Timo Soeterik

131 Development of a novel EPE nomogram Clinical usefulness A systematic analysis of the event status of patients who would fall above and below the risk threshold between 5% and 30% is provided in Table 5. At a risk threshold of 20%, a non-nerve sparing approach would be advised in 642/832 (77%) of the prostatic lobes with EPE. Nerve sparing would be recommended in 1980/2908 (68%) of all prostatic lobes without EPE. Risk thresholds ranging from 0% to 30% were regarded as clinically most relevant, for which net benefits of all four models are presented in Figure 3. All models can be regarded clinically useful for risk thresholds between 6% and 30%, as net benefits were found to be higher compared to the “treat all” and “treat none” approach. On external validation, DCA revealed relatively lower net benefits for model 1 compared with models 2,3 and 4, respectively. (Figure 3B and 3C). DISCUSSION Our analysis showed that the three nomograms (model 2, 3 and 4) based on clinical information combined with mpMRI staging information outperformed the nomogram without mpMRI staging information (model 1), in terms of AUC, calibration and net benefit. Among these three nomograms, discrimination and net benefit were comparable. However, model 2 outperformed both model 3 and 4 in terms of agreement between predicted and observed probabilities. Therefore, this nomogram should be the preferred tool for side-specific EPE risk prediction. Besides the fact model 2 outperformed all other models in terms of calibration, it was also the most minimalistic, since the model consists solely of three predictor variables (PSAD, highest biopsy ISUP grade and mpMRI T-stage). In addition, this model is applicable in a wide range of clinical situations, independent of prostate biopsy protocol used. The model can be accessed online at https://www.evidencio. com/models/show/2142. A common explanation for the miscalibration observed when a nomogram is applied in an external population is the case-mix severity. In this study, this is also the most likely cause of the systematic overestimation of the predicted EPE risk in validation cohort 2. As an example, suspicion of EPE on mpMRI was reported in 15% of the lobes in the development cohort, compared with 6% in validation cohort 2. In addition, the prevalence of EPE on final pathology was substantially higher among cases in the development cohort compared to validation cohort 2: 26% versus 16%. As stated previously, highest agreement between predicted and observed probabilities was achieved when using model 2. However, overestimation of the predicted risk was still observed when applied in validation cohort 2. Overestimation was predominantly observed for predicted risks >30%. This was possibly due to the fact that a substantial lower number of patients with relatively high risk for EPE were selected for RARP in validation cohort 2, compared to the development cohort. For example, only the relative proportion of patients with low 7