Abounding confounding


Abounding confounding

Sick Quitters and Healthy Drinkers

In recent decades a substantial number of reports in diverse populations [1-4] have shown consistently that light-to-moderate alcohol drinkers are at lower risk of coronary artery disease (CAD) than abstainers. Evidence for plausible biological mechanisms has appeared [5-7], and controlled experiments show explanatory effects of moderate drinking upon several intermediary markers of CAD. The apparent benefit from alcohol drinking is relatively specific for atherothrombotic conditions (primarily CAD), and frequency of drinking of moderate amounts may have extra benefit [8]. There is evidence that persons with an alcohol dehydrogenase polymorphism resulting in ‘slow metabolism’ of alcohol [9] may have more CAD benefit. While all of these aspects suggest a causal protective effect of light-to-moderate drinking on CAD, the absence of randomized trials with CAD outcome events leaves open the valid issue of possible bias by confounding. This issue continues to be a focus of analysis and controversy.

Few epidemiologic observations have been scrutinized for bias as thoroughly as the apparent CAD benefit from alcohol. Concerns about the explanation of the lower risk of CAD in alcohol drinkers have been expressed from the beginning. With neither precedent nor evident biological basis the authors of the first report [1] suggested eight possible indirect explanations before mentioning the possibility of a protective effect of alcohol. The ‘sick quitter’ hypothesis [10] appeared on the scene because some reports used as referent a combined lifelong abstainers and ex-drinkers group. If ex-drinkers disproportionately include persons with increased CAD risk, a spurious impression of benefit from light-to-moderate drinking might ensue. Prospective studies that separated ex-drinkers from lifelong abstainers and/or controlled for baseline CAD have refuted this hypothesis [2-4]. However, it is still cited and even the use of lifelong abstainers as a referent has been questioned [11] on the grounds that, because of incorrect classification, they actually include some past drinkers. It is now accepted that ex-drinkers should be a separate category in alcohol epidemiology analyses.

Another potential source of confounding is healthy life-style habits of light-to-moderate drinkers [12, 13]. The article by Naimi et al. [14] in this issue of Addiction falls into this area, with specific focus on the apparent extra benefit of more frequent moderate drinking. The data presented show that the more frequent moderate drinkers are less likely to be bingers and have healthier traits than light-to-moderate drinkers with fewer drinking occasions. The authors point out that frequent drinking may ‘simply be a marker for drinking safer amounts’. The observation is logical and indicates the need for careful control when studying the role of drinking frequency. Unfortunately, neither outcome data nor data about possible intermediate biological markers, such as blood lipids or blood pressure, are available. The presentation leaves unresolved the key issue of whether frequent drinking of safe amounts has more benefit than infrequent drinking of like quantities. The data lead the authors to a substantial discussion of public health considerations and with evident skepticism about the benefit of moderate drinking.

The Beverage Choice Issue: Wine, Liqor or Beer?

The hypothesis that CAD benefit is relatively specific for red wine was initiated by international comparisons suggesting less CAD mortality in wine-drinking countries than in beer- or liqor-drinking countries [15]. Such aggregate correlation data are obviously subject to confounding by many uncontrolled traits, but the presence of non-alcoholic antioxidant phenolic compounds with antioxidant and antithrombotic properties in red wine [5-8] lends indirect support to the concept. Epidemiologic data with CAD endpoints suggest that white wine, red wine and beer may all be related to lower CAD risk [3]. The issue is complicated by data indicating that wine drinkers have the most favorable CAD risk profile [16], thus creating another basis for confounding. The interesting and complex beverage choice issue has bearing upon this discussion for two reasons: (i) study of relative benefit of the beverage types is replete with confounding issues, and (ii) if non-alcoholic ingredients of wine are involved in health benefit, the role of alcohol would be less.

Confounding Can Work Both Ways

Objectivity mandates similar attention to possible confounding in apparently harmful relationships of light-to-moderate alcohol drinking. These include reports about common conditions, such as hypertension, cancer and stroke.

Cigarette smoking is correlated positively with both lighter and heavier drinking in most populations [1]. Thus, smoking represents an important life-style trait that might spuriously increase the apparent risk of many conditions. In the case of CAD it is likely that residual confounding from smoking diminishes apparent benefit. Actually, it is difficult to study the independent role of alcohol in conditions strongly related to smoking, such as lung cancer or chronic obstructive pulmonary disease, partly because the small proportions of non-smokers may disproportionately have atypical types of the diseases. The inverse alcohol–CAD relationship has been shown among never smokers [3].

Under-reporting alcohol intake is another source of bias with respect to adverse effects of light-to-moderate drinking. Under-reporting produces misclassifications in most—if not all—survey databases. We have reported an attempt to study this aspect in a cross-sectional analysis of alcohol-drinking and hypertension [17]. The results (Fig. 1) showed that higher prevalence of hypertension at 1–2 drinks per day was essentially limited to suspected under-reporters.

Figure 1.

Odds Ratios of hypertension (OR HTN) by drinking category among 105 378 white and black men and women and by likelihood of under-reporting among persons reporting 1–2 drinks per day. Adapted from data in Klatsky A. L. et al. J Stud Alcohol 2006; 67: 421–8

Upper part shows adjusted Odds Ratios (95% confidence interval) of hypertension (≥ 140/90 mmHg) by drinking category with never drinkers as referent among 105 378 white and black men and women. A threshold relationship is seen starting at 1–2 drinks per day

Lower part shows persons reporting 1–2 drinks per day at index data collection stratified according to likelihood of under-reporting alcohol intake. ‘Likely under-reporters’ reported ≥ 3 drinks per day on another occasion or had an alcohol-related diagnosis (death certificate, hospital discharge, outpatient) at any time. ‘Unlikely under-reporters’ never reported heavy intake and had no alcohol-related diagnosis


As studies of disease outcomes in alcohol epidemiology are likely to be indefinitely limited to observational studies, potential confounding will remain an issue with respect to both beneficial and adverse effects [18]. Substantial skepticism about optimal amounts of alcohol intake and thresholds for harm is warranted. Advice issues—both individual and public health—need to be tempered by objectivity and common-sense [19].

Declaration of interests

This work was supported by a grant from the Community Budget program of the Kaiser Permanente Medical Care Program in Oakland, CA, USA. Data collection in 1978–85 was supported by a grant from the Alcoholic Beverage Medical Research Foundation. No funding source played any oversight role in the analysis or in the preparation of the manuscript. Dr Klatsky is a member of the International Scientific Forum on Alcohol Research (ISFAR) and on the editorial board of Alcohol in Moderation (AIM), both of which are associated with the alcohol industry. Neither of these entities has supplied him or his institution with any financial support.