Risk assessment and prediction of violent and sexual recidivism in sex offenders: Long-term predictive validity of four risk assessment instruments

The long term predictive validity of four current risk assessment instruments (Static-99, HCR-20, SVR-20, PCL-R) was assessed in Germany on three different groups of subjects. The groups consisted of 73 sex offenders released from a prison-based special therapeutic unit (therapeutic sample), 15 individuals who did not complete this treatment (drop-outs), and another 46 subjects who were assessed in a psychiatric hospital having been accused of a sexual offence (assessment sample). The results of the three samples were then compared. The outcome was obtained by examining the national conviction registry. The mean follow-up time was 9 years (range: 1 - 340 months). The highest rate of recidivism was observed in the drop-out group, with only one individual not reoffending. The assessment group had slightly fewer reoffences compared to the treatment group, which had been selected as a high risk population. The treatment sample exceeded the assessment sample, especially concerning sexual reoffences, but less so concerning violent reoffences. Of all the assessment instruments and using the total sample, the Static-99 was the most efficient predictor of all, violent non-sexual and non-contact sexual recidivism. For the prediction of contact sexual recidivism the AUC of the HCR-20 and the SVR-20 was slightly better, but very similar to the Static-99. These differences between assessment instruments were, however, not statistically significant. The AUC for the Static-99 was smaller in the treatment group, but larger than for other assessment instruments. The Static-99 risk categories correlated significantly with the Kaplan - Meier survival functions.