This retrospective cohort followed 326,115 mothers with singleton pregnancies in a large Californian health system from their first delivery after January 2007 until diabetes was diagnosed outside pregnancy, membership ended, or December 2022. Across 442,225 pregnancies, 45,935 were complicated by gestational diabetes, of which 4,842 were recurrences. Median follow-up was 6.2 years.
A single episode of gestational diabetes was associated with more than five times the risk of later diabetes (hazard ratio 5.87, 95% CI 5.67-6.08). A recurrence took that to nearly ten times (HR 9.91, 95% CI 9.14-10.75). The ethnic pattern is the part that should change practice: Asian women had both the highest rate of gestational diabetes (20.18%) and the highest recurrence rate (53.28%), and the highest risk after recurrence (HR 11.74, 95% CI 9.64-14.30). More than half of Asian women who had gestational diabetes once had it again.
The care pathway this argues for is unglamorous and often broken: previous gestational diabetes → postpartum glucose check at 6-12 weeks → annual or biennial screening → weight and lifestyle support between pregnancies → early detection of dysglycaemia. In Indian practice these numbers are, if anything, conservative, given a population with higher gestational diabetes prevalence still. The obstetric record usually captures the current pregnancy well and previous ones badly, which is exactly the wrong way round for risk stratification.
- Ask about every previous pregnancy affected by gestational diabetes, not just the most recent one, and write the count in the notes
- Treat two or more affected pregnancies as a prediabetes-equivalent risk state and screen at that intensity
- Arrange the postpartum glucose check before discharge rather than leaving it to a later appointment that is often missed
- Use the interval between pregnancies for weight and lifestyle work; it is the only window in which recurrence is still preventable
- Flag Asian ethnicity specifically when planning screening frequency, given a recurrence rate above 50% in this cohort
The statistics, in plain English
The confidence intervals here are unusually tight, from 5.67 to 6.08 for a single episode, because the cohort is very large; that precision tells you the association is well measured, not that it is causal. Gestational diabetes status was treated as time-varying, meaning a woman contributed person-time as unexposed until her first affected pregnancy, which is the right way to avoid crediting the exposure with risk accrued before it happened. The main caution is detection: women with a history of gestational diabetes are screened more often, so some of the excess reflects earlier diagnosis rather than earlier disease. That cannot explain a hazard ratio near ten, but it probably inflates it somewhat.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for diabetes & endocrinology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free