Statistics from Altmetric.com
I agree with Yano that scientific integrity should be maintained and researchers be free from pressure from commercial and campaigning interests. I disagree that I have caused any “distortion of scientific findings” or “misrepresentation and misappropriation” of results, serious charges which I will show are unjustified. To clarify the situation I will start with the history preceding the study.
In 1981 Hirayama reported an increased lung cancer risk in non-smoking women married to smokers.1 Following this I demonstrated random misclassification of smokers as non-smokers, coupled with smokers tending to marry smokers, leads to an observed increased risk even when environmental tobacco smoke (ETS) has no effect.2,3 The tobacco industry then agreed to support a study in England of this “misclassification bias”, using cotinine to validate smoking, which was reported in 1987.4 I also reviewed evidence on misclassification,5 revealing the lack of useful data in Japan, where cultural differences might affect reporting of smoking.
Several tobacco companies therefore decided to fund a study there. The protocol, drafted by Proctor and discussed with me, included two major phases. Phase I, described in my paper,6 mainly concerned misclassification, using urinary cotinine/creatinine ratio (CCR) to validate smoking, but also collected data on dietary/lifestyle factors. Phase II mainly involved comparing different ETS markers. The study was organised in Japan by Yano, who discussed the design and findings with Proctor.
Proctor sent me the phase I data in 1992, but I have never seen the phase II data. I conducted statistical analyses and helped draft possible papers for publication. The exact history is now unclear to me, but discussions occurred between myself and Proctor, and Proctor and Yano, as to how best to present the findings. I felt then that Yano did not fully understand the complexities of misclassification. Eventually, after discussions with Yano, Proctor asked me to author a paper including suitable acknowledgements to Yano and for the funding.
Below I comment on Yano’s criticisms of my analyses.6
RELIABILITY OF CCR
I mainly used CCR to detect misclassified smokers. As Yano’s table 4 shows, current smokers had a median CCR almost 100 times that of non-smokers, with values > 100 ng/mg seen in 90% of smokers and only 9% of non-smokers. This counters Yano’s claim that CCR was an invalid marker of self reported smoking. Nicotine based markers are commonly used to detect misclassified smokers.7,8
I found little relationship of CCR in non-smokers to husband’s smoking, consistent with other evidence8 that CCR is a valid marker of smoking, but less good as a marker of ETS exposure. Yano appears to argue that CCR is invalid generally. My main conclusion, of high misclassification rates in Japanese women, depends only on CCR being a valid marker of active smoking.
SMOKERS WITH LOW CCR VALUES
Some current smokers have low values, possibly because they have not smoked recently, though the reasons for this are not fully understood.8 The percentage of self reported smokers among women with CCR < 100 ng/mg (8/298 = 2.7%) was much less than the percentage of self reported non-smokers among women with CCR > 100 ng/mg (28/98 = 28.6%). Also the bias to the lung cancer/spousal smoking relationship that results from smokers denying smoking much exceeds the corresponding bias resulting from non-smokers claiming to smoke.9 I therefore concentrated on the former misclassification.
Yano claims that misclassification of self reported smokers as non-smokers (8/78 = 10.3%) and of non-smokers as smokers (28/318 = 8.8%) are similar. This erroneously uses self report, not CCR, as the “gold standard”. One uses CCR to detect true smokers and hence calculate misclassification rates.
COMPARISON WITH WESTERN POPULATIONS
Not only this study, but also two other biomarker studies of Asian women, show misclassification rates substantially higher than seen in western populations.10
EFFECTS OF RANDOM MISCLASSIFICATION
Yano argues that random misclassification underestimates true associations. This is so for smoking by the spouse (used as the exposure variable) but not for smoking by the woman (used as the variable to select the non-smokers for analysis).2,3 The biasing effect of a given level of misclassification to the lung cancer/spousal smoking relationship is much less for spousal than for subject smoking.9
CONFOUNDING BY LIFESTYLE FACTORS
Meta-analyses show that, in non-smokers, lung cancer risk and ETS exposure are both associated with a poorer diet (less fruit and vegetables, more dietary fat) and poorer education, and that the resulting confounding effect is non-negligible.11 Earlier versions of my paper included a table showing that marriage to a smoker was generally associated with a poorer lifestyle, but not significantly, due probably to the small sample. After one journal rejected the paper I simplified it, concentrating on misclassification issues. However, my discussion section6 still cites some confounding results, overlooked by Yano.
Yano believes high kappas for repeat interview data indicate a reliable response. I would regard some subjects claiming in 1992 never to have smoked and in 1991 to have smoked as indicating some unreliability. Also, some subjects may have consistently denied smoking. There is much literature that demonstrates the inconsistency of reported smoking.7
I never intended to present all the available findings fully, concentrating on smoking misclassification in Japanese women. The phase II ETS data were only for non-smokers and would not have contributed to this. Yano has not demonstrated any distortion or misrepresentation, or that I “reached conclusions…totally at odds with the actual findings.” I stand by my conclusions.
Competing interests: Peter Lee is a long term consultant to the tobacco industry.
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.