Re: Multicenter case-control study of exposure to environmental tobacco smoke and lung cancer in Europe.

Boffetta et al. (1) claim smoking misclassification only has a minimal effect, of 1.01–1.02, on the lung cancer risk associated with spousal smoking, which they estimate from their multicenter study as 1.16. The bias is probably much larger and could explain most of the excess risk. In 1996, we (2) estimated the effect that misclassification bias had on spousal risk estimates from five U.S. studies of environmental tobacco smoke (ETS) and lung cancer, by use of a version of the complex Wells–Stewart procedure used by the U.S. Environmental Protection Agency (EPA) (3) that had been modified to avoid some mathematical errors. Assuming, as had EPA, that 1.07% of regular, 17.7% of occasional, and 10.8% of former smokers report they are never smokers and that the observed excess risks for misclassified occasional and former smokers are, respectively, 20% and 9% of those for regular smokers, the bias we estimated for all five studies was similar to that using an alternative procedure, which simply assumed the equivalent of 1.75% of ever smokers with average lung cancer risk denied smoking. We also reviewed evidence on misclassification rates in U.S. and Western European populations (4) and found that these rates were higher than EPA had assumed. We argued that, for bias estimation, an equivalent rate (ER) of 2.5% was better than one of 1.75%, but we noted that large between-study variation in misclassification rates suggested that the ER might plausibly lie in the range 1%–4%. Finally, we noted that correcting for bias using a 2.5% ER explains virtually all the lung cancer risk associated with spousal ETS exposure from 13 U.S. studies, reducing the meta-analysis estimate from 1.13 to 1.01. Boffetta et al. (1) assume an actual misclassification rate of 2% and that misclassified smokers have a relative risk of 2.0. Because the relative risk for average ever smoking is about 8–10 and the bias is approximately proportional to the excess risk, their assumptions imply an ER about 0.2%–0.3%, much lower than our estimated 2.5%. They suggest that we ignored the much-lower-thanaverage risk in misclassified smokers, but clearly we have not. They err in using misclassification rates that are far too low. This seems due, in part, to confusing misclassification rates expressed as a proportion of nonsmokers with those expressed as a proportion of smokers, in part to not combining all the sources of misclassified ever smokers and in part to not considering all available evidence on misclassification rates. We also note that misclassification rates of regular smoking are much higher in Asian than in U.S. or Western European women (5), perhaps explaining why spousal risks are elevated in Asian women despite their lower reported smoking frequency (2,5). Misclassification may bias estimates for workplace exposure. We have limited, unpublished data showing that concordance between smoking and working with a smoker is similar to that between smoking and spousal smoking. Although more data are needed, this suggests that the biases for workplace and spousal smoking may be similar.