Table 1.
Symbol allocation for quality characteristics of the G×E studies
| Method | Characteristic | − | −+ | + | Not applicable |
|---|---|---|---|---|---|
| All | Study type | Correlational | Case control | Randomized | |
| Sample size | < 1000 | 1000–2500 | > 2500 | ||
| Power calculation | No | – | Yes | ||
| Control for age and sexa | None | Descriptive | Statistical | Homogenous sample/age as predictor or outcome | |
| Control for ethnicityb | None | Descriptive | Statistical | Homogenous sample | |
| Control for rGEc | None | Descriptive | Statistical | Interventions/cohort effects | |
| Phenotype measures | Self-developed short survey | Validated survey/interview | Biological/combined measures | Interventions/cohort effects | |
| Haplotype | # of blocksd | 1–4 | – | > 4 | |
| # of genesd | 1–3 | – | > 3 | ||
| # of variantsd | < 5 | 5–10 | > 10 | ||
| Rationale for risk haplotypee | Debatable | – | Solid | ||
| Candidate | # of genesd | 1–3 | – | > 3 | |
| # of variantsd | < 5 | 5–10 | > 10 | ||
| Rationale for risk allele | Debatable | – | Solid | ||
| Polygenic score (PS) | Based on | Overlapping sample GWAS | – | Independent GWAS | |
| Discovery sample size | < 10,000 | 10,000–25,000 | > 25,000 | ||
| p value thresholdf | p < .0001 | – | p ≥ .0001 | ||
| Correspondence phenotypesg | Weak | Moderate | Strong |
aGenetic associations may vary in different age and sex groups (Kendler et al. 2008; The Wellcome Trust Case Control Consortium 2007)
bPopulation stratification resulting from ancestry differences can distort genetic association results (Price et al. 2006); statistical control using principal component analysis is preferable to control for these effects
cIn gene-environment correlation (rGE) genetic make-up influences to what environment an individual is exposed (only possible in non-randomized studies). These effects can muddle G×E findings (Rathouz et al. 2008a, b)
dInclusion of more genetic factors in the aggregate predictor was considered better. Cut-offs were based on commonly chosen numbers of variants for these studies
eThe rationale for defining which haplotype or allele was the active (risk/protective) allele was deemed less strong when it was based on the results of the main analyses in the same sample, rather than on theory or results from independent samples
fThis threshold most commonly concerns the p value for the association between the SNPs and the phenotype in the original GWAS. The lower this value, the fewer SNPs are included in the PS. We considered PS including only a few SNPs as less strong than PS including more SNPs, although the exact optimal threshold depends on several other study characteristics (Chatterjee et al. 2013; Dudbridge 2013)
gThe more similar the outcome variable is to the original GWAS phenotype on which the PS was based, the better the predictive value (Wray et al. 2014)