Breast Cancer Polygenic Risk Score Validation and Effects of Variable Imputation

Jeffrey J. Beck, John L. Slunecka, Brandon N. Johnson, Austin J. Van Asselt, Casey T. Finnicum, Cheryl Ageton, Amy Krie, Heidi Nickles, Kenneth Cowan, Jessica Maxwell, Dorret I. Boomsma, Eco de Geus, Erik A. Ehli, Jouke Jan Hottenga

Research output: Contribution to journalArticlepeer-review


Breast cancer (BC) is a complex disease affecting one in eight women in the USA. Advances in population genomics have led to the development of polygenic risk scores (PRSs) with the potential to augment current risk models, but replication is often limited. We evaluated 2 robust PRSs with 313 and 3820 SNPs and the effects of multiple genotype imputation replications in BC cases and control populations. Biological samples from BC cases and cancer-free controls were drawn from three European ancestry cohorts. Genotyping on the Illumina Global Screening Array was followed by stringent quality control measures and 20 genotype imputation replications. A total of 468 unrelated cases and 4337 controls were scored, revealing significant differences in mean PRS percentiles between cases and controls (p < 0.001) for both SNP sets (313-SNP PRS: 52.81 and 48.07; 3820-SNP PRS: 55.45 and 49.81), with receiver operating characteristic curve analysis showing area under the curve values of 0.596 and 0.603 for the 313-SNP and 3820-SNP PRS, respectively. PRS fluctuations (from ~2–3% up to 9%) emerged across imputation iterations. Our study robustly reaffirms the predictive capacity of PRSs for BC by replicating their performance in an independent BC population and showcases the need to average imputed scores for reliable outcomes.

Original languageEnglish (US)
Article number1578
Issue number8
StatePublished - Apr 2024


  • GWAS
  • breast cancer
  • genetic risk
  • genotype imputation
  • polygenic risk score
  • validation

ASJC Scopus subject areas

  • Oncology
  • Cancer Research


Dive into the research topics of 'Breast Cancer Polygenic Risk Score Validation and Effects of Variable Imputation'. Together they form a unique fingerprint.

Cite this