In order for the analyses to be consistent, the complete set of confounders was used for all five outcomes.
The final set of confounders was selected for all potential confounders that change odds ratios (OR) for any of the low lung function indicators by more than at least 10% in one country (Table 1).
The same set of confounders was considered for possible inclusion in the cross-sectional analyses of SCr.
The set of confounders was the same for each familial sero-marker, and consisted of: year of birth, total sibship size (as a proxy for parental socioeconomic status), and the other familial HBV sero-status measures.
Finally, a large set of confounders was available to control for, including dietary and lifestyle variables, and mis-reporting of energy intake.
We analyzed each ECG parameter in separate models adjusted for a set of confounders that minimized Akaike's information criterion (Akaike 1973).
The second model included all the other physical or psychosocial variables at the same time, still adjusting for the set of confounders.
First, all physical and psychosocial work environment factors were entered into the regression model one at a time adjusting only for the set of confounders (Model 1).
Adjustment for a set of confounders besides sex and age (model A) added little to the change in PR already observed on adjustment for age and sex alone (Table 2).
