The role of the comparison group in epidemiology and general limitations


  • Jurgen Rehm

    1. Social and Epidemiological Research (SER) Department, Centre for Addiction and Mental Health (CAMH), Toronto, ON, Canada
    2. Dalla Lana School of Public Health, University of Toronto, Toronto, ON, Canada
    3. Department of Psychiatry, Faculty of Medicine, University of Toronto, Toronto, ON, Canada
    4. PAHO/WHO Collaborating Centre for Mental Health & Addiction, Toronto, ON, Canada
    5. Epidemiological Research Unit, Klinische Psychologie & Psychotherapie, Technische Universität Dresden, Dresden, Germany
    Search for more papers by this author

I would like to start by congratulating Hans-Olav Fekjaer for his informative and thoughtful contribution on one of the most debated topics of alcohol epidemiology, the issue of the protective effect of alcohol consumption on disease incidence or prevalence [1]. To underline how debated the topic has been: a quick check at the beginning of March 2013 with PUBMED revealed 1669 hits for a search with the key words ‘alcohol’ and ‘beneficial effects’. A closer look revealed studies of different methodology, but most often cohort studies where alcohol consumption at one time-point was associated with lower incidence of morbidity or mortality after a certain time. What does this mean? First, we need to examine the control group: lower incidence compared to what? It has long been known that the usual control group of abstainers may comprise people who had quit drinking for health reasons (e.g. [2]). This may be overcome by using life-time abstainers, but as Fekjaer [1] noted, life-time abstainers may be different on other dimensions than just drinking [3]. One remedy may be to use samples from countries where abstention is more common, or the norm rather than the exception [1, 4], as most adults world-wide are currently abstainers [5, 6].

However, studies from countries where abstention is the norm do not solve the general problem of measurement bias in alcohol consumption [7]. Studies have shown that the measurement of abstention in general, and life-time abstention in particular, is problematic. For instance, in a nationally representative survey of the United States, results indicated that more than half of those who reported never having a drink of any alcoholic beverage in a particular survey reported drinking in previous surveys [4] and studies in low- and middle-income countries, where abstention is the social norm, will not lead to more reliable estimates, as social norms and resulting expectations tend to impact upon respondents' answering behaviour [8]. It is suspected that under-reporting of drinking as deviant behaviour is the main underlying reason why, in many countries with a very high prevalence of abstention, the mean drinking level among drinkers is also high [5].

Another way to find a better control group would be to associate reasons for abstention or life-time abstention with mortality, and first results here show that abstainers for moral or religious reasons, or out of family responsibility or for lack of being social, do not have any different mortality risks or life expectancies than light drinkers. The authors conclude that the route to long lives can be realized through either light drinking or abstention [9].

However, it should be stressed that it is misuse of epidemiology to try to reach definite answers to causal questions with epidemiology alone. I therefore very much agree with Fekjaer [1] that an answer on the question of alcohol as a universal preventive agent can only be given in considering simultaneously the results from different disciplines including biochemical effects (see [10], for similar reasoning). Thus, joint systematic reviews of the epidemiological and biochemical evidence from human and animal studies would be the way to go. Unfortunately, only in the field of determining carcinogenicity has science established such a procedure in the form of the monograph meetings of the International Agency of Research on Cancer (for the last meeting on alcohol see [11]). While it may be considered speculation to predict the outcome of such systematic scrutiny, I would predict that if such reviews were to be established for the disease conditions listed by Fekjaer [1], most of the claims of beneficial effects would disappear. The reason for this prediction is lack of good control in epidemiology coupled with lack of convincing biochemical pathways. This may not be true for all beneficial effects, however, and for the effect of alcohol on ischaemic events there may be some protection by regular and light drinking (see also the reasoning of [12]).

Even if this is the case, the detrimental effects of heavy drinking outweigh by far any beneficial effects (e.g. [13-15]), and the conclusions for alcohol policy should stress these harmful net consequences clearly. In other words, the alcohol policy framework as stated in the Global Strategy to Reduce the Harmful Use of Alcohol by the World Health Organization remains valid [16].

Declaration of interests