Commentary on Roerecke & Rehm (2012): The state of the science on moderate drinking and health—a case of heterogeneity in and heterogeneity out?



    1. Director, Department of Psychology—Centre for Addictions Research of BC, University of Victoria, Technology Enterprise Facility, Room 124 2300 McKenzie Avenue, Victoria, British Columbia BC, Canada V8P 5C2. E-mail:
    Search for more papers by this author

There are many admirable and impressive qualities in this comprehensive meta-analysis of the relationship between average daily alcohol consumption and risk of ischaemic heart disease [1]. Internationally accepted rules were followed in terms of specifying precisely how relevant studies were identified and sifted through to result with 44 articles finally included. Quality criteria were employed to remove some methodologically problematic studies, e.g. those which did not even control the age of their subjects or which enquired about their alcohol use over a period of no more than 2 weeks. Very clear and full descriptions are given of how the critical measures of average daily alcohol consumption were calculated and how adjustments were made to estimates from studies which included former drinkers in the reference group. This particular aspect is an improvement on the other recent meta-analysis by Ronksley et al. [2], who did not describe how their adjustments were calculated for a former drinker effect. I would also like to compliment the authors on their open discussion of the many limitations of this literature and their identification of large and significant heterogeneity in findings which limits the confidence in their meta-analysis.

Unfortunately, the technical sophistication does not, in my opinion, compensate completely for the generally impoverished state of this literature. While many methodological problems are acknowledged, I do not think they have been given sufficient or consistent weight. The paper starts out with a clear statement that there are protective effects from moderate drinking, goes on to acknowledge that these are controversial and then concludes that there is too much heterogeneity to be certain about any association, let alone causality. The best evidence they suggest for this comes from the literature on controlled laboratory studies [3]. Despite a statement that ‘strict inclusion criteria to identify high quality observational studies’ were applied to study selection, elsewhere in the paper the authors note how very poor were most of the studies that were included. Very few controlled for potential life-style confounding factors (e.g. not smoking) that might be correlated with both moderate drinking and health, very few controlled directly for the inclusion of former drinkers in the ‘abstainer’ reference group, virtually none control for heavy episodic drinking [4] and many assessed drinking at baseline over a relatively short time-period. Certainly, they made an adjustment for the former drinker effect, but to do this assumes that there are no other uncontrolled differences between studies which do and do not incorporate former drinkers in the abstainer group. Given the multitude of other problems in this literature, this is a highly uncertain assumption.

When examining methodological problems in the almost identical set of studies included in the Ronksley et al. [2] meta-analysis, even applying modest quality inclusion criteria results in the elimination of virtually all published studies [5]. Putting aside the problem of controlling for the effects of heavy episodic drinking [4], only two studies meet criteria for assessing adequately typical daily alcohol consumption at baseline, distinguishing former and occasional drinkers from both abstainers and moderate drinkers and controlling for a minimum set of life-style confounding factors (health and smoking status) [6,7]. These two studies reported mixed results: one found protective effects from moderate drinking in relation to cardiovascular disease (CVD) but not coronary heart disease (CHD) [6], the other found evidence for reduced CHD and CVD mortality but only for women with high alcohol consumption [7]. The most plausible alternative explanation to there being a causal association between moderate drinking and reduced risk of some diseases is the failure to control for confounding life-style risk factors. In a classic study, Naimi et al. [8] found that out of 30 potential risk factors for heart disease unrelated to drinking, abstainers were at significantly higher risk than were light or moderate drinkers on 27 risk factors. The plausibility of this alternative explanation is also made starkly apparent (although rarely remarked upon) by numerous studies reporting biologically implausible health benefits associated with moderate drinking [9]. Leading examples include the demonstration of significantly reduced risk of liver cirrhosis morbidity by moderate drinking males [10] and fewer birth complications for mothers who were light drinkers during their pregnancy [11]. There are also numerous other claims of health benefits from moderate drinking which should be scrutinized more closely, ranging from preventing ailments from dementia to the common cold.

What should be done to bring more clarity and certainty into this difficult area? Clearly, we need more than the present handful of high-quality studies and, ideally, randomized controlled trials if these should ever be possible. In the meantime, should we be just reporting significant and large heterogeneity in study results, or perhaps seek to conduct meta-analysis with increasingly strict quality criteria? This latter approach was used by Fillmore et al. [12], and is worthy of being re-visited in light of the many new studies published since that particular meta-analysis was conducted—Fillmore et al. concluded an absence of significant health benefits from moderate drinking among the few available high-quality studies.

Declaration of interests