Association of Multidisciplinary Care Team Engagement With Hemoglobin A1c Among Medicaid Enrollees With Diabetes: A Doubly Robust Machine Learning Analysis

OBJECTIVE To estimate the association between multidisciplinary care team engagement and hemoglobin A1c (HbA1c) among Medicaid enrollees with diabetes and to characterize treatment effect heterogeneity. RESEARCH DESIGN AND METHODS We performed a retrospective cohort study using linked claims, electronic health record, and care management data from a Virginia Medicaid managed care organization (2022–2025). Adults with diabetes identified for outreach with a baseline and follow-up HbA1c were included. The primary outcome was the change in mean HbA1c from baseline (6 months before index) to follow-up (mean of values 90–365 days after index), estimated by doubly robust debiased machine learning (DML) (n = 369). Heterogeneity was assessed via generalized random forest (GRF) and a treatment-by-baseline HbA1c interaction. RESULTS The DML-estimated effect on mean HbA1c was −0.30 percentage points (95% CI −0.58 to −0.01; P = 0.039; Cohen d = 0.24); a single-value analysis was concordant (−0.29; 95% CI −0.60 to 0.01). The estimated benefit increased by 0.32 percentage points per unit higher baseline HbA1c (P = 0.014); baseline HbA1c was the dominant heterogeneity predictor (GRF importance = 0.585), and, among people with baseline HbA1c ≥9%, the estimated reduction approached 1 percentage point. Estimates were directionally consistent across inverse probability of censoring weighting (IPCW) (−0.21), time-zero alignment (−0.39), and target trial emulation (−0.32). The intervention was projected cost-saving (net savings, $1,562 per person; 99.7% probability). CONCLUSIONS Care team engagement was associated with a 0.30–percentage point reduction in mean HbA1c, with substantially larger benefits among people with higher baseline glycemia, supporting targeted deployment toward Medicaid enrollees with uncontrolled diabetes.