The application and development of methods to combine and infer information from genetic epidemiological studies of cardiovascular and other complex traits

This thesis investigates methods to combine and infer information from genetic epidemiological studies. Three issues are explored, each in a distinct and self-contained chapter. Chapter 1 investigates how best to incorporate treatment information in genetic analyses of blood pressure. Different approaches to adjusting for treatment are compared in a number of simulated scenarios, and the approaches that utilise all the observed data are generally shown to perform best. One particular condition, however, causes these approaches to suffer bias. This is where a genetic variant (or some other factor) interacts with treatment. This chapter therefore urges caution in the interpretation of results from these studies, and suggests some possible approaches to identifying existing interactions with treatment. Chapter 2 concerns participant privacy in genome-wide association studies (GWAS). Recent methods claim to be able to infer whether an individual participated in a study, using only aggregate statistics from the study such as allele frequencies. In the past, these statistics have been freely published online. This chapter explores the full implications of these methods, by investigating their true capabilities and limitations. In addition, some modifications are proposed to one particular method, to demonstrate how it can be adapted for use in practice. This work finds that participant identification is possible in ideal conditions, but common characteristics of real studies may prevent any reliable application of these methods in practice. Chapter 3 proposes a new approach to synthesising data between studies. This approach – named “DataSHIELD” – guarantees identical results to an individual-level meta-analysis, while offering greater flexibility than the studylevel meta-analysis. DataSHIELD also potentially circumvents some of the laws that restrict data use, because it does not involve sharing any individual-level data between studies. This chapter outlines the principles underpinning DataSHIELD, and demonstrates its use in a simulated data example.