Critique 252: Dealing with previous alcohol exposure among “current non-drinkers” in epidemiologic studies testing the J-shaped curve; 3 December 2021

John U, Rumpf H-J, Hanke M, Mayer C.  Alcohol abstinence and mortality in a general population sample of adults in Germany: A cohort study.  PLOS Med 2021;18:e1003819.

Authors’ Abstract

Background:  Evidence suggests that people who abstain from alcohol have a higher mortality rate than those who drink low to moderate amounts. However, little is known about factors that might be causal for this finding. The objective was to analyze former alcohol or drug use disorders, risky drinking, tobacco smoking, and fair to poor health among persons who reported abstinence from alcohol drinking in the last 12 months before baseline in relation to total, cardiovascular, and cancer mortality 20 years later.

Methods and findings:  A sample of residents aged 18 to 64 years had been drawn at random among the general population in northern Germany and a standardized interview conducted in the years 1996 to 1997. The baseline assessment included 4,093 persons (70.2% of those who had been eligible). Vital status and death certificate data were retrieved in the years 2017 and 2018. We found that among the alcohol-abstinent study participants at baseline (447), there were 405 (90.60%) former alcohol consumers. Of the abstainers, 322 (72.04%) had met one or more criteria for former alcohol or drug dependence or abuse, alcohol risky drinking, or had tried to cut down or to stop drinking, were daily smokers, or self-rated their health as fair to poor. Among the abstainers with one or more of these risk factors, 114 (35.40%) had an alcohol use disorder or risky alcohol consumption in their history. Another 161 (50.00%) did not have such an alcohol-related risk but were daily smokers. The 322 alcohol-abstinent study participants with one or more of the risk factors had a shorter time to death than those with low to moderate alcohol consumption. The Cox proportional hazard ratio (HR) was 2.44 (95% confidence interval (CI), 1.68 to 3.56) for persons who had one or more criteria for an alcohol or drug use disorder fulfilled in their history and after adjustment for age and sex. The 125 alcohol-abstinent persons without these risk factors (27.96% of the abstainers) did not show a statistically significant difference from low to moderate alcohol consumers in total, cardiovascular, and cancer mortality. Those who had stayed alcohol abstinent throughout their life before (42; 9.40% of the alcohol-abstinent study participants at baseline) had an HR 1.64 (CI 0.72 to 3.77) compared to low to moderate alcohol consumers after adjustment for age, sex, and tobacco smoking. Main limitations of this study include its reliance on self-reported data at baseline and the fact that only tobacco smoking was analyzed as a risky behavior alongside alcohol consumption

Conclusions:  The majority of the alcohol abstainers at baseline were former alcohol consumers and had risk factors that increased the likelihood of early death. Former alcohol use disorders, risky alcohol drinking, ever having smoked tobacco daily, and fair to poor health were associated with early death among alcohol abstainers. Those without an obvious history of these risk factors had a life expectancy similar to that of low to moderate alcohol consumers. The findings speak against recommendations to drink alcohol for health reasons

Forum Comments

Most cohort studies have demonstrated a reduced risk of cardiovascular diseases, diabetes, and even dementia among light-to-moderate consumers of alcohol when compared with lifetime abstainers.  Many criticisms of such findings have been based on the inclusion of ex-drinkers in the referent category of non-drinkers.  This is a spurious assertion in that most of the recent, large cohort studies of alcohol and health have separated ex-drinkers from lifetime non-drinkers, so the complaint has no basis.  It has been recognized by epidemiologists for decades that using “current non-drinkers” as a referent group can be a problem if a sizable proportion of that group may have consumed alcohol in the past, and especially when many ex-drinkers were formerly very heavy drinkers or alcoholics.  

The present study from Germany provides important information about using current non-drinkers as a referent group: more than 90% of their ‘current non-drinkers” were ex-drinkers, the majority of whom were ex-heavy drinkers or had evidence of having had an alcohol use disorder.  (This is quite different from rates in some populations, especially in the USA, where the large majority of non-drinkers are usually lifetime abstainers, and their inclusion in the referent group makes little difference.)  However, in a situation where almost all of the current non-drinkers are ex-drinkers, it is essential that when comparing the health effects of alcohol among current drinkers, comparisons are made only with lifetime abstainers.  

Comments by individual Forum members:  Reviewer Ellison stated: “The present study clearly shows how previous alcohol excess, as well as other unhealthy lifestyle factors (e.g., smoking) as well as socio-economic factors (e.g., education, income, access to adequate medical care, use of illegal drugs) can influence the risk of many diseases.  When relating alcohol consumption to non-drinking for effects on the diseases of ageing it is essential not only to separate drinkers from non-drinkers but also to adjust for other risk factors or risk indications of disease and mortality.  It is realized that alcohol use/non-use is only one lifestyle factor that must be considered.  Even for lifetime abstainers, lower socio-economic factors, heavy smoking, obesity, lack of exercise, poor diet, etc. will have to be evaluated as well as being factors affecting risk.”  

Ellison continued: “Unfortunately, in defining their referent group, the authors included some ex-drinkers (in addition to lifetime abstainers), making it an unusual and inappropriate group used for comparisons with drinkers. They did not present the data necessary to determine if the risk of disease among moderate drinkers was different from that of lifetime true abstainers (which was presumably the key goal of the study).  Thus, their claim that the lower risk of cardiovascular disease demonstrated in most studies among moderate drinkers is unrelated to alcohol is not supported by their main analyses.

“The authors claim that other lifestyle factors that are more common among moderate drinkers may explain their lower risk of many diseases.  However, many papers have shown that even among subjects who have a good diet, are not obese, are active, and are non-smokers, i.e., those with a ‘healthy lifestyle,’ adding moderate alcohol to each group reduces the risk of disease further, often markedly (see Joosten et al for an excellent example).  The results of the present paper strengthen the point that you should never consider alcohol as the only exposure for a health outcome, but the authors greatly confused their results by mixing ex-drinkers and lifetime abstainers in the referent group.”  

Reviewer Finkel considered this to be such a flawed analysis that our Forum should not bother to review it.  Forum member Skovenborg argued: “This is not a good study, but we have seen similar poor studies of this type before, and for years and years these studies have been quoted by many public health officers and used as arguments for further restrictive measures related to alcohol consumption, such as downsized guidelines, reduced availability, etc.  So it is not the journalists that I have in mind but the public health departments and the politicians.  I think that one of the raisons d’etre of our Forum is to review misguided papers with a cool and polite pointing out where there are some fatal methodological flaws.”

Forum member De Gaetano wrote: “It is a common understanding that measurement error may bias the findings from poorly conducted observational studies, especially when the exposure is alcohol intake.  Inclusion of former drinkers in the non-drinker category is a particular example. In all our previous studies (not mentioned in their paper) and also in a very recent one (De Castelnuovo et al), we very carefully excluded from the category of abstainers not only former drinkers but also individuals for whom we could not clearly make the distinction between lifetime abstainers and former drinkers.  After having implemented such precautions, our data suggest that, in comparison with subjects with low-volume consumption of alcohol, not-drinking is associated with a no less than 10% higher mortality risk.

“We also examined the dose-response relationship between level of alcohol intake and total mortality, using 5 gr/day intake of alcohol as the reference value.  We could document a J-shaped curve showing a decreasing risk up to about 5 gr/day, while drinking 10 to 20 gr/day was associated with a statistically significant 10% increase in mortality risk.  Thus, in the absence of a substantial, appropriate life-long non-drinkers group, the low-volume drinking category may be an appropriate reference group to compare different volumes among drinkers.  In all cases, the results do not challenge the existence of a J-shaped phenomenon.”

Forum member Skovenborg expanded the discussion of the choice of a ‘referent group’ by the authors for their comparisons with drinkers.  “Lifelong abstainers were those who answered ‘No’ to the question whether they had ever in life drunk a glass of an alcoholic beverage.  With that rigorous definition the number of “lifetime abstainers” was only 42 subjects (!), only 9.40% of the current abstainers (n=447) and only 1.05% of the total number of study participants (n=4,028).  It is quite clear that the number of true lifetime abstainers was completely inadequate to test the hypotheses suggested by the authors; their further extensive discussion of this topic and their conclusions are not supported by their data.  

 “This is a classic example of results from a study grossly underpowered to convincingly prove a plausible protection by moderate alcohol consumption due to a very small number of participants in the analyzed subgroups.  Further, the authors have committed the cardinal sin of saying that non-significance is the same as ‘no effect’ (like the false conclusion drawn by Knott et al from results of a study with the similar lack of power).  Such examples made some statisticians call for a stop to the use of P values in the conventional dichotomous way (e.g., Rothman; Amrhein et al).  

“Further, it appears that this study did not have a protocol and no formal prospective statistical analysis plan.  They state that ‘all analyses were planned except for the analysis of subgroups presented . . . these analyses were added during the peer review process.’ Post hoc analyses of subgroups are not recommended and do not belong in good epidemiological practice.  

“The authors were actually aware of this problem, stating ‘These findings speak against health protective effects of alcohol consumption.’  However, it has to be kept in mind that after adjustment for age and sex, the HR was not statistically significant but larger than 1.  It cannot be precluded that ‘in larger samples, the HR might become significant.’  But what is their reaction to that insight?  ‘This result shows that the risk factors that we analyzed do not fully explain the increased likelihood of early death among abstainers. Residual confounding is likely.’  The claim of ‘residual confounding’ is made by the authors in spite of the conclusion in the abstract of their reference to Keys et al: ‘Quantitative bias analyses indicated that omitted confounders would need to be associated with ~ 4-fold increases in mortality rates for men and ~ 9-fold increases for women to change the result.’ 

“Due to the small number of lifetime abstainers (n=42) the hazard ratio for shorter time to death for this group ‘was not significantly different compared to low to moderate alcohol consumers after adjustments for age, sex and tobacco smoking: HR 1.64 (CI 0.72-3.77).’  A subgroup of healthy alcohol abstainers with no alcohol- or tobacco-related risk factors and very good/good self-rated health (n=79) did not show a statistically significant difference in mortality risk compared with low to moderate alcohol consumers: HR 1.88 (CI 0.97-3.62) adjusted for age and sex.’  These analyses do not support the conclusions of the authors.”  

Forum member Finkel stated: “I certainly support the review by our Forum of this paper, and agree that the points made by other investigators are convincing.  We are living in an age of published ‘dysinformation,’ which often must not be allowed to stand unchallenged.  It is too bad that media usually value sensation more than truth.”  Reviewer Goldfinger added: “Obviously the authors have ignored or discounted the many studies that have adequately controlled for confounding factors, such as being a sick quitter.  There seems to be little integrity in their conclusions.”

References from Forum critique

Amrhein V, Greenland S, McShane B.  Retire statistical significance.  Nature 2019;567:305-307.  doi: 10.1038/d41586-019-00857-9.

Di Castelnuovo A, Costanzo S, Bonaccio M, McElduff P, Linneberg A, Salomaa V, et al. Alcohol intake and total mortality in 142960 individuals from the MORGAM Project: a population based study. Addiction 2021. https://doi.org/10.1111/add.15593

Joosten MM, Grobbee DE, van der A DL, et al.  Combined effect of alcohol consumption and lifestyle behaviors on risk of type 2 diabetes. Am J Clin Nutr 2010;91:1777-1783.  DOI: 10.3945/ajcn.2010.29170

Kerr WC, Lui CK, Williams E, Ye Y, Greenfield TK, Lown EA.  Health Risk Factors Associated with Lifetime Abstinence from Alcohol in the 1979 National Longitudinal Survey of Youth Cohort. Alcohol Clin Exp Res. 2017; 41(2):388–98. Epub 2017/01/08. https://doi.org/10.1111/acer.13302 PMID: 28063241; PubMed Central PMCID: PMC5272800

Knott CS, Coombs N, Starnatakis E, Biddulph JP.  All cause mortality and the case for age specific alcohol consumption guidelines: pooled analyses of up to 10 population based cohorts.  BMJ 2015; 350:h384.    

Rothman KJ.  Six persistent research misconceptions. J Gen Intern Med 2014;29:1060-1064. PMID: 24452418 DOI: 10.1007/s11606-013-2755-z.

Forum Summary

Most cohort studies have demonstrated a reduced risk of cardiovascular diseases, diabetes, and even dementia among light-to-moderate consumers of alcohol when compared with lifetime abstainers.  This is often described as a “J-shaped curve,” a reduction in disease risk for light drinkers (compared with abstainers) but an increase in risk for heavy drinkers.  Many criticisms of the J-shaped curve have been based on claims that many studies include “sick quitters” in their referent group.  Such criticisms ignore most recent cohort studies in which only lifetime abstainers make up the referent group.  Epidemiologists have known for decades that including ex-drinkers in the referent group for evaluating health effects of alcohol is an error; such a group cannot be used for comparisons with moderate drinkers.

While the present study from a large cohort in Germany clearly shows how previous alcohol excess and other unhealthy lifestyle factors and socio-economic factors influence the risk of many diseases, it creates a referent group made up of mixing lifetime abstainers and some ex-drinkers, and that their analyses show that this group does not have a higher risk of mortality than light drinkers (thus, the protection of light drinking leading to a decrease in risk of disease is erroneous).  The authors of this paper apparently did this because they had such a small group of lifetime abstainers (only 42 subjects out of a cohort of 4,000!), a number grossly inadequate for comparisons with drinkers.  

Forum members agreed that it was a serious error for the investigators; instead of admitting that their numbers were too low in this group for reliable comparisons, it appears that the authors augmented the comparison group with a group of ex-drinkers (that they considered to not be heavy drinkers) and reported that their analyses do not support a reduced risk among light drinkers. Thus, their claim that the lower risk of cardiovascular disease demonstrated in most studies among moderate drinkers is unrelated to alcohol consumption.  However, this claim is based on an improper referent group used in their analyses.  Forum members considered this to be a deeply flawed analysis, and of no merit for use in setting drinking guidelines.

*          *          *          *          *          *  

Comments included in this critique have been provided by the following members of the International Scientific Forum on Alcohol Research: 

Giovanni de Gaetano, MD, PhD, Department of Epidemiology and Prevention, IRCCS Istituto Neurologico Mediterraneo NEUROMED, Pozzilli, Italy 

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

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

Tedd Goldfinger, DO, FACC, Desert Cardiology of Tucson Heart Center, University of Arizona School of Medicine, Tucson, AZ, USA

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

Creina Stockley, PhD, MSc Clinical Pharmacology, MBA; Principal, Stockley Health and Regulatory Solutions; Adjunct Senior Lecturer, The University of Adelaide, Adelaide, Australia

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

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

Fulvio Ursini, MD, Dept. of Biological Chemistry, University of Padova, Padova, Italy

Andrew L. Waterhouse, PhD, Department of Viticulture and Enology, University of California, Davis, USA