When is Alcohol Consumption Protective?

Review

Knott CS, Coombs N, Stamatakis 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.

Review

This study evaluated the association between alcohol consumption and all-cause mortality, as well as the relevance of age-specific limits for alcohol consumption. Alcohol use, particularly consumption of excessive amounts, is negatively associated with many acute and chronic diseases. The direct and indirect economic burden due to alcohol use is substantial.

Previous studies highlight associations that implied consumption of low to moderate amounts of alcohol may have a protective effect against cardiovascular disease and all-cause mortality, whereas consumption of higher amounts of alcohol increases mortality. However, the protective effects of alcohol observed in some studies may be the result of confounding, due in part to statistical analyses that classify heterogeneous non-drinking groups into a single reference cohort. For example, previous studies have categorized ‘former drinkers’ and ‘never drinkers’ into a single ‘non-drinker’ reference group. Therefore, it is unclear if low to moderate alcohol use truly confers a protective effect, and if so, whether this is applicable to older groups of men and women.

This study is unique in that the authors attempt to elucidate a possible protective effect of low-moderate alcohol consumption on all-cause mortality by selecting a more appropriate reference group for statistical analysis as well as by adjusting for multiple potential confounders.

Study Methods

Population-based data from the Health Survey for England was collected from an annual cross-sectional survey of non-institutionalized residents from the years 1998-2008. Data was then pooled for analysis, and linked to a national mortality registry in this observational study. The authors further stratified the analysis based on age (50-64, ≥ 65 years old) and by sex (male, female), and attempted to control for potential confounders such as personal, socioeconomic, and lifestyle factors.

Results and Outcomes

The main outcome measured was all-cause mortality, as recorded up through March, 2011.

The authors report both unadjusted and adjusted models based on their data. The unadjusted models demonstrated a protective effect of alcohol use across multiple age and sex groups. However, the protective effects were diminished upon controlling for potential confounders, such as socioeconomic status, smoking, BMI, economic activity, education, ethnicity, region, and marital status. When the authors defined the reference group as ‘never drinkers’, as opposed to ‘non-drinkers’ (which included former alcohol consumers), the protective effects of low-moderate alcohol use were limited mainly to women aged 65 years and older.

Conclusions

Previously reported beneficial associations between low alcohol consumption and all-cause mortality may be the result of a heterogeneous, inappropriately selected reference group for comparison and poor control for confounders (such as age, smoking status, socio-economic status etc.).

By using ‘never-drinkers’ as a reference group, as opposed to ‘non-drinkers’, this study finds the protective effects from low consumption of alcohol may be limited to women over 65 years old.

The use of inappropriate reference groups (typically ‘non-drinkers’, which is inclusive of former drinkers and never-drinkers) and poorly controlled confounders may explain, at least in part, previous findings which indicate a beneficial association between alcohol consumption and all-cause mortality.

Commentary

The authors report that beneficial associations previously observed between alcohol use and all-cause mortality may be the result of an inappropriately selected reference group, poor control of confounding factors, and selection biases. They report a beneficial effect specifically in older females, with other age and sex groups failing to demonstrate a clear benefit. This study is unique in its attempt to create a refined reference group for comparison in which former drinkers are excluded, and never-drinkers are included for comparison. This study also attempted to control for age and sex by discerning between men and women, aged 50-64 years old and those 65 years and older.

The strengths of this paper include the authors’ attempt to develop an improved reference group (never-drinkers verses non-drinkers) and control for more confounders than prior studies. They stratified their data by age and sex in an effort to demonstrate differences in the protective effect of alcohol consumption among these groups. Additionally, they report separate data after controlling for multiple potential confounders (adjusting for personal, socioeconomic, and lifestyle factors). The authors also report on two metrics for assessing drinking behavior: average weekly consumption and heaviest alcohol use in a single day.

The results of this study should be interpreted cautiously, as the findings are limited due to several factors including selection, reporting and recall biases. A selection bias (that which occurs when the population included is not consistent with your patients or the general population) should be considered, as patients who are chronically ill may not have participated as often in the study, and were potentially excluded from analysis. As with other studies of the effects of alcohol on health, excluding those with ill-health at baseline may positively select for a healthier reference group and lead to inaccurate analyses.

Reporting bias (when data included is collected in a subjective manner, like from patient’s memory) may be present because data is dependent on self-reported surveys. A tendency for recall bias has been found in similar studies, which suggest that participants may not accurately identify as ‘never drinkers’ and could therefore be misclassified. Additionally, the external validity of these findings needs to be considered when applying and generalizing these findings to populations outside of England. Subdividing data into multiple alcohol-consumption categories (perhaps in an attempt to illustrate a dose-response relationship) may have resulted in an underpowered data set and should be considered.

This paper highlights the need for more robust data sets with sufficient power to further demonstrate the relationship of low to moderate alcohol consumption on cardiovascular disease and all-cause mortality, as well as any differences between age and sex groups. Stronger data and improved analyses will help clinicians and medical societies make more evidence-based recommendations regarding the risks of alcohol consumption. Until more data is available, where confounders are well addressed, recommending low to moderate alcohol consumption should be reserved for older women.