Dr Emaediong Akpanekpo, Preeyaporn Srasuebkul, Azar Kariminia, Julian N. Trollor, John Kasinathan, Melanie Simpson, Anyiekere Ekanem, Tony Butler
Abstract:
Background: We validated a risk assessment tool for predicting violent reoffending among Australian justice-involved youth. The tool, initially developed in a youth custodial cohort using a Cox proportional hazards model, utilised socio-demographic, criminological, and clinical data.
Method: The validation cohort comprised 766 community-supervised youth (aged 10-17 years at index offence) in New South Wales, Australia. Predictor variables were obtained from data linkage of the Young People on Community Orders Health Survey with statewide data collections, including the Reoffending Database, Admitted Patients Data Collection, and Mental Health Ambulatory Data Collection. Violent reoffending was assessed at 1- and 2-year intervals. We evaluated the tool's predictive performance using discrimination and calibration metrics, and calculated sensitivity, specificity, positive predictive value, and negative predictive value at selected risk thresholds. Decision Curve Analysis was employed to assess the net benefit of the tool.
Results: The median age of validation cohort was 17 years (IQR: 16, 18). Cumulative incidence of violent reoffending at 1 and 2 years was 16% and 27%, respectively. The tool demonstrated good discrimination (AUC: 0.74 for 1-year and 0.75 for 2-year follow-up) and calibration, with null calibration-in-the-large values, an Expected: Observed event ratio of 1.00, and a calibration slope of 1.00. Higher net benefit was observed across a wide range of risk thresholds.
Conclusion: The risk assessment tool showed good predictive ability, outperforming previously validated tools among Australian justice-involved youth. Its robust performance in both custodial and community settings indicate versatility across different supervision contexts, supporting its potential for widespread implementation.