Real-world Implementation and Predictive Performance of Suicide Prediction Models: A Systematic Review and Meta-analysis

This systematic review and meta-analysis evaluates the real-world predictive performance of suicide prediction models. By synthesizing 20 evaluations from 9 studies, we found strong overall discrimination (pooled AUC 0.850) but extreme performance heterogeneity. Crucially, complex machine learning algorithms did not significantly outperform traditional logistic regression. The findings highlight that future implementation efforts should prioritize rigorous local validation and workflow integration over the search for a universally complex algorithm.