Presentation Authors: Karim Saba*, Marian S. Wettstein, Laura Lieger, Olivia Märzendorfer, Andreas M. Hötker, Olivio F. Donati, Cédric Poyet, Tullio Sulser, Daniel Eberli, Ashkan Mortezavi, Zurich, Switzerland
Introduction: Multivariable prostate cancer risk calculators (PC-RC) aim to guide the decision to undergo prostate biopsy by predicting the prevalence of clinically significant prostate cancer (csPCa) defined as Gleason score â‰¥3+4. A new generation of unvalidated PC-RCs incorporate information of multiparametric magnetic resonance imaging (mpMRI) to enhance their predictive performance. The aim of this study is to externally validate 3 novel PC-RCs (Radkte et al. [RC1], Mehralivand et al. [RC2] & van Leeuwen et al. [RC3]) and compare their predictive performance to 2 established mpMRI-naive PC-RCs (PBCG & PCPT2).
Methods: All men without a previous PCa diagnosis who underwent TRUS-mpMRI-fusion guided template prostate biopsy at our institution between 2014 and 2018 were considered for this study. All predictors were retrospectively collected. Probabilities of csPCa were calculated according to the logistic regression models of all PC-RCs. Discrimination, calibration and clinical usefulness were assessed by ROC analyses, calibration plots and decision curve analyses, respectively.
Results: After exclusion of 56 (11%) patients due to missing predictors, 468 men were eligible for final analyses, of whom 323 (69%) were biopsy naive and 145 (31%) had at least one prior negative prostate biopsy. MpMRI-TRUS-fusion guided template prostate biopsy detected csPCa in 193 (41%) patients. All mpMRI-RCs clearly outperform the established mpMRI-naive RCs, when it comes to discrimination (RC1: 0.73, RC2: 0.84, RC3: 0.83, PCBG: 0.68, PCPT2: 0.66). Calibration of RC2, PBCG and PCPT2 were considerably worse compared to RC1 and RC3 (predicted proportion of csPCa were 0.43, 0.68, 0.39, 0.29 and 0.12 for RC1, RC2, RC3, PBCG and PCPT2 respectively. RC3 added an incremental benefit over the other PC-RCs in a decision curve analysis across the range of clinically meaningful threshold probabilities to biopsy.
Conclusions: Incorporation of mpMRI as an additional predictor for PC-RCs leads to better predictive performance in comparison to established mpMRI-naive PC-RCs. Among the novel PC-RCs, the model of Van Leeuwen et al. (RC3) outperformed its competitors cumulatively with regards to discrimination, calibration and clinical usefulness within our external validation cohort.