How do we formulate low-risk drinking guidelines if zero consumption is lowest risk?
This thought-provoking paper by Naimi and colleagues  packs a great deal into a few short pages. The meat comprises elegant analyses which (i) demonstrate why drinking more frequently may not be better for the health of ‘moderate’ drinkers and (ii) illustrate further the serious confounding in uncontrolled observational studies of alcohol consumption and health. The discussion, however, also serves a generous helping of relish by raising sharp questions about whether health benefits from moderate drinking exist, at what dosage they may be optimal and what this all means for the formulation of low-risk drinking guidelines. Naimi et al.  conclude that such guidelines should advise drinkers to not exceed about half of one standard drink per day on average. I will make one comment on the empirical findings, and then suggest a difficulty with their broader argument about low-risk drinking guidelines.
In a restricted sample of people who on average drink within the US Dietary Guidelines, the authors report an inverse relationship between quantity of alcohol consumed on a drinking day and drinking frequency. This should not be surprising because, as they acknowledge, if the total consumption of a sample of drinkers is restricted in this way such a relationship is mathematically inevitable. What is important, however, is that this inevitable relationship has also influenced findings in other studies, which appear to show that increased frequency of drinking is associated with improved health (e.g. [2, 3]). Thus they illustrate why such empirical observations should not be interpreted as evidence to encourage people who drink a little to do so more often. It is also important that Naimi's sample of drinkers comprise not just a small subsample, but 87.2% of all current drinkers in this large representative US national survey. The so-called ‘moderate’ drinking category for average consumption per day used in such studies contains the great majority of people who drink and who evidence substantial heterogeneity in drinking patterns and health risk profiles. The additional finding, that infrequent drinkers have significantly elevated health risk factors, adds further to worries about systematic biases in longitudinal studies of alcohol and health. Just one such worry is the bias introduced by the common practice of including ‘occasional drinkers’ in the abstainer reference group , e.g. anyone who drinks less than once a month.
Naimi et al.  contribute to a groundswell of scepticism about the veracity of observed health benefits of light to moderate drinking thought to underlie the observation of a J-shaped risk relationship between level of drinking and risk of all-cause mortality (e.g. [4, 5]). Other recent contributions include (i) the demonstration of systematic bias early in the life span caused by young adults who choose to be totally abstinent also having poorer health and lower incomes ; and (ii) the finding of apparent health benefits associated with moderate drinking from a large and improbably diverse range of diseases, including even liver cirrhosis . Naimi et al.  also suggest that the lowest risk of all-cause mortality is at a consumption level of only 7 g for men and 5 g for women, involving mainly a frequency of drinking unlikely to confer significant physiological benefits. However, they argue further that this ‘nadir’ of drinking risk should also be the level recommended for low-risk drinking. One logical problem here is that if health benefits for moderate drinkers are discounted, we cannot really assert that the nadir of drinking risk is anything higher than zero consumption. It may be that, in the future, quantitative guidelines for low-risk drinking need to be established on the basis of broadly acceptable levels of risk in comparison with other common behaviours, such as driving cars or playing sport—along with advice about how to minimize risk through modifying drinking pattern and context.
Naimi et al.  refer to the Canadian low-risk drinking guidelines  as an example of upper limits being set inappropriately at a consumption level where risk of all-cause mortality is no higher than that of life-time abstainers—a group which, for the above reasons, may be biased towards negative health outcomes. In their defence, the meta-analysis of drinking and all-cause mortality upon which the Canadian guidelines were based primarily was selected as the best in a bad bunch of four attempts to control for abstainer biases . Furthermore, it is worth noting that these guidelines unusually also take drinking frequency as well as quantity into account (as recommended by Naimi et al.), and if followed it was estimated alcohol consumption in Canada would be halved . If health benefits from moderate alcohol consumption are rejected, then both the Naimi et al. and the Canadian approaches to setting low-risk guidelines are in equal trouble.
Declaration of interests