Commentary on Di Castelnuovo et al: Implications of using low volume drinkers instead of never drinkers as the reference group

In studies of alcohol and mortality, using low volume drinkers as the reference group has several advantages compared to using never drinkers as the reference group. Di Castelnuovo et al. are to be commended for a well done study about alcohol and mortality using data from a number of large cohort studies [1]. Overall, the findings and attendant limitations are similar to other large and well-performed observational studies. The issue about possible health benefits of low-volume alcohol consumption and scepticism about this body of evidence remain. Much of the scepticism revolves around comparisons of drinkers with non-drinkers. An interesting contribution of this paper is its exploration of the reference group in the supplementary materials. An appropriate reference group can minimize bias, whereas an inappropriate reference group can ‘contaminate’ all relative risk estimates. In the main analyses presented in the paper, the authors pursue the traditional route of using never drinkers (referred to as ‘lifetime abstainers’) as the reference group. However, the never drinker group has several limitations to consider. First, studies find that many or most self-reported never drinkers are in fact former drinkers (based on closer questioning or based on previous self-report among cohort participants with multiple waves of follow-up) [2, 3]. Inclusion of misclassified former drinkers in the never drinker reference group can result in under-estimates of risk among comparator drinking groups [4], particularly because some former drinkers have a history of heavy alcohol use and others are ‘sick quitters’ who are frail or in generally poor health. Second, never drinkers tend to have worse health and socioeconomic status compared to the general population and current drinkers, with differences observed from young ages onward [5-7]. This increases the likelihood of adverse confounding relative to drinking groups. Third, never drinking is a lifetime-based measure, which is distinct from cross-sectional levels of current drinking. An appropriate lifetime counterfactual scenario for a ‘never drinker’ is an ‘ever drinker’ (i.e. all those who have ever initiated alcohol consumption). The fourth limitation of the never drinking reference group is that simply excluding former drinkers from the non-drinking reference group selectively removes unhealthy persons (e.g. the sick quitters) whose poor outcomes should, from an intention-to-treat perspective, have accrued among the various drinking groups [8]. This bias could be addressed by re-allocating former drinkers into drinking groups based on their lifetime consumption levels [9]; however, most cohort studies assess consumption at only one point in time. To their credit, the authors performed a supplemental analysis to re-allocate former drinkers using imputation. This found significantly increased risks of all-cause mortality among those drinking 10.1 to 20 grams daily compared with drinking up to 10 grams of alcohol daily. A final limitation is that the number of never drinkers is often small relative to low-volume drinkers, and comparisons between drinking groups (e.g. low, medium and high) are made indirectly via their relationships to never drinkers. In this paper, using never drinkers as the reference group (Table 2), the hazard ratio confidence intervals for any-cause, cardiovascular and cancer mortality among those drinking 0.1 to 10 grams of ethanol daily and those drinking 10.1 to 20 grams per day overlapped, and risks for any-cause and cancer only appeared to increase above 20 grams. These data support the conclusion of the abstract that ‘Intake of more than 2 drinks per day was associated with an increased risk of total, cardiovascular and especially cancer mortality.’ Yet, when drinkers consuming 10.1 to 20 grams daily were compared to a reference group consuming 0.1 to 10 grams per day (Supplemental Table 5), confidence intervals no longer overlapped and total mortality risk increased significantly from 10.1 grams per day. For both reference groups, the risk differences based on the point estimates were very similar (~10%). A 10% increase in total mortality risk is substantial and greatly exceeds thresholds defining efficacy for drug trials or public opinion about acceptable levels of risk. Based on the use of low-volume drinkers as the reference group, the most important conclusion to draw from this study might be that consuming more than 1 drink per day is associated with increased mortality compared to drinking less. Using low volume drinkers as the reference group seems preferable scientifically as there are the aforementioned reasons to believe that the never drinker reference group results in systematic under-estimation of alcohol-related risk. In addition, if the purpose is for health guidance for drinkers, directly comparing risks among different drinking group seems the most relevant framing. The impetus to use low volume drinkers as the reference group is particularly important if failing to do so obscures significant increases in risk between drinking 1 versus 2 drinks daily. Reclassifying former drinkers into their drinking groups may further reveal significant differences in risk between drinking up to 1 drink per day versus drinking 1 to 2 drinks per day. Timothy Stockwell was previously a Senior Editor for Addiction and oversaw the review of the paper by Di Castelnuovo et al. [1].

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.