Abstract LB222: Performance of polygenic risk scores for GI cancer prediction in an academic biobank

Background: Genome wide association studies (GWAS) have identified hundreds of common, low risk genetic variants significantly associated with a number of gastrointestinal cancers, but the predictive ability of individual variants or a polygenic risk score (PRS) derived from multiple variants is still unclear especially among minorities. Specific Aims: To evaluate the discriminatory ability of PRS models to differentiate cancer cases vs. controls in patients of European (EUR) and African (AFR) ancestry in an academic biobank. Methods: We identified 445 cases of esophageal, colon, and pancreatic cancers and 12,565 cancer-free controls. Genome-wide significant variants were selected for each of the cancers, and Plink 1.9 was used to generate a PRS, an effect size weighted sum of specific cancer associated alleles for each population. To determine the discriminatory ability of PRS, we performed multivariate logistic regressions in R controlling for age, sex, and the first 10 principal components. Results: There were 285 cases of colon (48.4% AFR), 63 cases of esophageal (34.9% AFR), and 97 cases of pancreatic cancer (51.5% AFR) vs. 12,565 controls (48.5% AFR). Among the EUR individuals, the PRScolon was significantly associated with colon cancer [OR 1.25 (1.06-4.48, p=0.007)] (Table 1). The discriminatory ability of the model comprised of age, gender and principal components was 0.680-0.732 in the respective cancer cancers and the AUC minimally increased to 0.688-0.747 after inclusion of the PRS in the model. Among AFR individuals, the discriminatory ability was overall higher in the full model (AUC 0.755-0.812) but PRS increased the AUC less in AFR vs. EUR. Conclusion: Colon, esophageal, and pancreatic cancer PRS models have a moderate discriminatory ability to identify cases. However, the individual contribution of PRS to the model was small. Further studies are needed to determine additional genetic predictors of cancer risk and how best to incorporate PRS into gastrointestinal cancer risk prediction models. Citation Format: Louise Wang, Heena Desai, Anh Le, Ryan Hausler, Shefali Verma, Anurag Verma, Renae Judy, Abigail Doucette, Peter Gabriel, Scott Damrauer, Marylyn Ritchie, Daniel Rader, Rachel Kember, Kara Maxwell. Performance of polygenic risk scores for GI cancer prediction in an academic biobank [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr LB222.