Gestational diabetes mellitus (GDM) is commonly treated using a broadly similar clinical framework, although growing evidence suggests that affected individuals do not represent a single uniform group. A population-based cohort study published in Diabetes Care used machine learning-based analysis of clinical data to define GDM subtypes and evaluate how these patterns related to pregnancy outcomes and later diabetes development.

The study followed 37,544 individuals with GDM for as long as 12 years after delivery. The cohort was separated into discovery and validation sets in a 70:30 ratio. Using routinely collected sociodemographic, behavioral, and clinical measures, the analysis applied dimension reduction and clustering approaches to derive phenotype groups. Associations with outcomes were examined with covariate-adjusted modified Poisson and Cox regression models.

Four phenotype clusters emerged from the analysis. In the discovery set, 65.6% of individuals were classified as C1, 14.5% as C2, 12.0% as C3, and 7.8% as C4, with comparable proportions in the validation set. Relative to C1, which was characterized by later diagnosis, lower body mass index, and postload hyperglycemia, clusters C2 through C4 showed a greater risk of adverse perinatal events and incident postpartum diabetes. The highest-risk profile was C4, defined by earlier diagnosis, greater comorbidity burden, and higher glucose challenge test values.

Within C4, the adjusted relative risk was 1.43 (95% CI 1.19-1.72) for severe maternal morbidity and 1.53 (95% CI 1.41-1.66) for neonatal intensive care unit admission. The adjusted hazard ratio for postpartum diabetes was 4.32 (95% CI 3.94-4.73). Additional analysis within C1 identified three subclusters that differed in perinatal complication risk, although postpartum diabetes risk remained similar across them. These results support a more individualized approach to risk assessment in GDM.