Table 2. Summary of the variables used as predictors of BMI in each of the models.
| Model | BMI | Scores | Activity | Diet |
|---|---|---|---|---|
| Age and sex (sex, age × sex, age2, age2 × sex, age3, age3 × sex) | ||||
| East co-ordinate | — | — | — | |
| East co-ordinate × age | — | — | — | |
| East co-ordinate × sex | — | — | — | |
| FTO (rs1421085) | — | |||
| Activity variables (‘Number of days/week walk 10+ minutes', ‘Number of days/week moderate physical activity 10+ minutes', ‘Number of days/week vigorous physical activity 10+ minutes', and their interactions with age and sex) | — | — | ||
| Activity variables × FTO | — | — | — | |
| Activity score | — | — | ||
| Activity score × FTO | — | — | ||
| Diet variables (‘cooked vegetable intake', ‘non-oily fish intake', ‘oily fish intake', ‘processed meat intake', ‘poultry intake', ‘beef intake', ‘lamb/mutton intake', ‘pork intake', ‘cheese intake', ‘bread intake', ‘tea intake' and ‘frequency of added salt') | — | — | ||
| Diet variables × FTO | — | — | — | |
| Diet score | — | — | ||
| Diet score × FTO | — | — | ||
| Other variables (‘age','alcohol intake frequency',‘sleep duration','sleep duration2', ‘current regular smoker (yes/no)',‘Townsend deprivation index', ‘Hours watch TV', and their interactions with age and sex) | ||||
| Other variables × FTO | ||||
| Genotyping array | — | |||
| Genotyping array × FTO | — |
BMI, body mass index; CI, confidence interval; FTO, fat mass and obesity associated.
An ‘ × ' between two variables indicates an interaction effect. The ‘BMI' model is the model chosen by the cross-validation procedure in the non-genotyped sample (see Methods), and the ‘Scores' Model uses the coefficients fitted in the ‘BMI' model to construct the activity and diet scores. The ‘Activity' and ‘Diet' Models each have their relevant score variable replaced with the constituent variables of the score: ‘Activity score' replaced with ‘Activity variables' and so on. Note that to adjust for population structure in the models fitted in the genotyped samples, we added principal components in the British Sample, and we added random effects in a mixed model in the Diverse Sample (Methods).