Model performance by task type
| Outcome type | Mean validation performance—reported from study | Mean validation performance—NYU data | Mean performance difference between study validation performance and NYU original validation performance | Mean validation performance—NYU retrained | Mean performance difference between study validation performance and NYU retrained validation performance |
| Predicting deterioration (n=4) | Mean AUROC=0.82 (n=4; 0.75–0.88) | Mean AUROC=0.64 (n=3; 0.59–0.74) | Mean AUROC difference=0.18 (n=3; 0.14–0.26) | Mean AUROC=0.72 (n=4; 0.68–0.77) | Mean AUROC difference=0.10 (n=4; 0.01–0.17) |
| Predicting mortality (n=5) | Mean AUROC=0.93 (n=2; 0.88–0.98) | Mean AUROC=0.72 (n=2; 0.67–0.79) | Mean AUROC difference=0.31 (n=1) | Mean AUROC=0.77 (n=5; 0.74–0.93) | Mean AUROC difference=0.15 (n=2; 0.09–0.21) |
| Predicting either deterioration or mortality (n=3) | Mean AUROC=0.73 (n=3; 0.72–0.74) | – | – | Mean AUROC=0.72 (n=3; 0.71–0.73) | Mean AUROC difference=0.01 (n=2; −0.01–0.02) |
–=Value unavailable because authors did not provide feature weights when reporting model development.
AUROC, area under the receiver–operator curve; NYU, New York University.