Introduction: The MyProstateScore (MPS) test employs the long non-coding RNA PCA3 and the TMPRSS2:ERG (T2:ERG) gene fusion to detect clinically-significant (Grade Group [GG]=2) prostate cancer (PCa). To build upon MPS, we analyzed RNA-seq data from a PCa compendium, identifying 50 additional cancer- and high-grade cancer-specific markers. In total, 54 transcripts were measured by multiplex qPCR in 1504 urine biospecimens. The current study sought to develop and validate a novel urinary test for improved detection of GG=2 PCa. Methods: The 54-marker qPCR panel was assessed in a development cohort of men undergoing biopsy at the University of Michigan. The development cohort was divided into four subsets over ten iterations for glmnet modeling, identifying 17 markers independently contributing to discriminative accuracy. Acknowledging varying availability of patient-level clinical data (c) and prostate volume (v) in clinical practice, three novel MyProstateScore 2.0 models were developed: i) markers-only (MPS2), ii) markers and clinical data (MPS2c), and iii) markers, clinical data, and prostate volume (MPS2cv). The locked MPS2 models were provided to external analysts for validation in a blinded, multi-institutional NCI-EDRN biopsy cohort. Results: Model development included 761 men with PSA 3-10 ng/ml, of whom 293 (39%) were found to have GG=2 cancer on biopsy. The existing MPS test provided an area under the receiver operating characteristic curve (AUC) of 0.73, while the MPS2, MPS2c, and MPS2cv models yielded cross-validated AUC values of 0.78, 0.80, and 0.82, respectively. In the external validation cohort of 743 men with PSA 3-10 ng/ml, of which 151 (20%) were found to have GG=2 cancer, the existing MPS test yielded an AUC of 0.73. The MPS2, MPS2c, and MPS2cv models yielded AUC values of 0.75, 0.81, and 0.82, respectively. At test thresholds providing 95% sensitivity, MPS provided 17% specificity, while the novel MPS2cv model provided 38% specificity. Conclusions: Incorporating novel transcripts associated with high-grade cancer, we externally validated a 17-marker urinary panel for detection of GG=2 PCa. Compared to the clinically-available MPS test, the MPS2cv model improved diagnostic accuracy (AUC) by nearly 10% and increased specificity by 21% at clinically-actionable, highly-sensitive thresholds. These findings suggest that the MPS2 test meaningfully improves detection of GG=2 PCa relative to current diagnostic tests and could contribute to a more optimal contemporary diagnostic pathway. SOURCE OF Funding: This work was funded by the Prostate Cancer Foundation Young Investigator Award (20YOUN11), the Early Detection Research Network (U01CA214170, U2CCA271854), the NCI Prostate SPORE (P50CA186786), and an NCI Outstanding Investigator Award (R35CA231996). We would like to acknowledge the EDRN PCA3 cohort study investigators and coordinators for their contribution to this work.
