RIPTOSO: The development of a screening tool for adverse events during forensic-psychiatric inpatient treatments of offenders with schizophrenia spectrum disorders

Adverse events such as compulsory measures, absconding, illicit substance use, self-harm, aggressive behavior, and prolonged hospitalization pose significant challenges in forensic psychiatric inpatient care. This study introduces a machine learning-based tool to predict these events in patients with schizophrenia spectrum disorders (SSD) upon admission. Data from 370 court-mandated forensic inpatients treated at an academic center in Zurich, Switzerland, were retrospectively analyzed. Twenty-seven variables, available upon admission in clinical settings, were tested using six machine learning algorithms (support vector machines (SVM), logistic regression, naive Bayes, gradient boosting, fine trees, and neural networks). Predictive performance was assessed using metrics such as area under the curve (AUC) and balanced accuracy. SVM demonstrated the highest performance, achieving an AUC of 0.79 and a balanced accuracy of 69.8 %. These results suggest that the tool can identify patients at higher risk for problematic treatment courses, enabling earlier interventions and more efficient resource allocation. The simplicity of the model, based on routinely collected data, enhances its clinical applicability. However, validation studies in multi-center and international settings are essential to confirm its robustness and generalizability. This tool represents a promising step toward integrating machine learning into forensic psychiatry to improve treatment outcomes and patient safety.