Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Professor Eiji Yano raises a number of issues in his letter1 which responded to my commentary2 on his article3 about the Japanese spousal study, as does Chapman in his editorial.4 Here I reply to the main points raised.
Studies of environmental tobacco smoke (ETS) exposure and lung cancer commonly identify a group of self reported non-smoking women and then compare risk according to the smoking habits of the husband. If some true smokers are erroneously included among the female subjects, an apparent relationship of spousal smoking with lung cancer may be seen even when no true effect of ETS exists. This has been mathematically demonstrated (for example, Lee and Forey5), with attempts to correct for it made by major independent authoritative reviews of the evidence on passive smoking and lung cancer.6–8 The magnitude of the bias depends (among other things) on the extent to which women who smoke are misclassified as non-smokers. It can also be shown mathematically5 that a given rate of misclassification of smokers as non-smokers is a much more important cause of bias than is the same rate of the reverse misclassification, of non-smokers as smokers. Since such reverse misclassification is also implausible, adult women having little reason to claim erroneously to be smokers, the major reviews6–8 have all ignored its minor effects.
Given that in the Japanese spousal study (using a urinary cotinine/creatinine ratio (CCR) above 100 ng/mg as an index of true smoking) the reverse misclassification rate (8/298 = 2.7%) was much lower than the misclassification rate itself (28/98 = 28.6%), it becomes abundantly clear that reverse misclassification is not relevant to the passive smoking/lung cancer issue. It is difficult to understand why Yano places such emphasis on it.
Yano1 states that I am “confused with the calculation formula” and that my “definition of misclassification was obtained by dividing those with > 100 ng/mg CCR (n = 28) by self reported non-smokers (n = 318)”. It appears that Yano himself is confused. I had previously made it clear2 that the denominator should not be 318, but 98, the number of women with a CCR value indicative of smoking (or perhaps 106, if one also includes those women who claimed to smoke but had a CCR < 100 ng/ml).
The misclassification rate calculation is clearly based on CCR > 100 ng/mg validly indicating smoking. Such an assumption is widely used,9 though may be subject to some error, and was the best technique available at the time. Most smokers admit to smoking, so that self report has some validity as an indicator of true smoking status, but this does not help us estimate the magnitude of the misclassification bias. The observed lack of correlation in the Japanese spousal study between CCR in non-smokers (with CCR < 100 ng/mg) and other indices of ETS exposure suggests that inaccuracy in CCR measurement at low levels may be important. However, such inaccuracy may not be relevant to the misclassification rate calculation, which merely attempts to use CCR to distinguish smokers from non-smokers. Over half the self reported non-smokers with values over 100 ng/mg actually had values of 1000 ng/mg, and it would be very surprising indeed if errors in CCR measurement were so huge that these women were really non-smokers.
Though I would be happy to see results of further studies using up to date, state of the art chemical methods to detect nicotine metabolites in self reported non-smokers, the conclusion I reached in 1995 that misclassification rates are much higher in Japanese than in Western populations10 seems to be correct. I note that the existence of high misclassification rates in Asian women has in fact been independently confirmed.11
Yano states1 that I used his data without his consent. As far as I am aware, the data never belonged to Yano. The study had been funded by the industry which had carried out the cotinine analyses (blind of self reported smoking status). I had originally proposed that the study be done, following conduct of a similar study in England, which the industry supported at my request, the results of which I reported.12 The original intention had been for Yano to be a major author, but problems arose because his interpretation of the findings differed materially from mine, due to his misunderstanding of the complexities of misclassification. Discussions took place between Yano and Proctor, who played an important role in the planning and organisation of the study on behalf of the tobacco companies who funded it; I was told that these discussions led to Yano deciding not to be an author, and to his understanding that the work would be published by others.
I had assumed that Proctor would keep Yano informed about the status of the publication and was surprised Yano did not find out about the paper, published in 1995, until some seven or eight years later. Clearly, one of us should have kept him informed, and for this I apologise. In his original article,3 Yano states that “at no stage in my interactions with Proctor was Lee’s name or role ever mentioned”. This is surprising inasmuch as the study proposal stated that I would assist in reviewing the study design and in interpreting the data. Was Yano really unaware of the previous literature on misclassification of smoking, in which I figured prominently (see Lee2) when conducting a study, a major aim of which concerned the determination of misclassification rates?
Chapman4 considers that “it is hard to imagine a more flagrant example of attempted ghost authorship”. It is difficult to see why Chapman sees the publication as ghost authorship at all, when I proposed the study, helped in its design, and then published it. The study was a joint enterprise, as I saw it, and it is perfectly normal for some of the scientists involved in a study to write a draft for others to agree to. It would clearly have been better had a version acceptable to all, with Yano in the author list, been published. However, Yano’s failure to understand the mathematics of misclassification made this impossible. There was no agreement I am aware of that Yano had sole rights to authorship. Had I not published the paper10 it seems that the findings would never have appeared in the public domain at all. Did Yano also have sole rights to suppress the findings?
At the end of the day it is interesting that, though the evidence of high misclassification rates in Japanese women has been independently confirmed,11 the relevance of this to the ETS/lung cancer relationship has been ignored in recent major reviews of ETS and lung cancer (for example, Hackshaw et al,8 International Agency for Research on Cancer13). I have demonstrated the major biasing effect of this finding in detail elsewhere.14
Competing interests: Peter Lee is a long term consultant to the tobacco industry.