AI-MED

Access restricted to authorized researchers

AI-MED

Accelerating Improvement in Medical Education and Delivery

AI-MED is a research study that deploys conversational simulated standardized patients (SSPs) via large language models. It is developed and maintained by researchers at Harvard University for use in academic research partnerships.

David Duong
David Duong, MD, MPH Harvard Medical School
Program in Global Primary Health Care
Benjamin Daniels
Benjamin Daniels, MSc Harvard T.H. Chan School of Public Health
Department of Global Health and Population

Background

Standardized patients — trained actors presenting scripted clinical scenarios — are a key tool for measuring the technical quality of health care delivery (Daniels et al., 2023; Das et al., 2022). AI-MED extends this methodology into the digital domain, enabling at-scale deployment of interactive, pre-scripted patient simulations without the logistical constraints of in-person enumerators (Zhang et al., 2026).

Methodology

A provider opens a session and is presented with a simulated patient's initial complaint -- for example, "Doctor, I have been having a headache off and on for a while now." Providers then conduct natural-language clinical interviews with these LLM-driven patients generated from structured case vignettes, then submit diagnoses and other reactions through standardized survey instruments. Conversation transcripts and form responses are evaluated against clinical rubrics.