ENVIRONMENTAL TOBACCO-SMOKE AND LUNG-CANCER - A CRITIQUE OF A METAANALYSIS

Wald et al. (1986) conclude from a meta-analysis of ten case-control studies and three prospective studies that there is a significantly high odds ratio between spousal smoking and lung cancer in nonsmokers. Their reported 95% confidence limits for their estimated global odds ratio are 1.19 to 1.54. A re-analysis of the original reports revealed the heterogeneity of these studies. Because of different protocols they are not suitable for combination in a meta-analysis. As an example of this heterogeneity, the original authors use different classifications of exposure, forcing Wald et al. (1986) to degrade reported ETS exposure to a simple yes/no dichotomy. To demonstrate the subjective nature of a meta-analysis on these data, Wald's analysis was repeated, using GLIM with a normal error structure, and modified. The re-analysis initially gave the same results as those reported in the 1986 paper in the British Medical Journal. However, it was found that adjustment for country of origin (principally the USA versus the rest of the World) gives global 95% confidence limits of 0.96 to 1.41, thus negating Wald's conclusion. Similarly, the addition of one large US case-control study, not included by the original authors, gives a result which is just non-significant (95% confidence limits of 0.99 to 1.19). A recent and more general paper (Fleiss & Gross, 1991) considers the same meta-analysis (as presented by the National Research Council's 1986 Report on ETS). By restricting attention to US studies they find that the overall odds ratio for the 9 US studies has a 95% confidence interval of 0.95 to 1.30. As a result of their analysis, Fleiss and Gross conclude that there is insufficient evidence to claim an association between ETS and lung cancer in the United States.