To date, there are a limited number of studies looking at the forensic outcomes of high security patients who have been discharged via medium secure care [
27,
28]. In this study the 72 HSPH patients had similar characteristics to those described in other MSUs e.g. [
28,
41‐
44] in that they were predominately male with extensive forensic and psychiatric histories. In a pseudo-prospective study design we examined the predictive accuracy of the HCR-20 for outcomes following transfer from high to medium secure psychiatric care. As far as we know this is the first international study to look at the HCR-20 in this way as most studies have focused on either institutional or community violence [
12,
16,
29,
33,
35‐
37,
45‐
48]. It is also the first to report data on the validity of this measure at predicting a broader range of outcomes following transfer to lower levels of security in the UK or elsewhere. We predicted that high scores on the Historical Clinical Risk -20 scale would be predictive of poor outcome in medium secure services. We did indeed find that the HCR-20 score was a good predictor of failed transfer. The total score ROC AUC curve was 0.86 which is much higher than the modest to moderate ROCs reported in many previous studies [
9]. It is also noteworthy that it was the clinical and risk management subscales that contributed most to this effect. Studies have reported varying degrees of contribution from the dynamic subscales but the research evidence seems to suggest that the contribution of dynamic scales vary as a function of the stage of rehabilitation. In Gray's et al's [
33] pseudo prospective 2 year follow up study of patients discharged from medium security to the community only the Historical and Risk scales were predictive. The clinical scales did not show notable accuracy. They suggest that the lack of predictive accuracy in their sample may reflect the clinical stability of those deemed suitable for discharge to the community as well as the differences in follow up time. Our finding that the clinical and risk items both contribute significantly to the prediction of poor outcomes fits with our previous studies in medium secure samples [
16,
45] and also fits with the notion that the clinical items may be more robust predictors of negative outcomes if failure is also determined by clinical issues such as lack of response to medication. There are a number of studies that have compared the post discharge outcomes of patients and using the HCR-20 with Violence Risk Appraisal Guide [
49] and the Psychopathy Checklist Revised [
50] or Psychopathy Checklist- Screening Version (PCL;SV.[
51]) which are measures of psychopathy that have been shown to be predictive of post discharge violence [
52]. In one study [
53] 193 psychiatric patients were assessed using both the HCR-20 and The PCL: SV. At 2 year follow up, the AUCs for the HCR-20 ranged from 0.76-0.80 for a range of aggressive and threatening behaviors, but the PCL: SV had only moderate predictive power. Interestingly, the HCR-20 had incremental validity over and above the PCL: SV. Similar findings were noted in our previous prospective 24 week follow up study of patients discharged from medium secures and civil psychiatric settings work who had been assessed using the HCR-20, VRAG and PCL:SV[
45]. Here we found that the HCR-20 and PCL:SV were better predictors of violence post discharge than the VRAG, but in the regression analyses the HCR-20 (particularly the clinical and risk scales) had incremental validity over and above the PCL:SV [
45]. A Swedish retrospective study on 40 male forensic patients [
37] also found that the HCR-20 was highly predictive of violent recidivism and that the clinical and risk management scales predicted recidivism much better than the historical scale. Overall, our findings seem to suggest that the HCR-20 is a useful tool in predicting those who will fail in their rehabilitation. The broader literature also suggests that it has utility in predicting post discharge recidivism (particularly violent outcomes) for both forensic and correctional samples [
9]. There is a growing literature that suggests it has utility in predicting in-patient aggression and outcome [
35] although the findings have been less robust as in-patient aggression may be more associated with heightened affect and active psychotic symptoms in US studies [
12]. While there is now little doubt that structured risk assessment instruments outperform clinical judgment for the prediction of violent behavior and poor outcome for predominately male samples [
6,
11], there is relatively little data on female forensic or correctional samples. The vast majority of risk assessment studies in women have been based on psychopathy assessments [
54,
55] and there is limited data on the validity and utility of the HCR-20 in women [
56]. Some studies looking at gender differences in the HCR-20 do not note significant differences between men and women [
8,
14] however, work by de Vogel & de Ruiter [
57] showed that the HCR-20 total score demonstrated lower predictive accuracy for violent outcome in women compared to men. Given the observed gender differences future studies need to address this issue[
15].