Metformin in pregnancy was not linked to developmental problems at school age in 177,409 Australian children

Metformin is one of the most widely used medicines there is, and more and more women take it while pregnant. It passes through the placenta to the baby. What that means for a child years later has been hard to answer, because the children have to grow up first.

A study in JAMA Network Open, published on September 30, followed children to their first year of school. Among more than 177,000 children in the Australian state of Victoria, those whose mothers had taken metformin in pregnancy were more often flagged by their teachers as developmentally vulnerable. But their mothers were also heavier and far more likely to have diabetes. Once those differences were accounted for, the gap was gone.

Why metformin in pregnancy raises a question

Metformin is a tablet that lowers blood sugar, and its use in expectant mothers is growing. The paper calls it an oral hypoglycemic agent increasingly prescribed during pregnancy. The main reasons are gestational diabetes, which is diabetes that first appears in pregnancy, and type 2 diabetes. It is also given to some women with polycystic ovary syndrome, a hormonal condition.

The case for it rests on a trial from 2008. Researchers in New Zealand and Australia enrolled 751 women with gestational diabetes mellitus at 20 to 33 weeks of gestation and assigned them to metformin or insulin. Complications in the newborns were the same either way. The rate of the primary composite outcome was 32.0% in the group assigned to metformin and 32.2% in the insulin group, a composite being a combined count of several problems. And the women had a clear preference for tablets over injections: more women in the metformin group than in the insulin group stated that they would choose to receive their assigned treatment again (76.6% vs. 27.2%).

The worry is about what a trial lasting until birth cannot see. During pregnancy, metformin crosses the placenta, with the fetus exposed to similar medication concentrations as that in the maternal blood stream, raising potential concerns about fetal harm. There is a laboratory reason for attention to the brain in particular. In animal models, metformin influences embryonic cortical development, and it is uncertain whether this translates to clinical implications for humans. Cortical means relating to the cortex, the outer layer of the brain.

Earlier studies in children were reassuring but young. The authors of the new study pooled them in 2024. A total of 7 studies met the inclusion criteria, including a combined cohort of 14,042 children with 7641 children who were exposed and followed for up to 14 years of age. Most of the data, though, concerned children under five, an age at which, as the new paper points out, developmental delays may not yet be apparent.

How the study measured child development

The researchers, with Hannah Gordon as first author and Roxanne Hastie as senior author, joined three sets of records. Birth records for the state supplied the pregnancies. Pharmacy records supplied the exposure, defined as a metformin prescription dispensed between the start of pregnancy and birth. The paper lists the exposure in five words: metformin prescription dispensation during pregnancy. Dispensed means the medicine was actually collected, which is a step closer to its being taken than a prescription alone.

The outcome came from an unusual source. Pregnancy and birth data were linked with the Australian Early Development Census (AEDC), a standardized national assessment of childhood development completed every 3 years during the first year of full-time school (age 4-6 years). Each child’s teacher fills in the assessment. It covers five areas: physical health and well-being, social competence, emotional maturity, language and thinking skills, and communication and general knowledge.

A child counted as developmentally vulnerable if they fell in the bottom tenth on two or more of the five. The paper’s wording is scoring below the 10th percentile in at least 2 of the 5 assessed domains of the AEDC. By design, then, a sizable minority of all children meet the definition.

The measure has some predictive weight. According to the authors, developmental vulnerability on the AEDC provides an important estimation of emotional and behavioral problems throughout primary school and literacy and numeracy delays in high school.

From about 872,000 single births in the state between 2009 and 2020, the team found 177,409 children who had started school in a year when the census was run and could be matched to it. Of those, 1,095 had been exposed to metformin. Children born with congenital anomalies were left out.

What it found about metformin and development

	Exposed to metformin	Not exposed
Children	1,095	About 176,300
Rated developmentally vulnerable	199 (18.2%)	24,379 (13.9%)
Difference after adjustment	No clear difference: 3% lower, range 26% lower to 29% higher

Taken at face value, the exposed children did worse. Developmental vulnerability was observed in 199 children (18.2%) exposed to metformin, against 13.9% of the rest. That raw gap, a rate about 31% higher, was well beyond chance.

It did not survive adjustment. When the researchers weighted the comparison so that exposed and unexposed children had mothers with similar characteristics, the rate among exposed children was 3% lower, with a plausible range from 26% lower to 29% higher. The paper states the result as follows: metformin was not associated with developmental vulnerability among children exposed to metformin. The same held for each of the five areas taken separately.

The authors draw a cautiously worded conclusion, that the result may provide reassurance to clinicians and women considering metformin use during pregnancy.

Why the unadjusted gap in children exposed to metformin disappeared

The answer lies in who takes metformin. Women prescribed it in pregnancy were not a cross-section of mothers.

They were heavier. Women using metformin had a higher mean (SD) BMI of 32.3 (9.1) compared with women not using metformin (26.0 [5.8]). BMI is body mass index, a measure of weight for height, and a value above 30 falls in the obesity range.

They were much more likely to have the conditions the drug is prescribed for. Among women taking metformin, 39.2% had gestational diabetes, against 8.3% of the others. For type 2 diabetes the figures were 17.9% and 0.5%, and for polycystic ovary syndrome 22.8% and 2.3%.

And their pregnancies ended differently. Their babies were born preterm in 12.5% of cases, against 5.9%.

Each of those things is linked to how children develop, quite apart from any medicine. Statisticians call such a tangle confounding: the apparent effect of one thing is really the effect of something that travels with it. The raw comparison set the children of women with diabetes and obesity against the children of mostly healthy women. The adjusted comparison tried to set like against like.

The researchers tested their answer by throwing out the healthy women altogether. They repeated the analysis in the 16,121 children whose mothers all had gestational or type 2 diabetes. Again the raw figures favored the unexposed, 20.7% against 15.9%. Again adjustment removed the difference, leaving an estimate 8% higher with a range from 14% lower to 35% higher.

The authors also argue that any flaw in their adjustment would cut in one direction. Because metformin users were less healthy in ways that are hard to capture fully, leftover confounding should make the drug look worse than it is. In their words, its influence would have been expected to bias estimates toward a seemingly harmful association. On that reasoning, finding no association after adjustment is more reassuring than it first appears.

How it fits with earlier studies of metformin in pregnancy

It agrees with them, and reaches an older age than most.

The group’s own 2024 review of seven studies concluded that in utero exposure to metformin does not seem to be associated with adverse neurodevelopmental outcomes in children up to the age of 14 years. In utero means in the womb.

Two register studies from other countries had looked at related outcomes. One, from New Zealand, started from the same concern, that because metformin crosses the placenta, patients and clinicians are concerned with its long-term effect on child health. It compared children of women treated with metformin or insulin at a preschool health check and found no link between metformin and behavioral difficulties.

The other was a register-based cohort study from Finland covering children born over 13 years, nearly 4,000 of them exposed to metformin alone. Its long-term result was that exposure to metformin or combination treatment versus insulin was not associated with increased risk of long-term outcomes, a list that included problems of motor and social development. It did report one short-term difference. An increased risk of small for gestational age (SGA) was observed for metformin, meaning more babies were smaller than expected for the length of the pregnancy. In the new Australian data, the share of small babies was similar in the two groups.

On the body, as distinct from the brain, a 2024 meta-analysis found little difference too. Children at the age of 9 years born to mothers who were treated with metformin during pregnancy had similar BMI to those whose mothers had insulin.

Those earlier studies mostly compared metformin with insulin among women who had diabetes. The new one compares exposed children with all other children, whatever the reason their mothers were given the drug, which matters because the drug is increasingly used outside diabetes.

What the study cannot show about metformin and child development

A moderate effect. With 1,095 exposed children, the estimate is loose. Applied to the 13.9% of unexposed children rated vulnerable, the plausible range runs from roughly 10% to roughly 18%. The study argues against a large harm. It cannot exclude a modest one, or a modest benefit.

Dose, duration and timing. About two-thirds of the women were first dispensed metformin in the first three months of pregnancy, and 467 women (42.6%) were dispensed more than 1 metformin prescription during pregnancy. The authors had planned to ask whether longer or later exposure mattered. Those analyses, they report, were too small to give reliable answers.

Whether the tablets were taken. A dispensed prescription is not proof of use. Women who filled a prescription just before conceiving and kept taking it would have been counted as unexposed.

Unmeasured differences. No adjustment is complete. Self-reported race and ethnicity data were not available, and the reason each woman was prescribed metformin was inferred from her diagnoses.

The children who could not be linked. Only about one birth in five could be matched to a school assessment, in large part because the census runs only every third year. The matched and unmatched children were broadly alike, though matched mothers had less gestational diabetes.

Later childhood. A teacher’s rating at age five is one checkpoint. It says nothing about adolescence.

The comparison with untreated diabetes. The study asks whether metformin is linked to harm. It does not weigh that against the known harms of high blood sugar in pregnancy, which is the choice women and their doctors face.

How diabetes or another condition is best treated in a particular pregnancy depends on the woman’s health and blood sugar, which is assessed by her doctor. Do not stop or change a prescribed medicine without talking to your doctor.

In 177,409 Australian children assessed by their teachers in the first year of school, the 1,095 whose mothers were dispensed metformin in pregnancy were no more likely to be rated developmentally vulnerable once their mothers’ weight and diabetes were taken into account, in an observational study whose estimate was too loose to exclude a moderate effect and which could not examine dose or timing.