OBJECTIVE “Food Is Medicine” approaches are popular disease management strategies for patients experiencing food insecurity (FI) and diet-sensitive conditions; however, few studies consider heterogeneous treatment effects (HTE) across patient subpopulations. We estimate HTE of a prescription for produce (PRx) tested within a randomized controlled trial of patients with diabetes at risk of FI. RESEARCH DESIGN AND METHODS We use machine learning to estimate HTE of a PRx conducted within an integrated health system, academic medical center, and primary care network in the southeastern U. S. This trial included 2, 155 patients enrolled in the health system’s diabetes registry with at least one hemoglobin A1c (HbA1c) measurement in the prior year and at-risk status for FI. Participants received 80/month for 12 months on a debit card covering fruit and vegetable purchases. Outcomes included HbA1c change (post- minus preintervention) and any postintervention emergency department (ED) utilization. RESULTS For HbA1c, the largest negative standardized difference in HbA1c change between intervention arms was observed among participants with HDL cholesterol 50. 67 mg/dL and triglycerides 82. 5 mg/dL (estimate SE −0. 25% 0. 12, −2. 7 mmol/mol 1. 3; P = 0. 019). For ED visits, the largest reduction in probability of any postintervention visits between arms was observed among individuals with low outpatient (OP) utilization, nonprivate and non-Medicare insurance, LDL cholesterol 78. 5 mg/dL, and age 50. 5 years (estimate SE −0. 2 0. 05; P 0. 001). CONCLUSIONS Our findings suggest that social factors (e. g. , age and OP utilization) and clinical factors (e. g. , cholesterol) may impact PRx effectiveness for diabetes management.
Jordan et al. (Fri,) studied this question.