Timo Soeterik

125 Development of a novel EPE nomogram Model performance, external validation and clinical usefulness Performance of all models was assessed in the development cohort and two validation cohorts. Discrimination, which refers to the ability of the model to distinguish a prostate lobe with the endpoint (EPE) from a lobe without EPE, was quantified using the area under the receiver operating characteristic curve (AUC). 14 Model calibration, which refers to the agreement between observed endpoints and predictions, was assessed using calibration in the large and calibration slopes. 14 The net benefit per risk threshold was determined using decision-curve analysis. The net benefit is calculated as the proportion of “net” true positives (true positives corrected for the false positives weighted by the odds of the risk cut-off, divided by the sample size). 22 Missing data Missing data patterns were explored using response matrix and correlation plots. Missing data was assumed to be missing at random, as their missingness was related to the diagnostic work-up (e.g. selection of patients for mpMRI and biopsy protocols) of the hospitals. Missing data were handled by using multivariate imputation by chained equations including pooling using Rubin’s rules. 23 RESULTS Patient populations Overall, respectively 887 patients (development cohort), 513 patients (validation cohort 1) and 470 patients (validation cohort 2) were included. The values of EPE prevalence on prostatic lobe level of these cohorts were respectively 458/1774 (26%), 225/1026 (21%) and 148/940 (16%). Baseline characteristics on patient and prostatic lobe level are presented in respectively Table 1 and 2. Performance of the four multivariable logistic regression models in the development cohort At multivariable analyses, PSAD, DRE staging, mpMRI staging, ISUP grades 3-5, and percentage positive cores were all found to be significant predictors of EPE (Table 3). Model 4, which includes all available predictors, resulted in the highest AUC (0.82). The AUCs of the other three models ranged from 0.80 to 0.81 (Table 3). 7