In response to Spring and colleagues , we focus on the methodological issues raised in hope of shedding light and not generating heat. The field must decide on the right methodological approach.
Spring and colleagues assert that weight gain should be measured in the entire population of treated smokers. We agree that it would be valuable to collect data on weight in relapsed smokers. However, although sparse, available data indicate that smokers who relapse lose the weight they gained [2,3], but more data are needed before concluding that there is not a problem that needs management.
The key methodological issue is whether to combine data from relapsed and abstinent smokers in an intention-to-treat (ITT) analysis. We believe that this approach is inappropriate for assessing post-cessation weight gain because of the confounding effect of abstinence. Within the length of most studies, longer abstinence leads to greater weight gain. Interventions may increase, decrease or have no effect on smoking or weight. Imagine a trial with an intervention that reduced weight gain but increased abstinence compared with control: there will be more abstinent smokers in the intervention group than the control group, therefore the mean weight gain of the ITT population will be higher in those receiving the intervention than those in the control group even if the intervention limited weight gain. An intervention that decreases abstinence but does not affect weight gain will appear to produce less weight gain in the intervention group when presented as an ITT analysis. Furthermore, the average weight gain in an ITT population is hard to interpret—it is neither the typical weight gain of abstinent nor relapsed smokers.
ITT analysis implies that data are available on all participants and, when not available, data are imputed. In smoking cessation trials standard practice is to impute failure of abstinence in participants not followed-up and there are empirical data to support this . In trials of weight loss for obesity, researchers typically impute baseline observation carried forward or last observation carried forward for missing data. In both smoking cessation and weight loss trials, participants continue to attend clinic and follow-up while they strive to reach their goal. When they abandon their goals, they typically stop attending clinic. Data on clinic attendees only are therefore a biased estimate of the entire population's weight, and neither last observation carried forward or baseline observation carried forward seems appropriate. ITT analysis is therefore impractical without heroic efforts at follow-up or some basis on which to impute weight.
As we explained in our first response, we created subgroups for meta-analysis. We consider that an intervention aimed at reducing concern about weight gain might not have the same effect as one which restricts calories for weight loss. Nor do we feel that a very low-calorie diet, which suppressed hunger, can be considered equivalent to modest calorie restriction where hunger is much more likely. Given that going hungry increases smoking and urges to smoke , it seems important to examine the effects of interventions that might enhance abstinence, such as exercise , separately from those that might decrease it. Hall and colleagues randomized smokers to control, general advice about nutrition, calories and exercise or to the same kind of information plus an individualized exercise and calorie plan, an individual weight target which must not be breached and action if the target was breached . Like us, Hall and colleagues must have postulated that these two active interventions could affect outcomes differently. Perkins and colleagues' weight control intervention also incorporated individual targets with regular monitoring against the target and readjustment if the target was breached, and these are the two interventions we combined in our ‘individualized programme’ meta-analysis . Although there was some degree of tailoring in the intervention tested by Pirie and colleagues, there was no regular monitoring and reassessment according to individual need, and so we combined this with the Hall study in the ‘advice only’ meta-analysis . We should have been more clear in describing in our review that the key common element of the individualized interventions was the individual target, regular monitoring and readjustment if the target was breached. Regular monitoring and readjustment is incorporated into most general weight management programmes. A recent meta-regression showed that interventions to promote healthy eating and regular exercise that incorporated self-monitoring were significantly more effective than those that did not . In our review , we did not draw such strong conclusions as suggested by Spring and colleagues. Instead, we suggested that further research on weight control after cessation should incorporate these kinds of techniques, which most interventions have not. We are happy to discuss the methodological issues with researchers in this area because there is not yet consensus on appropriate methods, as this exchange demonstrates.