Revisiting the Association between Environmental Tobacco Smoke Exposure and Lung Cancer Risk: II. Adjustment for the Potential Confounding Effects of Fruit, Vegetables, Dietary Fat and Education

Analyses are described estimating the extent to which confounding by dietary variables and education might bias the reported association of smoking by the husband with lung cancer risk in non-smoking women. For each of the four variables fruit, vegetable and dietary fat con sumption and education, independent estimates of their relationships with lung cancer risk in non-smokers were combined by meta-analysis. The risks were estimated as 0.83 per standard deviation [SD, 95% confidence interval (CI) = 0.74-0.93, based on n = 9 studies] for fruit con sumption, as 0.84 per SD (Cl = 0.75-0.93, n = 11) for vege table consumption, as 1.17 per SD (Cl = 1.08-1.26, n = 7) for dietary fat consumption and as 0.92 per year (Cl = 0.89-0.96, n = 17) for education. For each of the same four variables independent estimates of the difference between non-smokers exposed or unexposed to envi ronmental tobacco smoke at home were also combined. Using unweighted means, the differences were esti mated as -0.1752 SDs for fruit, -0.0460 SDs for vegeta bles, +0.1750 SDs for dietary fat and -0.5174 years for education (n = 11, 17, 14 and 15). Using weighted means the differences were -0.0986 SDs for fruit, -0.0387 SDs for vegetables, -0.0464 SDs for dietary fat and -0.1023 for education. The weighted estimates are dominated by one very large study whose results, particularly for dietary fat and education, differed from those seen in the other studies. Using these data and methodology devel oped to account for confounding by multiple correlated variables, simultaneous adjustment for fruit, vegetables, dietary fat and education is estimated to reduce the increase in lung cancer risk associated with each 10 ciga rettes/day smoked by the husband from 10% (95% Cl 5- 15%) to 6% (1-11%) using unweighted means or to 9% (5-14%) using weighted means. Uncertainties in the esti mation procedure are discussed. Other sources of bias will be considered in later papers in this series.