Critique 220: An unusual analysis relating alcohol intake to mortality – October 16, 2018

Hartz SM, Oehlert M, Horton AC, Grucza RA, Fisher SL, et al. Daily Drinking Is Associated with Increased Mortality. Alcoholism: Clin & Experimental Research 2018; pre-publication. DOI: 10.1111/acer.13886.

Authors’ Abstract
Background: There is evidence that low-level alcohol use, drinking 1 to 2 drinks on occasion, is protective for cardiovascular disease, but increases the risk of cancer. Synthesizing the overall impact of low-level alcohol use on health is therefore complex. The objective of this paper was to examine the association between frequency of low-level drinking and mortality.
Methods: Two data sets with self-reported alcohol use and mortality follow-up were analyzed: 340,668 individuals from the National Health Interview Survey (NHIS) and 93,653 individuals from the Veterans Health Administration (VA) outpatient medical records. Survival analyses were conducted to evaluate the association between low-level drinking frequency and mortality.
Results: The minimum risk drinking frequency among those who drink 1 to 2 drinks per occasion was found to be 3.2 times weekly in the NHIS data, based on a continuous measure of drinking frequency, and 2 to 3 times weekly in the VA data. Relative to these individuals with minimum risk, individuals who drink 7 times weekly had an adjusted hazard ratio (HR) of all-cause mortality of 1.23 (p < 0.0001) in the NHIS data, and individuals who drink 4 to 7 times weekly in the VA data also had an adjusted HR of 1.23 (p = 0.01). Secondary analyses in the NHIS data showed that the minimum risk was drinking 4 times weekly for cardiovascular mortality, and drinking monthly or less for cancer mortality. The associations were consistent in stratified analyses of men, women, and never smokers.
Conclusions: The minimum risk of low-level drinking frequency for all-cause mortality appears to be approximately 3 occasions weekly. The robustness of this finding is highlighted in 2 distinctly different data sets: a large epidemiological data set and a data set of veterans sampled from an outpatient clinic. Daily drinking, even at low levels, is detrimental to one’s health.

Forum Comments
The perceived implications of the authors of the reported results of this study, including their title, raised concerns among most Forum reviewers. The reported conclusions of these authors are completely different from the results of most well-done large cohort studies: regular light to moderate consumers of alcohol who do not binge drink are consistently found to have lower cardiovascular and total mortality. As stated by reviewer Ellison: “As soon as I saw the title, my first goal was to see what methodology the authors used to get this result, which is different from what has been reported from essentially all epidemiological cohort studies.”

Effects of combining data from very different types of studies: Several factors became immediately obvious to most Forum members to explain why the present analyses came to different conclusions than did previous studies. The most important may be considering the data from the VA Study along with that of the National Health Interview Survey data; adding these together in one paper makes no sense. The VA study, as pointed out by the authors, was an evaluation of subjects treated clinically (many of whom may have been treated for alcohol-related disorders) in the VA system. There are a number of other differences between the two groups, the most important may be that not only was the VA study a clinical study, but it did not have data on previous drinking among non-drinkers, educational or occupational status, and no data on concomitant smoking. As would be expected, the mortality rates for the VA subjects were more than double those of the NHIS subjects. Hence, the inclusion of more than 90,000 VA subjects only weakens any conclusions made by the authors in those analyses in which the two studies are included. For providing data that could be used for setting drinking guidelines, the VA data are of no use.

Reviewer McEvoy pointed out that that the authors do give results separately for the two studies in some analyses. “I agree that the results from the VA data are uninterpretable because they were not able to control for the major confounder of smoking. Smoking and drinking are highly correlated, and since smoking is strongly associated with increased risk of mortality, without controlling for smoking one cannot say anything meaningful about the association of drinking and mortality. However the results from the National Health Interview Survey are worth considering in more detail. Survey respondents were not asked to provide any information on weekly frequency of drinking so an assumption is being made that drinking reported across the past 365 days was evenly distributed; this assumption may not be warranted. These factors, as well as the lack of consideration of under-reporting of alcohol, means that these data are not appropriate for determining recommended weekly amounts of alcohol intake.

Basing life-time exposure to alcohol on a single assessment of intake: Another factor that may relate to the unusual results of this study is the inadequate assessment of exposure to alcohol. The VA exposure data are based on a single assessment of alcohol intake “over the past year” obtained in 2008, when the subjects were 40 to 60 years of age. For the NHIS subjects, it is unclear how alcohol data were estimated from the several surveys carried out, but near the end of their Discussion the authors state that “alcohol use patterns were measured once”. Evidently, changes in alcohol intake during follow up were not evaluated.

Lack of control for smoking and other lifestyle factors: Forum member Van Velden pointed out potential problems with confounding variables in this study. “The problem with part of this observational study is that it did not take into consideration all the confounding factors such as smoking, exercise, BMI, socio-economic status and other lifestyle-related issues. Responsible alcohol consumption is part of an overall healthy lifestyle — it cannot be seen in isolation. There is no doubt that harmful alcohol consumption is bad for your health; this is usually a component of the problem among people with an overall atherogenic lifestyle. As already mentioned, under-reporting is always a problem; people just do not truthfully reveal their true alcohol consumption, and this has a serious negative influence on the results.” Added Teissedre, “In France Professor Mirouze in Montpellier had reported in the last century on the safe level of alcohol consumption, stating that an average of up to 28 g/day can be metabolized by an individual. I agree also that cofounding factors are not taken into account in this paper. Alcohol intake needs to be evaluated in a system of multivariate factors of lifestyle (diet, exercise, etc.).”

What measures best reflect alcohol consumption? Reviewer Ellison noted: “The authors collected data on the frequency of consumption, and give results on subjects drinking 3-4 times/week, 4-7 times/week, etc., but it is unclear how they combined the quantity of alcohol consumed per drinking occasion with the frequency of consumption. They construct a group (evidently from both studies combined) who reported 1-2 drinks/day about 3 times weekly and compared data for subjects consuming more or less. It has been clearly established that heavy alcohol consumption is bad for your health. The key analyses needed are those comparing risk among non-drinkers (preferably never drinkers) with light drinkers (say, up to no more than one drink/day), then evaluate the effects of 1 to 1.5, 1.5-2, 2.5-3.0, etc., drinks/day. Unfortunately, only one category is used here for what are generally considered light-to-moderate drinkers: 1 to 2 drinks/day. This limits the ability of the authors to carry out a precise estimate of the level of intake related to an increase, rather than a decrease, in the risk of mortality.”

Forum member Skovenborg noted: “I agree with the comments on the epidemiological analyses; further a serious flaw is the lack of plausible mechanism between cause (drinking 1-2 drinks more than 3 times a week) and effect (increased risk of all-cause mortality). One of Sir Austin Bradford Hill’s proposed set of nine criteria to provide epidemiologic evidence of a causal relationship between a presumed cause and an observed effect is a plausible mechanism. A recent review found no association between a moderate consumption (up to 15 g/day) and the incidence of the 20 most common cancer types in the western world (Hendriks & Calame).”

Skovenborg added: “In their introduction, the investigators mention Francis E. Anstie, the author of ‘Anstie’s Limit’ that refers to the daily amount of alcohol that the average drinking individual can consume without risk of deterioration of health. His counsel became a widely respected and quoted opinion in medical and insurance circles. A hundred years after his death (1874) the 25th edition of Dorland’s Illustrated Medical Dictionary lists the dictum as a rule used in connection with life insurance examinations: “the maximum amount of absolute alcohol taken daily without injury is 1 ½ ounces, equivalent to about 3 ounces of hard liquor, a pint of light wine, or 24 ounces of bottled beer or ale” (Baldwin). The limit proposed by Hartz et al, 1-2 drinks 3 times weekly, may not stand for the next 100 years.

“In spite of serious weaknesses of their analyses, the authors suggest that the guidelines for ‘healthy alcohol use’ should be lowered; however, the ‘1-2 drinks 3 times weekly’ guidelines may share the fate of the existing governmental standard drink definitions and low-risk alcohol consumption guidelines in 37 countries, ranging from 10 g per day (Bosnia and Herzegovina, Croatia and India) to 56 g per day (Chile). It seems unlikely that every guideline is correct, as Kalinowski & Humphreys conclude.” Other reviewers emphasized what has been well described by Harding & Stockley: many factors, in addition to average alcohol intake, must be considered when setting drinking guidelines for an individual person or country.

Effects of under-reporting of alcohol intake: Essentially all epidemiologic research on alcohol consumption is based on the self-report of alcohol by subjects. If all subjects report less than they actually consume, one option would be to simply push the reported intake upward, so that 2 drinks/week turns into 3 or more drinks/week; this would have marked implications in setting “safe” drinking guidelines for the public. Adjusting for under-reporting is a difficult problem.

Luckily, Arthur Klatsky and his colleagues have found that subjects who are probably the main “under-reporters” can be identified from other data collected in large cohort studies. Using data within their very large Kaiser-Permanente datasets, it was possible to identify subjects who on one or more occasions stated that they consumed 1 to 2 drinks/day but also had references in their data that they had been treated for excessive drinking, problems with alcohol, or alcohol-related diseases such as alcoholic cirrhosis. They found that supposedly “moderate” drinkers who were thusly estimated to be “under-reporters” of alcohol on this basis showed increased risks of certain cancers, while subjects whose overall data did not indicate that they were prone to be under-reporting their intake had no increase in the risk of cancer (Klatsky et al). Those authors conclude: “The apparent increased risk of cancer among light-moderate drinkers may be substantially due to underreporting of intake.”

Forum member Keil stated: “I found that an ounce of alcohol translates to 28 grams, which means that the daily non-harmful intake of alcohol would be about 42 grams. One Maß or Mass at the Oktoberfest in München means 1 liter of beer, containing exactly 40 grams of alcohol. So the Bavarians seem to be right if they stay with one Maß. Normally a Maßkrug is not filled properly, which means that the Bavarian beer police must intervene against the fraudulent pouring of beer in the Maßkrug. You see, different people have different problems. Why do I mention all this? Well, underreporting is a problem at many levels.”

Failure to comment on the much higher risk of mortality among non-drinkers: Reviewer McEvoy pointed out that the data show a large protective effect of drinking when compared with not drinking. “The NHIS study clearly shows the often-reported U or J shaped association of alcohol with mortality; especially the protective association of drinking on cardiovascular mortality over never drinking is striking, particularly among women.

“Strengths of the NHIS portion of the paper include a large (n= 340,668) epidemiological sample, and control for several important potential confounders (smoking, health, exercise, SES). Weaknesses are the single time point assessment of alcohol intake, and the problem with converting self-reported yearly amount of alcohol intake into weekly amounts, as mentioned. Although I disagree with some of the authors’ interpretations of the results, I think this study does contribute to the literature showing that moderate alcohol intake is associated with reduced risk of mortality; unfortunately it cannot inform on what the limits of moderate intake are, and at what level of intake associations change to harmful. It does make the important point that alcohol-associated risks and benefits will vary based on an individual’s risk of cancer and cardiovascular disease.” Ellison agreed: “Their data show a clear J- or U-shaped curve. However, the authors focus only on the 1-2 drinks/day 3 times/week and say that those drinking more had higher mortality, without focusing on the much greater risk (higher risk than all drinkers except those reporting 7 or more drinks/day) for the non-drinkers.”

References from Forum critique

Baldwin AD. Anstie’s alcohol limit: Francis Edmund Anstie 1833-1874. Am J Pub Hlth 1977l:67:679-681.

Harding R, Stockley CS. Communicating through government agencies. Ann Epidemiol 2007;17:S98-S102.

Hendriks HFJ, Calame W. The Contribution of Alcohol Consumption to Overall Cancer Incidence in the Western World: A Meta-Analysis. J Nutr Health Sci 2018;5:308.

Kalinowski A, Humphreys K. Governmental standard drink definitions and low-risk alcohol consumption guidelines in 37 countries. Addiction 2016;111:1293-1298. doi: 10.1111/add.13341.

Klatsky AL, Udaltsova N, Li Y, Baer D, Nicole Tran H, Friedman GD. Moderate alcohol intake and cancer: the role of underreporting. Cancer Causes Control 2014;25:693-699. doi: 10.1007/s10552-014-0372-8.

Forum Summary

The authors of this treatise on alcohol consumption and mortality combined, for an unclear reason, results from two, very different studies: one from the Veterans’ Health Administration, based on outpatient clinical medical records, and the other from a national survey in the USA [The National Health Interview Survey (NHIS)]. Unfortunately, the VA data did not adjust for tobacco use or other important lifestyle habits, and are not useful in judging the effects of alcohol consumption on mortality.

While the NHIS survey included data on potential confounding, both it and the VA study based their analyses on a single estimate of alcohol consumption. The authors then created a variable that they stated was associated with the lowest risk of mortality and compared results from such alcohol intake with data from subjects reporting less or more alcohol. They combined data on all subjects reporting 1 to 2 drinks/week, so did not have the ability to provide a precise estimate of the association between low levels of intake and mortality risk. They did not comment on the effects that under-reporting of alcohol would have on their studies (and most investigators agree that the self-reported level of alcohol consumption is usually an under-estimate of actual consumption). Further, they did not point out the very much higher risk of mortality of non-drinkers, compared with moderate drinkers, demonstrated in their data.

To the extent that people accurately reported their past drinking status (never vs former), this study should put to rest concerns that protective associations of drinking versus non-drinking arise from inclusion of former drinkers into non-drinking groups. Never drinkers had higher risk of mortality than former drinkers in this study.

There are considerable data from many well-done cohort studies that have repeated assessments of alcohol intake over many decades and the subsequent risk of mortality. Such studies provide very clear and consistent results indicating a J-shaped curve: lower risk of mortality for light and moderate drinkers than for non-drinkers (even lifetime abstainers) and some increase in risk for heavy drinkers. These are the studies that can provide reliable information upon which drinking guidelines for different individuals and populations can be based.

* * * * * * *

Comments on this critique by the International Scientific Forum on Alcohol Research were provided by the following members:

David Van Velden, MD, Dept. of Pathology, Stellenbosch University, Stellenbosch, South Africa

Pierre-Louis Teissedre, PhD, Faculty of Oenology–ISVV, University Victor Segalen Bordeaux 2, Bordeaux, France

Arne Svilaas, MD, PhD, general practice and lipidology, Oslo University Hospital, Oslo, Norway

Creina Stockley, PhD, MSc Clinical Pharmacology, MBA; Health and Regulatory Information Manager, Australian Wine Research Institute, Glen Osmond, South Australia, Australia

Erik Skovenborg, MD, specialized in family medicine, member of the Scandinavian Medical Alcohol Board, Aarhus, Denmark

Linda McEvoy, PhD, Department of Radiology, University of California at San Diego (UCSD), La Jolla, CA, USA

Fulvio Mattivi, MSc, CAFE – Center Agriculture Food Environment, University of Trento, via E. Mach 1, San Michele all’Adige, Italy

Ulrich Keil, MD, PhD, Professor Emeritus, Institute of Epidemiology & Social Medicine, University of Muenster, Germany

Harvey Finkel, MD, Hematology/Oncology, Retired (Formerly, Clinical Professor of Medicine, Boston University Medical Center, Boston, MA, USA)

R. Curtis Ellison, MD, Professor of Medicine, Section of Preventive Medicine & Epidemiology, Boston University School of Medicine, Boston, MA, USA