Maternal Gut Microbiome as a Predictor of Insulin Therapy Requirement in Gestational Diabetes

Gestational diabetes mellitus (GDM) often requires pharmacological intervention beyond lifestyle modification to achieve optimal glycemic control. This study aimed to develop machine learning models that integrate clinical and gut microbiome data to predict the need for insulin therapy (IT) in women with GDM.

We characterized 205 pregnant women with GDM from the Genetic and Epigenetic Mechanisms of Developing Gestational Diabetes Mellitus and its Effects on the Fetus study, collecting clinical parameters, lifestyle questionnaires, self-monitored blood glucose records, and gut microbiome profiles based on 16S rRNA gene sequencing. Gradient-boosting models were trained to predict IT, basal insulin (BI), and prandial insulin (PI) requirements. Model discrimination was assessed using repeated stratified five-fold cross-validated area under the curve-receiver operating characteristic (AUC-ROC) (nested cross-validation). Feature importance and interpretability were evaluated with SHapley Additive exPlanations and permutation analyses. Differential microbial abundance was analyzed by ANCOM-BC2 (analysis of composition of microbiomes with bias correction, version 2), and metabolic pathways were predicted via PICRUSt2.

Women requiring insulin were older and had higher pre-pregnancy body mass index (BMI), fasting plasma glucose, 1-hour oral glucose tolerance test glucose, and glycated hemoglobin than diet-treated women (P < .05 for all). Adding microbiome data improved AUC-ROC for IT prediction from 0.63 (95% CI = 0.43, 0.83) to 0.70 (0.50, 0.89), BI from 0.77 (0.59, 0.95) to 0.82 (0.65, 0.99), and for PI from 0.69 (0.50, 0.88) to 0.70 (0.51, 0.89). Key influential features included glycemic markers, BMI, and microbial taxa, such as Phascolarctobacterium faecium, Alistipes ihumii, Cloacibacillus evryensis, Ruthenibacterium lactatiformans, and Methanosphaera stadtmanae, and the predicted microbial metabolic pathway PWY-5823.

Our findings demonstrate that integrating gut microbiome characteristics with clinical data improves the prediction of insulin treatment needs in GDM, particularly for BI initiation.