Abstract P5-09-05: A model with polygenic risk score and mammographic density predicts interval cancers

Abstract

Abstract Introduction: Interval breast cancers present with clinical symptoms following a normal screening mammogram. They are associated with unfavorable biological features and with dense breasts. Models predictive of aggressive phenotypes may facilitate tailored screening for women at elevated risk of interval cancers. Polygenic risk scores (PRS) represent the cumulative effects of multiple single nucleotide polymorphisms (SNPs) and can be used to risk-stratify women. In prior reports, PRS is preferentially associated with screen-detected rather than interval cancers. We investigated methods to refine the PRS to preferentially predict interval cancers, and tested the performance of the PRS in joint models with mammographic breast density (MBD). Methods: We used data from 1058 breast cancer cases from The Cancer Genome Atlas (TCGA) as the discovery set for our PRS. We selected 107 SNPs from genomewide association studies of breast cancer risk for testing against tumor status at last follow-up in TCGA. Presence of tumor indicated recurrence, progression, or positive margins after resection. Women with tumor present at <100 days of follow-up were excluded. Suggestive associations (p<0.2) were used to construct a PRS, calculated as the sum across all SNPs of the per-allele log-odds ratio multiplied by the number of risk alleles for each SNP. We tested the performance of the PRS in a nested case-control dataset with 471 cases (102 interval cancers, 369 screen detected) and 496 controls from the California Pacific Medical Center Research Institute cohort. Logistic regression was used to evaluate the association between PRS, MBD and interval cancers. Area under the receiver operating characteristic (AUROC) curve was used to measure discrimination. Results: Of 107 SNPs, 23 had suggestive associations with presence of tumor at last follow-up in TCGA. The 23-SNP PRS discriminated between women with interval cancers and controls, with AUROC 0.57 (95% CI 0.51-0.63). With the inclusion of MBD in the model, the AUROC was 0.68 (95% CI 0.62-0.74). Women in the highest PRS quintile had an unadjusted 2.07-fold odds (95% CI 1.05-4.07) of developing interval cancers compared with women in the lowest quintile; adjustment for MBD did not change the point estimate. The PRS also discriminated between women with interval and screen-detected cancers, although the findings did not reach statistical significance (AUROC 0.55, 95% CI 0.48-0.61). With the inclusion of MBD in the model, the AUROC was 0.63 (95% CI 0.57-0.69). Discussion: A PRS associated with presence of tumor at last follow-up was independently predictive of interval cancers relative to controls. Models with PRS and MBD discriminated between interval and screen-detected cancers, although MBD provided most of the predictive power. Our findings are limited by the size and low number of recurrences in TCGA. It is possible that tumor status largely reflects treatment received, and may only partially represent the biological pathways of interval cancers. Our results suggest that SNPs may potentially identify women at risk for developing interval breast cancer, although further validation is required. Citation Format: Shieh Y, Hu D, Huntsman S, Ma L, Gard CC, Leung JWT, Tice JA, Cummings SR, Kerlikowske K, Ziv E. A model with polygenic risk score and mammographic density predicts interval cancers [abstract]. In: Proceedings of the 2016 San Antonio Breast Cancer Symposium; 2016 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2017;77(4 Suppl):Abstract nr P5-09-05.