Model Performance and Validation

Calibration was rarely assessed, with only one of ten prediction models reporting calibration methods and results. Only one model used internal validation and only one model underwent external validation. The externally validated scoring system had a validation sensitivity of 75.0% and specificity of 98.1%. The review did not meta-analyse AUROC because the included studies were heterogeneous across patients, sources, departments, settings, and regions. The included prediction models reported AUROC values from 0.76 to 0.95, indicating apparently good discrimination.