A hierarchical machine learning model for predicting self-harm and suicidal behavior in hospitalized patients with schizophrenia using clinical history and nursing observations

1

The study developed a two-layered machine learning framework to identify high-risk schizophrenia inpatients for self-harm and suicidal actions.
2

A total of 477 schizophrenia patients were reviewed, with 159 classified as high-risk based on documented self-harm or suicidal episodes.
3

Key predictors of self-harm included previous self-harm, hopelessness/depression, younger age, and higher educational level.
4

The optimized static and dynamic models achieved AUCs of 0.7564 and 0.8531, respectively, with their fusion improving performance to AUC 0.9048.
5

This hierarchical model enhances the detection of high-risk patients, supporting timely preventive strategies in psychiatric settings.

Frontiers In Psychiatry

by Liang, Chen , Meng, Xianfeng , Duan, Ying , Yang, Wei , Zhu, Gang , Wang, Jinhuan , Sun, Ying , Wang, Mingtao , Liu, Miao , Sun, Chenhui , Hu, Kunyuan , Shao, Wei , Ren, Jintao , Shao, Xiaojun , Zhang, Yang
April 24, 2026

1