Source context
Objective This study aimed to develop and validate a dynamic risk assessment tool for individuals serving community sentences that accounts for the effects of acute adverse events and desistance from crime. Hypotheses Dynamic risk prediction models that incorporate updated data on mental health relapses, incidents of victimization, and desistance from crime will produce more accurate risk stratification for reoffending than models lacking dynamic measurement of risk factors. Method We analyzed a national cohort of 59,676 individuals given community sentences in Sweden, of whom 23,879 (45%) had prior psychiatric diagnoses and 18,546 (31%) had substance use disorder diagnoses. Model development tested prespecified criminal history, sociodemographic, and clinical risk factors. Employing landmarking methods for time-to-event data, we modeled the effects of new health care episodes during community supervision, changes in a supervised individual's circumstances, and the impact of crime desistance. We validated the model in a geographically distinct population. Results During follow up, 18,307 (31%) were reconvicted, 4,416 (7%) committed a violent offense, and 5,381 (9%) were hospitalized with a psychiatric diagnosis. The model demonstrated strong calibration and discrimination performance (c-index = 0.74 for violent reoffending, c-index = 0.69 for general reoffending). It also outperformed comparison models that did not incorporate dynamic data. The final model was translated into an online risk calculator (OxMore). Conclusions Implementation of dynamic models could lead to more accurate risk stratification for individuals under community supervision, including those with psychiatric and substance use disorders, potentially improving resource allocation, and linkage to interventions that reduce recidivism rates. (PsycInfo Database Record (c) 2026 APA, all rights reserved).
How this agent works
It searches only this paper’s licensed source text, returns matching passages, and refuses to invent an answer when the record is silent.
Original journal Law and human behavior
Published 2025-12-04
DOI 10.1037/lhb0000641
Open record PMC7618717 The original authors and American Psychological Association do not endorse this journal or its agent. Verify findings in the canonical article before making consequential decisions.