A Layered Multi-Expert Framework for Long-Context Mental Health Assessments

TL;DR

This paper presents SMMR, a layered multi-expert framework that improves long-context mental health assessments by combining multiple models to reduce hallucinations and increase accuracy.

Contribution

Introduces SMMR, a novel layered framework leveraging multiple models and specialized experts for more reliable long-context mental health evaluations.

Findings

SMMR outperforms single-model baselines in accuracy and F1-score.

SMMR reduces hallucinations and captures clinical nuances.

Demonstrates effectiveness on DAIC-WOZ and 48 case studies.

Abstract

Long-form mental health assessments pose unique challenges for large language models (LLMs), which often exhibit hallucinations or inconsistent reasoning when handling extended, domain-specific contexts. We introduce Stacked Multi-Model Reasoning (SMMR), a layered framework that leverages multiple LLMs and specialized smaller models as coequal 'experts'. Early layers isolate short, discrete subtasks, while later layers integrate and refine these partial outputs through more advanced long-context models. We evaluate SMMR on the DAIC-WOZ depression-screening dataset and 48 curated case studies with psychiatric diagnoses, demonstrating consistent improvements over single-model baselines in terms of accuracy, F1-score, and PHQ-8 error reduction. By harnessing diverse 'second opinions', SMMR mitigates hallucinations, captures subtle clinical nuances, and enhances reliability in high-stakes mental health assessments. Our findings underscore the value of multi-expert frameworks for more trustworthy AI-driven screening.