Relationships between alcohol consumption and health outcomes are especially difficult to estimate because individuals quit drinking or reduce their consumption in response to health problems of many types. Because most studies of alcohol-related health and mortality outcomes have been based on cross-sectional surveys, seldom measuring life-course consumption, Liang and Chikritzhs’ paper  makes an important contribution by demonstrating that people quit or reduce their drinking in many cases because of health concerns and by quantifying the number of such people on a one year basis in a general population sample of Australians.
A major limitation of this otherwise excellent paper is that it considers only indictors of reduction or quitting, taken together during the prior 12 months, and separately quitting at an earlier but unspecified time. Ascertaining the amount of reduction is also an important consideration as the drinker who has reduced might still be drinking heavily or may conversely have been a light drinker already before reducing to an even lighter volume. Also, quitting and reducing may be qualitatively different behaviors. Furthermore, drinking pattern, as just implied, may be particularly important in the medium volume range [2,3], and receives little consideration here. One needs to recall that these behaviors are self reported. Even reports of lifetime abstention may ‘fluctuate’ in repeat surveys  and self-reported reduction, while possibly accurate in many cases, might also in some be an indication of intent more than an actuality when a serious health condition has been recently diagnosed. It will be good in future work to seek details about when the diagnosis occurred and the time course of the health conditions, not just whether they occurred in the prior 12 months. Also, given the multi-stage, stratified area design, we are concerned that there may be substantial design effects due to clustering. The authors are unclear about whether the analyses took proper account of the clustered sampling in the estimation of standard errors.
Despite these inherent limitations, this important paper makes several major contributions. First, the work adds to the evidence that reporting quitting or reducing drinking is related to poorer health status and having diagnoses of several health problems including diabetes, heart disease or hypertension and anxiety. As the authors correctly mention, health-related influences on reducing or refraining from drinking are a major source of bias in alcohol-related health outcome studies that lack a life-course perspective in measuring exposure, particularly in prospective cohort studies. In effect, absent information about prior or current health influences on drinking, one cannot gauge the influence of Shaper's sick quitter effect  on outcome risk estimates for alcohol. For an Australian general adult population, these results provide some useful indications of the extent of the measurement bias for studies where current health status measurements are unavailable. For mortality outcomes especially, but also for morbidity, the population is highly selected by the ages where most outcomes occur. The authors discuss returning former drinkers to lifetime drinker categories by analogy with ‘intent to treat’ analyses. A recent paper using the US National Alcohol Survey  made a very similar argument, that former drinkers are a type of drinker, not a type of abstainer. That paper utilized lifetime drinking categories in a retrospective cohort design and also examined relationships with current drinking categories controlling for lifetime drinking measures. Using this framework and controlling for a variety of psychosocial, health and demographic risk factors through propensity score matching, lifetime moderate drinkers (who reported no 5+ drinking days in retrospective decades measures) were shown to have a reduced risk of diabetes, but not heart problems or hypertension as compared to lifetime abstainers; heavy drinkers had an elevated risk of hypertension. Like the present article, these results confirmed the importance of sick quitter effects, but also found a protective relationship between moderate drinking and diabetes based on lifetime drinking categories. Of course, the possibility of bias from unmeasured life-course confounders remains.
What then can we do going forward? We can ask questions about prior drinking at baseline to determine key aspects of life-course drinking (e.g. if ever drank, age of onset; any heavy drinking and for how long this continued; history of serious alcohol-related problems or alcohol dependence; date/s and reasons for quitting). We should also conduct methodological studies to understand weaknesses of retrospective measures and how to get the best possible information of this type in cross-sectional surveys. Although periodic longitudinal research on life-course measurement reliability studies has been reported, more such studies are needed to judge the reliability using repeated measures. Longitudinal studies over long time periods would also be important because time-varying covariates, including health conditions, could be included. Typical cohort studies measure alcohol at one time, with reporting durations covering drinking during a day, a week, a month or a year . Implicitly, this average volume is then extrapolated to cover the person's whole life, from teen years to the period following the interview until the end point in question. Even when we only have the ‘current drinking’ information, is this the best estimate of the person's drinking pattern at all of those times? Perhaps even in such designs we should be modeling drinking change of subgroups, defined by specific characteristics, based on longitudinal studies. Most importantly, we believe that a life-course perspective including how alcohol consumption patterns change over time, and major reasons for such changes, should be the primary perspective for future studies of alcohol-related health risks.