Model performance measures, stratified by type of prediction study.

Numbers are column percentages, with absolute numbers in parentheses. The percentages sometimes do not add up to 100% because development studies commonly reported more than one performance measure or validity assessment.aImpact studies (n = 3) were excluded since all items were not applicable. Additionally, two external validation studies were excluded because they evaluated risk stratification tools that did not provide predicted probabilities (the Manchester triage system [74] and predictive life support tools [75]). Hence, almost all items were not applicable. Hence, for this table total n = 66 studies.bThe predictive performance (e.g., C-statistic, calibration, or net reclassification index) of the prediction model as estimated from the same data from which the model was developed.AUC-ROC, area under the receiver operation characteristic curve; NRI, net reclassification index.