External Validation and Comparison of Magnetic Resonance Imaging-based Risk Calculators for Detection of Clinically Significant Prostate Cancer: A Multi-institutional Study Incorporating Transrectal and Transperineal Biopsy Approaches.
TL;DR
MRI-based models' performance remained consistent across subgroups, including age, ancestry, biopsy approach (transrectal or transperineal), clinical setting (academic or community), Prostate Imaging-Reporting and Data System (PI-RADS; v2.0 or v2.1), and MRI-fusion technique (cognitive or software).
Abstract
Background
AND
Objective
Magnetic resonance imaging (MRI)-based risk calculators are widely used to inform prostate biopsy decision-making; however, they were predominantly developed for transrectal biopsies. Their performance using the transperineal approach remains unclear. This study externally validated five MRI-based risk calculators across a diverse cohort undergoing either approach.
Methods
This retrospective cohort study included 4168 men with pre-biopsy MRI followed by prostate biopsy between October 2016 and December 2023 across four Canadian academic and community hospitals. Five MRI-based (Mehralivand, Kinnaird, Patel, Wang, and Peters) and one non-MRI-based (Prostate Biopsy Collaborative Group) risk calculators were compared. The primary outcome was the presence of clinically significant prostate cancer (csPCa; Grade Group ≥ 2). Model performance was characterized using area under the receiver-operating-characteristic curve (AUROC), calibration plots, and net benefit. KEY
Findings
AND
Limitations
Overall, 1797 of 4168 men (43%) had csPCa. Seventy-four percent were of Caucasian ancestry. Biopsies were transperineal in 1379 cases (33%); 32% of patients were biopsy-naïve, 30% had a prior negative biopsy, and 38% had prior Grade Group 1 disease. The MRI-based calculators' performance was similar but worse than originally reported (AUROC = 0.78-0.80 vs 0.80-0.89) although better than that of the non-MRI-based model (AUROC = 0.67, p < 0.001). MRI-based models' performance remained consistent across subgroups, including age, ancestry, biopsy approach (transrectal or transperineal), clinical setting (academic or community), Prostate Imaging-Reporting and Data System (PI-RADS; v2.0 or v2.1), and MRI-fusion technique (cognitive or software). The Kinnaird, Patel, and Peters models showed superior calibration and net benefit.
Conclusions
AND CLINICAL IMPLICATIONS We did not see evidence of a difference in the performance of MRI-based risk calculators regardless of biopsy approach or fusion technique.