Table 2.

The power (%) of detecting a non-zero GxE interaction or joint exposure effect using different sampling designs and likelihood approaches. Results are based on 1000 replicates, each including a cohort of 1000 subjects from which 250 are selected for retrospective genotyping. Unbalanced longitudinal outcome is considered, with a monotone missing pattern of 10% random dropouts at each follow-up visit and up to 5 measurements for each subject. Genotype assumed to be independent of personal exposure, with a prevalence of 0.2. β_E = −1.5 and β_GE = −1.5.

Sampling scheme	Sampling variable	UUL		IPWL		CCL		FCL
		βGE	βE + βGE	βGE	βE + βGE	βGE	βE + βGE	βGE	βE + βGE
Random	-	22	70	-	-	-	-	-	-
E-enriched	Ei	33	89	-	-	-	-	35	94
OLS-based	η̂0i	2	80	56	81	25	90	34	92
E + OLS	(Ei, η̂0i)	7	42	34	75	29	97	46	98
BLUP-based	â0i	18	69	55	84	32	95	33	95
E + BLUP	(Ei, â0i)	2	25	36	77	41	95	48	96