Multi-gene risk score for prediction of clinical outcomes in treatment-naïve metastatic castrate-resistant prostate cancer.

Background: To determine the performance of a multi-gene copy number variation (MG-CNV) risk score in metastatic tissue and plasma biospecimens from treatment-naïve metastatic castration-resistant prostate cancer (mCRPC) patients for prediction of clinical outcomes.

Methods: The mCRPC tissue and plasma cell-free DNA (cfDNA) biospecimen sequencing results obtained from publicly accessed cohorts in dbGaP, cBioPortal, and an institutional mCRPC cohort were used to develop a MG-CNV risk score derived from gains in AR, MYC, COL22A1, PIK3CA, PIK3CB, NOTCH1 and losses in TMPRSS2, NCOR1, ZBTB16, TP53, NKX3-1 in independent cohorts for determining overall survival (OS), progression-free survival (PFS) to first-line androgen receptor pathway inhibitors (ARPIs). The range of the risk scores for each cohort was dichotomized into "high-risk" and "low-risk" groups and association with OS/PFS determined. Univariate and multivariable Cox proportional hazards regressions were applied for survival analyses (P < .05 for statistical significance).

Results: Of 1137 metastatic tissue-plasma biospecimens across all cohorts, 699/1137 were treatment-naive mCRPC (235/699 metastatic tissue; 464/699 plasma-cfDNA), and 311/1137 were matched tissue-cfDNA pairs. In multivariable analysis, the MG-CNV risk score derived from metastatic tissue or in cfDNA was statistically significantly associated with OS with high score associated with short survival (hazard ratio = 2.65, confidence interval = 1.99 to 3.51; P = 1.35-11) and shorter PFS to ARPIs (median PFS of 7.8 months) compared with 14 months in patients with low-risk score.

Conclusions: A molecular risk score in treatment-naïve mCRPC state obtained either in metastatic tissue or cfDNA predicts clinical survival outcomes and offers a tumor biology-based tool to design biomarker-based enrichment clinical trials.