Brian Gin | Searcharxiv

arXiv2026

"Crash Test Dummies" for AI-Enabled Clinical Assessment: Validating Virtual Patient Scenarios with Virtual Learners

Background: In medical and health professions education (HPE), AI is increasingly used to assess clinical competencies, including via virtual standardized patients. However, most evaluations rely on AI-human interrater reliability and lack a measurement framework for how cases, learners, and raters jointly shape scores. This leaves robustness uncertain and can expose learners to misguidance from unvalidated systems. We address this by using AI "simulated learners" to stress-test and psychometrically characterize assessment pipelines before human use. Objective: Develop an open-source AI virtual patient platform and measurement model for robust competency evaluation across cases and rating conditions. Methods: We built a platform with virtual patients, virtual learners with tunable ACGME-aligned competency profiles, and multiple independent AI raters scoring encounters with structured Key-Features items. Transcripts were analyzed with a Bayesian HRM-SDT model that treats ratings as decisions under uncertainty and separates learner ability, case performance, and rater behavior; parameters were estimated with MCMC. Results: The model recovered simulated learners' competencies, with significant correlations to the generating competencies across all ACGME domains despite a non-deterministic pipeline. It estimated case difficulty by competency and showed stable rater detection (sensitivity) and criteria (severity/leniency thresholds) across AI raters using identical models/prompts but different seeds. We also propose a staged "safety blueprint" for deploying AI tools with learners, tied to entrustment-based validation milestones. Conclusions: Combining a purpose-built virtual patient platform with a principled psychometric model enables robust, interpretable, generalizable competency estimates and supports validation of AI-assisted assessment prior to use with human learners.

cs.HC↗

arXiv2019

A Dyadic IRT Model

We propose a dyadic Item Response Theory (dIRT) model for measuring interactions of pairs of individuals when the responses to items represent the actions (or behaviors, perceptions, etc.) of each individual (actor) made within the context of a dyad formed with another individual (partner). Examples of its use include the assessment of collaborative problem solving, or the evaluation of intra-team dynamics. The dIRT model generalizes both Item Response Theory (IRT) models for measurement and the Social Relations Model (SRM) for dyadic data. The responses of an actor when paired with a partner are modeled as a function of not only the actor's inclination to act and the partner's tendency to elicit that action, but also the unique relationship of the pair, represented by two directional, possibly correlated, interaction latent variables. Generalizations are discussed, such as accommodating triads or larger groups. Estimation is performed using Markov-chain Monte Carlo implemented in Stan, making it straightforward to extend the dIRT model in various ways. Specifically, we show how the basic dIRT model can be extended to accommodate latent regressions, multilevel settings with cluster-level random effects, as well as joint modeling of dyadic data and a distal outcome. A simulation study demonstrates that estimation performs well. We apply our proposed approach to speed-dating data and find new evidence of pairwise interactions between participants, describing a mutual attraction that is inadequately characterized by individual properties alone.

stat.AP↗