Abstract
BACKGROUND: Prostate cancer (PCa) is the most common solid organ cancer in men and the fifth leading cause of cancer-related deaths globally. PSA helps identify men at risk but has low specificity and has resulted in unnecessary biopsies. The PROSTest, a novel machine learning-based 27-gene mRNA liquid biopsy assay, was developed to detect PCa. We evaluated its utility as a stratification tool in symptomatic men undergoing biopsy or surgery for PSA > 2 ng/mL. METHODS: Of 123 men assessed, 105 (85%) met eligibility criteria (age > 55 years, PSA > 2 ng/mL, symptomatic) and underwent image-guided biopsy or surgery. Blood samples for PROSTest were collected prebiopsy, and RNA-stabilized samples underwent RNA isolation and cDNA production. PCR results were analyzed using a machine learning algorithm, generating a 0-100 score with a cutoff of 50 for a binary (positive/negative) readout. The performance of PROSTest was against PSA using AUROC and evaluated for Gleason Grade (GG) 1 versus GG2-5 patients. RESULTS: Median age was 68 years (55-86 years); median PSA was 8.2 ng/mL (IQR: 7.2-92 ng/mL). PCa was diagnosed in 65 men (62%) (27 GG1; 38 GG2-5). PROSTest was positive in 63/65 (97%) of those with PCa and in 2/40 (5%) without PCa. Sensitivity was 97%, specificity 96%. PROSTest outperformed PSA (AUROC: 0.99 vs. 0.61, p < 0.0001). GG2-5 had significantly higher (p < 0.0001) PROSTest scores (92 ± 3). CONCLUSIONS: PROSTest demonstrated superior sensitivity and specificity compared to PSA for detecting prostate cancer across all Gleason grades. Additionally, it showed potential for distinguishing GG2-5 from GG1 + BPH, which could help guide clinical decision-making and reduce unnecessary biopsies. By leveraging a machine learning-based approach, PROSTest may offer a more accurate and less invasive diagnostic tool for prostate cancer stratification. However, larger prospective studies are needed to validate these findings and further define its clinical utility.