Evaluating the performance of existing tools to predict clinically significant prostate cancer in men with indeterminate lesions on biparametric MRI and development of a novel multiplex model: a prospective cohort study

Background: Indeterminate lesions on prostate MRI, such as PI-RADS 3, present a clinical challenge due to their equivocal nature, complicating biopsy decisions in men undergoing testing for prostate cancer. Understanding the predictive capacity of biomarkers and risk calculators is critical to improve clinical decision-making and reduce unnecessary biopsies. Methods: In this prospective cohort study, men with PI-RADS 3 findings on biparametric MRI (bp-MRI) who underwent combined biopsy (fusion targeted and systematic) in the STHLM3-MRI randomised clinical trial (first- and second-rounds) and at Capio St Göran's Hospital (Capio PCC), Sweden were included, representing screening-by-invitation, repeat screening, and clinical practice cohorts, respectively. Data collection occurred between Feb 5th, 2018, and Mar 4th, 2020, for STHLM3-MRI first-round screening, between Nov 10th, 2021, and Feb 20th, 2023 for second-round screening, and between Jan 7th, 2017 and June 30th, 2023 for Capio PCC. The data was collected directly from the participating laboratories using standardized reporting forms, medical charts, and additional study-specific data collection forms filled by patients. The primary outcome was detection of clinically significant prostate cancer (csPCa; ISUP ≥2) in men with PSA ≥3 ng/mL confirmed by the combined biopsy. The predictive capacity of the evaluated biomarkers (PSA density, the Stockholm3 test, prostate volume, MRI lesion volume ratio, and Stockholm3 density), as well as seven risk calculators, was assessed via the area under the curve (AUC) computed using logistic regression. Sensitivity and specificity of detecting csPCa and high-grade prostate cancer (ISUP ≥3) were reported. Complete-case analysis was performed for men with complete data on their PSA, prostate volume, Stockholm3 test, MRI lesion volume, findings on the digital rectal examination, family history of prostate cancer, and previous biopsy. The findings were contrasted to the analysis from the imputed dataset. Findings: Of the 6554 men included into the three cohorts, 1187 received PI-RADS score of 3 on the bp-MRI, and 1146 underwent combined biopsy. Of them, 900 had PSA ≥3 ng/mL, and 656 men were included in the complete-case analysis (169 from STHLM3-MRI first-round, 72 from the second-round, and 415 from Capio PCC). Overall, 370/900 men (41%) and 258/656 men (39%) had ISUP ≥2, but only 75/900 (8%) and 50/656 (8%) had ISUP ≥3. PSA density, tested risk calculators, and probability tests had low-to-moderate AUC (range 0.50–0.73; PSA density range 0.58–0.66, Stockholm3 range 0.59–0.67, lesion volume ratio range 0.54–0.63), and performed similarly across individual cohorts and the combined dataset in the complete-case and imputed dataset analysis. For detection of ISUP ≥2 based on STHLM3-MRI first-round, PSA density at 0.10 had a sensitivity of 69% (56%, 80%), specificity of 49% (39%, 58%), and missing 27% (6%, 61%) of ISUP ≥3, while a PSA density of 0.15 had a sensitivity of 37% (25%, 50%), specificity of 84% (76%, 90%), missing 45% (17%, 70%) of ISUP ≥3. The best-performing model based on STHLM3-MRI included age, prostate volume, Stockholm3 density and MRI lesion ratio, and reduced prostate biopsies by 33% (26%, 40%) while maintaining 98% (91%, 100%) sensitivity to detect ISUP ≥2 cancer, specificity of 50% (41%, 60%) and AUC of 0.82 (0.76, 0.87). Meanwhile, the best-performing model based on the complete-case combined dataset included age, prostate volume, PSA density, and Stockholm3 density, and reduce prostate biopsies by 26% (23%, 30%) with a sensitivity of 90% (85%, 93%), specificity of 36% (31%, 41%), and AUC of 0.70 (0.66, 0.74). Interpretation: Current risk-stratification tools and individual biomarkers perform suboptimally for guiding biopsy decisions in men with PI-RADS 3 lesions. The findings highlight the limitations of relying on PSA density alone and emphasize the need for caution in clinical recommendations. However, multiplex models might offer possibility to reduce unnecessary biopsies while maintaining high sensitivity for clinically significant prostate cancer detection. These findings should be externally validated and evaluated for cost-effectiveness. Funding: STHLM3-MRI clinical trial is funded by the Swedish Cancer Society (Cancerfonden), the Swedish Research Council (Vetenskapsrådet), the Swedish Research Council for Health Working Life and Welfare (FORTE), the Strategic Research Programme on Cancer (StratCan), Hagstrandska Minnesfonden, Region Stockholm, Svenska Druidorden, Åke Wibergs Stiftelse, the Swedish e-Science Research Centre, the Karolinska Institutet, and Prostatacancerförbundet.