Timo Soeterik

132 CHAPTER 7 suspicion of (extensive) EPE, on mpMRI or DRE were selected for RARP. Whereas patients with a high risk of (extensive) EPE, were more likely to be treated with radiation therapy. This assumption is supported by positive predictive value (PPV) rates for EPE established by DRE and mpMRI. The PPV for EPE assessed by DRE was respectively 54% in validation cohort 2, whereas this was 73% in the development cohort. PPV of mpMRI for EPE detection was 48% in validation cohort 2, compared to 63% in the development cohort. Interestingly, model discrimination was found to be higher for model 2, 3 and 4 when applied in validation cohort 1, compared with the development cohort (0.83 vs. 0.80, 0.81 and 0.81). These differences might be explained by the heterogeneity of the patient cohort used for model development. As mentioned previously, a large proportion of the patients undergoing RARP at CWH underwent diagnostic staging work-up elsewhere. Owing to the referral pattern, there was a larger variation in used prostate biopsy protocols, mpMRI readings and histopathological biopsy evaluation as patients came from different hospitals. However, we assume that the multi-centre nature of this cohort enabled accurate model estimation leading to robust tools which can be applied in different patient settings. Another explanation for the observed improved discrimination could stem from the fact a large prospective prostate biopsy trial (4M study) was ongoing in validation cohort 1 during the study period. 24 As of part of the protocol, higher rates of patients underwent MRI target biopsy as well as concomitant systematic biopsies in a protocolled trial setting, potentially leading to more accurate tumour sampling. To our knowledge, two other nomograms for prediction of side-specific EPE including mpMRI features have been developed previously. One of these was derived using data of 264 consecutive men undergoing RP between 2012 and 2015. The authors reported excellent model discrimination (AUC: 0.86) and excellent calibration. 25 The model, however, includes a number of complex features, which may not always be readily available in a real-world clinical setting, such as ESUR classification for EPE and capsule contact length on MRI. In addition, this model has not yet been externally validated and thus the performance when applied in other populations remains unclear. The other nomogram, developed by Martini et al, was based on data from 589 patients who underwent RARP between February 2014 and October 2015. The authors reported excellent discrimination in terms of AUC (0.82) and high agreement between predicted and observed probabilities. 26 Sighinolfi et al also externally validated The Martini model . 27 In this external validation study, moderate-low discriminative ability (AUC 0.68) and low sensitivity (20%) and specificity (54%) at the 20% cutoff were reported. 27 In another recently published external validation study, good discrimination in terms of AUC (0.78) but poor calibration, even after model updating, were reported. 28 What our study adds to this previous work is that we have shown that our developed nomogram not only provides accurate EPE risk prediction in the development cohort, but also when applied in external patient populations. Implementation of tools that facilitate shared decision-making may improve quality of prostate cancer care, as active involvement of patients is associated with less decision conflict and less decision regret. 29 Our proposed nomograms can facilitate this,