A psychosis-specific prediction model for violent reoffending following discharge from forensic psychiatric care – FoVOx-P

Introduction

In many jurisdictions, forensic psychiatric care is primarily reserved for individuals with schizophrenia-spectrum disorders. While several violence risk assessment tools exist, many were developed for heterogeneous psychiatric or prison populations [1]. The original FoVOx was developed to estimate violent reoffending risk after discharge from forensic psychiatric care, but was based on broader diagnostic groups rather than psychosis-specific populations [2]. Accurate risk estimation is essential for safe discharge planning, resource allocation, and the reduction of disproportionate detention. However, traditional clinical judgment often shows limited reliability, highlighting the need for evidence-based tools tailored to the clinical profiles of patients with psychotic disorders in specialised forensic settings.

Aims

The aim of this study was to develop and internally validate a psychosis-specific risk prediction model (FoVOx-P) for violent reoffending following discharge from compulsory forensic psychiatric care. The model was designed to be applicable for schizophrenia-spectrum disorders, especially for jurisdictions where forensic services are restricted to this population, utilizing routinely collected clinical and sociodemographic variables.

Methods

Using Swedish national administrative and health registers, we identified a psychosis-restricted subsample of 646 individuals (events = 124) sentenced to forensic psychiatric care and subsequently discharged. The primary outcome was a conviction for a violent crime committed within 12 or 24 months of total discharge from forensic services. We employed a Cox proportional hazards regression model, and variable selection was conducted using the Akaike Information Criterion (AIC) to identify the optimal combination of predictors [3]. The candidate predictors included sociodemographic data, criminal history, diagnostic categories, and pharmacological treatments. Internal validation was performed using bootstrap resampling (500 samples) to estimate optimism-corrected performance metrics, including Harrell’s C-index, time-dependent Area Under the Curve (AUC), and calibration slopes. Overall accuracy was assessed using Brier scores.

Results

The AIC-based selection identified eight key predictors for the FoVOx-P model: age at discharge, previous violent crime, alcohol use disorder, other substance use disorders, antisocial personality disorder, clozapine treatment, antidepressant medication, and absence of previous psychiatric inpatient care. Younger age, previous violent offending (HR 3.16; 95% CI 1.38–7.23), comorbid substance misuse, and antisocial traits (HR 3.20; 95% CI 2.01–5.11) were significantly associated with elevated risk. Conversely, clozapine treatment (HR 0.24; 95% CI 0.10–0.60) and antidepressant use (HR 0.63; 95% CI 0.38–1.04) were associated with reduced risk. FoVOx-P achieved a global C-index of 0.76 (SE = 0.02) and an optimism-corrected C-index of 0.73. Time-dependent AUCs were 0.77 (95% CI 0.70–0.85) at 12 months and 0.79 (95% CI 0.74–0.85) at 24 months. The optimism-corrected calibration slope was 0.82, suggesting moderate overfitting, though calibration remained satisfactory across all risk deciles. Brier scores were 0.061 at 12 months and 0.093 at 24 months.

Conclusions

The FoVOx-P model provides a robust, data-driven approach to risk assessment specifically for patients with psychosis in forensic settings. By incorporating treatment-related variables such as clozapine and antidepressants, the model aligns with clinical evidence regarding pharmacological risk reduction. This psychosis-specific tool offers a transparent alternative to categorical risk labels, supporting more precise and individualized discharge planning.

References

1. Wolf, A., Fanshawe, T.R., Sariaslan, A., Cornish, R., Larsson, H., Fazel, S., 2018. Prediction of violent crime on discharge from secure psychiatric hospitals: A clinical prediction rule (FoVOx). Eur. Psychiatry 47, 88-93.
2. Ogonah, M.G.T., Seyedsalehi, A., Whiting, D., Fazel, S., 2023. Violence risk assessment instruments in forensic psychiatric populations: a systematic review and meta-analysis. Lancet Psychiatry 10 (10), 780-7-9.
3. Cavanaugh JE, Neath AA. The Akaike information criterion: Background, derivation, properties, application, interpretation, and refinements. WIREs Comput Stat 2019; 11: e1460.95% CI