Aims:
Gestational diabetes mellitus (GDM) is recognized as a heterogeneous disorder with varying degrees of impairment in insulin secretion and sensitivity. We aimed to identify heterogeneous metabolic clusters using routine clinical measures and to evaluate their associations with obstetric and neonatal outcomes.
Methods:
We analysed 6,334 participants from five centres of the Hyperglycaemia and Adverse Pregnancy Outcome (HAPO) Study, including 1,084 (17.1%) women with GDM according to IADPSG criteria. Homeostatic model assessment 2 estimates of β-cell function (HOMA2-β) and insulin resistance (HOMA2-IR) were derived from blinded 75-g 2-h OGTT. K-means clustering was applied based on maternal age, ethnicity-adjusted BMI z-score, HOMA2-β, and HOMA2-IR, with the optimal number of clusters determined by gap statistics. Relative risks (RRs) for adverse pregnancy outcomes were estimated using robust Poisson regression, and predictive performance was assessed using the area under ROC curves.
Results:
Five reproducible clusters were identified, with 1,675 (26.4%), 1,587 (25.1%), 1,261 (19.9%), 1,177 (18.6%), and 634 (10.0%) women in Clusters 1-5. Clusters 1-3 mainly comprised women with normal BMI, with Cluster 1 representing the oldest group (mean [SD]: 34.0 [3.2] years) and Cluster 2 the youngest (24.7 [3.0] years). Cluster 1 showed the most favourable metabolic profile with lower fasting glucose (median [IQR]: 4.4 [4.2-4.7] mmol/L), whereas Cluster 5 had the highest glucose (4.8 [4.5-5.1] mmol/L) and BMI (36.1 [32.2-40.4] kg/m2). Compared with Cluster 1, women in Clusters 2-5 showed increasingly higher risks of adverse pregnancy outcomes, with the highest risk observed in Cluster 5. The risk of preeclampsia increased by 55% (RR 1.55, 95% CI 1.12-2.14), 81% (1.81, 1.30-2.52), and 262% (3.62, 2.61-5.01) in Clusters 3-5, respectively. The clustering approach demonstrated better predictive performance compared with conventional GDM classification (AUROC for preeclampsia: 0.668 vs 0.552; p<0.001).
Conclusion:
Data-driven clustering based on routine clinical variables revealed metabolic heterogeneity in GDM and improved risk stratification beyond conventional diagnosis, supporting more individualized pregnancy care.
Acknowledgement:
This study was partially supported by a grant from the RGC Area of Excellence Scheme (M-401/24-R). CZ acknowledge support from the Hong Kong PhD Fellowship Scheme.