Goulden R. Moderate Alcohol Consumption Is Not Associated with Reduced All-cause Mortality. Am J Med 2015;pre-publication. http://dx.doi.org/10.1016/j.amjmed.2015.10.013.
BACKGROUND: A large body of research suggests that light or moderate alcohol consumption is associated with reduced all-cause mortality. However, concerns remain that the observed relationship is due to selection bias, misclassification of ex-drinkers, or residual confounding.
METHODS: The association between alcohol consumption and all-cause mortality was analyzed using Cox regression. The analysis was performed using data from the Health and Retirement Study, a longitudinal cohort of 24,029 individuals from a nationally representative sample of US adults aged more than 50 years. Drinking level was based on alcohol consumption measured at 3 points over the 4 years before the start of follow-up. Occasional drinkers—those who reported drinking on at least 1 occasion, but always less than once per week—served as the reference category. There was extensive adjustment for sociodemographic variables, health status, and functional status.
RESULTS: During 206,966 person-years of follow up 7,902 individuals died. No level of regular alcohol consumption was associated with reduced all-cause mortality. The hazard ratio and 95% confidence interval in fully adjusted analyses was 1.02 (0.94-1.11) for <7 drinks/week, 1.14 (1.02-1.28) for 7 to <14 drinks/week, 1.13 (0.96-1.35) for 14 to <21 drinks/week, and 1.45 (1.16-1.81) for ≥21 drinks/ week.
CONCLUSIONS: Moderate alcohol consumption is not associated with reduced all-cause mortality in older adults. The previously observed association may have been due to residual confounding
The large majority of observational epidemiologic studies, including recent ones with excellent control of confounding, have shown that moderate consumers of alcohol have lower risk of total mortality. Much of such protection has been thought to relate to the inverse association between moderate alcohol intake and cardiovascular disease, an association supported by thousands of experiments in animals and limited randomized clinical trials in humans.
The present analysis does not support most previous studies, which led Forum members to seek to determine why the results of this study are different. The two main concerns about the paper and its interpretation were the author’s failure to consider the effects of under-reporting of alcohol intake when choosing the referent group, and including in the analyses adjustments for factors that have been shown to be mechanisms by which alcohol affects the risk of disease and mortality.
Specific comments by Forum members on the paper: Forum member Zhang stated: “I found that the approach that author took to adjust for confounders difficult to understand. From the description of the subjects, one can see that alcohol drinkers, especially light to moderate drinkers, had better risk profiles except for smoking. The majority of people do not begin to drink at or after age 60; thus, even though comorbidities were assessed at baseline, this does not mean adjusting for these cormorbidities is correct, i.e, it does not adjust for mediators. Further, age- and sex -adjusted hazard ratios were confounded by smoking, whereas no model appropriately adjusted for smoking effect because it is always added into a regression model with other comorbidities. It would be very interesting to see the effect of alcohol consumption on all-cause mortality if the model only adjusted for real confounders, i.e., age, sex, smoking, education, ethnicity, and perhaps exercise.”
Imputing missing data for alcohol consumption among subjects: Forum member De Gaetano and associate Augusto Castelnuovo had a major concern that the author inputted missing values for alcohol. “In our opinion a method for imputing missing data may only be acceptable for covariates, not for the exposure of interest. Subjects with missing values for alcohol should have been removed from analysis. In our meta-analyses on alcohol and health, we excluded abstainers who were former drinkers (Di Castelnuovo 2002, 2006)(Costanzo).” Other reviewers stated that as long as the percentage of subjects missing data on alcohol consumption was low, imputing their data would probably not change results very much. Zhang stated: “As far as the imputation of data for alcohol among subjects not providing data, it is also okay to input the exposure as well for a defined cohort. If missing is more than 20% we might worry about why the percentage is so high. So I think that the imputation approach is fine.” Forum members Djoussé and Barrett-Connor agreed that the imputation procedure used in this study was satisfactory. The latter commented: “I do believe the statistics are correct. Multiple imputation is ok, although not perfect. Perhaps the author could improve the analyses by repeating the analyses with and without multiple imputation and see how the results differ. He could still do this.”
Under-reporting of alcohol intake; using “occasional drinkers” as the referent group: Reviewer Ellison wondered “Is there a problem with ‘under-reporting,’ so that the ‘occasional drinkers’ are really the ‘light-to-moderate’ drinkers? As data on lifetime abstainers were also available, they could serve as alternate referent group. (The results provided by the author in the Appendix show that the relative risk of mortality for moderate drinkers, when compared with lifetime abstainers, are very similar to those of most previous studies: nondrinkers, even lifetime abstainers, show higher risk of mortality than what are referred to in this study as “occasional drinkers.”
Reviewer Djoussé commented: “Under-reporting of alcohol could lead to major misclassification where <1 drink/week is used for ‘occasional drinkers.’ The use of average intake from 3 time points is naïve, as the author assigned equal weights to reports over the three occasions of 0-0-1 vs. 1-0-0 (0=no alcohol, 1 =<1/week); in both scenarios, the average will be the same, but the author is not certain that a 0-0-1 is a person who is beginning to drink alcohol and may be consuming up to 1 or 2 drinks/day in the next few years. For cardiovascular disease, recent consumption is most important.”
Reviewer Skovenborg commented: “In general, comparison studies in the alcohol literature have shown that self-reported alcohol consumption accounts for only 40-60% of alcoholic beverages sold as measured by sales and tax data (Midanik); thus, in this as in all observational studies, there may well be considerable under-reporting of alcohol intake. Of more importance, there may also be selective under-reporting. Stockwell et al found alcohol consumption to be underestimated significantly more by low-risk more than by high-risk drinkers (76.25 ± 0.34% versus 49.22 ± 3.01%). Spirits consumption was underestimated by 65.94% compared with sales data, wine by 38.35%, and beer by 49.02%.”
Skovenborg continued: “No valid arguments have been published against the use of lifetime abstainers as the referent group with state-of-the-art control for possible confounders; however, you need a reasonable number of lifetime abstainers to form a referent group. For example, among the 1536 men in the Italian rural cohorts of the Seven Countries Study, there was only 38 nondrinkers (Farchi et al). In the present analysis, however, there were 8,427 nondrinkers, making up 38.9% of the cohort. Included in the Appendix is a table indicating that when nondrinkers made up the referent group, the “occasional drinkers” had a RR of 0.74 (CI 0.69-0.80) and subjects reporting < 7 drinks/week had a RR of 0.78 (CI 0.72-0.86). These risk ratios are almost exactly the same as those shown in most previous well-done cohort studies.
“A reported level of ≥0.5 – <6.0 g alcohol/day is a level for which a physiological effect of alcohol is not deemed plausible, but this level was associated with a significant increased risk of breast cancer-specific death in a German study (Vreiling et al). This finding suggests that underreporting of intake (and/or the effects of drinking pattern) could be playing a role in the results of the present study.”
Forum member Keil also commented on the choice of the referent group: “I think that one of the first investigators to use persons with minor alcohol intake (0.1-5.0 g/day) as a control group was Eric Rimm. He published a paper with data from the Health Professionals Study in 1991 in the Lancet, where he also used abstainers as a control group and compared the two analyses. When the alcohol consumption groups 5.1-10.0 g and higher up to > 50.0 g were compared with the abstainer group and separately with the minimal alcohol intake group, the results were practically identical. It is understandable that journals like to publish controversial results, but when a paper is at odds with more than thirty years of research findings, the editors of the journal should request better data quality and better reasoning.”
Adjusting for factors in the pathway of alcohol’s effects: There was particular concern that the author may have “thrown out the baby with the bath water” by adjusting for diabetes and coronary disease when relating alcohol intake to mortality. In other words, there is the possibility that the the author “over-corrected” for confounding, and adjusted for factors that are in the pathway of alcohol’s effects on mortality. This would attenuate (or even completely erase) an inverse association between alcohol consumption and death from coronary heart disease, which is the leading cause of total mortality in the USA.
Skovenborg cites an example: “In a German study (Holahan et al), moderate wine consumption was inversely related to the risk of coronary heart disease (CHD) after adjustment for multiple potential confounders; however, the inclusion of risk factors of CHD that might be affected by alcohol consumption [HDL cholesterol, apolipoproteins A-I and A-II, lipoprotein(a), and fibrinogen] in the multivariable models markedly attenuated the inverse association between alcohol consumption and CHD risk.” Studies by Keil et al and by Brenner et al also showed how inclusion of factors that are mechanisms for alcohol’s effect attenuates the estimated overall effect of alcohol.
Added Forum member Djoussé: “An inverse association between alcohol and CVD mortality is well justified by biologic mechanisms; further, I think that the author should have access to deaths from CVD. If this is true, I am not sure why he chose not to present such data. One could argue that accidents and natural disaster can cause deaths that are not related to alcohol or any other exposure, but including all deaths in one bag limits the interpretation of the data.”
Limitations to author’s Discussion: Forum member Barrett-Connor stated: “I was confused about the review of the world literature, which used different definitions of alcohol intake, amount and type of alcohol (very different in different countries and cultures and economies, and even religions). I believe we could criticize this paper/review based on differing definitions of alcohol and its use—the way people drink in the UK is different than the way people drink in the United States. Until about 10 years ago, people in the UK drank in the pub and not at home. This has almost completely changed in the last 10 -15 years.”
Barrett-Connor concluded: “Given the amazingly large sample and remarkably complete follow up, I cannot completely refute the findings from this study, even though they are not what I expected to see based on most previous research. We need to figure out how to distinguish between alcohol use and alcoholism, and this paper does not do it. That may be where all these disagreements are manifest.” Reviewer Lanzmann-Petithory added: “I agree with the remarks of others regarding misclassification due to likely underreporting, subjects with missing data on alcohol and those with binge drinking not being excluded, lack of consideration of type of beverage or pattern of drinking, unjustified choice of referent group, over-adjustment that annuls the correlation, very biased discussion and conclusions, etc. However, as the author shows in the Appendix, he has observed the same as everybody else when using nondrinkers as the referent group: a ‘J-shaped’ association between light-to-moderate drinking and mortality.”
References from Forum review
Brenner H, Rothenbacher D, Bode G, März W, Hoffmeister A, Koenig W. Coronary heart disease risk reduction in a predominantly beer-drinking population. Epidemiology 2001;12:390-395.
Costanzo S, Di Castelnuovo A, Donati MB, Iacoviello L, de Gaetano G. Wine, beer or spirit drinking in relation to fatal and non-fatal cardiovascular events: a meta-analysis. Eur J Epidmiol 2011;26:833-850. DOI 10.1007/s10654-011-9631-0
Di Castelnuovo A, Rotondo S, Iacoviello L, Donati MB, de Gaetano G. Meta-analysis of wine and beer consumption in relation to vascular risk. Circulation. 2002;105:2836–2844.
Di Castelnuovo A, Costanzo S, Bagnardi V, Donati MB, Iacoviello L, de Gaetano G. Alcohol dosing and total mortality in men and women: an updated meta-analysis of 34 prospective studies. Arch Intern Med 2006;166:2437–2445.
Farchi G, Fidanza F, Giampaoli S, Mariotti S, Menotti A. Alcohol and survival in the Italian rural cohorts of the Seven Countries Study. Int J Epidemiol 2000;29:667-671.
Holahan CJ, Schutte KK, Brennan PL, North RJ, Holahah CK, Moos BS, Moos RH. Wine consumption and 20-year mortality among late-life moderate drinkers. J Stud Alcohol Drug 2012;73:80–88.
Keil U, Chambless LE, Döring A, Filipiak B, Stieber J. The Relation of Alcohol Intake to Coronary Heart Disease and All-cause Mortality in a Beer-drinking Population. Epidemiology 1997;8:150–156.
Midanik L. The validity of self-reported alcohol consumption and alcohol problems: a literature review. Br J Addict 1982;77:357-382.
Rimm EB, Giovannucci EL, Willett WC, Colditz GA, Ascherio A, Rosner B, Stampfer MJ. Prospective study of alcohol consumption and risk of coronary disease in men. Lancet 1991;338(8765):464-468.
Stockwell T, Zhao J, Macdonald S. Who under-reports their alcohol consumption in telephone surveys and by how much? An application of the ‘yesterday method’ in a national Canadian substance use survey. Addiction 2014;109:1657-1666.
Vrieling A, Buck K, Heinz J, Obi N, Benner A, Flesch-Janys D, Chang-Claude J. Pre-diagnostic alcohol consumption and postmenopausal breast cancer survival: a prospective patient cohort study. Breast Cancer Res Treat 2012;136:195–207. DOI 10.1007/s10549-012-2230-2.
Most observational studies have found that moderate drinkers, in comparison with nondrinkers, tend to have lower risk of all-cause (total) mortality; this is probably related primarily to a reduction in the risk of cardiovascular disease, the leading cause of death among the elderly. This large study has conflicting findings, as the author claims that the present analyses do not demonstrate protection against mortality from light-to-moderate drinking. In this study, what were termed “occasional drinkers,” rather than nondrinkers, were used as the comparison group.
Forum members had two main concerns about this study that warrant an investigation of the author’s conclusions: there was no consideration of under-reporting of alcohol intake when declaring “occasional drinkers” as the referent group, and (2) the inclusion, and adjusting for as “confounders,” several factors that are actually mechanisms by which alcohol has been shown to reduce mortality.
The first concern could have led to many light drinkers being included in the referent group, and thus not evaluated for a potential protective effect of light drinking on mortality. In fact, presented only in the Appendix to the paper are data showing that when nondrinkers make up the referent group, consumers of 1-7 as well as 7-14 drinks per week show significant 20-30% reductions in the risk of mortality; these findings are very similar to those of most previous epidemiologic studies.
The second, and perhaps more important concern, is that some of the mechanisms by which moderate alcohol intake may lead to lower mortality, such as reducing the risk of diabetes and coronary heart disease, were “adjusted” for in the analyses. This would attenuate or even erase any true reduction in risk of mortality from moderate drinking.
Some Forum members also were concerned that some subjects were missing data on alcohol consumption but, rather than excluding them, an estimated value was imputed for them. Further, data on the pattern of drinking (regular moderate versus binge drinking) or on the type of beverage consumed were not included in the evaluation.
Forum members conclude that the results of this study will obviously be considered in conjunction with other scientific data when seeking to judge the relation of alcohol intake to mortality. However, concerns about the analysis raise questions about the conclusion of the author of no protective effect of alcohol on mortality, a finding that conflicts with the results of most previous studies.
* * * * * *
Comments on this critique by the International Scientific Forum on Alcohol Research were provided by the following members:
Elizabeth Barrett-Connor, MD, Distinguished Professor, Division of Epidemiology, Department of Family Medicine and Public Health and Department of Medicine, University of California, San Diego, La Jolla, CA USA
Giovanni de Gaetano, MD, PhD, Department of Epidemiology and Prevention, IRCCS Istituto Neurologico Mediterraneo NEUROMED, Pozzilli, Italy
Luc Djoussé, MD, DSc, Dept. of Medicine, Division of Aging, Brigham & Women’s Hospital and Harvard Medical School, Boston, MA, USA
R. Curtis Ellison, MD, Professor of Medicine & Public Health, Boston University School of Medicine, Boston, MA, USA
Harvey Finkel, MD, Hematology/Oncology, Boston University Medical Center, Boston, MA, USA
Ulrich Keil, MD, PhD, Professor Emeritus, Institute of Epidemiology & Social Medicine, University of Muenster, Germany
Dominique Lanzmann-Petithory,MD, PhD, Nutrition/Cardiology, Praticien Hospitalier Hôpital Emile Roux, Paris, France
Erik Skovenborg, MD, specialized in family medicine, member of the Scandinavian Medical Alcohol Board, Aarhus, Denmark
David Van Velden, MD, Dept. of Pathology, Stellenbosch University, Stellenbosch, South Africa
Yuqing Zhang, MD, DSc, Clinical Epidemiology, Boston University School of Medicine, Boston, MA, USA
In addition, a comment was invited from Professor Augusto Di Castelnuovo, Department of Epidemiology and Prevention, IRCCS Istituto Neurologico Mediterraneo NEUROMED, Italy.