Improving Consistency in Colorectal Cancer Genetic Risk Assessment and Return of Results Through an Integrated Polygenic Risk Score

Abstract

Background: Published colorectal cancer (CRC) polygenic risk scores (PRS) may demonstrate similar population-level predictive performance but often assign substantially different risk classifications to the same individuals. Such discordance presents challenges for clinical implementation and return of results (RoR). Methods: We developed a stable PRS for CRC (stPRS-CRC) by integrating 28 published PRS using PRSmix in the All of Us (AoU) development cohort (n=321,703). Performance was evaluated in a held-out AoU validation cohort (n=80,423) and independently validated in the Genomic Health Initiative (GHI) (n=17,545). CRC associations were assessed using logistic regression, comparing individuals in the highest PRS decile with the remaining 90%. Individual-level concordance in high-risk classification was evaluated using mean Jaccard indices. Results: Published CRC PRS showed substantial discordance in individual-level high-risk classification despite generally consistent associations with CRC. stPRS-CRC integrated information from multiple published PRS and demonstrated CRC associations comparable to PGS003852, the strongest-performing individual PRS. The odds ratios for stPRS-CRC were 1.94 (95% CI, 1.79-2.11) in AoU development, 2.08 (1.76-2.45) in AoU validation, and 2.08 (1.55-2.79) in GHI. Mean Jaccard indices using stPRS-CRC versus PGS003852 as anchors, each compared with the same 27 published PRS, were 0.29 versus 0.20, 0.43 versus 0.25, and 0.27 versus 0.19, respectively (all P<0.001). Conclusions: Integrating multiple published CRC PRS significantly improved individual-level concordance in high-risk classification while maintaining population-level predictive performance. stPRS-CRC may provide a more consistent framework for clinical genetic risk assessment and ROR.