Critique 132: Pattern of alcohol consumption and cause of death in a large European prospective study — 21 January 2014
Bergmann MM, Rehm J, Klipstein-Grobusch K, et al (38 authors). The association of pattern of lifetime alcohol use and cause of death in the European Prospective Investigation into Cancer and Nutrition (EPIC) study. Int J Epidemiol 2013;42:1772-1790
Banks E. Commentary: Lifetime alcohol consumption and mortality: have some, but not too much. Int J Epidemiol 2013;42:1790–1792; doi:10.1093/ije/dyt218
Stockwell T, Chikritzhs T. Commentary: Another serious challenge to the hypothesis that moderate drinking is good for health? Int J Epidemiol 2013;42:1792–1794; doi:10.1093/ije/dyt217
Background There is limited evidence for an association between the pattern of lifetime alcohol use and cause-specific risk of death.
Methods Multivariable hazard ratios were estimated for different causes of death according to patterns of lifetime alcohol consumption using a competing risks approach: 111 953 men and 268 442 women from eight countries participating in the European Prospective Investigation into Cancer and Nutrition (EPIC) study were included. Self-reported alcohol consumption at ages 20, 30, 40 or 50 years and at enrolment were used for the analysis; 26 411 deaths were observed during an average of 12.6 years of follow-up.
Results The association between lifetime alcohol use and death from cardiovascular diseases was different from the association seen for alcohol-related cancers, digestive, respiratory, external and other causes. Heavy users (> 5 drinks/day for men and > 2.5 drinks/day for women), regardless of time of cessation, had a 2- to 5-times higher risk of dying due to alcohol-related cancers, compared with subjects with lifetime light use (≤ 1 and ≤0.5 drink/week for men and women, respectively). Compared with lifetime light users, men who used < 5 drinks/day throughout their lifetime had a 24% lower cardiovascular disease mortality (95% confidence interval 2-41). The risk of death from coronary heart disease was also found to be 34–46% lower among women who were moderate to occasionally heavy alcohol users compared with light users. However, this relationship was only evident among men and women who had no chronic disease at enrolment.
Conclusions Limiting alcohol use throughout life is associated with a lower risk of death, largely due to cardiovascular disease but also other causes. However, the potential health benefits of alcohol use are difficult to establish due to the possibility of selection bias and competing risks related to diseases occurring later in life.
These analyses are from the very large prospective EPIC study that apparently collected alcohol intake only on one occasion by questionnaire; at the time, subjects were asked to retrospectively also provide an estimate of their alcohol intake at 20, 30, 40, and 50 years of age. While data on type of beverages consumed was collected in this study, this paper does not give beverage-specific results. The average follow-up for death was 12.6 years. Causes of death were ascertained from a variety of sources, but the specific causes given could not be validated.
As for the assessments of alcohol consumption, Reviewer Thelle comments that “The assessment of alcohol intake in the EPIC study is neither worse nor better than that of other studies; but, because of its size, the influence of random misclassification is likely to be smaller.” He believes that the authors are presenting a balanced view and have done what is possible to account for confounding; but it is still there. Thelle continues: “The EPIC is a heterogeneous study with different recruitment procedures, and some countries have been excluded in this analysis. Whether this could have influenced the results is debatable, as heavy drinking is more prevalent in some of the excluded populations, e.g., Norway and Sweden, with a large number of deaths.”
Reviewer Ellison stated that while the analyses are well done, it is disturbing that only a single notation in the text is made on the effects of alcohol consumption on total mortality. A figure showing effects of alcohol on total mortality is provided only in the supplemental material on the internet. It shows a striking J-shaped curve, with lower mortality for men who consume alcohol than for abstainers, up to about 45 g/d (about 4 “typical drinks”). For women, the estimated risk for women drinkers never goes above the risk level of non-drinkers, indicating a greater risk of total morality for abstainers than for women who consume alcohol at all levels. Given the known problems with ascertaining disease-specific causes of death (which was determined by a variety of methods in this study) and the inability to validate such diagnoses in the present study, it would seem appropriate to include total mortality data in the main paper.
Reviewer Finkel agrees that the authors should have reported total mortality effects of alcohol (given that assessments of total mortality are equally valid in all cohorts). He was also concerned that the investigators relied on the long-term memory of subjects to determine how much subjects drank in the past, then used that as hard data to support their conclusions.
Forum member Thelle comments: “The lack of end-point verification is difficult to assess. Modern cancer treatment is associated with an increased risk of CVD. Add to this the tendency to ascribe death of otherwise unknown cause in a cancer patient to the cancer. A number of these patients may well have died from unrecognized heart disease.” Thelle is less concerned about the lack of emphasis of the authors on total mortalithy, but points out that access to health-care systems may be skewed, with people who are better off getting the best care, and these are the people more likely to be light to moderate drinkers. The authors have attempted to adjust for this by including education and other factors, but residual confounding remains as a real concern.
Forum member Skovenborg was less concerned about the methodology. He did state that “the EPIC Study is prone to information bias regarding alcohol consumption, as are all population studies. This is an inborn error of cohort studies and that is why you should treat all results from cohort studies with timely caution.” Skovenborg continues: “However, the definition of categories of alcohol use in this paper is inconsistent, using the ‘recommended upper limits of daily alcohol consumption’ of one drink/day for women and two drinks /day for men in one part of the text, then categorizing heavy alcohol use as >5 drinks/day for men and >2.5 drinks/day for women elsewhere. There is also an inconsistency between the title of the paper, referring to the ‘pattern of lifetime alcohol use,’ and the fact that no information was available regarding binge drinking, frequency of consumption, or whether beverages were consumed with meals.” Thus, the paper is missing some of the most important aspects of drinking pattern.
Forum member Djoussé had several concerns: “There may well be misclassification of alcohol reporting in the data; especially, with an average age of 50 years at enrollment, people may not necessarily recall how much they drank when they were 20 and 30 years of age (information used to construct lifetime drinking patterns).” He was also concerned that in the key figures in the paper, “moderate drinking” is defined as up to 60 g/day for men and up to 30 g/day for women; “These upper limits exceed what most consider to be ‘moderate drinking.’ Further, there is a lack of adjustment for comorbidity, and when controlled for, the authors relied on self-reported information that was restricted to a few chronic conditions. Stratified analysis on one factor at a time does not replace simultaneous adjustment for comorbidity. Hence, misclassification and confounding by indication are possible threats to validity of the reported findings.”
Forum member Zhang had two major concerns about this paper: “I am not quite sure why the authors focus on ‘lifetime alcohol use’ (even though what they calculate is not a good estimate of lifetime alcohol consumption at all) rather than estimates of current use obtained at enrolment for the previous 12 months. From the data presented, it is very obvious that light-moderate alcohol consumption at enrolment was associated with a lower risk of all cause-specific mortality in men and most cause-specific mortality in women (except death from the digestive system, which is a little odd, and death from external causes).”
Zhang also pointed out a statement in the paper that “However, these lower relative risks applied only to those study participants considered free of disease at time of enrolment into the study,” implying that such an association was not observed among subjects who had chronic diseases (e.g., cancer, myocardial infarction, stroke, diabetes mellitus, or hypertension). “These findings are not surprising at all. It is unclear how many subjects suffered from each of these four diseases at enrolment, but including such subjects could raise the possibility of what is known as ‘collider stratification bias.’ As an example, and assuming that light-to-moderate alcohol consumption is associated with a lower risk of MI, including subjects who had already had a MI would tend to dilute the effect of moderate drinking on the risk of death among subjects with MI if there had been alcohol consumption before the occurrence of an MI.”
Forum member Lanzmann-Petithory was worried that the authors were very selective in the references they decided to include, giving inadequate attention to the many studies showing the considerable cardiovascular effects of moderate drinking. Indeed, the adverse effects of alcohol on cancer seem to be emphasized much more than the effects on reducing cardiovascular disease. Reviewer Van Velden stated: “In Afrikaans we say: “Van weë die bos, sien jy die bome nie! (Because of the dense forest, you do not see the trees!).”
“Political” or “ideological” rather than “scientific” arguments: Several Forum reviewers reading this paper were reminded of previous instances in which it appeared that conclusions were based more on “political” or “ideological” rather than “scientific” arguments. A notable example of this problem, from the 1970s, was the suppression by the National Institutes of Health of solid data from the Framingham Heart Study showing beneficial health effects of moderate drinking (Seltzer CC. “Conflicts of Interest” and “Political Science.” J Clin Epdemiol 1997;50:627-629). As described by Seltzer, the NIH would not approve publication of the report until “all references to alcohol had been removed,” arguing that a finding of lower risk of coronary disease among drinkers would be “socially undesirable in view of the major health problem with alcoholism that already exists in this country.” The instructions from NIH at that time added that the article should maintain “the conclusion of no significant relationship of alcohol to the incidence of coronary artery disease.”
Reviewer Mattivi stated: “After a close look at the heavily incomplete choice of the literature cited by the authors, and at the tendency to highlight only some aspects while obscuring others, this seems to me a paper that could be read starting from the conclusions and backward. Since this is possibly the way it was conceived.” Forum member Goldfinger said: “It is interesting that the authors of this paper conclude that limiting alcohol intake has salutary effects rather, than coming out to say that moderate drinking has salutary effects. They make the assumption, thus, that man has the natural tendency to excess and alcoholism.”
Reviewer Keil had strong reservations about the conclusions of the authors. He noted the statement in the paper: “The apparent health benefits of low to moderate alcohol use found in observational studies could therefore in large part be due to various selection biases and competing risks, which are related to both lifetime alcohol use and risk of disease, usually occurring later in life.” Keil states: “In my opinion it is not honest to repeat this argument of bias and confounding and competing risks again and again, because over the decades most of these contentions have been refuted many times; for quite a while, we know that a protective effect of alcohol on the cardiovascular system and on all-cause mortality is very probable.”
Keil suggests that the authors should re-read the excellent paper by Maclure from 1993 (Maclure M. Demonstration of deductive meta-analysis: ethanol intake and risk of myocardial infarction. Epidemiologic Reviews 1993;15:328-351). He states: “I think it was the epidemiologist Lewis Kuller who coined the term ‘circular reasoning,’ meaning that some people come up again and again with the same arguments, and never examine data that may refute them. We get more and more papers telling us that the protective effect of alcohol on CVD and all-cause mortality is cofounded and biased, but these statements are not backed by good intellectual reasoning, but by strong ideology.”
Reviewer Skovenborg agrees: “The selection of evidence to support a pre-determined view — cherry-picking — is not science; it is politics. Many of us would prefer to see politics move in the evidence-based direction of science, and not science move in the direction of politics. Scientists are also interested parties; they stand to gain from policy taking one direction rather than another, and the virus of ‘noble-cause corruption’ has infected the mind of many scientists today backed by strong ideology.” Forum member James adds: “In this paper there is evidence of bias contained both in the mode of presentation of the results and in the discussion.” He adds: “The accompanying opinion piece by the Australian epidemiologist Banks, while not admissible as ‘new evidence,’ certainly makes the point I feel should be considered.”
Forum member Waterhouse was even more concerned about the interpretation of their data by the authors of the present paper: “This ideological attack on science should be addressed. I would suggest that one approach is to organize a team of epidemiologists and applying consistent analytical techniques (e.g., as demonstrated by the Maclure study) to all recently published meta-analyses. In other words, a critical review of meta-analyses to reveal any bias in data selection, interpretation, etc. It might be useful to include an analysis of funding sources to reveal whether or not funding source is related to the outcome, as I am always surprised that investigators funded by organizations with a clear anti-alcohol bias always state that they have “no conflict.’”
Commentaries on the present paper published in the same issue of the journal: It is interesting that the two attached Commentaries came up with divergent opinions. Banks concluded: “For the age group included in the EPIC study (i.e., older adults), accounting for prevalent illness, the risks of death were lowest in men with lifetime patterns of alcohol consumption of >2–24 g/day and consumption at enrolment of 15–20 g/day; corresponding levels for women were >1–12 g/day and around 10 g/day. The evidence indicates that it is advisable to avoid heavy drinking throughout life. If taken as causal, these findings are consistent with most public health advice about alcohol, except that most advice recommends an upper limit to alcohol consumption, but does not actually encourage drinking. In fact, the evidence goes further than this and indicates that, in later life, on average and bearing in mind the priorities and risks of specific individuals, drinking at least some alcohol, but not too much, is likely to minimize the overall risk of death.”
On the other hand, the commentary by Stockwell and Chikritzhs concludes: “Bergmann et al have already taken some important steps such as removing former drinkers from the lifetime abstainer group and, unusually, also presenting results separately for individuals who previously drank at a low versus higher levels. Even former low-level drinkers had elevated risks of death for most outcomes compared with lifetime abstainers.” They continue: “Bergmann et al’s use of a reference group of occasional drinkers also partly avoids problems with former drinker bias. Unlike lifetime abstainers, occasional drinkers should not be biased towards poorer health, nor will they be drinking enough to receive any theoretical health benefits (e.g., drinking less than one drink per week). However, beyond these considerations, much further thought needs to be given to the problem of systematic bias in longitudinal research due to, for example, the inclusion or exclusion of former drinkers, occasional drinkers and individuals in poor health at baseline. Until such time as these issues are resolved ‘a healthy dose of scepticism’ is warranted for the hypothesis that light/moderate alcohol consumption is beneficial to health.”
The arguments in this latter commentary are difficult to follow, as the authors point out a number of ways Bergmann et al have dealt with possible confounding (excluding former drinkers, using occasional drinkers as the reference group, dealing with poor health, etc.), yet end up with the old argument (largely discredited by most recent papers, and by data in the current paper) about bias in longitudinal research. As stated by one Forum member, “The authors of this commentary acknowledge that light to moderate consumption of alcohol is associated with an apparent lower risk of death, and their statement that the question of causality is related to the possible effects of selection bias, confounding and competing risks is hardly controversial. However, applying the bias and confounding challenges quoted by Stockwell and Chikritzhs in their Commentary to other lifestyle factors would also nullify the effects of factors such as exercise, a healthy diet, keeping a normal weight, etc.”
A large group of investigators participating in the European Prospective Investigation into Cancer and Nutrition (EPIC) study have reported on the association of alcohol consumption with disease-specific mortality over a 12-year period among a very large number of men and women. Alcohol intake over the past year (as well as estimates of earlier intake) was reported at baseline. A variety of approaches were used to identify deaths in the participating cohorts, but validation of the cause of death was not possible. No data were reported on certain key aspects of drinking pattern (frequency of alcohol intake, binge drinking, with meals, etc.) and beverage-specific effects were not reported.
The authors report that the risk of death from cardiovascular diseases was lower, and from cancer and certain other causes of death were higher, among drinkers than among their referent group of “lifetime light users” (men reporting ≤1 drink/week and women reporting ≤ 0.5 drinks/week). However, their graphs of alcohol intake and risk of total mortality (not included in the paper but in supplementary data they provide on the internet) show strong J-shaped curves for increasing alcohol intake for both men and women. For men, the risk among alcohol consumers of up to about 48 gr/day (the equivalent of approximately four “typical drinks”) was lower than that of the lifetime light users that made up the referent group; at higher intake, the risk increased above that of light drinkers. For women, the risk of total mortality for drinkers remained lower than that of the referent group at all reported levels of intake. Despite these findings, the entire paper focuses only on the increased risk of death from alcohol for certain diseases, and almost completely ignores the net, overall effects on total mortality.
Forum reviewers considered that while the analyses were done correctly, there were major weaknesses in the estimation of “lifetime alcohol intake,” no data on the pattern of intake, no validation of coexisting diseases and, especially, no validation of the specific cause of death, even though this was the primary outcome of the study. All of these factors weaken the strong assertions made by the authors in their conclusions.
Forum members were concerned that the authors seemed to obscure the total effects of light-to-moderate drinking (lower risk of all-cause mortality), and emphasize only the harmful effects. Some reviewers considered that the presentation appeared to be based more on ideology than on an unbiased assessment of the data, and one suggested that the paper appears to have been written starting from pre-conceived conclusions, then finding data to support them.
It is interesting that two Commentaries published with this article came to divergent views. One by Stockwell and Chikritzhs emphasized potential bias and confounding that cause concern when relating alcohol to mortality in observational studies. While they then pointed out how well the present authors adjusted for many of these factors, they nevertheless concluded their commentary repeating how such factors are difficult to control and stating that “ . . . a healthy dose of skepticism is warranted for the hypothesis that light/moderate alcohol consumption is beneficial to health.” (They do not mention that data on alcohol and mortality are far more robust that data supporting generally accepted beliefs of the beneficial effects of physical exercise, eating a healthy diet, maintaining a lean body mass, etc.)
On the other hand, the commentary on the present paper by Banks considered the data presented quite sound, and concluded: “If taken as causal, these findings are consistent with most public health advice about alcohol, except that most advice recommends an upper limit to alcohol consumption, but does not actually encourage drinking. In fact, the evidence goes further than this and indicates that, in later life, on average and bearing in mind the priorities and risks of specific individuals, drinking at least some alcohol, but not too much, is likely to minimize the overall risk of death.”
* * *
Comments on this review by the International Scientific Forum on Alcohol Research were provided by the following members:
Luc Djoussé, MD, DSc, Dept. of Medicine, Division of Aging, Brigham & Women’s Hospital and Harvard Medical School, Boston, MA, USA
Dag S. Thelle, MD, PhD, Senior Professor of Cardiovascular Epidemiology and Prevention, University of Gothenburg, Sweden; Senior Professor of Quantitative Medicine at the University of Oslo, Norway
Fulvio Ursini, MD, Dept. of Biological Chemistry, University of Padova, Padova, Italy
Harvey Finkel, MD, Hematology/Oncology, Boston University Medical Center, Boston, MA, USA
Dominique Lanzmann-Petithory,MD, PhD, Nutrition/Cardiology, Praticien Hospitalier Hôpital Emile Roux, Paris, France
Yuqing Zhang, MD, DSc, Epidemiology, Boston University School of Medicine, Boston, MA, USA
David Van Velden, MD, Dept. of Pathology, Stellenbosch University, Stellenbosch, South Africa
Erik Skovenborg, MD, Scandinavian Medical Alcohol Board, Practitioner, Aarhus, Denmark
Arduino A. Mangoni, PhD, Strategic Professor of Clinical Pharmacology and Senior Consultant in Clinical Pharmacology and Internal Medicine, Department of Clinical Pharmacology, Flinders University, Bedford Park, SA; Australia
Fulvio Mattivi, PhD, Head of the Department Good Quality and Nutrition, Research and Innovation Centre, Foundazione Edmund Mach, in San Michele all’Adige, Italy
Tedd Goldfinger, DO, FACC, Desert Cardiology of
Ulrich Keil, MD, PhD, Institute of Epidemiology and Social Medicine, University of Münster, Münster, Germany
Arne Svilaas, MD, PhD, general practice and lipidology, Oslo University Hospital, Oslo, Norway
Oliver James, MD, formerly Head of Medicine, University of Newcastle, UK; now Emeritus Professor of Hepatology, University of Newcastle, UK
R. Curtis Ellison, MD, Section of Preventive Medicine & Epidemiology, Boston University School of Medicine, Boston, MA, USA