The real significance of the Britton et al. paper is overlooked by the authors themselves. Alcohol intake varies in the same individuals over time, declining with increasing age in many countries [2–7]. This decline has been shown to occur among those individuals most vulnerable to morbidity and mortality [8–10]. People most likely to be stable in their drinking—the light or moderate drinkers—tend to have the most favourable health characteristics [11,12]. Britton et al. acknowledge these findings by stating that it is well established that the non-drinking group often includes former drinkers who terminated consumption due to ill health.
The Britton et al. results are not surprising, in that they show a ‘drift’ down in consumption after the middle years of life among those who have been ‘heavier’ drinkers at one time or another. An understanding of this ‘drift’ requires multiple measurements associated with health changes over time. The authors might have contributed substantially to understanding whether it is the declining drink, the illness/medication or the interactions of these that predict premature mortality/morbidity.
Britton et al. provide but four characteristics of group variability (smoking, age, employment grade and alcohol consumption in the last week). Unfortunately, the authors do not supply the characteristics relating to health status, serious disease and medication use that might account not only for the decline in the drinking, but also for mortality/morbidity. If these characteristics are not factored into the analysis, we have no understanding of what is actually driving the elevated risk—the drinking, the onset of illness or both.
Despite this major limitation, this paper should be welcomed because cohort studies associating alcohol use with disease outcomes have lagged far behind the remainder of the alcohol field, the latter taking into account variation in drinking behaviour. In recent years greater attention has been paid to these matters by cohort studies, but they are few (e.g. [13–15]). They illustrate that drinking behaviour is complex and may be implicated with respect to disease incidence and, in fact, may be critical—but only in part—to understanding how drinking might affect outcome.
Previously, alcohol studies have put great emphasis on differentiating between these patterns, recognizing that they had profound implications for social problems, accidents and injuries. Knupfer's 1989  attempt to classify drinkers into, among other groups, a variety of ‘heavy’ drinkers found that when the bar was set low, the association between ‘heavy’ drinking and frequent drunkenness and serious problems was missed. This required the combination of 10 general population cross-sectional data sets. Like most studies, Britton et al. lack an adequate sample size. As anyone performing these studies will testify, the choice of analytical strategy is difficult, the analysts fast run out of cases and serious analytical compromises must be made. We encourage those analysts in command of multiple measurement point studies to combine their raw data so that these problems may be reduced.
The study will also be welcomed by those emphasizing that ‘binge drinking’ is predictive of premature death and there is sufficient evidence to suggest that this may be the case for some disease outcomes; but this should not be the central point. Rather, investigation should have been devoted to answering why the decline in drinking among the highest variation group happened at all. In fact, the premature mortality or morbidity may not have been ‘caused’ by the occasional past binge drinking but, rather, by the onset of disease.
There are additional problems. First, non-drinkers over multiple measurements were excluded from the analysis, but the variation in drinking before the onset of consistent abstinence should have been assessed because probable ill health may have contributed to elevated risk. Secondly, civil servants are not well suited to study variation because they do not represent the general population and sample loss (37%) may have been due to vulnerable subjects dropping out of the study. Undoubtedly, respondents retained in this non-representative sample are biased in favour of those with continuing good health.
This study reminds us all that a number of confounders, including the onset of ill health, may explain these findings [11,17], and until these are properly accounted for our efforts will remain tentative and incomplete. Britton et al. may still redeem their own efforts by performing additional analyses on their data to address the critical question of whether illness drove alcohol consumption down among the high variability group or vice versa. We challenge the authors to consider that the culprit here may not be the ‘declining drink’ among the highest variability group at all but, rather, other life and health changes experienced by this group which are associated with drinking style .
The importance of this missing analysis is critical. By not accounting for onset of illness/medication or interactions with drinking changes that predict premature death and morbidity from heart disease, the authors have failed to identify the probable cause of both the decline in drinking and the increased mortality/morbidity risk.