Table 3.

Apparent and validation performance of prediction model.

Performance Statistics	Apparent performance in original sample	Internal validationa			External validationb
		Bootstrap performance	Test performance in original sample	Optimism corrected
C statistic	0.9147	0.91739	0.91157	0.90887	0.8176
Calibrationc	Slope: 1.0254 Intercept: 0.0326	Slope: 1.0244Intercept: 0.0150	Slope: 0.9272 Intercept: −0.1874	Slope:0.9282	Slope: 0.5986 Intercept: −1.4804

^aPrediction model was internally validated in AECOPD patients admitted between 2015 and 2017. The estimates (95% confidence interval) of apparent performance of prediction model in original sample: 0.9147 (0.8850, 0.9444) for C statistic, 1.0254 (0.7276, 1.3233) for calibration slope and 0.0326 (−0.7834, 0.8486) for intercept. The apparent performance in the model derived from bootstrap sample was compared with the test performance obtained when applying the model to the original sample. Differences in apparent and test performances across all models were averaged to estimate the overall optimism for C statistics and slope. The optimism-corrected C statistics and slope were 0.90887 and 0.9282, respectively. The corrected slope (0.9282) was also shrinkage factor to adjust the original prediction model.

^bExternal validation was performed in AECOPD patients admitted between 2018 and 2019. At external validation, the calibration slope reflects the combined effect of overfitting on the development data (2015–2017) and true differences in effects of predictors. Estimates (95% confidence interval) of prediction model performance were 0.8176 (0.7487, 0.8865) for C statistic, 0.5986 (0.2409, 0.9563) for calibration slope, and −1.4804 (−2.8037, −0.1571) for intercept, respectively.

^cIntercept and slope of calibration plot were estimated in a logistic regression model with in-hosptial death events as outcome and linear predictor as the only independent variable.