Key Takeaways
- A population-level Australian cohort study of 177,409 children found no evidence that metformin exposure during pregnancy was associated with greater developmental vulnerability at school entry after researchers adjusted for major differences between exposed and unexposed pregnancies.
- Before adjustment, 18.2% of metformin-exposed children met the study’s definition of developmental vulnerability compared with 13.9% of unexposed children. However, mothers prescribed metformin differed substantially in diabetes, body mass index, polycystic ovary syndrome, and other characteristics.
- After adjustment, the relative risk was 0.97 (95% CI, 0.74 to 1.29), while the adjusted risk difference was -0.23 percentage points (95% CI, -4.01 to 3.54). These estimates did not show a statistically clear increase in risk.
- Researchers also found no statistically clear association in any of five individual developmental domains, including language and cognitive skills, social competence, and emotional maturity.
- Results remained broadly consistent when researchers restricted the analysis to pregnancies involving gestational or type 2 diabetes and when they examined first-trimester metformin exposure.
- The observational study does not prove that metformin has no effect on child development, and it did not examine every possible long-term outcome. Pregnant patients should not start, stop, or change metformin based on this study alone.
Introduction
A large Australian study found no evidence that metformin exposure during pregnancy was associated with increased developmental vulnerability at school entry after adjustment for measured differences between pregnancies.
Metformin is widely used to lower blood glucose and may be prescribed during pregnancy for conditions including diabetes. Because the medication crosses the placenta, researchers have continued to investigate whether exposure before birth could have consequences that become apparent later in childhood.
A large study published September 30, 2026, in JAMA Network Open provides reassuring evidence on one important part of that question.
Researchers followed 177,409 children in Victoria, Australia, through their first year of full-time school. They found no evidence that children exposed to metformin during pregnancy were more likely to show developmental vulnerability at school entry after accounting for substantial differences between women who were and were not prescribed the medication (Gordon et al., 2026).
The distinction between the raw and adjusted results is especially important. At first glance, developmental problems appeared considerably more common among metformin-exposed children. But women receiving metformin had very different medical profiles, including much higher rates of gestational diabetes, type 2 diabetes, and polycystic ovary syndrome.
Once researchers attempted to account for those differences, the apparent association disappeared.
What the Study Examined
Gordon and colleagues conducted a population-level cohort study using linked health, pharmaceutical, pregnancy, and childhood-development records from Victoria, Australia.
The investigators began with singleton births from January 2009 through December 2020 that could be linked with childhood developmental data. Their eligibility criteria also excluded children with congenital anomalies and certain pregnancies involving uncertain diabetes diagnoses or insulin use without a recorded diabetes diagnosis.
The final cohort contained 177,409 children. Of those, 1,095 had been exposed to metformin during pregnancy and 176,314 had not (Gordon et al., 2026).
Metformin exposure was identified through Australia’s Pharmaceutical Benefits Scheme. A pregnancy was considered exposed when at least one metformin prescription was dispensed between the calculated first day of the mother’s last menstrual period and delivery.
That definition is stronger than relying on a prescription being written, but it still cannot establish whether the medication was actually taken or how consistently it was used.
How Development Was Assessed
Researchers assessed development using the Australian Early Development Census, a standardized assessment completed by teachers during a child’s first year of full-time school.
The assessment covered five areas: physical health and well-being, social competence, emotional maturity, school-based language and cognitive skills, and communication skills and general knowledge.
The study’s primary outcome was developmental vulnerability, defined as scoring below the 10th percentile in at least two of those five domains (Gordon et al., 2026).
What Researchers Found
The unadjusted numbers initially appeared concerning.
Among children exposed to metformin during pregnancy, 199 of 1,095, or 18.2%, met the definition of developmental vulnerability. Among children without recorded metformin exposure, 24,379 of 176,314, or 13.9%, met the definition.
That corresponded to an unadjusted relative risk of 1.31, suggesting 31% higher relative risk among exposed children (Gordon et al., 2026).
But interpreting that number as evidence that metformin harmed development would be misleading.
The two groups differed considerably before statistical adjustment.
Women prescribed metformin had a mean body mass index of 32.3 compared with 26.0 among those without metformin exposure. Gestational diabetes was recorded in 39.2% of the metformin group compared with 8.3% of the unexposed group. Type 2 diabetes affected 17.9% versus 0.5%, while polycystic ovary syndrome affected 22.8% versus 2.3%.
Metformin-exposed pregnancies also had higher rates of preterm birth and cesarean delivery (Gordon et al., 2026).
These differences matter because the health conditions that lead someone to receive metformin may themselves be related to pregnancy and childhood outcomes. This creates a major challenge when trying to separate a possible medication effect from the characteristics of the patients receiving the medication.
The Association Disappeared After Adjustment
The researchers used inverse probability of treatment weighting combined with regression adjustment to account for measured differences between the groups.
After adjustment, metformin exposure was not associated with greater developmental vulnerability.
The adjusted relative risk was 0.97 (95% CI, 0.74 to 1.29).
The adjusted risk difference was -0.23 percentage points (95% CI, -4.01 to 3.54) (Gordon et al., 2026).
Importantly, the relative risk of 0.97 should not be interpreted as evidence that metformin protects child development. The confidence interval crossed 1 and was compatible with both lower and higher risk.
The appropriate conclusion is that the study did not identify a statistically clear association after adjustment.
No Clear Difference Across Five Areas of Development
The researchers also examined each developmental domain separately.
After adjustment, metformin exposure was not associated with a statistically clear change in risk for physical health and well-being, social competence, emotional maturity, school-based language and cognitive skills, or communication skills and general knowledge (Gordon et al., 2026).
The adjusted relative risks ranged from 0.84 for emotional maturity to 1.09 for school-based language and cognitive skills, with all confidence intervals including the null value.
The consistency across several developmental measures is reassuring. However, it does not establish that metformin is incapable of affecting any aspect of childhood development.
What Happened When Researchers Focused on Diabetes?
One of the study’s central challenges is that metformin is not randomly prescribed. Women who receive it differ from women who do not, and diabetes is one of the most important reasons for treatment.
The researchers therefore performed a sensitivity analysis restricted to 16,121 pregnancies involving either gestational diabetes or type 2 diabetes.
Within this group, 20.7% of metformin-exposed children met the developmental-vulnerability definition compared with 15.9% of unexposed children.
Again, the unadjusted association appeared concerning, with a relative risk of 1.30.
After adjustment, however, the relative risk fell to 1.08 (95% CI, 0.86 to 1.35), with no statistically clear association between metformin exposure and developmental vulnerability (Gordon et al., 2026).
The same general pattern appeared across the five individual developmental domains.
Restricting the analysis to pregnancies involving diabetes cannot eliminate all differences between treated and untreated patients. However, the results provide additional evidence that the higher crude rate observed in metformin-exposed children should not automatically be attributed to the medication itself.
What About Metformin During the First Trimester?
Timing matters when researchers investigate medication exposure during pregnancy because fetal organs and tissues develop at different stages.
The investigators therefore separately examined first-trimester exposure.
Among children exposed to metformin during the first trimester, 17.2% were classified as developmentally vulnerable compared with 13.9% of children without metformin exposure.
After adjustment, first-trimester exposure was not associated with a statistically clear increase in developmental vulnerability. The adjusted relative risk was 1.07 (95% CI, 0.77 to 1.48) (Gordon et al., 2026).
Again, the confidence interval is important. The result should not be described as proving that first-trimester exposure is harmless. Rather, the analysis did not detect evidence of increased risk for the developmental outcome measured in this study.
How It Compares With Earlier Research
The new findings are broadly consistent with previous research examining neurodevelopment after prenatal metformin exposure.
A 2024 systematic review and meta-analysis evaluated seven studies involving 14,042 children, including 7,641 who were exposed to metformin during pregnancy (Gordon et al., 2024).
The meta-analysis found no statistically clear association between prenatal metformin exposure and neurodevelopmental delay during infancy or at ages 3 to 5. It also found no significant differences in pooled motor or cognitive scores (Gordon et al., 2024).
The new study substantially expands the number of children examined and adds a standardized developmental assessment at school entry.
It does not, however, settle every question about prenatal metformin exposure. Neurodevelopment is only one dimension of childhood health. Gordon and colleagues note that evidence concerning later growth and cardiometabolic outcomes remains less certain (Gordon et al., 2026).
What the Results May Mean
For patients who need metformin during pregnancy, the study provides a measure of reassurance about developmental outcomes at school entry.
It does not show that metformin improves development, nor does it establish complete long-term safety.
Instead, the central finding is narrower: after accounting for measured differences between exposed and unexposed pregnancies, researchers found no evidence that metformin dispensing during pregnancy was associated with increased developmental vulnerability at school entry.
The study also demonstrates why raw percentages from observational research need context.
If researchers had stopped at the unadjusted numbers, metformin exposure would have appeared to be associated with substantially greater developmental risk. But the patients receiving metformin were much more likely to have medical conditions that distinguished them from the comparison group.
Statistical adjustment cannot perfectly reproduce a randomized trial, and unmeasured differences may remain. In this study, however, adjustment substantially changed the estimated association.
Pregnant patients should not stop, reduce, start, or otherwise change metformin solely because of this study. The reason for treatment, blood glucose control, other medical conditions, pregnancy history, and alternative treatment options all matter. Medication decisions during pregnancy should be made with the clinician managing the pregnancy and the condition for which metformin was prescribed.
Limitations
The study’s observational design is its most important limitation. Metformin exposure was not randomly assigned, so differences between treated and untreated pregnancies could remain even after statistical adjustment.
The indication for metformin was not directly available. Researchers used maternal health conditions to help identify likely reasons for treatment and incorporated relevant variables into their analyses, but residual confounding remains possible.
Race and ethnicity data were also unavailable. The researchers considered maternal country of birth as a possible proxy but did not include it in the final models because they considered it an inadequate measure of race and ethnicity in the Australian population.
Medication exposure was based on pharmacy dispensing. A dispensed prescription does not prove that a patient took the medication, and the researchers could not determine the dose actually taken.
Only 467 women in the exposed group, or 42.6%, had more than one metformin prescription dispensed during pregnancy. Despite the large overall cohort, some planned analyses involving multiple prescriptions and treatment initiated during the second or third trimester did not have sufficient statistical power for the researchers’ preferred adjusted models (Gordon et al., 2026).
Some women could also have filled metformin shortly before conception and continued taking it during early pregnancy without another pregnancy-period dispensing. Those pregnancies could have been misclassified as unexposed.
The developmental assessment has boundaries as well. The Australian Early Development Census is a standardized measure of development at school entry, but it does not capture every possible neurological, psychiatric, behavioral, educational, or metabolic outcome.
The study therefore cannot determine whether prenatal metformin exposure affects outcomes that emerge later in childhood or adolescence.
Finally, children with congenital anomalies were excluded. The findings should not be extended to outcomes or populations that were not represented in the analysis.
Final Thoughts
The study provides substantial new evidence on a question that matters to patients who use metformin during pregnancy.
Among 177,409 Australian children, developmental vulnerability at school entry initially appeared more common following prenatal metformin exposure. But women receiving metformin had markedly different health profiles from women who did not receive it.
After researchers accounted for measured differences, metformin exposure was not associated with increased developmental vulnerability overall or in any of the five individual developmental domains. Similar results appeared when researchers focused on pregnancies involving gestational or type 2 diabetes and when they examined first-trimester exposure (Gordon et al., 2026).
Those findings are reassuring, but they are not equivalent to proving that metformin has no long-term effects. The study was observational, exposure was based on dispensing records, medication dose was unavailable, and development was assessed at one stage of childhood.
The most defensible conclusion is therefore a measured one: this large study found no evidence of increased developmental vulnerability at school entry associated with metformin exposure during pregnancy after adjustment for measured confounding.
Funding and Disclosures
The 2026 study received funding from the Norman Beischer Medical Research Foundation and the Australasian Diabetes in Pregnancy Society. Several investigators also received Australian government or National Health and Medical Research Council scholarship and fellowship support. The researchers reported that the funders had no role in the study design, data collection or management, analysis, interpretation, manuscript preparation, or decision to submit the work for publication (Gordon et al., 2026).
Stephen Tong reported National Health and Medical Research Council grants and personal fees from Diamedica Therapeutics, Eli Lilly, Beech Biopharma, and Avilar Therapeutics outside the submitted work. He also reported involvement in randomized clinical trials evaluating metformin for the treatment or prevention of preeclampsia. Roxanne Hastie reported grants from the National Health and Medical Research Council and Medical Research Future Fund during the study. No other disclosures were reported (Gordon et al., 2026).
The 2024 systematic review reported no conflicts of interest. Its authors reported Australian government and National Health and Medical Research Council scholarship or research support, with the funders having no role in the research or publication decision (Gordon et al., 2024).
References
Gordon, H., Hiscock, R. J., Price, S., Shub, A., Roddy Mitchell, A., Atkinson, J., Walker, S., Forsythe, A., Tong, S., Lindquist, A., & Hastie, R. (2026). Metformin exposure in pregnancy and childhood developmental outcomes. JAMA Network Open, 9(9), e2637054. https://doi.org/10.1001/jamanetworkopen.2026.37054
Gordon, H. G., Atkinson, J. A., Tong, S., Mehdipour, P., Cluver, C., Walker, S. P., Lindquist, A. C., & Hastie, R. M. (2024). Metformin in pregnancy and childhood neurodevelopmental outcomes: A systematic review and meta-analysis. American Journal of Obstetrics & Gynecology, 231(3), 308–314.e6. https://doi.org/10.1016/j.ajog.2024.02.316