DOI: 10.1111/dom.71246 ISSN: 1462-8902

Linking Metabolic Subgroups and Adverse Pregnancy Outcomes in GDM : A Complementary Analytic Approach of Logistic Regressions and Machine Learning

Yanjing Zeng, Jing Zhou, Zhengbin Ou, Yun Li, Junping Fan, Nan Wang, Chunxia Cheng, Hui Yang, Man Ping Wang, Jie Dong, Jia Guo

ABSTRACT

Aims

Associations between metabolic heterogeneity in women with gestational diabetes mellitus (GDM) and adverse pregnancy outcomes have often been explored without considering interrelated metabolic biomarkers, limiting accurate predictions of outcomes. This study aimed to identify metabolic heterogeneity across distinct classes of biomarkers in GDM women and link this metabolic heterogeneity with adverse pregnancy outcomes through a complementary analytic approach.

Methods

We conducted a retrospective cohort study using medical records of GDM women from 2018 to 2023 in Central South China. We identified metabolic heterogeneity of GDM using unsupervised k ‐means clustering based on oral glucose tolerance test (OGTT) 0, 1 and 2 h, uric acid (UA), triglycerides (TG), high‐density lipoprotein (HDL) and low‐density lipoprotein (LDL). We then linked this metabolic heterogeneity to adverse pregnancy outcomes through a complementary analytic approach, including logistic regression for group traits and machine learning for single features.

Results

The first approach categorised 2246 GDM women into four distinct metabolic subgroups: reference (36%, relatively normal and average biomarker levels), fasting hyperglycemia (23.3%), hyperuricemia and hypertriglyceridemia (34.9%) and combined metabolic dysregulation (5.8%). The complementary analytic approach yielded consistent results. The hyperuricemia and hypertriglyceridemia subgroup showed higher odds of preterm delivery (OR = 1.86, 95% CI = 1.15–3.01). The combined metabolic dysregulation subgroup had the highest risks of preterm delivery (OR = 3.45, 95% CI = 1.79–6.65), insulin treatment (OR = 14.67, 95% CI = 8.72–24.66) and hypertensive disorders of pregnancy (OR = 2.72, 95% CI = 1.48–5.01). SHAP analysis showed that 57%–71% of the subgroup‐defining biomarkers ranked among the top 15 predictive features.

Conclusions

This study confirms the metabolic heterogeneity in GDM women, supporting synergistic effects among metabolic factors and the need for integrated metabolic management. The identified metabolic subgroups may facilitate risk stratification in real‐world settings by integrating biomarkers from multiple metabolic pathways. Women in the hyperuricemia and hypertriglyceridemia subgroup, as well as those in the combined metabolic dysregulation subgroup, urgently require interventions to improve pregnancy outcomes.

More from our Archive