The bootstrap method was used to perform an internal validation of the PO and ECR models and to obtain bias-corrected estimates of predictive accuracy as described by Harrell and coworkers (14). Bias may result from overfitting the models. For both models, 1000 bootstrap replications were used to estimate and correct for optimism in various statistical indices (Table 2.2). The bias-corrected indices were similar for both models. For the original D%y values the optimism from overfitting was estimated to be 0.02 for both models, resulting in bias-corrected estimates of predictive discrimination of 0.80 and 0.82 for the PO and ECR models, respectively. The intercept and slope were closer to zero and one, respectively, for the ECR model than for the PO model, and the maximum calibration error (E^ax) was slightly smaller for the ECR model (Table 2.2).