Even with a dataset as large as the one used here, the low 12-month prevalence of suicide attempts made it impossible to develop stable risk indices separately for each participating country. However, in order to get a rough indication of the performance of the summary risk indices in particular countries we calculated AUCs to assess concordance between predicted probability of an attempt and actual attempts. The predicted probability estimates were generated from separate equations for planned and unplanned attempts in developed and developing countries. Borrowing standard verbal descriptors used to characterize the strength of κ,50 the AUC estimates were substantial (>.80) for nearly 30% of the replications, moderate (>.70–.80) for an additional 25% of replications, and fair for another 25%. Prediction accuracy was considerably greater, though, in developed countries (nearly 50% substantial and 70% either moderate or substantial) than developing countries (less than 20% substantial and 40% either moderate or substantial) and somewhat greater for planned attempts (nearly one-third substantial and nearly 60% either moderate or substantial) than unplanned attempts (about one-fourth substantial and nearly 50% either moderate or substantial). (Detailed results are available on request.)
