Statistical analyses

The predictive accuracy of the scales was examined using the following measures: sensitivity (the proportion of people who repeated self-harm and were correctly identified by the scale as high risk), specificity (the proportion of people who did not repeat self-harm and were correctly identified by the scale as low risk), positive predictive value (the probability that the person identified as high risk went on to repeat self-harm), negative predictive value (the probability that the person identified as low risk did not repeat self-harm), positive likelihood ratio (the increased likelihood of a high risk scale result in a person who repeated self-harm vs. one who did not), negative likelihood ratio (the decreased likelihood of a low risk scale result in a person who repeated self-harm vs. one who did not) and diagnostic odds ratio (the odds of a high risk scale result in a person who repeated self-harm vs. one who did not). Receiver operator characteristic (ROC) curves, which show sensitivity on the y-axis and 1 minus specificity on the x-axis for all possible scale thresholds were plotted [24]. The area under the curve (AUC), based on the published cut-off points for each scale, was also calculated. The AUC represents the overall proportion of cases correctly predicted by the test; an AUC of 0.5 would suggest the test does not perform any better than chance while an AUC of 1.0 indicates every case is predicted correctly. Chi-square tests were used to examine differences in the AUC between subgroups. Stata V.13.1 and OpenEpi were used for the analyses.