Letter regarding ‘Does a bit of alcohol turn off inflammation and improve health?’

Age Ageing (2016) doi: 10.1093/ageing/afw146 First published online: August 29, 2016

Sir, we wish to express caution over the interpretation of Shah and Paulson's analysis [1] in the accompanying editorial by Bektas et al. [2], which further prompted a BGS blog (1 September 2016) on the topic.

Bektas et al., we feel mistakenly attributes causality to the protective association reported by Shah et al. Whilst Shah et al. present an elegant mediation analysis, one cannot escape the fact that this is based upon cross-sectional self-reported data and as such is prone to reverse causality, recall bias, confounding and further biases as outlined below. And, crucially, Bektas’ message contravenes recent national guidelines on alcohol consumption.

Cross-sectional data do not provide information on temporal relationships and therefore directions of causality regarding alcohol consumption, measures of frailty and other co-variates/mediators assessed at the same time, cannot (and should not) be inferred, even if the statistical analyses are faultless. It is quite possible that as individuals age (and their CRP rises), they self-regulate their alcohol intake in response to their perception of their changing health status; i.e. reverse causality due to the ‘sick quitter effect’ (individuals quitting or never starting to consume alcohol due to underlying ill-health) [3]. Thus associations are seen because of ‘the effect of the drinker, not the effect of the drink’.

Self-reported alcohol consumption is notoriously prone to measurement error, a limitation rightly acknowledged by Shah et al. But those who are frailer, with poorer health, may well be less keen (or more forgetful) to acknowledge their alcohol consumption in a health survey. This (non-differential) recall bias can act to induce a false or exaggerated association between frailty and lower alcohol consumption.

Most epidemiological studies that use conventional analytical approaches are prone to residual confounding, and the present one is no exception. For example, socio-economic position is not adjusted for in these analyses, yet ‘middle-class’ moderate drinking is an established phenomenon, and those who are more affluent are likely to be healthier, less frail, have less co-morbidity and hence lower CRP.

The original Health and Retirement Study was set up in 1992 in Michigan (http://hrsonline.isr.umich.edu/); this current analysis included just 3,229 of 11,349 (28.5%) adults aged 65+ years available in the 2008 wave of the study. Analyses conducted on a small subset of the eligible sample can introduce biases such as selection bias specifically in the form of survival bias; whilst the ‘fit’ alcohol drinkers survive into old age, those with co-morbidities or with higher levels of alcohol consumption do not. This limits the internal (and external) validity of study findings.

So in this, like in other observational studies, assessing the long-term causal effects of alcohol drinking is notoriously difficult. Decades of observational data showing J-shaped relationships between alcohol intake and disease risk, particularly cardiovascular disease [4], fuelled by confirmation bias, resulted in alcohol policies such that individuals were recommended to drink in moderation, due to putative cardioprotective effects. Critically, randomised controlled trials to investigate the long-term effects of alcohol drinking are not feasible for reasons including lack of suitable and ethical interventions and extended duration (and hence cost and likely high loss to follow-up).

Complimentary approaches for causal inference that could mimic the effects of randomised allocation to alcohol exposure acting over prolonged periods of time can be used to provide guidance on the reliability of existing data. Mendelian randomisation (MR) is one such approach: an instrumental variable analysis exploiting genotypes as instruments [5]. Specifically, variants exist in alcohol dehydrogenase (ADH) and aldehyde dehydrogenase (ALDH) genes that are associated with faster or slower alcohol metabolism and different patterns of accumulation of metabolites, and which in turn can confer protection from excessive alcohol consumption (e.g. a variant in ADH1B) or even result in virtual intolerance to alcohol in the most extreme of cases (e.g. homozygosity for the *2 variant in ALDH2, i.e. ALDH2*2/*2 individuals) [6]. Such MR analyses can provide causal estimates of life-long exposure lasting for at least as long as someone's history of alcohol use, by comparing individuals according to their relatively alcohol-prone or alcohol-adverse genotype. Some of these MR studies have shown causal effects of prenatal alcohol exposure on neurodevelopment of the child [7, 8], and for even light alcohol exposure having adverse consequences on cardiovascular health [9].

These more compelling MR analyses, from which one can draw causal inference, have since been taken into consideration in a review of national alcohol consumption guidelines commissioned by the UK Chief Medical Officer. The Guidelines Development Group's revised advice (January 2016) states that in pregnancy the safest approach is not to drink at all, and recommends the same ‘low risk limit’ for adult men and women, acknowledging that even below 14 units/week alcohol consumption is not ‘safe’ [10].

In conclusion, much as we may wish to believe a little wine is good for us, we do not wish for geriatricians reading these articles, to conclude that the advice they should give to their patients should encourage a moderate alcohol intake. We feel this would be a misinterpretation of these data, would pose a risk to patients and would contravene the recently revised UK alcohol guidelines, which considered the totality of evidence on alcohol effects in a holistic paradigm.

Conflict of interest

None declared.

References