• Smoking cessation outcomes;
  • smoking status imputation;
  • tobacco research methodology

The issue of what to do when participants drop out of tobacco cessation trials may seem open to a number of solutions. Smolkowski et al.[1] have set out to look for a better answer than the currently recommended approach of assuming that dropouts have resumed tobacco use. They compare several methods of divining the tobacco use status of those lost to follow-up and recommend the use of different imputation algorithms derived separately for each trial. While the enterprise of arriving at better ways of imputing missing values is attractive, we think that this kind of exercise cannot deliver the answers needed and that the commonly used assumption that the dropouts have relapsed remains the best option generating the least amount of error.

Imputations other than ‘missing = relapsed’ are critically dependent upon the assumption that treatment failures and treatment successes have equal probability of responding or not responding to follow-up. Such evidence as we have suggests strongly that this is rarely the case.

Successful quitters are usually keen to let the treatment providers know of their success. They often feel grateful for the help received, are receptive to the request to help with data collection, they know that their success will please the treatment providers, know they will be praised and feel good about their outcome. In contrast with this, treatment failures feel embarrassed, typically blame themselves for letting the treatment providers down, expect that their report will disappoint those who were trying to help them and feel typically that if they do report back, they will have to present elaborate justifications of their relapse and promises for the future. As a result of this, it is rare for a successful quitter to avoid making their success known; and it is common for treatment failures to drop out of treatment and associated trials and stop providing further data, or do so only after repeated prompts. This difference cannot be addressed by missing value imputation algorithms.

In our clinical and research practice with smokers, we have encountered a number who were lost to follow-up and re-entered treatment some time later. They admit invariably that they avoided follow-up because of their return to smoking. In the only study that we are aware of that has looked at this phenomenon systematically, Foulds et al.[2] invested extra effort to track down the first 50 non-responders in a trial of nicotine skin patches. Not a single one of them was abstinent. Variables such as their pre-treatment stage of change or their answers to other questionnaire measures were irrelevant.

The last observation carried forward (LOCF) approach, which is one among several used by Smolkowski et al.[1], is an extreme example of the problems that can arise. People keep providing data while still abstinent, and stop doing so when they relapse. In stop smoking programmes where people are seen 24 hours after their quit date, the LOCF assumption would lead to claims of a nearly 100% long-term sustained success rate, despite the fact that only some 15% of smokers achieve such a goal.

The assumption that non-responders have relapsed avoids the myriad methodological and logical problems the authors identified in their report, and is unlikely to lead to any substantial misclassification. The Russell Standard [3], which seeks to unify the reporting of smoking cessation outcomes to allow cross-study comparisons, requires that smokers lost to follow-up are classified as continuing to smoke. A similar principle would apply to smokeless tobacco users.

It is also important to note that the Russell Standard includes guidance on reporting sustained abstinence, as opposed to the much softer and less meaningful 7 days of not smoking, and the rationale for biochemical validation of self-reported abstinence. The reported trial used soft options on both these counts. Although in this type of trial biochemical validation could be a challenge, the lack of data on sustained abstinence poses problems in integrating its findings with the body of more rigorous research.


  1. Top of page
  2. Declarations of interests
  3. References