Comprehensive Analysis of Multiple Cohort Datasets Deciphers the Utility of Germline Single Nucleotide Polymorphisms in Prostate Cancer Diagnosis.

Prostate cancer (PCa) susceptibility is a polygenic trait. We aimed to examine the controversial diagnostic utility of single nucleotide polymorphisms (SNPs) for PCa. We analyzed two datasets collected from Europeans and one from Africans. These datasets were generated by the genome-wide association studies, i.e. CGEMS, BPC3 and MEC-Africans, respectively. ~550,000 SNPs, including 61 risk markers that constitute a panel termed MK-61, were commonly genotyped. For each dataset, we augmented the MK-61 panel to generate a MK-61+ one by adding several thousands of SNPs that were moderately associated with PCa occurrence in external dataset(s). We assessed the diagnostic utility of both panels by measuring their predictive strength for PCa occurrence with AUC statistics. We calculated the theoretical AUCs using quantitative genetics model-based formulae and obtained the empirical estimates via ten-fold cross-validation using statistical and machine learning techniques. For the MK-61 panel, the 95% confidence intervals of the theoretical AUCs (AUC-CI.95) were 0.578-0.655, 0.596-0.656 and 0.539-0.596 in the CGEMS, BPC3 and MEC-Africans cohorts, respectively. For the MK-61+ panels, the corresponding AUC-CI.95 were 0.617-0.663, 0.527-0.736 and 0.547-0.565. The empirical AUCs largely fell within the theoretical interval. A promising result (AUC = 0.703, FNR =0.354, FPR = 0.353) was obtained in the BPC3 cohort when the MK-61+ panel was used. In the CGEMS cohort, the MK-61+ panel complemented PSA in predicting the disease status of PSA >= 2.0 ng/ml samples. This study demonstrates that augmented risk SNP panels can enhance PCa prediction for males of European ancestry, especially those with PSA >= 2.0 ng/ml.