Santa Claus, the Tooth Fairy, and purported lifetime nondrinkers: Ramifications for observational evidence about alcohol and health

A recent publication in ACER by Ekholm et al. provides important documentation on the hazards of observational studies using "never drinkers" as a control group in studying the relationship between alcohol and health (Ekholm et al., 2024). For observational studies of relationships between alcohol use and health outcomes, and particularly those between “moderate” drinking and health, there are three major methodological limitations. The first is confounding—compared with those who don't drink (including former drinkers and never drinkers) those drinking low average amounts, and particularly those who don't engage in binge drinking, have many favorable lifestyle characteristics that are unrelated to alcohol consumption and thus may reflect wellbeing but are not its genesis (Naimi et al., 2005; Ng Fat & Shelton, 2012). The second is selection bias—this is driven by the fact that most cohort studies used to study the effects of alcohol enroll subjects in middle and later ages. By that point, “moderate” drinkers represent a select population who have not become heavy drinkers or alcohol dependent, not died from an alcohol death or become unavailable for cohort enrollment from an alcohol-caused disability in young and early middle ages, and not stopped drinking due to an alcohol-related problem or due to other causes (Naimi et al., 2017). The third major limitation involves the misclassification of exposure to alcohol. This can be through inadequately measuring a respondent's total or average consumption, failing to assess patterns of consumption (e.g., binge drinking), measuring consumption at only one point in time (which may be an inadequate proxy for lifetime exposure), or misclassifying drinker exposure groups (e.g., misclassifying former drinkers as never drinkers, also referred to as lifetime abstainers). The third of these prodigal biases, misclassification, should not be ignored nor buried in the shadow of its more prominent, misbegotten scientific siblings. Of misclassification errors, the most critical is the so-called “former drinker bias” in which ex-drinkers are misclassified as never drinkers. This relates not just to the degree of exposure to ethanol (as would be the case with misclassification of alcohol consumption levels among drinkers), but to whether there was even exposure to alcohol at all. In developed countries, most current nondrinkers are in fact former drinkers, many of whom have had considerable exposure to ethanol, may have stopped drinking because of alcohol-caused ill health, or who are in remission from an alcohol use disorder (John et al., 2021; Rosansky & Rosenberg, 2020). Others may stop drinking for reasons that are unrelated to alcohol such as declining health or increasing frailty; these persons are often termed “sick quitters” (Rosansky & Rosenberg, 2020; Shaper, 2011; Shaper et al., 1988). This is an example of reverse causation—in this case, poor health is causing drinking cessation, as opposed to a lack of drinking causing poor health. In either case, former drinkers (who stop drinking due to alcohol-caused outcomes or for other reasons) are generally unhealthy and thus contaminate the reference group of never drinkers with unhealthy former drinkers. As such, former drinker bias yields falsely increased risk for presumed never drinkers and minimizes apparent risks for all consumption levels to which they are compared. To combat this problem, efforts are made to differentiate former drinkers from never drinkers (Fillmore et al., 2007). Studies that fail to do so adequately may be excluded from or downgraded in meta-analyses. Despite general awareness among researchers that former drinkers should be excluded from the nondrinking reference group, a recent study found that 70% of systematic reviews and meta-analyses failed to do so, and those that did found increased risk of mortality from alcohol use compared with studies that attempted to do so (Sarich et al., 2024). It should be noted that while it is worse to leave former drinkers in the nondrinking reference group, simply excluding former drinkers or treating them separately results in residual bias since their poor outcomes should ideally accrue to drinking groups based on their lifetime consumption (Liang & Chikritzhs, 2013). But how effective are even well-intentioned and scientifically diligent efforts to distinguish former drinkers from never drinkers? Furthermore, how much alcohol exposure is experienced by former drinkers who are misclassified? The recent publication by Ekholm et al. (2024) helps address these questions in several important ways. First, it adds to the small but important literature documenting the unreliability of self-reported never drinker status in developed countries (Bell & Britton, 2015; Caldwell et al., 2006; Callinan et al., 2019; Rehm et al., 2008); this literature finds that roughly half or more of those who report no lifetime drinking previously reported past alcohol use. Second, this study found that 40% of misclassified former drinkers report past binge drinking. In terms of their exposure to alcohol, this demonstrates that misclassified former drinkers are dissimilar from lifetime abstainers (i.e., a substantial fraction of former drinkers drank more than tiny amounts, and in ways that clearly increase risk). Third, compared with other studies this study examined the reliability of information over a relatively short (~1-year) time frame. This demonstrates that reliability cannot necessarily be mitigated through short recall periods. Furthermore, the fact that such large discrepancies were reported within such a short period suggests that in some cases this unreliability may be akin to describing oneself as a “born-again virgin” (White, 2016) in which one's conception of alcohol-related chastity may be more aspirational than actual. Fourth and finally, Ekholm et al. linked participants with inconsistent responses about lifetime nonconsumption to health registry data, finding that 2.4% of such persons had a health condition that was fully alcohol-attributable (e.g., alcoholic liver disease and alcoholic cardiomyopathy); this prevalence was twice that among current drinkers! This is remarkable given that such conditions are the proverbial tip-of-the-iceberg when considering the far broader range of conditions that may be alcohol-related (i.e., where alcohol is causally associated with a given condition but not the sole possible cause), because registry data likely under-estimate prevalence of such conditions through a relative infrequency of health contacts including among heavy drinkers, the fact that the registry did not capture diagnoses during early study years, and due to nonspecific diagnostic coding (e.g., reporting cardiomyopathy as opposed to alcoholic cardiomyopathy). Linking survey reports to health registry data also helps clarify whether claims of previous exposure to alcohol among those who subsequently reported no lifetime consumption are not themselves entirely a product of poor self-report in a group with known inconsistencies about their alcohol use. What's the upshot of all of this? This study should serve as another strong caution against using purported never drinkers as a reference group in observational alcohol studies. While we and others have called for this change (Naimi et al., 2022; Rehm et al., 2008) and recent, influential studies were restricted to drinkers or included former drinkers and nondrinkers but used low-volume consumers as the reference group (Ortolá et al., 2024; Wood et al., 2018), breaking this methodological habit has proven difficult. Perhaps it's our human inclination to “throw good money after bad”—we've been investing so heavily in this for approach so long, there simply MUST be a way to fix it—let's try harder! But alas it appears that as a viable epidemiological class the lifetime abstainer—like Santa Claus or the Tooth Fairy—doesn't really exist after all. The final concern related to framing alcohol-related risk in relation to never drinking status has to do with the relevance to public health practice and guidelines development. To suggest, or to have the public infer, that alcohol might be used as a therapeutic agent would require higher levels of evidence than that from observational data, particularly since genetic (aka, Mendelian) randomization studies do not support protective effects of alcohol consumption for ischemic heart disease, ischemic stroke, diabetes, hypertension, or dementia (Biddinger et al., 2022; Chen et al., 2008; Holmes et al., 2014; Lankester et al., 2021; Lawlor et al., 2013; Millwood et al., 2019; Zheng et al., 2024). Furthermore, observational studies of persons who happen to be drinking at a given level are not particularly relevant to questions of whether a nondrinker should purposefully initiate consumption or purposefully continue to drink into old age to improve health (Greenfield & Kerr, 2014). More relevant is a scientific and communications focus regarding different risk levels among drinkers, rather than risks associated with nondrinking status. To bring this issue to greater attention, we somewhat humorously propose a new diagnosis related to the repeated, unquestioning use of a scientifically inadequate reference group: epidemiological lifetime abstainer use disorder (ELAUD). As in the case with other use disorders, scientists working on observational alcohol investigations need first to acknowledge the problem in defining subjects as lifetime abstainers and take action by using low-volume drinkers as the reference group, even if it makes comparisons with previous work somewhat more difficult. With respect to observational data, using purported never drinkers to define the risk of alcohol consumption leads to systematic underestimates and has meaningful negative consequences for public health and clinical practice. The authors have no conflicts of interest to report.