Gu et al. [2009] used a type of latent components (factor) analysis called supervised statistical learning approach for multivariate analysis (SLAM) for longitudinal data and generalized estimating equations to account for familial correlation. They used data from the Offspring Cohort at Exams 1, 3, 5, and 7. The primary phenotype of interest was coronary heart disease (CHD) status, and the data on ten variables including CHD endophenotypes (body mass index, total cholesterol, high-density lipoprotein cholesterol, triglycerides, systolic blood pressure (SBP), diastolic blood pressures, and fasting glucose) and environmental covariates (age at visit, cigarette smoking, and alcohol use) were used for the latent-component analysis. They identified several genes, including two well known CHD candidate genes (SCNN1B and PKP2) with potential time-dependent G×E interactions, and several others including a novel cardiac-specific kinase gene (TNNI3K), with potential G×E interactions independent of time and marginal effects.
