As with cohort studies, Di Castelnuovo et al. found that alcohol consumption of 0–10 g/day led to significant reductions in all-cause mortality. However, there is a need to consider factors which affect internal and external validity, as these factors may lead to the observed protective J-shaped curve. The article by Di Castelnuovo et al. outlines the alcohol relative risks (RRs) for all-cause and cause-specific mortality based on a collective re-analysis of data from 16 cohort studies [1]. Like other similar cohort studies, Di Castelnuovo et al. found that consumption of between 0 and 10 g of alcohol per day (g/day) led to significant reductions in all-cause mortality and cardiovascular deaths. As it is currently under debate if a J-shaped curve for alcohol and health exists [2, 3] this study's findings, as well as the results of other cohort studies with similar methods examining public health and clinical practices, should be analyzed within the context of each study's internal and external validity. The sample size (n = 142 960) of the Di Castelnuovo et al. study is unique, as it provides power to detect small effect sizes. However, despite its sample size, there was insufficient power to reach conclusions regarding the RR of 0.98 for people who consumed 10 to < 20 g/day. Therefore, larger cohort studies are required to determine the significance of these RRs. The internal validity of studies which measure RRs close to the null (RR = 1) is particularly important, as small biases can affect the directionality of the results. As with numerous other cohort studies, Di Castelnuovo et al.'s findings are subject to measurement and design limitations. These limitations, mentioned in the article, include a lack of information on drinking patterns, using cross-sectional measurements for alcohol use, inconsistent life-time abstention, using broad measurements for confounders (e.g. daily cigarette use), survivor bias (i.e. unhealthy drinkers dying before cohort recruitment), former drinker bias (drinkers who experience negative health affects abstaining from alcohol and healthy drinkers continuing to drink) and the cardio-effects of alcohol differing for people with different ALDH2 genotypes [4-6]. Such limitations should be addressed when designing future cohort studies. The external validity of the study by Di Castelnuovo et al. is important to consider. Di Castelnuovo et al. examine the impact of alcohol on deaths, and did not examine the impact of alcohol on premature mortality, morbidity or total health loss. The scope of the analysis should be considered when comparing the results of this study to other studies. For example, the Global Burden of Disease study modelled total health loss and found that no level of alcohol consumption provides a health benefit [3]. While data on mortality are easier to communicate than is the total health loss, such as via disability-adjusted life years (DALYs) lost, the DALYs lost provide more comprehensive estimates of the health burden [7]. As with other similar cohort studies, Di Castelnuovo et al.'s results are affected by representivity bias (caused by design and response biases). For example, the southern Finland/alpha-tocopherol, beta-carotene cancer prevention (ATBC) study restricted recruitment to men aged 50–69 years. Additionally, women, people who drink less, are older in age and have higher socio-economic status are often more likely to agree to participate in studies [8, 9]. These factors bias the representation of cohort deaths (upon which all-cause mortality RRs are based) towards those which are more common in people of older ages, women and people of higher socio-economic status. This is particularly important, as alcohol is the leading risk factor for death among people aged 15–49 years (due mainly to injuries) [3, 10], and there is emerging evidence that socio-economic status may be an effect modifier with RRs being higher for people of lower socio-economic status [11]. Therefore, efforts should be made in future cohort studies to achieve population representation, either through design or weighting. Lastly, different approaches are used to model the impact of alcohol consumption on health (i.e. using all-cause mortality or summing cause-specific mortality). The use of all-cause mortality probably accounts for causes of death which are causally related to alcohol consumption, but where evidence of causality has not yet been established (versus the Global Burden of Disease study, which would exclude these causes of death [3]). However, all-cause mortality estimates are also more susceptible to biases due to the inclusion of causes of death which have a spurious association with alcohol. The study by Di Castelnuovo et al., as well as other cohort studies which examine the effect of alcohol on all-cause mortality, provide data which are critical to public health. However, efforts should be made to avoid potential biases when designing future cohort studies, especially in studies which examine the effects of low-dose alcohol consumption on health. None. None Kevin Shield: Conceptualization. Jurgen Rehm: Conceptualization.
Read full abstract
Jul 19, 2024
Chon Kit Pun + 7
Open Access
