Abstract
BACKGROUND: Prostate-specific antigen (PSA) remains the standard biomarker for prostate cancer (PCa) detection, but its limited specificity-particularly in the 3-10 ng/mL range-leads to overdiagnosis and unnecessary biopsies. Including multiparameteric MRI (mpMRI) helps to reduce the number of PSA-false-positive biopsies, but there is still a need for noninvasive diagnostics that improve risk stratification and reduce unnecessary interventions. METHODS: PROSTest is a peripheral blood-based assay that quantifies a 27-gene signature in the androgen receptor (AR) signaling pathway in addition to 3 housekeeping genes (HKGs). PCR results are fed into a proprietary machine learning (ML) algorithm to produce a numerical score on a scale of 0-100 with a clinically validated cutoff of 50 for a final binary readout; positive or negative for likelihood of cancer on biopsy. In this report, the diagnostic performance of PROSTest was evaluated in 1,894 male subjects including 970 individuals with actionable results (PSA ≥ 3.0 ng/mL). We focused on two PSA-graded intended-use populations: subjects aged ≥ 45 years with PSA 3-10 ng/mL (n = 467), and those with PSA > 10 ng/mL (n = 503). PSA and PROSTest were conducted on all subjects. RESULTS: In the 970 cohort, adding the PROSTest achieved an AUC of 0.96 and was significantly more accurate than PSA alone for differentiating PCa from benign prostatic disease (Chi(2) = 134.1, p < 0.0001). In the 467 subjects with PSA 3-10 ng/mL, the PROSTest achieved an AUC of 0.94, with 94% sensitivity and 91% specificity. The PPV for PROSTest was 91.6% (95%CI: 88.3-94.1%) and NPV was 95.6% (95%CI: 91.6-97.7%). The blood-based gene expression profiling correctly identified 435 of 467 subjects (93.1%) and was significantly more accurate than PSA alone where only 271 of 467 (58.0%) with high PSA had PCa (Chi(2) = 155.9, p < 0.0001). The sensitivity of the assay for detecting PCa was 97.0% (263/271). In the 503 subjects with PSA > 10 ng/mL, PROSTest yielded an AUC of 0.93 versus 0.76 for PSA (z = 5.3, p < 0.0001). Despite the very high levels of PSA ( > 10 g/mL), 63 (20 BPH and 43 non-BPH controls) out of the 503 (12.5%) subjects were negative for PCa. PROSTest sensitivity was 93.6% (412/440) and the accuracy was 92.8% (467/503). The PPV for PROSTest was 98.1% (95%CI: 96.4-99.0%) and NPV was 66.3% (95%CI: 57.6-74.0%). If PROSTest was used, it would have precluded 55 of the 63 (87.5%) PSA-falsely driven biopsies. CONCLUSIONS: PROSTest demonstrates improved stratification value relative to PSA and could significantly reduce PSA-driven false positive biopsies. Out of 259 non-PCa subjects biopsied based on high PSA levels, applying PROSTest could potentially eliminate 227 biopsies (87.6%). PROSTest superior NPV was not confounded as a tradeoff for the PPV as PROSTest exhibited a sensitivity of ~95% (675/711 PCa detected).