Revisiting the association between environmental tobacco smoke exposure and lung cancer risk - IV. Investigating heterogeneity between studies

There is highly significant (p < 0.001) heterogeneity between estimates from 47 studies of the lung cancer risk in lifelong non-smoking women associated with each 10 cigarettes per day smoked by the husband or with whether or not the husband smokes. Sources of this heterogeneity were investigated by relating risk estimates to 15 different study characteristics using multiple logistic regression analysis. There was a marked tendency for risk estimates to be higher in the I I studies that had not adjusted for age, and there is a strong case for removing such studies from meta-analyses as being of unacceptable quality. Studies which report close-response results for smoking by the husband also report higher risks, demonstrating that estimates of the risk per 10 cigarettes per day smoked by the husband based only on those studies will be biased upward. There is also evidence that risk estimates tend to be lower in large studies, in studies published in the 1990s, in studies not requiring histological confirmation of all cases and in studies where the proportion of proxy respondents was no higher in cases than in controls, though these associations are not independent. Variation in risk by study characteristics largely explains the apparently low relative risk reported in one large Chinese study, arguing against the view that it is an outlier and should be excluded from meta-analyses. Copyright (C) 2002 S. Karger AG, Basel.