Interpretable machine learning model to predict aggressive behavior in first-episode schizophrenia

This study aimed to predict the risk of aggressive behavior in patients with first-episode schizophrenia using time-to-event machine learning models.

This study used a cohort study design of patients newly diagnosed with schizophrenia between December 2019 and September 2022 at Beijing An Ding Hospital. Demographic questionnaires and social-psychological scales were utilized in the baseline survey, and patients were followed for 24 months post-diagnosis to assess aggressive behavior. The data in this study were randomly separated into a training (70 %) set and a testing (30 %) set. Three representative survival models were developed to predict aggressive behavior in first-episode schizophrenia patients, using indicators such as the concordance index (C-index), Brier score, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), area under the precision-recall curve (AUPRC) and the area under the time-dependent receiver operating characteristic curve (time-dependent AUC) to assess performance. Shapley Additive Explanation (SHAP) was used to determine feature importance rankings.

Based on the inclusion and exclusion criteria, a total of 216 patients were enrolled in the study, with aggressive behavior observed in 36 cases (16.67 %) during the follow-up period. Among the three models, the Random Survival Forest model demonstrated the highest predictive performance, with a C-index of 0.79, an average time-dependent AUC of 0.91, and an average AUPRC of 0.59. The differences in predictive performance among the three models at each time point were statistically significant (p < 0.05). The SHAP method identified high impulsivity, higher average monthly personal income, larger household size, being unemployed, and lower EPQ-L scores as the top five contributors influencing model decision-making.

The Random Survival Forest model effectively predicted aggressive behavior in first-episode schizophrenia patients, with impulsivity and average monthly income as the most significant factors.