OBJECTIVE: To assess the diagnostic performance of semen RNA-based biomarkers for detecting prostate cancer and differentiating cancer grade groups. METHODS: In a multi-center prospective study, semen samples were collected from men prior to undergoing prostate biopsy. RNA was extracted and sequenced to generate exome-wide gene expression profiles. Differentially expressed genes were selected and used to train a machine-learning classifier designed to detect prostate cancer with ≥95% sensitivity. The finalized model was locked and independently evaluated in a validation cohort. Histopathological diagnosis served as the reference standard. Model performance was further analyzed in relation to ISUP Grade Group classification. RESULTS: Of 301 enrolled participants, 279 samples met quality criteria and were included in model development (training set: n = 199; validation set: n = 80). The median age and PSA level were 62 years and 5.70 ng/mL, respectively. In the validation cohort, the classifier achieved an AUC of 0.90, sensitivity of 92%, and specificity of 69%, with no significant performance difference compared to the training cohort. Importantly, high-risk cancers (ISUP Grade Group ≥3) were ruled out with a negative predictive value of 96% in validation. CONCLUSION: This study demonstrates that non-invasive, high-accuracy prostate cancer tests can be developed using semen samples collected at home. The proposed test could potentially eliminate 60-70% of unnecessary biopsies, representing a substantial improvement in prostate cancer testing and risk stratification.
Whitney et al. (Wed,) studied this question.