Critique 141: A world-wide study of alcohol consumption and the risk of myocardial infarction — 3 July 2014
Leong DP, Smyth A. Teo KK, McKee M, Rangarajan S, Pais P, Liu L, Yusuf S, on behalf of the INTERHEART investigators. Patterns of Alcohol Consumption and Myocardial Infarction Risk: Observations from 52 Countries in the INTERHEART Case-Control Study. Circulation 2014 (pre-publication) DOI: 10.1161/CIRCULATIONAHA.113.007627.
Background—While moderate alcohol use is associated with protection against myocardial infarction (MI), it is not known whether this effect is generalizable to populations worldwide. It s also uncertain whether differences in the pattern of alcohol use (and in particular heavy episodic consumption) between different regions negates any beneficial effect.
Methods and Results—We included 12,195 cases of first MI and 15,583 age- and sex-matched controls from 52 countries. Current alcohol use was associated with a reduced risk of MI (compared to non-users, adjusted odds ratio 0.87; 95% CI 0.80-0.94, p=0.001), however the strength of this assocation was not uniform across different regions (region-alcohol interaction p<0.001). Heavy episodic drinking (≥6 drinks) within the preceding 24 hours was associated with an increased risk of MI (odds ratio 1.4; 95% CI 1.1-1.9, p=0.01). This risk was particularly elevated in older individuals (for age >65 years, odds ratio 5.3; 95% CI 1.6-18, p=0.008).
Conclusions—In most participants, low levels of alcohol use are associated with a moderate reduction in the risk of MI, however the strength of this association may not be uniform across different countries. An episode of heavy drinking is associated with an increased risk of acute MI in the subsequent 24 hours, particularly in older individuals.
The authors report on data collected in the INTERHEART Study, a large international collaborative project of 12,461 individuals with a first myocardial infarction (MI) and 14,637 age- and sex-matched controls from 52 countries in Asia, Europe, the Middle East, Africa, Australia, North and South America. Included were nations with very divergent lifestyles, religions, degrees of development, and drinking habits. The authors attempted to use multivariable analysis to provide a “world-wide” overview of the relation of alcohol consumption to the risk of MI, using a case-control design. They also attempted to determine how alcohol consumption shortly prior to (in the 24 hours before) the occurrence of a MI related to the short-term risk of MI (by comparing alcohol intake in that period with the reported intake 24-48 hours prior to the MI, using a case-crossover design.
The title says “patterns” of use, yet the investigators really have no data on drinking patterns of individuals (e.g., regular moderate drinking versus binge drinking), as they state (“Other limitations were that the quantity and type of alcohol consumed by participants were not recorded.”). Unfortunately, they listed anyone who had one or more drinks within the past 12 months as an “alcohol user” in most comparisons (“Alcohol use was defined as the consumption of ≥ 1 alcoholic beverage within the previous 12 months.”). Thus, key results are only a comparison of “any alcohol use” versus “none.” Further, when they do present some results by frequency of drinking, they are able to report the highest alcohol frequency group only as drinking “> 4x/week,” as they had no information on the amount of alcohol consumed. Thus, in addition to using a case-control analysis, which prevents determination of a causal relation, the combination of data from such divergent populations to judge the overall effect of alcohol on risk of MI raises serious problems.
Forum member Lanzmann-Petithory was concerned about the absence of data on amount or type of alcohol consumed. As stated by the authors: “Other limitations were that the quantity and type of alcohol consumed by participants were not recorded.” She comments: “In my opinion, the absence of discrimination of type of alcohol, especially wine, could be an important confounding factor of the study and explain part of the differences between the countries.”
One aspect of the analyses in this paper was to judge the effects of heavy drinking in the 24 hours before a MI. The authors compared reported drinking in the 24 hours before the MI with drinking in the 24-48 hour period before the MI, in a case-crossover approach, which is probably reasonable. They report an increased risk of MI for heavy drinking in the 24 hours preceding the MI; they state: “There was no excess risk of MI if any alcohol was consumed within the hazard period (odds ratio 1.0; 95% CI 0.91-1.2, p=0.7). In contrast to the neutral effect of any alcohol consumption during the hazard period, heavy drinking (≥6 drinks) during the hazard period was associated with a significant elevation in the risk of MI (OR 1.4; 95% CI 1.1-1.9, p=0.01).” Thus, the data reported in this paper indicate a marked increase in the short-term risk of MI from heavy drinking, especially for older subjects and subjects from South Asia. It is unclear what to make of these strong findings, as some previous studies have shown an opposite effect. Even the authors have questions about their conclusions, as they state: “Thus there remains uncertainty over the risk of MI in the period immediately following alcohol intake.”
Reviewer Lanzmann-Petithory points out that changes in platelet function in the 24-48 hours after binge drinking could play an important role in the occurrence of a subsequent MI. Such changes have been well described by Renaud, Ruf, and their associates (Renaud SC, Beswick AD, Fehily AM, Sharp DS, Elwood PC. Alcohol and platelet aggregation: the Caerphilly Prospective Heart Disease Study. Am J Clin Nutr 1992;55:1012-1017)(Ruf JC, Berger JL, Renaud SC. Platelet rebound effect of alcohol withdrawal and wine drinking in rats. Relation to tannins and lipid peroxidation. Arterioscleros Thromb Vasc Biol 1995;15:140-144). These effects are not discussed in the present paper.
Reviewer Zhang had some other problems with the paper: “Adjusting for HDL-cholesterol is problematic, as HDL is one of the mechanisms by which alcohol affects the risk of MI. Further, if the study was to assess risk factors for MI, one of the strongest risk factors, smoking, was measured rather crudely. For their estimates of effects of alcohol just prior to a MI, the use of alcohol consumption in the 25-48 hour period before the MI as the ‘control period’ is fine if we assume that the triggering effect of alcohol on MI is short, say less than 24 hours.”
Forum member Djoussé commented: “Alcohol assessment is not well described, as reporting only the times per week it was consumed does not say anything about the quantity of alcohol consumed. A subject who reported alcohol intake of, say, 2 times per week could be consuming only 2 drinks per week or 12 or more. The paper does not seem to differentiate those habits. This will lead to misclassification of exposure, and no amount of adjustment can help.” Djoussé also commented on the statistical approach used. “Despite a large number of cases (12,195), the number of categories considered with all variables included in their statistical models is very large. This might have introduced a sparse data bias (as described by Rothman & Mosquin, Ann Epi 2013;23:43-48). For example, 2 alcohol categories * 10 geographic regions * 3 age groups *18 other variables *2 (if all of them were dichotomized) gives approximately 2,160 cells to fill; this suggests cells with an average of 5 cases/cell, assuming equal distribution (we know that will not happen in real life).”
Difficulties in combining data from such diverse populations: Only 1% of the non-drinkers in this study came from W. Europe and 1% from North America, while more than 70% of the non-drinkers came from the Middle East or Asia. With such diversity (e.g., almost all the abstainers were from one area), it is not possible to adequately “adjust” for such regional differences in statistical analysis. The authors realized that they were attempting to answer questions about alcohol and MI using very divergent populations; hence, they warn against using their results to make overall conclusions about the association. They state: “The associations observed between alcohol use and MI may be accounted for by unmeasured confounders such as genetic differences between populations, variation in alcohol type or preparation, and heterogeneity of social context. The present analysis should therefore prompt further research to clarify the nature of the association between alcohol use and MI.”
They admit many other limitations of their study as well, in one place stating: “In addition, as with any survey of alcohol use, there is likely to be a social desirability bias; respondents in countries where alcohol use is taboo may be more likely to underestimate their consumption compared with countries where drinking is socially acceptable. Recall bias, in which there is differential recollection of alcohol use depending on case or control status, may also have influenced our findings. Selection bias may have resulted from the approach used to identify controls since these participants represent a population with contact with a hospital.” Unfortunately, the conclusions presented in the text and in the abstract do not mention such problems.
Contributions of this analysis to our understanding of the relation of alcohol consumption to MI: Stated Forum member Keil: “This paper is based on a case-control analysis, with all the well-known weaknesses of this study type. Such studies are useful for rare conditions, when they may be the only study type possible for gaining knowledge. In the field of research into the etiology of cardiovascular diseases, however, we have an abundance of large, high quality, prospective studies. Our current knowledge on alcohol consumption and health has been derived from these large prospective cohort studies, which not always but quite often have provided detailed and even repeated assessments of alcohol exposure data. Why then do these authors fall back in quality and try to answer questions on alcohol consumption and MI using an inappropriate data set, when these questions have already been answered by analysing data sets with much higher quality and much less proneness to bias and confounding?”
Reviewer Finkel commented: “The serious liabilities in the methodology and reasoning of this paper make it impossible to use its results to improve our knowledge about the health benefits and adversities of drinking, particularly with regard to myocardial infarction.”
Reviewer Waterhouse stated: “In the discussion of the results, the authors admit to major shortcomings in their data and clearly state that these shortcomings limit their ability to reach definitive conclusions, but these doubts disappear in the abstract and conclusions. Thus the public and press, unable to digest the body of the paper, are left with a very misleading impression, and will not be aware that the conclusions trumpeted in the abstract are truly compromised by weak data.”
The authors report on data collected in the INTERHEART Study, a large international collaborative project of individuals with a first myocardial infarction (MI) and age- and sex-matched controls from 52 countries in Asia, Europe, the Middle East, Africa, Australia, North and South America. Included were nations with very divergent lifestyles, religions, degrees of development, and drinking habits. The authors attempted to provide a “world-wide” overview of the relation of alcohol consumption to the risk of myocardial infarction (MI). However, they were forced to use a case-control design, a type of study with many opportunities for bias and one that prevents determination of a causal relation. Further, the lack of detailed data on the amount of alcohol consumed by individuals weakens their results.
While the authors state that their study shows that low levels of alcohol use are associated with a moderate reduction in the risk of MI, they point out differences in results in different countries. Forum members considered that these large differences severely limit their ability to use such a dataset to provide overall conclusions on alcohol and MI. For example, only 1% of the non-drinkers in this study came from W. Europe and 1% from North America, while more than 70% of the non-drinkers came from the Middle East or Asia. With such diversity (e.g., almost all the abstainers were from one area), it is not possible to adequately “adjust” for such regional differences in analysis.
The authors also studied how alcohol consumption shortly prior to (in the 24 hours before) the occurrence of a MI relate to the short-term risk of MI. While their results show an increased risk for heavy drinking immediately prior to an MI (a result that differs from that of some previous studies), they admit analytic problems, stating: “Thus there remains uncertainty over the risk of MI in the period immediately following alcohol intake.”
Previous large prospective cohort studies (a type of epidemiologic study that decreases the risk of recall bias regarding alcohol intake and certain types of confounding) have clearly shown an inverse relation between moderate alcohol consumption and MI. Most of the previous studies have provided much more detailed data on alcohol exposure than did the present study. The authors realized many of these problems, and warn against using their results to make overall conclusions about the association between alcohol and MI from their paper. They state: “The associations observed between alcohol use and MI may be accounted for by unmeasured confounders such as genetic differences between populations, variation in alcohol type or preparation, and heterogeneity of social context. The present analysis should therefore prompt further research to clarify the nature of the association between alcohol use and MI.” Unfortunately, the conclusions presented in the abstract and in the text do not mention such problems.
Some Forum members worried that many readers of the paper will be left with a very misleading impression of the relation of alcohol consumption to myocardial infarction, and will not be aware that the conclusions trumpeted in the abstract are compromised by weak data. Overall, members considered that the present paper adds little to our knowledge about the risks and benefits of alcohol in relation to cardiovascular disease.
* * *
Comments on this critique 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
Ulrich Keil, MD, PhD, Institute of Epidemiology and Social Medicine, University of Münster, Münster, Germany
Yuqing Zhang, MD, DSc, Epidemiology, Boston University School of Medicine, 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
Harvey Finkel, MD, Hematology/Oncology,
Erik Skovenborg, MD, Scandinavian Medical Alcohol Board, Practitioner, Aarhus, Denmark
Dominique Lanzmann-Petithory,MD, PhD, Nutrition/Cardiology, Praticien Hospitalier Hôpital Emile Roux, Paris, France
R. Curtis Ellison, MD, Section of Preventive Medicine & Epidemiology, Boston University School of Medicine, Boston, MA, USA