Discriminative accuracy of genomic profiling comparing multiplicative and additive risk models

R Moonesinghe, MJ Khoury, T Liu, Cecile Janssens

Research output: Contribution to journalArticleAcademicpeer-review

10 Citations (Scopus)

Abstract

Genetic prediction of common diseases is based on testing multiple genetic variants with weak effect sizes. Standard logistic regression and Cox Proportional Hazard models that assess the combined effect of multiple variants on disease risk assume multiplicative joint effects of the variants, but this assumption may not be correct. The risk model chosen may affect the predictive accuracy of genomic profiling. We investigated the discriminative accuracy of genomic profiling by comparing additive and multiplicative risk models. We examined genomic profiles of 40 variants with genotype frequencies varying from 0.1 to 0.4 and relative risks varying from 1.1 to 1.5 in separate scenarios assuming a disease risk of 10%. The discriminative accuracy was evaluated by the area under the receiver operating characteristic curve. Predicted risks were more extreme at the lower and higher risks for the multiplicative risk model compared with the additive model. The discriminative accuracy was consistently higher for multiplicative risk models than for additive risk models. The differences in discriminative accuracy were negligible when the effect sizes were small (<1.2), but were substantial when risk genotypes were common or when they had stronger effects. Unraveling the exact mode of biological interaction is important when effect sizes of genetic variants are moderate at the least, to prevent the incorrect estimation of risks. European Journal of Human Genetics (2011) 19, 180-185; doi:10.1038/ejhg.2010.165; published online 17 November 2010
Original languageUndefined/Unknown
Pages (from-to)180-185
Number of pages6
JournalEuropean Journal of Human Genetics
Volume19
Issue number2
DOIs
Publication statusPublished - 2011

Research programs

  • EMC NIHES-01-64-03
  • EMC NIHES-02-65-01

Cite this