Timo Soeterik

144 CHAPTER 8 RESULTS Study population A total of 1062 patients fulfilled the inclusion criteria. Overall, 301 (28%) patients had histologically confirmed LNI. The median number of lymph nodes removed was 20 (interquartile range 13-25). A total of 21 patients (2%) had one or more covariates missing including DRE T-stage ( N = 11), mpMRI T-stage ( N = 2), and biopsy data ( N = 8). Baseline characteristics of the study cohort are presented in Table 1. Model performance using DRE or mpMRI T-stage Initial validation included use of DRE T-stage. Discrimination in terms of AUC was 0.71 (95% CI 0.70 – 0.72) for the MSKCC and 0.73 (95% CI 0.72 – 0.74) for the Briganti nomogram. Mean predicted probability for LNI was respectively 24% for the MSKCC and 21% for the Briganti nomogram, where the observed LNI rate was 28% (Table S2). On visual exploration of calibration plots, we also observed systematic underestimation of the predicted risk of both nomograms, particularly for risk thresholds between 0% and 30% (Figures 1a and 1b). FIGURE 1. Calibration plots of both nomograms based on DRE T-stage (A and B) and mpMRI T-stage (C and D). MSKCC, Memorial Sloan Kettering Cancer Centre. A histogram displayed at the top of each calibration plot shows the distribution of predicted risks for pN0 and pN1 cases. pN0 is indicated by the 0 (top side of the histogram), and pN1 is indicated by the 1 (bottom side of the histogram)