Rigorous methodology is needed to analyze and interpret observational data on the use and effectiveness of smoking cessation AIDS.

Abstract

In the current issue, Clare et al.1 present data from 23,402 respondents (of which 18% were recent ex-smokers) to four waves of an Australian household survey to address two research questions: (a) whether use of smoking cessation aids differed by indicators of socioeconomic status; (b) whether smoking status differed by use of smoking cessation aids. Advantages of a study using such household data include large statistical power and high level of external validity. However, the internal validity of a study that relies on relatively crude, observational data can be limited. The biggest threat to the internal validity of an observational study—especially a cross-sectional study—on the use and effectiveness of smoking cessation aids is potential bias due to confounding. The most important confounding factor in this respect is tobacco dependence. A simple analysis of the association between socioeconomic status and use of smoking cessation aids is confounded because socioeconomic status is associated with dependence,2 and smokers who are more dependent are more likely to use pharmacotherapy to treat their dependence.3–6 Similarly, an analysis of the association between use of smoking cessation aids and abstinence is confounded because, again, smokers who are more dependent are more likely to use pharmacotherapy to treat their dependence, and smokers who are more dependent have a lower chance of achieving abstinence.7 An analysis that does not adequately adjust for differences in dependence would lead to an underestimation of the effect of pharmacotherapy. Clare et al.1 adjusted their analysis of the association between smoking cessation aids and abstinence from smoking only for several demographic and socioeconomic variables (Table S2). Thus, their results on the relative effectiveness of the various smoking cessation aids are difficult to interpret. Adjusting the analysis of cross-sectional data for the confounder tobacco dependence is challenging because, dependence measures that include an estimate for cigarettes smoked per day are not valid in respondents who are recent ex-smokers like in the study by Clare et al.1 A better method, recently introduced, uses ratings of urges to smoke for this purpose.8