Machine Learning-Based Classifier for Identifying Inpatients With Schizophrenia at High Risk of Suicide

The suicide rate of individuals with schizophrenia is higher than the general population. In clinical practice, it is essential to identify patients with schizophrenia who are at an elevated risk of suicide. However, previous studies may not fully account for potential factors that could influence the suicide risk among schizophrenia patients. Our study leverages machine learning to identify predictive variables from a broad range of indicators.

Cross-sectional.

A total of 131 patients with schizophrenia were recruited at the Mental Health Center of West China Hospital from August 2021 to July 2022. We collected complete blood analysis, thyroid function, inflammatory factors, childhood trauma experiences, psychological impact related to the Coronavirus Disease 2019 epidemic, sleep quality, psychological distress, income level and other demographic data. We utilised machine learning algorithms to predict the suicide risk of patients with the above features. The Shapley values were used to illustrate important predictive variables of suicide risk.

We gathered important variables for predicting suicide risk of patients with schizophrenia, such as the Nurses' Observation Scale for Inpatient Evaluation factor, neutrophil count, psychological impact during Coronavirus Disease 2019 epidemic, prolactin level and plasma thromboplastin component level.

The features identified in this study are anticipated to aid in the clinical identification of suicide risk in individuals with schizophrenia in the future. This study also promoted improvements in the suicide prediction model among patients with schizophrenia.

This study identified key predictive variables for suicide risk in schizophrenia patients using machine learning. Our findings will enhance clinical tools for assessing suicide risk in schizophrenia, potentially leading to more effective prevention strategies. This advancement holds promise for improving suicide prevention efforts and tailoring interventions to individuals' specific risk profiles.

STROBE Statement (for cross-sectional studies).

None.