

Quick answer: A September 29, 2026 Med systematic review included 10 observational studies covering 2,118,215 pregnancies, with 8,325 reported as exposed to a GLP-1 receptor agonist around conception. Nine studies entered meta-analysis. The pooled estimates did not show clearly detectable increases across the outcomes assessed, but the authors explicitly concluded that heterogeneity, residual confounding, and limited datasets prevent those results from establishing safety.[^med]

What did the 2026 GLP-1 pregnancy safety meta-analysis find?

Across the pooled analyses, researchers did not identify a clearly detectable increase in miscarriage or intrauterine death, congenital malformations, preterm birth, hypertensive disorders of pregnancy, gestational diabetes, fetal-growth outcomes, or excessive gestational weight gain.

The estimated odds ratios were close enough to 1.0, or imprecise enough, that the analyses did not show clear evidence of increased risk:

| Outcome | Datasets and total sample | Pooled odds ratio (95% CI) | What the result means | |---|---:|---:|---| | Miscarriage or intrauterine death | 2 datasets; 1,138 | 1.30 (0.76–2.23) | No clear difference detected; the interval remains compatible with lower or higher risk | | Congenital malformations | 4 datasets; 577,620 | 0.92 (0.71–1.19) | No clear difference detected | | Preterm birth | 4 datasets; 1,079,725 | 1.05 (0.80–1.38) | No clear difference detected; substantial heterogeneity | | Hypertensive disorders of pregnancy | 4 datasets; 1,078,629 | 0.92 (0.76–1.11) | No clear difference detected; substantial heterogeneity | | Gestational diabetes | 3 datasets; 1,078,367 | 0.94 (0.74–1.21) | No clear difference detected; substantial heterogeneity | | Small for gestational age or fetal-growth restriction | 3 datasets; 1,069,975 | 0.99 (0.78–1.25) | No clear difference detected | | Large for gestational age | 2 datasets; 1,069,229 | 1.04 (0.90–1.21) | No clear difference detected | | Excessive gestational weight gain | 2 datasets; 2,882 | 0.90 (0.52–1.57) | No clear difference detected; substantial heterogeneity |

The review also reported a lower pooled estimate for preeclampsia: OR 0.87 (95% CI 0.78–0.98). That result came from only two datasets, with one contributing nearly 90% of the statistical weight. The authors treated it as hypothesis-generating, not evidence that GLP-1 exposure protects against preeclampsia.[^med]

Why is the 2.1-million headline easy to overread?

The 2,118,215 figure describes everyone represented across the systematic review, not 2.1 million GLP-1-exposed pregnancies and not 2.1 million observations for every outcome. Only 8,325 pregnancies were reported as exposed, about 0.4% of the headline population.

The usable denominator changed again for each outcome. Congenital-malformation analysis involved 577,620 participants across four datasets. The miscarriage or intrauterine-death analysis involved just 1,138 participants across two datasets. Excessive gestational weight gain involved 2,882 across two datasets.

This is normal in meta-analysis: each source study measures different outcomes and contributes different data. It is still crucial context. A press headline can inherit the largest denominator even when the question a reader cares about rests on a much smaller evidence pool.

Evidence-literacy rule: Always ask three denominator questions: How many pregnancies were included overall? How many were actually exposed? How many contributed to the specific outcome being discussed?

Why does “no detectable increase” not prove safety?

A null meta-analysis says the available studies did not distinguish the pooled estimate from no association; it does not prove that the true effect is exactly zero. Several features of this review limit what its null findings can exclude.

1. All included studies were observational. Researchers adjusted for measured differences, but unmeasured or imperfectly measured factors can still influence outcomes.
2. Only nine of the 10 studies entered meta-analysis. One was excluded to avoid potentially overlapping cases.
3. Some outcomes relied on two datasets. That makes pooled results more sensitive to the design and weight of individual studies.
4. Exposure windows differed. The review accepted exposure within six months before a positive pregnancy test, while source studies varied in timing and duration.
5. Indications differed. Diabetes, obesity, baseline metabolic risk, prior weight change, glycemic control, and healthcare engagement can all affect pregnancy outcomes independently of medication exposure.
6. Heterogeneity was substantial for several outcomes. The reported I² values were 85% for preterm birth, 82% for hypertensive disorders, 79% for gestational diabetes, and 81% for excessive gestational weight gain.
7. Dose and specific agent were not consistently assessable. Older prescribing patterns may not represent today's broader use or higher-dose exposure patterns.

The congenital-malformation sensitivity analysis deserves another caution. Restricting exposure to 90 days before conception through the end of the first trimester produced the same estimate as the primary analysis—but it used the same four studies and participants. It was not an independent confirmation.[^med]

What does residual confounding mean here?

Residual confounding means exposed and unexposed groups may differ in ways that statistical adjustment does not fully remove. That problem is especially difficult when the medication indication is itself tied to pregnancy risk.

People receiving GLP-1 therapies may have type 2 diabetes, obesity, or both. Those conditions, their severity, metabolic changes before conception, other medicines, access to care, and how exposure was recorded can each affect observed outcomes. A pooled association can therefore reflect some mixture of drug exposure, underlying health, treatment selection, and healthcare behavior.

This is why the apparently lower preeclampsia estimate should not be converted into a benefit claim. The result could reflect preconception metabolic changes or baseline differences rather than a pharmacologic effect during pregnancy.

For a broader example of why observational associations do not establish causality, see our analysis of a large GLP-1 and tuberculosis study.

How does the Med review compare with the Lancet consensus review?

The contemporaneous Lancet Obstetrics, Gynaecology & Women’s Health review reached a similarly cautious conclusion on congenital anomalies but reported an early-pregnancy-loss association that is difficult to interpret.[^lancet]

In the Lancet review:

• no increase in congenital anomalies was observed when GLP-1 receptor agonists were discontinued during the first trimester: RR 1.02 (95% CI 0.96–1.08), four studies, 42,144 exposed participants;
• early pregnancy loss was more common in the pooled analysis: RR 1.31 (95% CI 1.26–1.35), two studies, 41,046 participants;
• the early-loss outcome combined miscarriage with termination in the largest contributing study;
• that one dataset supplied about 99% of participants in the early-loss analysis;
• the authors graded the evidence as very low certainty.

Those definitions matter. A combined endpoint of miscarriage or termination cannot be read as a miscarriage estimate because termination may reflect personal choice, prenatal findings, clinician counseling, or other factors. When almost all participants come from the study using that combined outcome, the pooled number largely inherits that study's limitations.

The two reviews are therefore less contradictory than the raw numbers suggest. Both found no congenital-anomaly signal in the data they analyzed. Both relied on non-randomized evidence. Both left major uncertainty. The difference in early-loss findings reflects, at least partly, different included studies, populations, exposure windows, and outcome definitions.

What does this evidence say—and not say—about GLP-1 pregnancy safety?

The evidence offers limited reassurance about inadvertent exposure around conception or early pregnancy, but it does not establish safety for use during pregnancy. Regulatory status and observational findings answer different questions.

GLP-1 receptor agonists are not approved for use in pregnancy. That position is not reversed by a meta-analysis that failed to detect clear increases in selected outcomes. Regulatory decisions consider the total evidence package, including animal data, product-specific information, exposure timing, pharmacology, and gaps in human evidence—not just whether one pooled analysis crossed a statistical-significance threshold.[^fda]

The current evidence is especially sparse for:

• exposure that continues through the second or third trimester;
• breastfeeding and infant exposure;
• outcomes beyond birth, including long-term child growth, metabolism, and neurodevelopment;
• specific agents and doses analyzed separately;
• newer and higher-dose use patterns;
• organ-specific congenital anomalies and uncommon outcomes.

This article does not provide start, stop, washout, contraception, fertility, dosing, product-selection, or pregnancy-management instructions. Those questions require individualized clinical context that these pooled observational data cannot supply.

The bottom line

The strongest defensible reading is narrow: current observational evidence did not detect a clear increase in several studied outcomes after reported periconceptional GLP-1 receptor-agonist exposure. The overall dataset was large, but the exposed group was much smaller, outcome-specific denominators ranged widely, some analyses depended on two datasets, and residual confounding cannot be removed.

“No detectable increase” is useful evidence. It is not a safety certificate.

For context on how different evidence sources answer different safety questions, read GLP-1 Side Effects: What Trials, Labels, and Online Communities Can—and Cannot—Tell You and GLP-1 Peptides vs “Research Peptides”: What People Mean and Why It Matters.

Frequently asked questions

Does the 2026 meta-analysis prove GLP-1 medicines are safe in pregnancy?

No. It found no clearly detectable increase in the outcomes analyzed, but the authors explicitly said the observational evidence does not establish safety.

Were 2.1 million pregnancies exposed to GLP-1 receptor agonists?

No. The review included 2,118,215 pregnancies overall, of which 8,325 were reported as exposed around conception.

Did every outcome use the full 2.1-million dataset?

No. Outcome-specific samples varied substantially. The miscarriage or intrauterine-death analysis included 1,138 participants, while congenital malformations included 577,620.

Did the reviews find a congenital-anomaly signal?

Neither the Med review nor the contemporaneous Lancet review found a clearly increased pooled congenital-anomaly risk in the data analyzed. That does not rule out smaller effects, agent-specific risks, rare outcomes, or risks from exposure later in pregnancy.

Why is the Lancet early-pregnancy-loss result hard to interpret?

Its largest contributing study combined miscarriage with termination, and that dataset contributed about 99% of participants to the pooled early-loss analysis. The result therefore cannot be interpreted as a clean estimate of miscarriage risk.

This article is for general education and evidence literacy. It is not medical advice or individualized pregnancy care.

[^med]: D’Antonio F, Flacco ME, Manzoli L, Samara A, Khalil A. “GLP-1 receptor agonist exposure in the periconceptional period and adverse obstetric outcomes: A systematic review and meta-analysis.” Med. Published online September 29, 2026. doi:10.1016/j.medj.2026.101267. Open-access full text. [^lancet]: Dib S, Savage MJ, Adams D, et al. “Preconception, pregnancy, and postnatal use of GLP-1 receptor agonist-based therapies in women with type 1, type 2, or gestational diabetes: a systematic review, meta-analysis, and international expert consensus statement.” The Lancet Obstetrics, Gynaecology & Women’s Health. Published online September 29, 2026. doi:10.1016/S3050-5038(26)00266-9. Repository abstract. [^fda]: U.S. Food and Drug Administration. WEGOVY (semaglutide) Prescribing Information. 2026. Current FDA label. Product labels are product-specific; this citation illustrates why regulatory labeling should not be replaced by class-level observational reassurance.