Development and external multicenter validation of a multimodal nomogram integrating MRI findings and PSA biomarkers for individualized risk assessment of clinically significant prostate cancer

Background Accurate identification of clinically significant prostate cancer (csPCa; Grade Group ≥2 [GG2+]) remains challenging because prostate-specific antigen (PSA) lacks specificity and magnetic resonance imaging (MRI) alone may lead to unnecessary biopsies. We developed an interpretable model combining routinely available MRI and PSA-related variables. Methods This retrospective multicentre study included a development cohort from Center 1 (n = 260) and external validation cohorts from Centers 2 (n = 91) and 3 (n = 29). Multivariable logistic regression integrated age, PI-RADS score, free-to-total PSA ratio, and PSA density (PSAD). Performance was evaluated using discrimination, calibration, reclassification, bootstrap internal validation, and decision curve analysis. External validation used the fixed model without refitting. Results In the development cohort, 112 patients (43.1%) had GG2+ disease. The model achieved an apparent AUC of 0.872 (95% CI 0.823–0.914) and an optimism-corrected AUC of 0.864, outperforming PSAD alone (AUC 0.807; ΔAUC = 0.065, 95% CI 0.012–0.120; P = 0.013). Calibration was satisfactory (Hosmer Lemeshow P = 0.797). The model improved reclassification (continuous NRI = 0.789; IDI = 0.206) and provided greater net benefit than the comparator strategies at the evaluated threshold probabilities of 20%–30%. At a 25% threshold, it avoided 17 biopsies per 100 patients without additional missed GG2+ cancers compared with PSAD ≥0.15 ng/mL/cm 3 . External AUCs were 0.832 (95% CI 0.733–0.917) and 0.880 (95% CI 0.690–1.000). Conclusion The MRI–PSA model showed favorable discrimination and satisfactory development-cohort calibration, with potential clinical utility for GG2+ prediction. Prospective validation in representative non-Chinese populations and diverse healthcare systems is required before broader clinical adoption.