Revisiting evidence on lung cancer and passive smoking: adjustment for publication bias by means of "trim and fill" algorithm.

Meta-analyses are subject to bias because smaller or non-significant studies are less likely to be published, and most meta-analyses do not consider the effect of publication bias on their results. To assess the true risk, we revisited a famous meta-analysis including 37 studies on lung cancer and passive smoking, and adjusted for publication bias by means of the "trim and fill" algorithm. The adjusted relative risk of lung cancer in non-smokers who lived with a smoker from the 44 studies including the 7 filled ones was 1.19 (95% confidence interval 1.08-1.31, p = 0.0004), and the estimate of excess risk fell from 24 to 19%.