The association between smoking during pregnancy and mortality from the third trimester to age 1 year is presented as odds ratios with 95 percent confidence intervals. Other variables accounted for in the analyses were categorized as seen in table 1. These variables were evaluated in logistic regression analyses in two ways. They were included one at a time, and if they changed the measure of association between smoking and mortality by more than 10 percent, they remained in the final model (28). Subsequently, all of the covariates in table 1 were included at the same time based on a priori information that they might potentially confound the results. All covariates were entered as a number of dummy variables equal to the number of categries minus one. Effect modification was evaluated by stratified analyses and by logistic regression with interaction terms. The association between smoking and infant mortality was also evaluated by Cox regression analyses. However, because the results were similar to those from logistic regression analyses, they are not presented. The
attributable fraction was calculated from the formula: attributable fraction = proportion of exposed in the popu- lation × (odds ratio – 1)/proportion of exposed in the pop- ulation × (odds ratio – 1) + 1 (29).