A narrative review of alcohol consumption as a risk factor for global burden of disease

Table 1 gives an overview of the main CRAs conducted since 1996 on two main outcomes: all-cause mortality as measured in the number of deaths, and burden of disease as measured in DALYs (for a definition of DALYs see [3, 4]).

While Table 1 seems to indicate a substantial rise in both alcohol attributable mortality and disease burden, this is not the case. Any CRA estimates depend very much on methodology, and the majority of the variation between the first estimate for 1990 [12] and the other estimates can be explained by the following factors (see also [16, 17]:

These questions will be discussed separately under different headings below.

Having clarified this, there have been efforts to estimate real changes using the same methodology for comparisons in global or regional alcohol-attributable burden of disease between 1990 and 2010 [10]; between 1990 and 2013 [11, 18] or between for all years from 1990 to 2014 [19, 20].

Alcohol has been causally linked to more than 230 ICD 10 three digit disease categories [17], including about 40 that would not exist without alcohol (such as alcohol dependence or alcoholic liver cirrhosis; for a complete list see [21]). However, this does not mean, the various global CRA efforts are including all of these disease categories. First, global health statistics are not that detailed. For most of the population worldwide, there are no vital registries with cause of death information. The World Health Organization (WHO) estimates that vital registries fail to cover about two thirds, or 38 million out of 56 million annual deaths globally [22]. For the rest, i.e., the majority of cause of deaths globally, the basis are verbal autopsies and standardized algorithms to analyze these verbal autopsies and to scale up these results to wider regions [23]. Obviously, the resulting cause of death categories are fewer and broader than the categories of the ICD, as it is impossible to determine detailed cause of deaths via verbal autopsy [24]. As indicated above, burden of disease is composed of mortality and disability. For disability, the data situation is worse than for mortality [25], and estimates are developed based on estimated prevalence of disease categories [26] and disability weights; a disability weight is a factor that reflects the severity of the disease on a scale from 0 (perfect health) to 1 (equivalent to death) [27].

Currently, there are two different organizations which produce estimates for global data on health outcomes: the Institute for Health Metrics and Evaluation (IHME) [28] and the WHO [29]. Methodologies for estimating deaths are overlapping but different, resulting in different estimates on cause of death by categories (e.g., [30]). Secondly, different disability weights are applied for non-fatal outcomes, resulting in different estimates of burden of disease in DALYs ) [31‐33]. However, the number of disease categories are the same between both organizations. For CRAs this means, that the number of deaths and DALYs, where attributable fractions are applied to, are different based on which underlying statistics are used. One of the main differences specifically for alcohol consumption as a risk factor are alcohol use disorders, which are defined differently by IHME and WHO, and where different disability weights have been applied (IHME restricted to dependence, WHO including harmful use of alcohol with a non-zero weight). A reconceptualization on heavy drinking over time may help overcome these differences [34].

The availability of only broader categories on a global level has consequences for CRA. Diseases and causes of death like alcoholic cardiomyopathy or alcohol-induced chronic pancreatitis are relevant in many countries (for alcoholic cardiomyopathy: [35]; for pancreatitis: [36]), but there are no global statistics. As a consequence, the impact of alcohol can only indirectly be assessed by using larger categories such as cardiomyopathy or pancreatitis, for which global estimates exist. Unfortunately, often risk relations are missing for such larger categories: this is the case for cardiomyopathy, as there is neither a systematic review nor a meta-analyses for the impact of alcohol, and thus, this category will show no alcohol-attributable cases, even though we know, there are cases of alcoholic cardiomyopathy. For alcohol-induced chronic pancreatitis, the CRA has to rely on the larger category of pancreatitis and meta-analyses on the impact of alcohol on this category [37, 38] to be included into alcohol-attributable deaths or burden of disease. As a consequence, the alcohol-attributable disease and cause of death categories boil down to a much lower number of less than 25. In sum, global estimates on alcohol-attributable mortality and disease burden rely only on selected large disease categories. This eliminates most categories, which are 100 % alcohol-attributable by definition, except for alcohol use disorders and fetal alcohol syndrome (for some background on burden attached for alcohol use disorders: [39, 40]; for fetal alcohol syndrome [41]), as well as smaller partially attributable categories of disease or causes of death.

The number of alcohol-attributable disease categories in CRAs over the past decades has been increasing for three reasons. First, while the overall number is still small, more disease categories have been included into the global statistics (both as cause of death and as burden of disease in DALYs). To give two examples relevant for alcohol, pancreatitis and cardiac arrhythmias were added for the CRA of GBD study 2010 [17, 24]. Second, evidence on the causal impact of alcohol consumption became stronger and more convincing for certain disease categories, and third, better models for quantification of such causal impact were established.

Table 2 gives an overview on the relative risk estimates used for the WHO Global status report on alcohol and health [9] for all countries except for Russia and surrounding countries (for graphical displays of the dose-response relationships between average volume of consumption and outcomes see [83]; for the estimates used for Russia and surrounding countries see [84]).

While new meta-analyses on the risk relations between level of consumption and various disease/mortality outcomes appear regularly, this does not change the burden estimates dramatically. Take breast cancer as an example: there have been more than 100 single studies and 16 systematic reviews with meta-analyses over the past 20 years [85]. However, the main conclusions on relative risk did not change: there is a clear dose-response relationship with no protective effect for any level of drinking compared to lifetime abstainers [85]. Even drinking as low as one drink on average is associated with increased risk for breast cancer [85, 86]. The quantification for the different levels of alcohol-attributable risk for breast cancer had been quite similar over the years [85], as they had been for cancer in general (e.g., [87], which has been used to date for the CRAs, and [47]; which will be used in the future).

The field of cardiovascular outcomes has been less stable, in part, because two dimensions of alcohol consumption need to be taken into consideration, average volume and patterns of drinking, in particular heavy drinking occasions (in general: [88]; for cardiovascular in particular see [16, 89]), and because there are much fewer underlying studies (for some endpoints less than 10 studies; see the underlying studies on heavy drinking occasions and ischemic heart disease: [74, 90]), and because part of the effect is on acute drinking and part on chronic drinking with different methodologies which are hard to reconcile in meta-analytic approaches (acute drinking risks have been mainly measured via case-crossover studies such as [91]; chronic risks have been mainly measured in cohort studies such as [92]). Another problem is the fact that risk curves differ between fatal and non-fatal outcomes for many endpoints such as stroke [75] or ischemic heart disease [89]. Overall this makes the estimation of alcohol-attributable mortality and burden of disease challenging, and in almost each new CRA, a different approach has been used. As well, newest calculations in the WHO European Region suggest, that relatively small changes in exposure resulted in marked changes of cardiovascular mortality over the past 25 years [20].

Relative risks for other outcomes can be classified as in between cancer and cardiovascular disease in their complexity. For many outcomes, fatal and non-fatal risk relations differ. Thus, it has been found that for liver cirrhosis, the risk curves are steeper (more exponential) for mortality compared to non-fatal outcomes [70]. The explanation is simple: it takes quite a lot of alcohol consumption to cause liver cirrhosis often via different stages of liver disease [93], but once liver cirrhosis is established, no matter of what etiology, relatively small amounts of alcohol may be fatal [70, 94]. Heavy drinking occasions may play an additional role here, but we do not have enough data to quantify this relationship [95].

Similarly, for injury, acute alcohol use has been linked to more severe and in particular to fatal injuries [16, 96]. For CRA this means that different risk relations have to be used for mortality and for the non-fatal outcomes.

One of the main problems with the relative risks is that it is assumed, that they are biological constants and the same for all countries, and thus it does not matter, that the current estimates are derived from meta-analyses of select cohort studies from a few high income countries with a limited variation of drinking patterns (see above and [97], box on “Methodological issues relevant to studies of alcohol-related morbidity and mortality). Unfortunately, this assumption is not correct. It has been shown that the usual relative risks derived differ from the analogous risks found in Russia or surrounding countries [98, 99], and thus would lead to underestimates of alcohol-attributable burden [100]. In the most recent CRAs after 2010, this has been acknowledged and country specific relative risks have been used [84]. We suspect that different relative risks could be necessary for other countries with high per capita consumption per drinker (see [9], for a listing of countries), or for countries with irregular heavy-drinking patterns (e.g., countries with festive drinking such as Mexico; [101, 102]). Unfortunately, as of now, we have no evidence base to implement country- or region-specific relative risks for these patterns. Another problem is genetic predisposition, which sometimes interacts with alcohol consumption to produce different risks. As an example, the flushing gene [103], which is clearly associated with higher risks for alcohol-attributable cancer with an acetaldehyde pathway [104], clearly indicated different relative risk estimates for countries where this genetic constellation is prevalent (such as China, Japan and South Korea).

In sum, global estimates of risk relations tend to only minimally change for outcomes with many underlying studies such as cancer outcomes. This changes for cardiovascular outcomes, where more than one exposure dimension is relevant, and where there are few studies to quantify the risk relations. In future, more country- or region-specific relative risk estimates will be necessary to include genetic variability and more extreme drinking patterns, which are not measured in most cohort studies (see the box on “Methodological issues relevant to studies of alcohol-related morbidity and mortality” and [97] for further discussion of the limitations of cohort studies).

While the overall methodology for CRAs has been fairly stable (for a description see [105, 106]) based on original epidemiological concepts of the 1980s [107, 108], there are important shifts in details:

In the above, we described the methodology to conduct a CRA for alcohol consumption. In this methodology, instantaneous effects are assumed: i.e., exposure to alcohol consumption in a certain year is assumed to result in changes in mortality for this year. This clearly is a simplification, as there is usually a lag time between alcohol consumption and disease outcomes [126]. Moreover, individual consumption may vary, and many risk relations assume more or less a constant consumption over time [127]. Future CRAs should address this problem and take into consideration the lag time between consumption and outcomes (e.g., using methodology such as [128, 129]). However, such a step would need a reconfiguration of the conceptual model for all CRAs for all risk factors, as one of the main objectives of any CRA is to be comparative between risk factors and time.

In addition we expect the following methodological changes for future CRAs for alcohol:

As indicated above, all CRAs resulted in marked burden of disease caused by alcohol consumption. Two dimensions were identified as important to cause harm: overall level of consumption and patterns of drinking [88]. Policies need to address both dimensions.

There are effective and cost-effective policies to lower overall level of consumption in societies [140, 141], such as the so-called ‘best buys’, i.e., increase of taxation leading to increases in price of alcoholic beverages, decrease in availability, and ban of marketing and advertisement [142, 143]. However, these policies do not to seem to be too popular with governments, and in fact alcohol has become more available and affordable in most parts of the world over the past decades (e.g., in the European Union, see [144]). Other potential ways to decrease overall level of alcohol consumption would be a decrease in alcoholic strength, which is technically possible for all beverages, and which could be achieved via government regulation, taxation or industry initiatives [145]. It should be noted however, that a reduction in overall level of alcohol consumption does not necessarily mean a reduction in total alcohol-attributable mortality or burden of disease, as monitoring of the last 25 years for the WHO European Region has shown [20]. In addition, it has to be assured, that the heaviest drinkers do not increase their drinking (e.g., via treatment [146]), and that patterns of drinking do not get worse (see also [20]).

Regarding patterns of drinking, there are other promising policies such as minimum pricing [147, 148], and specific policies to decrease heavy drinking occasions in certain situation, such as in participation in traffic [136] or while operating machinery at the workplace [149]. Obviously, harm would be minimized, if in such situations abstinence was the norm.

Finally, the composition of alcohol-attributable burden of disease and mortality will have different implications for policy [150]. A high proportion of traffic injury could be reduced with specific measures for drink driving such as introduction and enforcement of a per se law regarding blood alcohol concentration, or reduction of the blood alcohol concentration threshold in existing laws [140, 151]. On the other hand, high alcohol-attributable intentional injury will ask for specific measures such as measures against binge drinking or per se laws on criminal prosecution [150]. To give one final example concerning chronic disease: high levels of alcohol-attributable liver disease mortality point to high overall level of consumption [152, 153], or to relatively high level of consumption combined with other etiological factors for liver disease such as HIV (as even comparatively small levels of alcohol consumption may cause liver mortality in people with liver cirrhosis no matter which etiology – see above for further detail and [20], for examples). Reductions of overall alcohol consumption, no matter how achieved, will lead to reductions in alcohol-attributable liver mortality [20].