An MRI-based predictive model (MRI-PM) for estimating the likelihood of clinically significant prostate cancer (csPCa) was created and independently validated across institutions in the Barcelona metropolitan region, accompanied by an online risk calculator (RC) that enables users to define their preferred csPCa probability cut-off. The model was derived from a cohort of 1486 men undergoing 3-tesla multiparametric MRI (mpMRI) and subsequent targeted and/or systematic biopsies at a single academic center. External validation was performed in 946 men evaluated with the same imaging and biopsy workflow at two additional academic hospitals. CsPCa was identified in 36.9% of the development cohort and 40.8% of the validation cohort (p = 0.054). Incorporation of the model increased the diagnostic AUC of mpMRI from 0.842 to 0.897 in the development dataset (p < 0.001) and from 0.743 to 0.858 in the validation dataset (p < 0.001). Using a 15% probability threshold would reduce biopsy procedures by 40.1%, with 5.4% of the 36.9% csPCa cases going undetected. For individuals with PI-RADS <3, biopsy would be recommended in 4.3% of cases, enabling detection of 32.3% of the 4.2% existing csPCa. Among those with PI-RADS 3, biopsy reduction would reach 62%, at the cost of missing 28% of the 12.4% csPCa. For PI-RADS 4, only 4% of biopsies would be avoided, and 0.6% of the 43.1% csPCa would remain undetected. In PI-RADS 5, 0.6% of biopsies would be avoided with no missed csPCa from the existing 42.0%. Although the Barcelona MRI-PM performed well overall, its clinical impact differed according to PI-RADS category. Allowing customization of csPCa probability thresholds in the RC may support more flexible external validation and potentially enhance performance of MRI-PMs within specific PI-RADS strata.
