- A LightGBM model accurately predicted hospitalisation-related severe relapse in schizophrenia (AUC 0.95, relapse recall 0.77, F1 0.74) on independent test set.
- SHAP analysis identified medication persistence and specialist recommendation for medication adherence training as the top predictors of relapse risk.
- Sustained medication, improved occupational and social functioning, and targeted community interventions associated with lower predicted severe relapse risk.
JMIR Med Inform. 2026 Oct 28;14:e86991. doi: 10.2196/86991.
ABSTRACT
BACKGROUND: Schizophrenia is a severe mental disorder characterized by high disability and relapse rates, posing significant challenges to public safety and community-based mental health management.
OBJECTIVE: This study aimed to develop and validate a model for identifying hospitalization-related severe relapse risk among individuals with schizophrenia and to determine the key factors associated with this risk.
METHODS: A total of 44,804 electronic records of community-managed individuals with schizophrenia were extracted from an information management system between January 1, 2022, and December 31, 2023. The dataset included demographic information, clinical records, interview data, policy support, and management indicators to construct a comprehensive feature system. A Light Gradient Boosting Machine (LightGBM) machine learning algorithm was used for model development, with cost-sensitive learning applied to address class imbalance. Model interpretability was enhanced using Shapley additive explanations (SHAP) to identify the most influential predictors of relapse risk.
RESULTS: After screening, a total of 39,818 community-dwelling individuals with schizophrenia were included in this study, of whom 3717 (9.33%) experienced hospitalization-related severe relapse. The dataset was randomly divided using stratified sampling into a training set (n=31,854, 80%) and an independent test set (n=7964, 20%). Among the 6 machine learning models evaluated, LightGBM showed leading and balanced performance, yielding an area under the receiver operating characteristic curve of 0.95, an area under the precision-recall curve of 0.79, a relapse-class recall of 0.77, a precision of 0.71, and an F1-score of 0.74 in the independent test set. SHAP-based interpretability analysis identified medication persistence (23.30%), specialist recommendation: medication adherence training (19.74%), number of follow-ups (7.38%), age (3.72%), and educational attainment (3.58%) as the most influential predictors. Higher hospitalization-related severe relapse risk was generally associated with interrupted or refused medication use, more frequent follow-ups, younger age, and higher educational attainment, whereas sustained medication use, better occupational functioning and productive work, greater social participation, and better social functioning were associated with lower predicted risk.
CONCLUSIONS: This study developed a hospitalization-related severe relapse risk identification model that integrates multidimensional factors reflecting the clinical and functional complexity of schizophrenia. The findings provide an evidence-based foundation for risk identification, targeted intervention, and the optimization of community mental health services to reduce hospitalization-related severe relapse risk and support long-term recovery.
PMID:42950201 | DOI:10.2196/86991
Share Evidence Blueprint
Save to Google Notes
Search Google Scholar
Save as PDF
⭐ My Revision List
