Table 2.

Performance metrics with corresponding 95% confidence intervals for each model.

Metric	Estimate (95% confidence interval) [95% prediction interval]
	Cox proportional hazards model	XGBoost	Neural Network
Harrell’s C index	0.802 (0.787 to 0.817) [0.766 to 0.839]	0.723 (0.689 to 0.756) [0.628 to 0.817]	0.650 (0.516 to 0.784) [0.202 to 1.000]
Calibration slope	0.980 (0.897 to 1.062) [0.778 to 1.182]	1.180 (1.056 to 1.305) [0.781 to 1.580]	1.855 (−0.945 to 4.654) [−7.552 to 11.261]
Calibration-in-the-large	−0.020 (−0.103 to 0.062) [−0.222 to 0.182]	0.180 (0.056 to 0.305) [−0.219 to 0.580]	0.855 (−1.945 to 3.654) [−8.552 to 10.261]
Royston & Sauerbrei’s D	1.880 (1.768 to 1.993) [1.629 to 2.131]	–	–
Royston & Sauerbrei’s R2	46.0% (43.1% to 48.9%) [39.3% to 52.7%]	–	–

For the Cox and XGBoost models, these were estimated using random-effects meta-analysis following internal–external cross-validation, which also provided a 95% prediction interval.