Perspective-taking training with AI-generated virtual patients for primary care clinicians to enhance care for black patients with chronic pain: a pilot feasibility and acceptability study

Chronic non-cancer pain disproportionately impacts Black adults in the USA. Primary care clinicians (PCCs) receive limited training in empathy and perspective-taking, likely contributing to treatment disparities. We assessed the feasibility and acceptability of a novel virtual training, featuring AI-generated patient videos, designed to improve PCCs' perspective-taking skills and reduce treatment disparities. We recruited PCCs from two health systems in the Midwestern US. The training included a didactic module on pain disparities and perspective-taking, delivered by an AI-generated Black "peer" clinician, along with videos of photorealistic digital Black patients sharing their experiences living with chronic pain. PCCs practiced perspective-taking via a typed-response interface. Feasibility was assessed using time-on-task and completion rates; acceptability via system usability scores and written feedback. Using electronic health records data, we conducted an exploratory patient-level analysis comparing pain medication prescribing (three months before versus after the intervention date) for Black versus non-Black patients of intervention PCCs against a control group of 19 high-prescribing PCCs from the same health systems who did not complete the training. Logistic regression models estimated overall pain prescribing changes and initiation of new medications across nine drug classes. Fifteen PCCs (5 from health system #1; 10 from health system #2) completed the training between December 2024 and February 2025. The training proved feasible and acceptable: average completion time was approximately 6 min, and 73.3% would recommend it. Exploratory analyses included 1,991 patients (628 intervention; 1,363 control). In exploratory models, intervention patients had lower odds of a change in pain drug class than controls (odds ratio (OR) = 0.59; 95% confidence interval (CI) 0.45–0.79). An intervention-by-race interaction was observed for new opioid prescriptions (OR = 0.17; 95% CI 0.02–0.78), indicating that the association between the intervention and new opioid initiation differed for Black versus non-Black patients. Clinician-facing virtual perspective-taking training with AI-generated virtual patients is feasible and acceptable within primary care workflows. Exploratory prescribing estimates are hypothesis-generating and support progression to a fully powered cluster-randomized trial of the training's impact on prescribing equity.