Generative AI-assisted clinical interviewing of mental health

Area of Science:

• Digital Health
• Artificial Intelligence in Medicine
• Psychiatry

Background:

• Traditional mental health assessments rely on clinician interviews, which are costly, time-consuming, and lack standardization.
• Large language models (LLMs) present an opportunity to develop AI systems that can simulate these interviews, addressing current limitations.
• Rigorous validation of AI tools for clinical interviews is scarce, highlighting a need for empirical evidence.

Purpose of the Study:

• To develop and evaluate an AI assistant, TalkToAlba, for conducting clinical interviews aligned with DSM-5 criteria.
• To assess the accuracy, reliability, and user experience of AI-powered diagnostic interviews compared to established methods.

Main Methods:

• An AI assistant conducted diagnostic interviews with 303 participants reporting various mental health disorders (MDD, GAD, OCD, PTSD, ADD/ADHD, ASD, ED, SUD, BD) and healthy controls.
• A secondary AI system analyzed transcripts to verify diagnostic criteria and justify conclusions.
• AI interview performance was compared against established rating scales using metrics like Cohen's Kappa, sensitivity, and specificity.

Main Results:

• The AI-powered interview demonstrated superior agreement, sensitivity, and specificity in identifying self-reported disorders compared to traditional rating scales.
• The AI system showed reduced co-dependencies between diagnostic categories.
• Participants reported high levels of empathy, relevance, understanding, and support during AI-powered interviews.

Conclusions:

• AI-powered clinical interviews are accurate, standardized, and person-centered tools for assessing common mental disorders.
• These AI tools offer scalability, low cost, and a positive user experience, positioning them as valuable complements to traditional diagnostics.
• The findings support the potential for widespread adoption of AI-driven interviews in mental healthcare delivery.