Risk stratification for early biochemical recurrence of prostate cancer in the era of multiparametric magnetic resonance imagining-targeted biopsy.

ABSTRACT

BACKGROUND: Multiparametric magnetic resonance imaging (MRI) and MRI-targeted biopsy are nowadays recommended in the prostate cancer (PCa) diagnostic pathway. Ploussard and Mazzone have integrated these tools into novel risk classification systems predicting the risk of early biochemical recurrence (eBCR) in PCa patients who underwent radical prostatectomy (RP). We aimed to assess available risk classification systems and to define the best-performing. METHODS: Data on 1371 patients diagnosed by MRI-targeted biopsy and treated by RP between 2014 and 2022 at eight European tertiary referral centers were analyzed. Risk classifications systems included were the European Association of Urology (EAU) and National Comprehensive Cancer Network (NCCN) risk groups, the Cancer of the Prostate Risk Assessment (CAPRA) score, the International Staging Collaboration for Cancer of the Prostate (STAR-CAP) classification, the Ploussard and Mazzone models, and ISUP grade group. Kaplan-Meier analyses were used to compare eBCR among risk classification systems. Performance was assessed in terms of discrimination quantified using Harrell's c-index, calibration, and decision curve analysis (DCA). RESULTS: Overall, 152 (11%) patients had eBCR at a median follow-up of 31 months (interquartile range 19-45). The 3-year eBCR-free survival rate was 91% (95% confidence interval [CI] 89-93). For each risk classification system, a significant difference among survival probabilities was observed (log-rank test p < 0.05) except for NCCN classification (p = 0.06). The highest discrimination was obtained with the STAR-CAP classification (c-index 66%) compared to CAPRA score (63% vs. 66%, p = 0.2), ISUP grade group (62% vs. 66, p = 0.07), Ploussard (61% vs. 66%, p = 0.003) and Mazzone models (59% vs. 66%, p = 0.02), and EAU (57% vs. 66%, p < 0.001) and NCCN (57% vs. 66%, p < 0.001) risk groups. Risk classification systems demonstrated good calibration characteristics. At DCA, the CAPRA score showed the highest net benefit at a probability threshold of 9%-15%. CONCLUSIONS: The performance of risk classification systems using MRI and MRI-targeted information was less optimistic when tested in a contemporary set of patients. CAPRA score and STAR-CAP classification were the best-performing and should be preferred for treatment decision-making.