Timo Soeterik

171 General Discussion and Future Perspectives (Chapter 7 & Chapter 8). However, inherent to use of prediction tools in general, assessment of performance of these models in external patient populations remains important to ensure their safety. This was emphasized by the results of our study described in Chapter 6, showing disagreement between predicted and observed rates of EPE when using the Martini EPE nomogram for the prediction of side-specific EPE in an external hospital population. These findings were concordant with those reported by Sighinolfi et al. 11 Based on the findings of these studies we conclude that the Martini nomogram may not be a suitable prediction tool to predict side-specific EPE in patients undergoing RARP. The major strength of our study in which we propose an alternative side-specific EPE nomogram, is that we used two separate hospital patient populations for external validation (Chapter 7). This nomogram is the first side-specific EPE prediction tool including mpMRI information, showing good performance when applied in other hospital populations. Therefore, this tool should be recommended for side-specific EPE risk prediction in clinical practice. The nomogram is currently publicly accessible as an online web calculator. 12 Although our side-specific EPE nomogram has shown to be an accurate tool, two important limitations of the study should be mentioned. First, there was a wide variety of biopsy protocols used in the development cohort. Overall 56% were diagnosed using systematic biopsy, 16% using MRI target biopsy without concomitant systematic biopsy and 28% underwent target biopsy and concomitant systematic biopsy. Due to the increased adoption of MRI-guided target biopsy, it can be expected that sampling error may further decrease, leading to more precise biopsy ISUP grading. Therefore, future updating of the nomogram is crucial to ensure the applicability in the contemporary era of clinical practice. It is advised that updating occurs by using data of a more contemporary cohort including men all receiving target and concomitant systematic biopsies. The second limitation is the lack of central review of MRI, which could have improved the diagnostic accuracy of MRI for the detection of EPE. 13 However, variations in MRI reading (quality and interobserver variability) are inherent to real-world clinical care, and central review is mostly not incorporated into daily clinical practice. Also, by using a multi-centre hospital population for model development, these variations were already partially accounted for. In Chapter 8, we reported that use of mpMRI T-stage improved model discrimination of two established nomograms used for the prediction of pelvic lymph node invasion (LNI) in patients with primary prostate cancer, compared with DRE T-stage. These results are important, since studies evaluating the added value of using mpMRI T-stage for LNI risk prediction are scarce. Based on the study’s main results, we can conclude that mpMRI T-stage can safely replace DRE T-stage as input parameter for nomogram-based LNI risk prediction. Use of mpMRI T-stage led to relatively higher probabilities of LNI, due to migration of patients initially staged as cT1c or cT2 disease by DRE, to cT2 and higher. When using mpMRI T-stage, lower rates of pelvic LNI would be missed (higher sensitivity), compared with DRE T-stage. However, this was at the cost of performing 10