Timo Soeterik

111 External validation of the Martini Nomogram Model performance The area under the receiver operating characteristic curve (AUC) was 0.78 (95% CI 0.75 – 0.82), indicating fair discriminative ability of the model in the study population. Calibration-in-the large showed substantial underestimation of mean predicted risk: 24% vs. 32% mean observed risk. The underestimation of present EPE of the model was confirmed in a graphical matter using the calibration curve (Figure 2). As shown in both calibration plot and table, the underestimation of the predicted risk was systematic and was present in 12 of 13 groups (Table 3). The outcome of the Hosmer-Lemeshow test was in line with these findings regarding calibration of the model, as it was statistically significant ( p < 0.001); indicating poor model fit. FIGURE 3 AND TABLE 4 - Calibration plot and table of the updated Martini nomogram Group Predicted risk (%) Patients (N) Mean predicted (%) Mean observed (%) 1 7 – 8 64 7 9 2 8 – 9 60 8 8 3 9 – 10 60 9 13 4 10 – 19 60 13 8 5 19 – 22 61 20 13 6 22 – 25 61 24 21 7 25 – 28 62 27 34 8 28 – 32 64 30 33 9 32 – 40 57 35 30 10 40 – 49 63 44 35 11 49 – 57 59 52 61 12 57 – 69 61 63 72 13 69 – 87 60 79 73 6