Page last updated: Tuesday, March 13, 2007
Research suggests systematic error in prospective studies relating to moderate alcohol use and reduced mortality
Source: Fillmore KM, Kerr WC, Stockwell T, Chikritzhs T, Bostrom A. Moderate alcohol use and reduced mortality risk: Systematic error in prospective studies. Addiction Research and Theory 2006; preprint.

Comments: R. Curtis Ellison, MD, Luc Djoussé, MD, ScD, Kenneth J. Rothman, DrPH, Yuqing Zhang, MD, ScD, Institute on Lifestyle & Health, Boston University School of Medicine.

Summary: The authors of this article state that the majority of prospective studies on alcohol use and mortality risk indicate that abstainers are at increased risk of mortality from both all causes and coronary heart disease (CHD). This meta-analysis of 54 published studies tested the extent to which a systematic misclassification error was committed by including as “abstainers” many people who had reduced or stopped drinking, a phenomenon associated with ageing and ill health. The studies judged to be error free found no significant all-cause or cardiac protection, suggesting that the cardiac protection afforded by alcohol may have been over-estimated. Estimates of mortality from heavier drinking may also be higher than previously estimated.

Comments: There have been a large number of “meta-analyses” evaluating the relation of alcohol consumption to the risk of CHD. Essentially all previous ones have concluded that, despite the high probability of some residual confounding (inherent in observational studies), the overwhelming evidence supports an inverse association between light-to-moderate alcohol intake and CHD. Such a finding has been strongly supported by hundreds of intervention studies in humans showing how alcohol can block the development of atherosclerosis, reduce most cardiovascular risk factors, and improve ventricular and endothelial function; further, studies in animals have shown dramatic lowering of risk of the development of atherosclerosis and myocardial infarction from the administration of alcohol. For heavy or abusive drinking, in addition to the known adverse results on a number of non-cardiac conditions (cirrhosis, increase in accidents, etc.), some epidemiologic studies suggest that there may also be an increase in risk of heart disease, but this may be primarily from cardiomyopathy, arrhythmias, etc., and not necessarily from atherosclerotic heart disease.

The present paper has made a number of assumptions and carried out analyses that do not support a protective role of moderate drinking against coronary disease. The primary reason that we question the results of the study is that there have been, especially in recent years, a number of prospective epidemiologic studies that have avoided the errors used by the authors to exclude studies from their analyses; i.e. they have used “lifetime abstainers” as the referent group, and have included occasional drinkers within specific “low-intake” categories, and not mixed them into categories of abstainers or regular consumers of small amounts. One could use such studies to test the hypotheses the authors present and avoid the two types of errors that the authors have sought to evaluate: former drinker misclassification error (failure to separate former drinkers . . . from complete abstainers) and occasional drinker misclassification error (failure to separate occasional drinkers . . . from complete abstainers). A review of all recent studies that have avoided the two errors described above, and accounted for the pattern of drinking as well, is needed to fully test the hypotheses proposed by the authors of the present paper. Only 2 studies for cardiovascular mortality and 7 studies for total mortality were used as a basis for the final conclusions of the authors in this paper, and we do not believe that they were in any way “representative” of the general population.

There are a number of specific concerns that we have about the analyses in the present paper. These are outlined briefly below:

(1) In the analyses the authors appear to have included studies that did not adjust for smoking. Including an indicator variable that “adjusts” for this failure is inadequate, since smoking is such an important risk factor for CHD, and it is well know that cigarette smoking is much more common in people who consume alcohol.

(2) The decline of alcohol consumption with age appears to be irrelevant as long as there is control for age. The authors do not provide any support for their statement that “Statistical controls for age in these studies are insufficient to control for a strong bias toward less healthy individuals being more likely to reduce or quit drinking.” While in the past physicians may have advised their patients with many diseases to stop alcohol consumption, the reverse is beginning to happen for subjects with coronary disease: more and more, such patients are being encouraged by their physicians to consume small amounts of red wine or other beverage on a regular basis. Further, a number of prospective studies have repeated assessments of alcohol intake, so changes over time in the intake of alcohol can be appropriately evaluated in the analysis.

(3) It is unclear why the authors chose the “light” and “moderate” categories to be so broad, given that they are assuming a J-shaped relation. Their category of “light” drinking combines subjects reporting alcohol intake ranging from one drink/month up to 24 g/day (equivalent to 2 typical drinks), which is up to the maximum limit advised for men, and includes up to double the officially recommended limit of 1 drink/day for women. The “moderate” range goes up to almost 4 drinks/day. In many studies, maximum benefits are seen in the range of ½ to 1 drink/day for women and 1 to 2 drinks/day for men. Creating such broad categories may dilute any putative effects of moderate drinking. The authors could have used many more categories (given the large number of subjects) and perhaps used a spline analysis to determine the shape of the association across models without pre-specified levels. This may have helped avoid exposure misclassification. If there is any benefit among light drinkers, combining them with near-abstainers or with only occasional or heavier drinkers could mask the true relation between alcohol and CHD.

(4) There are problems in adjusting for intermediate variables in seeking to judge the effects of alcohol on CHD. For example, about 15% of studies adjusted for HDL cholesterol, considered to be the most important biological mechanism of alcohol’s effects on CHD. Other factors adjusted for in the analysis included diabetes and body mass index, both of which have been postulated as intermediary factors. It would be important to see the results both before and after adjustments for such factors.

(5) There are ample data to permit stratified analyses, such as according to geographic location, whether or not smoking was adequately adjusted for, and quality of alcohol intake data. A number of studies had information on the pattern of drinking (frequency, quantity, binge drinking, etc.), which has been demonstrated to markedly modify the health effects of alcohol intake. Combining all studies into a single analysis may be inadequate to answer the questions posed by the authors.

The very few studies that were not excluded for total mortality included two reports from one small study of African-Americans initially aged 18-23 and two reports from Japan. There was one paper of Americans aged 18 and older from the general US population (Rehm et al, 2001, Am J Epidemiol) and one of middle-aged adults from the UK (Wannamethee & Shaper, 1997, Int J Epidemiol). The largest study included was from a population survey in the US (Liao et al, 2000, Am J Epidemiol) in which the authors concluded: “Data from these representative US cohorts demonstrated that less than 2 drinks per day for men and less than 1 drink per day for women are associated with the lowest all-cause mortality.” For CHD mortality, there were only two studies included in the analysis. Apparently the authors have based their conclusions only on these few studies, which may or may not reflect the general situation. A more deductive meta-analytic approach, such as that used by Maclure (1993, Epdemiol Rev), may be a preferable approach to utilize the very large amount of data available on this topic.

(6) The analysis inappropriately used significance testing for inference, and has ignored quantification for possible biases, e.g., from misclassification or residual confounding.

In summary, while we welcome analyses on this topic, we are concerned that the above described problems in methodology, and especially limiting the number of studies included in the final analysis to so few, raise questions about the conclusions of the authors. We agree with the authors that there is a need for further analyses limiting the referent group to lifetime abstainers and with appropriate categorization of intake (especially including pattern of drinking) to provide better estimates of alcohol’s effects on CHD. There should soon be enough well-done studies available to permit such an analysis.

Just looking over the papers that our Institute has critiqued in the last few months, we find examples where neither type of “error” studied by Fillmore et al was present. For example, the recent report on a large group of older adults by Mukamal et al1 was able to separate lifetime abstainers from former drinkers, and separated occasional drinkers from regular light drinkers. These authors demonstrated reductions in the risk of a variety of CHD outcomes from moderate drinking. In another recent paper on older people by Tolvanen et al,2 data permitted the separation of ex-drinkers from lifetime abstainers. In their analyses total mortality was highest in the ex-drinkers and lifetime abstainers and 30-40% lower in current drinkers. In a very-large population study in California by Klatsky et al,3 information was available to identify lifetime abstainers, and the investigators were able to separate occasional drinkers from regular light drinkers. As in most other studies, these investigators showed that consumption of 1 to 2 drinks/day was associated with 40% less heart failure associated with coronary artery disease.

Lay Summary: The authors have made the assumption that inadequate control for ex-drinkers and for occasional drinkers (including them in the same category as lifetime abstainers) has led to bias in the vast majority of scientific papers showing less coronary disease (CHD) among moderate drinkers. We have found a number of problems in the methodology in this paper that leads us to question their conclusion that the reduced CHD rate among moderate drinkers is due to bias. The number of well-done studies that avoid the misclassification problems described by the authors is rapidly increasing, and recent such studies have shown almost uniformly that moderate drinking is associated with lower rates of CHD.


1. Mukamal KJ, Chung H, Jenny NS, Kuller LH, Longstreth Jr WT, Mittleman MA, Burke GL, Cushman M, Psaty BM, Siscovick DS. Alcohol consumption and risk of coronary heart disease in older adults: The Cardiovascular Health Study. J Am Geriatr Soc 2006;54:30–37.

2. Tolvanen E, Seppä K, Lintonen T, Paavilainen P, Jylhä. Old people, alcohol use and mortality. A ten-year prospective study. Aging Clin Exp Res 2005; 17:426-433.

3. Klatsky A, Chartier D, Udaltsova N, Gronningen S, Brar S, Friedman GD, Lundstrom RJ. Alcohol drinking and risk of hospitalization for heart failure with and without associated coronary artery disease. Am J Cardiol 2005; 96:346–351?

no website link
All text and images © 2003 Alcohol In Moderation.