Page 48 - Contributed Paper Session (CPS) - Volume 7

CPS2027 Olayan A. et al.
Table 3 below shows an increasing linear trend observed between the
participants’ risk scores (in quintiles) and the corresponding HRs when the
st
lowest scoring category (i.e. 1 quintile) was the reference. Overall the
st
reoffending rates ranged from 5.0 per 100 PY (95% CI: 4.5, 5.6) (in 1 quintile)
th
to 23.6 per 100 PY (95% CI: 21.9, 25.4) (in 5 quintile) (for the development
st
dataset) and 5.7 per 100 PY (95% CI:4.9, 6.7) (in 1 quintile) to 25.0 per 100 PY
th
(95% CI:22.5, 27.8) (in 5 quintile) (for the validation dataset).
Table 3:Incidence rates (95% CI) at quintiles of the risk score

Overall (overall incidence rate: 9.6 per 100 PY (9.3,9.9)

Development data: Validation data:
Incidence rate (95% CI) Incidence rate (95% CI)
Overall 9.6 (9.2,10.0) per 100 PY 9.5 (9.0,10.1) per 100 PYǂ
ǂ
Score Incidence rate (95% Adjusted HR Incidence Adjusted HR
1
CI) (95% CI) p- rate (95% CI) p-
value (95% CI) value
1 5.0 (4.5,5.6) 1 5.7 (4.9,6.7) 1
st
quintile
2 6.9 (6.3,7.7) 1.38 <0.001 6.2 (5.4,7.0) 1.13 0.238
nd
quintile (1.19,1.61) (0.92,1.38)
3 9.1 (8.5,9.7) 1.81 <0.001 9.3 (8.1,10.6) 1.63 <0.001
rd
quintile (1.59,2.06) (1.33,1.99)
4 14.8 (12.8,17.0) 2.65 (2.22,3.17 <0.001 10.7 (9.4,12.1) 1.87 <0.001
th
quintile ) (1.53,2.27)
5 23.6 (21.9,25.4) 4.26 (3.73, <0.001 25.0 3.96 <0.001
th
quintile 4.87) (22.5,27.8) (3.29,4.76)
We have performed additional analyses in order to measure the
performance of the risk prediction models using different cut-points of the
total risk score in the development and validation datasets (Table 4). The
purpose of these analyses was to determine the optimum cut-point to classify
those at highest risk of the reoffending with statistically acceptable robustness.
Discriminative powers of the risk scores was only 0.57% and 54% for
development and validation data sets, respectively. However,
Table 4:Performance of the risk scoring algorithm for different cut points:
Overall
Development Data Validation Data
W =57% (95%CI: 56%, 58%) AUC =54 % (95%CI: 53%, 55%)
ǂ
ǂ
Median (IQR) Mean (SD) Median Mean (SD)
(IQR)
Risk score 9 (7-12) 10 (5) 10 (7-13) 11 (5)
Cut points † Sensitivity Specificity Sensitivity Specificity
≥4 98% 5% 99% 3%
≥8 84% 30% 94% 11%
≥9 81% 34% 91% 16%
≥10 72% 43% 89% 19%
≥11 69% 47% 87% 21%
≥12 58% 56% 83% 25%
≥14 56% 58% 78% 29%

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