Artikel
Quantification of the contribution of genetic variants in association analysis with survival outcome: three methods in comparison
Suche in Medline nach
Autoren
Veröffentlicht: | 6. September 2007 |
---|
Gliederung
Text
Introduction: Quantifying the contribution of genetic variants, i.e. alleles, genotypes, or haplotypes, to disease outcome or to quantitative traits is of great interest in order to interpret the impact of findings from genetic association analyses. In linear regression, the contribution of the genetic variants to the model can be quantified by the proportion of the phenotype’s variance that is explained by the variant, R². However, it is difficult to define a comparable criterion for survival analysis, due to the censored observations. Thus, it is difficult to judge the contribution of genetic variants to survival outcome.
Methods: A variety of answers is currently available for this topic. Our requirements for an appropriate criterion include: (a) limitation to the range [0;1] for interpretation as percentage of variation explained by the variant, (b) robustness against censoring percentage, and (c) values increasing with the associated hazard ratio. In simulation studies with a variety of scenarios (varying genotype frequencies, censoring percentages, and hazard ratios), criteria based on three different approaches have been compared with respect to the above stated requirements: (1) residuals formulated by means of the cumulative hazard, (2) variation of individual survival curves, (3) Schoenfeld residuals which measure the difference of observed and expected covariate values.
Results: The first approach (residuals based on cumulative hazard) was highly dependent on censoring percentage and showed a tendency to systematically exceed the desired range of values. The second approach (variation of survival curves) had a tendency to low values. Our requirements were best fulfilled by the criterion based on Schoenfeld residuals (approach 3). For this criterion, we also show how extended models adjusted for environmental variables or gene-environment interactions can be judged.
Conclusion: We present a powerful tool for judging the contribution of genetic variants as well as gene-environment interactions within the survival context.