Project · Randomized controlled trial
AI Clinical Reasoning Tutor RCT in progress
Can a well-designed AI tutor measurably improve clinical reasoning in pre-clerkship medical students? We built one, calibrated it with faculty, and are testing it in a randomized controlled trial.
The tutor
The AI Clinical Reasoning Tutor is a single AI agent that plays three roles at once for each faculty-designed case:
- Simulated patient — the student takes a history in natural conversation;
- Electronic health record — the student requests vitals, labs, and imaging as they would in a chart;
- Faculty tutor — the agent coaches the student's differential, probes their reasoning, and withholds answers until the student has committed to their own.
Cases are authored and calibrated by faculty. Notably, our faculty-calibration work found that most faculty initially over-scored a sample student workup; after structured calibration, scorers converged — an important reminder that measuring reasoning requires calibrating the measurers, not just the students.
The trial
- Design
- Randomized controlled trial: AI tutor cases versus traditional written homework, delivered periodically through the pre-clerkship Foundations of Practice course (Class of 2028).
- Status
- IRB approved February 2026; the trial is currently in progress.
- Primary outcomes
- Clinical reasoning measured with the Revised-IDEA rubric (Schaye et al., 2022) and a validated patient-note scoring rubric (Park et al., 2017).
- Secondary outcomes
- OSCE performance, student experience surveys, and platform learning analytics.
- Funding
- UCLA Innovation Grants (Catalyst), Dr. Wang as Principal Investigator.
Platform
Year 1 development ran on UCLA Health's HIPAA-compliant internal AI platform. The current phase runs on an academic clinical-AI simulation platform developed at Stanford, adding multi-model support, voice and multimedia simulation, and student- and course-level analytics, in use at universities on four continents.
Why it matters
Most published AI-tutor work in medical education is descriptive or observational. A randomized design with validated reasoning outcomes is what the field needs to move from "students like it" to "it changes how they reason."