Table 1.

Comparison between methods that are described in the text that can be used to identify GxE. Sub-approaches are presented with over-arching approach that they draw from. Methods are presented with use case situations where they are generally considered useful.

Approach	Sub-Approaches	Use Cases	Citation
FAMILY-BASED	generalized estimating equations	Pedigree data is available and exposure mis-specification is a concern	Basson et al. (2016) [161], Sitlani et al. (2016) [162]
	heirarchical linear model	Pedigree data is available and type I error is a concern
	linear mixed effects model	Pedigree data is available and type I error is a concern
CASE-CONTROL	Penalized method with least absolute deviation loss function	When large genome-wide data is available and hierarchical “main effects, interactions” structure is a concern	Wu et al. (2018) [165]
	Similarity-based regression	When large genome-wide data is available and rare-variants with binary phenotypes are being investigated	Zhao et al. (2015) [124]
	linear mixed model	When large genome-wide data is available and multiple exposure are being investigated	BIOS Consortium (2016) [166]
	Parametric bootstrap	Removes need for permutation tests when large genome-wide data is available	Gauderman et al. (2017) [159], Coombes et al. (2018) [167]
CASE-ONLY	Traditional	Increases precision when independence between exposure and genetics can be assumed	Piegorsch et al. (1994) [169]
	Multiple maximum-likelihood	Increases precision and relaxes independence assumption	Umbach and Weinberg (1997) [170], Chatterjee and Carrol (2005) [173], Mukherjee and Chatterjee (2008) [171]
	Bayesian
2-STEP	Likelihood ratio to traditional	Increases power and reduces multiple testing correction in situations where traditional case-control or case-case only approaches would be appropriate	Murcray et al. (2008) [171], Pare (2010) [174], Kooperburg and LeBlanc (2008) [175]
	Levene’s test to traditional
	Marginal effects to traditional
	Modified Pare et al.	Robust in situations with with multiple exposure and reduce type I error versus other 2-step approaches	Zhang et al. (2016) [160]
	Combined Pare and Kooperburg
GENE-SET ANALYSIS (GSA)	Traditional	Increases power versus more traditional approaches	Biernacka et al. (2012) [176]
	With similarity regression	GSA when there are multiple covariates and opposite effects that may cancel each other out are a concern	Tzeng et al. (2013) [180]
	GESAT	Established method for user friendly GSA	Lin et al. (2013) [180]
META-ANALYSIS	NA	Situations where investigators want to combine data from multiple studies to identify possible gene-environment interactions	Shi et al. (2017) [168]