Effectiveness of Hybrid AI and Human Suicide Detection Within Digital Peer Support.

Authors

Siddharth Shukla, Prachet Balaji, Ilayda Ozsan McMillan, Marvyn R Arévalo Avalos, Harpreet Nagra, Zara Dana

Published in

Journal of clinical medicine. Volume 15. Issue 5. Mar 03, 2026. Epub Mar 03, 2026.

Abstract

Background: Suicidality continues to rise, while mental health services face obstacles of access, availability, and affordability. Digital peer support (DPS) may help bridge these gaps and facilitate early identification of suicidal ideation (SI). Objective: This study examined (1) the effectiveness of a hybrid solution combining a proprietary AI-based SI detection with real-time human moderation within DPS, (2) distribution of SI, (3) active SI referral, (4) linguistic differences in SI, (5) sentiment changes among users, and (6) the effects of peer SI disclosure. Methods: We retrospectively analyzed 169,181 live-chat transcripts encompassing 449,946 user visits (January-December 2024) from a DPS provider, Supportiv. Passive and active SI were identified using a hybrid AI and human moderator solution with post hoc LLM verification. Sentiment analysis and ANCOVA compared changes in sentiment across three propensity-matched user groups: passive SI users, non-SI users exposed to peer SI, and non-SI users not exposed to SI. Results: SI occurred in 3.19% of live chats. The AI model identified SI faster than humans (in 77.52% passive and 81.26% active cases), with 90.3% agreement. Moderators followed up 71.3 s after AI alerts and referred 5472 active SI users (1.21%) to crisis care. All users significantly benefited from DPS, with reductions up to 29.3% in depression, 26.8% in loneliness, 25.3% in despair, and 22.3% in helplessness, with optimism increasing up to 40.4%. Conclusions: AI-integrated, human-moderated DPS offers scalable and effective support for high-risk populations. The proprietary SI detection AI model accurately detects suicidality, allowing for human-moderated DPS to improve the mental well-being of users with and without SI, and maintains peer safety.

PMID:
41827346
Bibliographic data and abstract were imported from PubMed on 15 Sep 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.