Longitudinal data methods for evaluating genome-by-epigenome interactions in families

Area of Science:

• Genetics
• Epigenetics
• Biostatistics

Background:

• Longitudinal measurement is vital in health research for understanding disease and trait progression.
• It enables effective analysis of correlated responses within clustered data, such as family studies.
• Evaluating methods for genome-by-epigenome interactions with longitudinal outcomes is essential.

Purpose of the Study:

• To assess and compare three statistical methods for analyzing genome-by-epigenome interactions.
• To investigate the performance of these methods when applied to longitudinal health data from family studies.
• To determine the most effective approach for handling correlated longitudinal outcomes in genetic and epigenetic research.

Main Methods:

• Employed linear mixed-effect models, generalized estimating equations, and quadratic inference functions.
• Utilized 200 simulated posttreatment replicates to test pharmacoepigenetic effects.
• Compared the impact of adjusting for baseline outcome versus using pre-to-post change scores.

Main Results:

• Adjustment for baseline outcome significantly enhanced statistical power and Type I error control compared to pre-to-post change scores.
• All three modeling approaches showed similar statistical power.
• Marginal models required bias correction for accurate analysis.

Conclusions:

• Linear mixed-effect models, generalized estimating equations, and quadratic inference functions are viable for analyzing genome-by-epigenome interactions with longitudinal data.
• Bias correction is necessary for marginal models.
• Quadratic inference functions showed a slight decrease in power compared to the other two methods, but overall performance was comparable.