Table 1.
Methodological Issue | Suggested Solution | Reference (Lead author |
---|---|---|
Modelling the G × E cross-product terms | Include an additional coefficient to model non-linear genetic effects (β 4G2), and a second to account for non-linear interaction effects (β 5G2 × E) | Aliev, Bavav Genet, 2014 |
Comparing biological frameworks (e.g. diathesis-stress model vs. differential susceptibility framework) | Adjust the parameters in the regression equation to compare alternate theoretical frameworks | Belsky, Psychol Bull, 2009 Widamen, Psychol Methods, 2012 |
Selection of interaction scale (e.g. additive vs. multiplicative) | Consider the application of the interaction test a priori. Additive scales have been recommended for identifying heterogeneous effects across subgroups in public health settings, while multiplicative scales are suggested for studying disease etiology | Ottman, Prev Med, 1996 |
Confounding of the G × E interaction term | Include all covariate × gene and covariate × environment interaction terms | Keller, Biol Psychiatry, 2014 |
Shared heritability between the outcome and covariates | Avoid the inclusion of heritable covariates that are associated with the gene variant being tested | Aschard, Am J Hum Genet, 2015 |
Correlation between the gene variant under study and the interacting environmental factor | Directly analyze the relationship between the interacting gene variant and environmental exposure to ensure that they are not correlated | VanderWeele, Am J Epidemiol, 2013 |
Variations in gene expression/silencing, and changing the heritability of BMI throughout development | Use a repeated measures design or include a G × E × Time term if the sample size is sufficient | Liu, Environ Health, 2012 |
Changing heritability of BMI throughout development | Use existing gene × age interactions to identify variants with differential effects across the lifespan | Elks, Front Endocrinol, 2012 Winkler, PLoS Genet, 2015 |
Measurement error associated with the environmental exposure and outcome | Consider more accurate measurement tools or repeated measures in favour of large sample sizes with less accurate measures | Wong, Int J Epidemiol, 2003 |