Precision Prostate Cancer Screening with a Polygenic Risk Score

Prostate cancer (PC) is the second-most common type of cancer and the fifth-leading cause of cancer-related death in men worldwide. Genome-wide association studies have identified numerous genetic variants (SNPs) independently associated with PC. The effects of such SNPs can be combined into a single polygenic risk score (PRS). Stratification of men according to PRS could be applied in secondary prevention. We aimed to construct a PRS model and to develop a pipeline for personalized prostate cancer screening. Previously published PRS models for predicting the risk of prostate cancer were collected from the literature. These were validated on the Estonian Biobank (EGC) consisting of a total of 16,390 quality-controlled genotypes with 262 prevalent and 428 incident PC cases and on 209 634 samples in the UK Biobank with 3254 prevalent cases and 6959 incident cases. The best performing model was selected based on the AUC in prevalent data and independently validated in both incident datasets. Using Estonian PC background information, we performed absolute risk simulations and developed individual risk-based clinical follow-up recommendations. The best-performing PRS included 121 SNPs. The C-index of the Cox regression model associating PC status with PRS was 0.641 (SE = 0.015) with a hazard ratio of 1.65 (95% confidence interval 1.51 - 1.81) on the incident EGC dataset. The model is able to identify individuals with more than a 3-fold risk increase. The risk of an average 45-year old could be attained by individuals between the ages of 41 and 52. A 41-year old male on the 95th percentile has the same risk as an average 45-year old but by age 55, he has attained the same genetic risk as an average 68-year-old. PRS is a powerful predictor of prostate cancer risk that can be combined with current non-invasive practices of PC screening. We have developed PRS-based recommendations for personalized PSA testing. Our approach is easily adaptable to other nationalities by using population-specific background data of other genetically similar populations.