|Saturday, February 23|
|CS17 Theme 3: Prediction and Analytics #4||
Sat, Feb 23, 10:45 AM - 12:15 PM
A Novel Approach for Improving Predictive Accuracy of Prognostic Models for Violent ReoffendingView Presentation *Constantinos Kallis, Queen Mary, University of London
Keywords: prognostic model, predictive accuracy, AUC
A number of prognostic methods are currently implemented for serious offenders prior to their release from prison to predict their likelihood of being convicted for violence after release. The predictive accuracy of these methods is moderate for the UK population of serious offenders and remains unknown for key personality disorder diagnostic groups.
We identify methodological problems related to the construction of existing prognostic methods for violent reoffending. We also create a novel prognostic model that will improve predictive accuracy.
The overall AUC for our prognostic model is 0.76 (95% C.I. 0.74, 0.79, p < 0.001) and it is higher and significantly different when compared to the corresponding AUCs for the other prognostic methods. For cases with primary psychiatric (Axis I) disorders, the AUC is 0.75 (95% C.I. 0.71, 0.78, p < 0.001) and for those with Antisocial Personality Disorder (ASPD), the corresponding AUC is 0.72 (95% C.I. 0.68, 0.75, p < 0.001).
By constructing a prognostic model for violent reoffending that uses only highly predictive items for a specific outcome, it is possible to improve significantly the predictive accuracy of prognostic methods.