Risk stratification for prostate cancer screening using PSA and prostate volume.

e17000 Background: Prostate-specific antigen (PSA) testing is widely used for prostate cancer screening but is limited by low specificity, leading to unnecessary imaging and biopsies. In this retrospective study across seven Brazilian laboratories, we developed a logistic regression model incorporating total PSA, free PSA, age, and PSA density (PSA/Prostate Volume), and compared its performance with PSA alone. Methods: We analyzed PSA data from 2,978 men aged 45-75 years who underwent prostate magnetic resonance imaging (MRI) or biopsy within six months of blood collection, collected between September 2010 and October 2024 across seven Brazilian diagnostic laboratory networks (Fleury, Labs a+, Grupo Pardini, a+ Medicina Diagnóstica, Felippe Mattoso, Weinmann and LAFE). Cases (190) were defined by biopsy-confirmed prostate cancer, and controls (2,788) by benign MRI findings (PI-RADS 1-2) or negative biopsy. Prostate volume was extracted from MRI or ultrasound reports. Data were pooled into a single dataset, randomly split into training (70%) and test (30%) sets, and used to train a logistic regression model with bootstrap-based performance estimation. Results: Total PSA, free PSA, PSA density (p < 0.001), and age (p < 0.05) were significantly higher in men with prostate cancer. Incorporation of these variables into a logistic regression model yielded an average AUC of 0.77 ± 0.04, with sensitivity of 0.80 ± 0.06, specificity of 0.65 ± 0.02, accuracy of 0.66 ± 0.02, balanced accuracy of 0.72 ± 0.03, negative predictive value of 0.98 ± 0.01, and positive predictive value of 0.13 ± 0.02. Compared with PSA alone using a cutoff of 4 ng/mL, which yielded a sensitivity of 0.80 ± 0.06 and a specificity of 0.46 ± 0.02, the model was associated with a 35.1% relative reduction in false-positive classifications. Conclusions: Our machine learning model, integrating PSA-derived parameters and PSA density, demonstrates potential as a risk stratification tool to identify men at elevated risk for prostate cancer, potentially optimizing the use of MRI and biopsy resources, as prostate volume can be estimated using lower-cost ultrasound. This strategy may be particularly advantageous in resource-limited settings, where access to advanced imaging and invasive diagnostic procedures is limited. External validation across diverse populations is required to support clinical implementation and confirm model generalizability and effectiveness.