Timo Soeterik

128 CHAPTER 7 Performance of the fourmodelswhen applied in two external patient cohorts Overall, higher AUCs were observed when applied in validation cohort 1 compared with those observed in cohort 2 (Table 4). As shown in Table 4. and Figure 1., discrimination and calibration of all models were both excellent when applied in cohort 1. In cohort 2, model 2 had the best performance with both fair AUC and relatively highest agreement between predicted and observed probabilities (Figure 1 and 2). In cohort 2, substantial miscalibration was observed for the other three models (Figure 2A, 2C and 2D). TABLE 4. Performance of all models when applied in two external cohorts Validation cohort 1 Validation cohort 2 Model 1 Model 2 Model 3 Model 4 Model 1 Model 2 Model 3 Model 4 AUC (95%CI) 0.80 (0.68-0.88) 0.83 (0.71–0.90) 0.83 0.71–0.90) 0.83 (0.72–0.90) 0.77 (0.62–0.87) 0.77 (0.64–0.87) 0.78 (0.64–0.88) 0.78 (0.64–0.88) FIGURE 1. Calibration plots of all 4 models when applied in validation cohort 1 A. Model 1 B. Model 2 C. Model 3 D. Model 4