Page 49 - Contributed Paper Session (CPS) - Volume 7

CPS2027 Olayan A. et al.
≥16 52% 60% 76% 30%
≥20 41% 64% 72% 31%
≥24 35% 66% 67% 33%
ǂAUC: Area Under the Curve for score as continuous variable; †cut-points identified based on the
distribution of the score
High cut-points resulted in high sensitivity with poor specificity across all
models. For example, the sensitivity of the scoring tool was estimated to be
98% for a cut-point ≥4 (5% specificity), while a cut-point ≥10 yielded 72%
sensitivity and 43$ specificity in the development set; sensitivity and specificity
were estimated as 89% and 19% when the analyses were repeated using
validation dataset. Based on these diagnostic measures, the cut-point ≥ 10 was
selected to classify individual as being at high-risk for the reoffending.

4. Discussion and Conclusion
The result from our study show that a combination of a set of risk factors
can be used to develop risk prediction models which can quantify an
individual’s risk of the reoffending with acceptable statistical accuracy. We
developed and validated a 6-item risk prediction model, using data from 7,743
individuals from a data-linkage study in New South Wales. Most of the risk
factors included in this study have been well recognised and consistently
reported to have a strong association with re-offending, with broader
implications. In the current study, having not contact with mental health
service after the first offence had the highest impact on the reoffending with
the highest risk scores. Those diagnosed with Schizophrenia and related
psychoses or Substance related psychosis, their age was less than 20 years at
the time of the first offence, not diverted outcome from the first offence were
also at risk of the reoffending. However, to improve the Discriminative powers,
we require to include more covariates in the data set such as education status,
number of previous conviction and their types.
In conclusion, our risk predication models can potentially identify
individuals at high risk of the reoffending by targeting very specific profiles
and conditions. This approach may have significant implications for justice
health system and health treatment planning within the clinical setting. Risk
prediction tools such as the one developed here could be seen have several
possible applications in local health care setting, justice system setting and
clinical research setting.

36 | I S I W S C 2 0 1 9