Evidence synthesis for genetic association studies

One explanation for variation in health outcomes, for example why some people get heart disease and others do not, is that genetic variants carried by some proportion of the population increase their risk of disease. However recent intense efforts to find these genetic variants have been only partially successful, apparently because the genetic effects are small and so can only be detected in vary large studies, or by combining or synthesising information across many smaller studies. This evidence synthesis has been very successful in other fields of biomedicine, but application to genetic studies is difficult and new methods are needed. This application will begin to develop such methods.

We will develop methodology for the meta-analysis of genetic association studies. Specifically, we will:

1) Develop methods for the analysis of candidate genes studies where either, or both, the SNP sets and genetic models used may differ between studies. As a motivating example we consider the relationship between plasma levels of C-reactive protein (CRP) and the CRP gene. An extension will consider the integration of existing candidate gene studies with (possibly multiple) whole genome association studies.

2) Assess the possibility of investigating G * E / G*G interaction in a meta-analytic framework. The additional value of individual patient data to detect such interactions will be assessed, and the value of integrating partial IPD with more extensive study level data determined. The latter in particular requires considerable methodological development. As a motivating example we examine the influence of mean serum folate on the association between the MTHFR/C677T polymorphism and homocysteine.

The methods developed will be based on hierarchical models, mainly fitted in the Bayesian framework using MCMC. Throughout, the sensitivity of our results to modelling assumptions will be carefully assessed. Programs developed during the project will be made freely available.