The Empirically Grounded Adaptive Virtual Patient for Psychotherapy Training: Disclosure That Responds to Therapist Micro-Skills

2026-06-08

AI Summary

This study addresses the limitations of existing virtual patients in psychotherapy training, which either rely on rigid scripts or lack behavioral controllability when directly employing large language models, thereby hindering effective practice of therapists’ microskills such as empathy and exploration. To overcome this, the authors propose an Adaptive Virtual Patient (AVP) framework that, for the first time, integrates a structural equation model—fitted from nearly 2,000 hours of real therapeutic dialogues—to dynamically modulate the patient’s level of self-disclosure in response to the therapist’s empathic and exploratory behaviors. The system further employs parameterized control over a large language model to generate clinically plausible and contextually adaptive responses. In evaluations across 80 sessions (1,033 turns), AVP demonstrated a significant increase in disclosure levels correlated with improvements in therapists’ microskills, outperforming prompt-engineering baselines; ablation studies identified exploratory behavior as the key adaptive signal.

Abstract

Simulated patients offer a scalable way to train psychotherapy micro-skills such as empathic responding and exploratory probing, but current systems either follow fixed scripts or rely on LLMs that drift unpredictably over long sessions. We present the Adaptive Virtual Patient (AVP), which adapts its disclosure behavior -- from guarded, through moderate openness, to full disclosure -- in response to trainee skill. The AVP is grounded in a structural equation model fit to nearly 2{,}000 hours of real-world psychotherapy transcripts, which quantifies how therapist empathy and exploration shift a patient's openness over time. An LLM generates the AVP's utterances conditioned on a disclosure level that the dynamics module updates each turn. In an evaluation with 20 clinicians and trainees over 80 sessions (1{,}033 turns), the AVP's disclosure rises in response to therapist empathy and exploration, while a prompt-only baseline stays flat; ablations confirm that the empirically motivated parameterization outperforms alternatives, with exploration carrying most of the adaptive signal.

Problem

• virtual patient
• psychotherapy training
• micro-skills
• disclosure adaptation
• large language models

Innovation

• Adaptive Virtual Patient
• psychotherapy training
• micro-skills
• structural equation model
• large language model