An AI-Based Behavioral Health Safety Filter and Dataset for Identifying Mental Health Crises in Text-Based Conversations

TL;DR

This study introduces a new AI-based safety filter for mental health crisis detection in text conversations, demonstrating superior performance over existing content moderation tools across multiple datasets.

Contribution

The paper presents the Verily Behavioral Health Safety Filter (VBHSF), a novel AI tool with high sensitivity and specificity for identifying mental health crises, validated on clinician-labelled datasets.

Findings

VBHSF achieved high sensitivity (0.990) and specificity (0.992) on the Verily dataset.

It outperformed existing guardrails like NVIDIA NeMo and OpenAI Omni Moderation in sensitivity.

VBHSF demonstrated robust, generalizable performance across different datasets.

Abstract

Large language models often mishandle psychiatric emergencies, offering harmful or inappropriate advice and enabling destructive behaviors. This study evaluated the Verily behavioral health safety filter (VBHSF) on two datasets: the Verily Mental Health Crisis Dataset containing 1,800 simulated messages and the NVIDIA Aegis AI Content Safety Dataset subsetted to 794 mental health-related messages. The two datasets were clinician-labelled and we evaluated performance using the clinician labels. Additionally, we carried out comparative performance analyses against two open source, content moderation guardrails: OpenAI Omni Moderation Latest and NVIDIA NeMo Guardrails. The VBHSF demonstrated, well-balanced performance on the Verily Mental Health Crisis Dataset v1.0, achieving high sensitivity (0.990) and specificity (0.992) in detecting any mental health crises. It achieved an F1-score of 0.939, sensitivity ranged from 0.917-0.992, and specificity was >= 0.978 in identifying specific crisis categories. When evaluated against the NVIDIA Aegis AI Content Safety Dataset 2.0, VBHSF performance remained highly sensitive (0.982) and accuracy (0.921) with reduced specificity (0.859). When compared with the NVIDIA NeMo and OpenAI Omni Moderation Latest guardrails, the VBHSF demonstrated superior performance metrics across both datasets, achieving significantly higher sensitivity in all cases (all p < 0.001) and higher specificity relative to NVIDIA NeMo (p < 0.001), but not to OpenAI Omni Moderation Latest (p = 0.094). NVIDIA NeMo and OpenAI Omni Moderation Latest exhibited inconsistent performance across specific crisis types, with sensitivity for some categories falling below 0.10. Overall, the VBHSF demonstrated robust, generalizable performance that prioritizes sensitivity to minimize missed crises, a crucial feature for healthcare applications.