Patients with traumatic brain injury (TBI) are at a greater risk of subsequent violent victimisation, and a prognostic model can assist in identifying those at highest risk who can benefit from targeted interventions. We aimed to develop and internally validate a clinical prediction model to estimate the risk of violent victimisation following TBI in adolescence.
We investigated a cohort of adolescents aged 10 to 24 exposed to TBI between 2009 and 2023, using data from a linked register, covering 86% of the population in Wales, the Secure Anonymised Information Linkage (SAIL) Databank. We fitted a multivariable Cox regression model for the association between predictors and time to violent victimisation identified in medical records, with optimism-corrected bootstrapping for internal validation. Key performance measures, including discrimination and calibration, were examined at 1 and 3 years post-TBI.
The cohort included 34 092 adolescents, of whom 332 (1.0%) were violently victimised within 1 year and 701 (2.1%) within 3 years of the TBI. The final model included a range of predictors including calendar age, sex at birth, substance misuse, psychiatric conditions, neurological conditions, conduct disorder, learning difficulties and a history of victimisation or self-harm. The clinical prediction model showed good calibration and moderate discrimination at 1 (area under the curve (AUC)=0.72) and 3 years (AUC=0.67) post-TBI.
This brief, scalable and evidence-based prediction model showed moderate predictive performance at internal validation. External validation is necessary to test the model’s transportability.
A large population-based sample was used to identify risk factors for violent victimisation and develop a novel clinical prediction model for use in adolescents with TBI.