Population-level risks of alcohol consumption by amount, geography, age, sex, and year: a systematic analysis for the Global Burden of Disease Study 2020

1Alcohol use accounted for 1·78 million (95% uncertainty interval [UI] 1·39–2·27) deaths in 2020 and was the leading risk factor for mortality among males aged 15–49 years (Bryazka D, unpublished). The relationship between moderate alcohol use and health is complex, as shown in multiple previous studies.,,,,, Alcohol consumption at any level is associated with health loss from several diseases, including liver cirrhosis, breast cancer, and tuberculosis, as well as injuries.,,, At the same time, some studies have found that consumption of small amounts of alcohol lowers the risk of cardiovascular diseases and type 2 diabetes.,, As a corollary, the amount of alcohol that minimises health loss is likely to depend on the distribution of underlying causes of disease burden in a given population. Since this distribution varies widely by geography, age, sex, and time, the level of alcohol consumption associated with the lowest risk to health would depend on the age structure and disease composition of that population.,,

2Two quantities are crucially relevant when formulating effective, evidence-based guidelines and alcohol-control policies: the theoretical minimum risk exposure level (TMREL), which represents the level of consumption that minimises health loss from alcohol for a population, and the non-drinker equivalence (NDE) level, which measures the level of alcohol consumption at which the risk of health loss for a drinker is equivalent to that of a non-drinker. The majority of studies to date consider one or a small subset of health outcomes associated with alcohol consumption at a time, although several broader systematic meta-analyses have been done.,,,, Findings from these studies vary in their estimates of the TMREL. Several studies have found evidence of a J-shaped relationship between alcohol use and all-cause mortality.,, However, others have reported that the all-cause or attributable cause burden weighted TMREL of alcohol is zero standard drinks per day., Uncertainty about the effect of alcohol on all-cause health loss results from differences in the relative disease composition between studies, conflicting studies on individual health outcomes, differences in study covariates and methods, estimation of drinking patterns, as well as issues relating to selection bias.,

3Evidence before this study The risks of moderate alcohol use on health outcomes have been widely studied and debated for many years. Studies have considered the health impacts associated with alcohol consumption through a variety of approaches, ranging from exploring the effects on a single disease, to considering multiple health outcomes, to using all-cause mortality as an outcome. Several systematic reviews have also been published on this topic, and in recent years several publications have used Mendelian randomisation to explore the association between alcohol use and health outcomes. Overall, the findings have varied, which partly contributes to this topic being controversial and a subject of debate. Several studies have found evidence of a J-shaped relationship between alcohol use and all-cause mortality or burden; in other words, at low levels of consumption, alcohol lowers the risk of all-cause mortality, whereas above some threshold it increases the risk. However, other studies, including a publication by the GBD 2016 Alcohol Collaborators in The Lancet in 2018, have reported that the level of alcohol consumption that minimises health loss is zero standard drinks per day. The apparent contradiction in findings across existing studies highlights the significance of continuing to study this topic and updating the evidence base as more information becomes available. Importantly, few previous studies analysing the effects of alcohol consumption on all-cause mortality have considered how the relationship between alcohol use and health is contingent on background rates of disease. We did a systematic review of the literature in which we searched PubMed and previous published meta-analyses using search terms such as “alcohol” and “drinking behavior”, terms concerning study outcomes such as “risk”, “odds ratio”, and “hazard ratio”, and terms concerning the specific causes included in the study, such as “ischemic heart disease” or “tuberculosis”. We searched for studies published up to Dec 31, 2019; the search was limited to English language publications. Added value of this study In this systematic analysis for the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2020, we estimated levels of alcohol consumption that minimise health loss using updated systematic reviews and meta-regressions, building on results from GBD 2016 and incorporating region-specific background rates of diseases and injuries within our assessment. To the best of our knowledge, this is the first study to consider the implications of background rates of disease on levels of alcohol consumption that minimise health loss. We updated the previously published systematic review and meta-regressions to consider all published studies through to December, 2019, reporting on the association between alcohol and the six alcohol-attributable health outcomes accounting for the highest number of global disability-adjusted life-years. We found insufficient evidence for an association between alcohol use and one of these outcomes and subsequently omitted it from further analysis. This analysis has yielded updated relationships on the relative risk of mortality for five causes, at various levels of alcohol consumption, which we combined with relative risk estimates from GBD 2016 for an additional 17 outcomes. We used this information, along with information on the burden of disease from these 22 diseases and injuries, to estimate the level of alcohol consumption that minimises health loss separately for each age group, sex, year, and region. In addition to estimating the level of consumption that is associated with minimising health loss, known as the theoretical minimum risk exposure level (TMREL), we also estimated the level of alcohol consumption at which the risk to health for a drinker is equivalent to that of a non-drinker—a quantity we refer to as the non-drinker equivalence. Implications of all the available evidence Our results are consistent with previous findings at the global level, and at the same time the more nuanced analysis done in this study strongly suggests that statements, guidelines, and recommendations on the optimal level of alcohol consumption need to take into consideration the background rates of diseases and injuries for each population. We provide clear evidence that the level of alcohol consumption that minimises health loss varies significantly across populations and remains zero or very close to zero for several population groups, particularly young adults. At the same time, small amounts of alcohol consumption are associated with improved health outcomes in populations that predominantly face a high burden of cardiovascular diseases, particularly older adults in many world regions. Given these findings, we recommend a modification of existing policy guidelines to focus on emphasising differential optimal consumption levels by age, rather than the current practice of recommending different consumption levels by sex. This study highlights the importance of prioritising interventions targeted at minimising alcohol consumption among young adults.

4Evidence before this study

5The risks of moderate alcohol use on health outcomes have been widely studied and debated for many years. Studies have considered the health impacts associated with alcohol consumption through a variety of approaches, ranging from exploring the effects on a single disease, to considering multiple health outcomes, to using all-cause mortality as an outcome. Several systematic reviews have also been published on this topic, and in recent years several publications have used Mendelian randomisation to explore the association between alcohol use and health outcomes. Overall, the findings have varied, which partly contributes to this topic being controversial and a subject of debate. Several studies have found evidence of a J-shaped relationship between alcohol use and all-cause mortality or burden; in other words, at low levels of consumption, alcohol lowers the risk of all-cause mortality, whereas above some threshold it increases the risk. However, other studies, including a publication by the GBD 2016 Alcohol Collaborators in The Lancet in 2018, have reported that the level of alcohol consumption that minimises health loss is zero standard drinks per day. The apparent contradiction in findings across existing studies highlights the significance of continuing to study this topic and updating the evidence base as more information becomes available. Importantly, few previous studies analysing the effects of alcohol consumption on all-cause mortality have considered how the relationship between alcohol use and health is contingent on background rates of disease. We did a systematic review of the literature in which we searched PubMed and previous published meta-analyses using search terms such as “alcohol” and “drinking behavior”, terms concerning study outcomes such as “risk”, “odds ratio”, and “hazard ratio”, and terms concerning the specific causes included in the study, such as “ischemic heart disease” or “tuberculosis”. We searched for studies published up to Dec 31, 2019; the search was limited to English language publications.

6Added value of this study

7In this systematic analysis for the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2020, we estimated levels of alcohol consumption that minimise health loss using updated systematic reviews and meta-regressions, building on results from GBD 2016 and incorporating region-specific background rates of diseases and injuries within our assessment. To the best of our knowledge, this is the first study to consider the implications of background rates of disease on levels of alcohol consumption that minimise health loss. We updated the previously published systematic review and meta-regressions to consider all published studies through to December, 2019, reporting on the association between alcohol and the six alcohol-attributable health outcomes accounting for the highest number of global disability-adjusted life-years. We found insufficient evidence for an association between alcohol use and one of these outcomes and subsequently omitted it from further analysis. This analysis has yielded updated relationships on the relative risk of mortality for five causes, at various levels of alcohol consumption, which we combined with relative risk estimates from GBD 2016 for an additional 17 outcomes. We used this information, along with information on the burden of disease from these 22 diseases and injuries, to estimate the level of alcohol consumption that minimises health loss separately for each age group, sex, year, and region. In addition to estimating the level of consumption that is associated with minimising health loss, known as the theoretical minimum risk exposure level (TMREL), we also estimated the level of alcohol consumption at which the risk to health for a drinker is equivalent to that of a non-drinker—a quantity we refer to as the non-drinker equivalence.

8Implications of all the available evidence

9Our results are consistent with previous findings at the global level, and at the same time the more nuanced analysis done in this study strongly suggests that statements, guidelines, and recommendations on the optimal level of alcohol consumption need to take into consideration the background rates of diseases and injuries for each population. We provide clear evidence that the level of alcohol consumption that minimises health loss varies significantly across populations and remains zero or very close to zero for several population groups, particularly young adults. At the same time, small amounts of alcohol consumption are associated with improved health outcomes in populations that predominantly face a high burden of cardiovascular diseases, particularly older adults in many world regions. Given these findings, we recommend a modification of existing policy guidelines to focus on emphasising differential optimal consumption levels by age, rather than the current practice of recommending different consumption levels by sex. This study highlights the importance of prioritising interventions targeted at minimising alcohol consumption among young adults.

10Importantly, no study to date has examined the variation in the theoretical minimum risk of alcohol consumption by geography, age, sex, and time, conditioned on background rates of disease. National dietary guidelines on low-risk drinking, such as those in the USA, UK, France, and Australia, base recommendations on studies of the risk of alcohol use on all-cause mortality and some cause-specific outcomes.,,, This complicates interpretation of the risk of alcohol use on mortality, given three aspects of all-cause mortality. First, causal pathways between alcohol use and cause-specific outcomes can differ, creating multiple confounding structures that are not readily adjustable when embedded within models analysing the effects of alcohol use on all-cause mortality. Second, all-cause mortality includes non-causally related outcomes, further increasing the threat to internal validity for evidence produced from analysing the effects of alcohol use on all-cause mortality. Third, and most importantly for the present study, the composition of causes within all-cause mortality can differ substantially between populations, changing the proportional risk of mortality due to alcohol use across these populations., In tandem, these features limit the applicability of determining minimum risk exposures on the basis of observational data on alcohol use and all-cause mortality.

11In this study, we used the distribution of causes of disability-adjusted life-years (DALYs) in each population, along with alcohol consumption patterns from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2020, to estimate the TMREL and NDE for each region, age group, sex, and year from 1990 to 2020. Using these estimates, we quantified the proportion of the population consuming alcohol in amounts exceeding these thresholds by location, age, sex, and year, serving as a guide for targeting alcohol control efforts.

12This manuscript was produced as part of the GBD Collaborator Network and in accordance with the GBD Protocol.

13GBD is the most comprehensive effort to date to understand the changing health challenges around the world. In the most recent revision, GBD 2020, estimates were produced for the mortality and health burden from 287 causes of death, 370 diseases and injuries, and 88 risk factors in 204 countries and territories by 5-year age groups and sex from 1990 to 2020. As part of GBD 2020, we estimated the TMREL and NDE of alcohol consumption for 21 regions by 5-year age group, sex, and year for individuals aged 15–95 years and older from 1990 to 2020 (Bryazka D, unpublished). Using the comparative risk assessment framework, we also quantified the population consuming alcohol in harmful amounts, by 5-year age group, country or territory, sex, and year. In the following sections, we provide an overview of our methods. This study adheres to the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) statement.

14As part of GBD, a previous systematic literature review and meta-analysis was published in 2018 that included 592 cohort and case-control studies across 23 outcomes associated with alcohol use. These outcomes included ischaemic stroke, intracerebral haemorrhage, ischaemic heart disease, hypertensive heart disease, atrial fibrillation and flutter, lip and oral cavity cancer, nasopharynx cancer, other pharynx cancer, oesophageal cancer, larynx cancer, colon and rectum cancer, breast cancer, liver cancer, type 2 diabetes, cirrhosis and other chronic diseases of the liver, pancreatitis, idiopathic epilepsy, tuberculosis, lower respiratory infection, transport injuries, unintentional injuries, self-harm, and interpersonal violence. As part of this previous meta-analysis, dose–response relative risk curves for each of these outcomes were estimated through use of a Bayesian meta-regression tool, DisMod ODE.,

15For GBD 2020, we updated this review for the six alcohol-attributable outcomes that accounted for the greatest number of global DALYs: ischaemic heart disease, ischaemic stroke, intracerebral haemorrhage, type 2 diabetes, tuberculosis, and lower respiratory infection. Through the update, we included 71 additional studies. After evaluating all available evidence, we found insufficient evidence for a relationship between alcohol use and lower respiratory infection. Based on these results, we removed this as a risk–outcome pair for GBD 2020 and from this analysis, resulting in 22 remaining relative risk curves. Further details of the systematic review, including search strings, inclusion criteria, Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagrams, and relative risk curves are provided in appendix 1 (pp 18–47).

16Using the updated data for these five outcomes, we revised the relative risk curves associated with each outcome using the meta-regression Bayesian, regularised, trimmed (MR-BRT) meta-regression tool. MR-BRT is a tool that is well suited to the complex task of estimating the dose–response risk association between alcohol and health because it is does not enforce a log-linear functional form, instead parameterising the log relative risk as a B-spline (Zheng P, Institute for Health Metrics and Evaluation, personal communication). It uses an ensemble approach for knot selection of splines based on level of exposure, and incorporates unexplained between-study heterogeneity into the uncertainty of the relative risk estimates. To adjust for aspects of study design that contribute to bias in relative risks, we included covariates for study reference group, adjustment for sick quitter bias, sex, age, population representativeness, outcome reporting method, exposure measurement timing, geographical representativeness, outcome measure (incidence versus mortality), and adjustment for confounders in risk estimation. The MR-BRT tool uses a generalised Lasso approach to select the most relevant bias covariates to adjust for in the final model. A full list of the confounders tested and included in each of these five models is summarised on in appendix 1 (p 15). Consistent with the previous systematic review, we utilised a reference group of non-drinkers. We estimated parameter uncertainty using 1000 draws from the posterior distribution, sampled at 1 g intervals of pure alcohol consumption between 0 g and 100 g per day. Further details of the meta-regression approach are available in appendix 1 (pp 14–16).

17The TMREL and NDE are based on aggregate, burden-weighted relative risk curves across health outcomes associated with alcohol use. Burden was quantified with DALY rates for each region, age, sex, and year obtained from GBD 2020 (Bryazka D, unpublished). DALYs are the sum of years of life lost (capturing the effect of premature mortality) and years lived with disability (capturing the effect of morbidity). For each region, age, sex, and year, we produced all-attributable cause relative risk curves as a weighted average of cause-specific relative risk curves, with weights based on the share of the overall DALY rates from each cause. The step-by-step process and formula for computing the weighted all-attributable cause curves are provided in appendix 1 (p 16). Using these estimates, we computed the TMREL and NDE by region, age, sex, and year. Uncertainty in the relative risk curve, based on 1000 draws of each cause-specific relative risk curve and 1000 draws of DALY rates used for weighting, was propagated to the estimates of TMREL and NDE. All estimates are presented to three significant figures. An example of a weighted all-attributable cause alcohol relative risk curve, for all 22 alcohol associated causes combined, is shown in figure 1.

18Since alcohol use contributes to the DALY rates that are used as weighting factors when constructing the TMREL and NDE, we did a sensitivity analysis that utilised risk-deleted DALY rates as alternative weights. We generated risk-deleted DALY rates by multiplying the DALY rate of each cause by the complement of the cause-specific population-attributable fraction due to alcohol (Bryazka D, unpublished). Additionally, our weighted attributable-cause relative risk curves were based on only 22 of 24 health outcomes since no relative risk curves could be computed for alcohol use disorder or alcoholic cardiomyopathy due to the paucity of studies on dose–response relative risks. To assess whether inclusion of these two outcomes could potentially affect the TMREL and NDE levels, we did a second sensitivity analysis in which we generated conservative hypothetical relative risk functions for alcohol use disorder and alcoholic cardiomyopathy and re-computed TMREL and NDE levels that reflect all 24 alcohol-associated outcomes. Additional details of the sensitivity analyses are presented in the appendix (p 17).

19To estimate the proportion of the population consuming alcohol in excess of the NDE, estimates of alcohol consumption in units of grams of pure ethanol consumed per day, on average, by current drinkers for 204 countries and territories, by age, sex, and year, were obtained from GBD 2020 (Bryazka D, unpublished). Briefly, this process combines supply-side data, household survey data, and administrative data, which allows us to adjust for under-reporting due to self-report bias in surveys, account for unrecorded alcohol consumption, and adjust for consumption among tourists. Current drinkers were defined as individuals who had consumed at least one standard drink in the past 12 months. To facilitate interpretation, we report estimates in terms of standard drinks per day, where one standard drink is defined as 10 g of pure ethanol, consistent with previous GBD publications (Bryazka D, unpublished). Further details on estimation of the prevalence of alcohol use and alcohol consumption have been published previously.

20The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

21The distribution of DALYs arising from outcomes associated with alcohol by GBD super-region, age, and sex for 2020 are shown in figure 2. The TMREL and NDE by region, age, and sex for 2020 are shown in figure 3. Overall, we found that the TMREL remained low regardless of geography, age, sex, or time, varying between 0 (95% UI 0–0) and 1·87 (0·500–3·30) standard drinks per day. As a result of the differences in the cause distributions across world regions, both the TMREL and NDE varied by region. The TMREL and NDE did not vary significantly by sex or year. There was significant variation in the TMREL and the NDE across ages, with younger age groups having much lower TMREL and NDE levels than older adults. In 2020, the TMREL varied between 0 (0–0) and 0·603 (0·400–1·00) standard drinks per day among individuals aged 15–39 years and between 0·114 (0–0·403) and 1·87 (0·500–3·30) standard drinks per day among individuals aged 40 years and older. The NDE varied between 0·002 (0–0) and 1·75 (0·698–4·30) standard drinks per day among individuals aged 15–39 years and between 0·193 (0–0·900) and 6·94 (3·40–8·30) standard drinks per day among individuals aged 40 years and older. This result was mainly driven by differences in the major causes of death and disease burden across ages, as seen in figure 2. Overall, we did not observe any significant differences in the TMREL and NDE between males and females in any age group. In all super-regions, among individuals aged 15–39 years, injuries accounted for the majority of alcohol-related DALYs in 2020. Globally, in this age range, all injuries accounted for 66·3% (95% UI 65·1–67·5) of alcohol-related DALYs for males and 47·9% (46·0–49·8) of alcohol-related DALYs for females; transport injuries comprised 25·9% (25·0–27·0) of alcohol-related DALYs among males and 12·7% (12·0–13·4) among females, self-harm comprised 11·7% (10·1–13·3) of alcohol-related DALYs among males and 12·3% (10·8–13·8) among females, and interpersonal violence comprised 12·4% (11·8–13·0) of alcohol-related DALYs among males and 6·70% (5·90–7·69) among females. The TMREL among males aged 15–39 years in 2020 was 0·136 (0–0·400) standard drinks per day. Among females aged 15–39 years in 2020, the TMREL was 0·273 (0–0·500) standard drinks per day. The NDE was 0·249 (0–1·00) standard drinks per day among males and 0·546 (0–1·30) standard drinks per day among females. The differences in the TMREL and the NDE between females and males were not statistically significant.

22In individuals aged 40–64 years, the health outcomes contributing to the alcohol-related burden shifted to chronic health conditions, including cardiovascular disease and cancer. In this population, ischaemic heart disease comprised 24·1% (95% UI 23·0–25·3) of alcohol-related DALYs among males and 19·5% (18·0–21·0) among females, and intracerebral haemorrhage comprised 10·3% (9·61–10·9) of alcohol-related DALYs among males and 11·7% (10·7–12·8) among females, whereas injuries, such as transport or unintentional injuries, remained significant sources of burden, comprising 23·0% (21·7–24·4) of alcohol-related DALYs among males and 16·7% (15·3–18·3) of alcohol-related DALYs among females. Health outcomes for which moderate alcohol use is associated with a lower risk constituted an increasing portion of the cause distribution in this age group, resulting in a higher TMREL and NDE than in individuals aged 15–39 years. The global TMREL among individuals aged 40–64 years in 2020 was 0·527 (0·400–1·00) standard drinks per day among males and 0·562 (0·400–0·800) standard drinks per day among females. The global NDE in 2020 was 1·69 (0·800–3·20) standard drinks per day among males and 1·82 (1·00–3·10) standard drinks per day among females. As in the younger age group, the differences in the TMREL and the NDE between females and males aged 40–64 years were not statistically significant.

23Among individuals aged 65 years and older, the major causes of disease burden were cardiovascular diseases. In 2020, ischaemic heart disease was responsible for 31·5% (95% UI 30·3–32·7) of all alcohol-related DALYs among males and 29·7% (28·2–31·2) among females, intracerebral haemorrhage was responsible for 11·6% (10·9–12·4) of all alcohol-related DALYs among males and 10·9% (10·1–11·8) among females, and ischaemic stroke was responsible for 14·2% (13·5–14·9) of all alcohol-related DALYs among males and 16·0% (15·2–16·7) among females. As a result, in this population the TMREL was higher than in the younger age groups and was estimated to be 0·636 (0·500–1·00) standard drinks per day among males and 0·656 (0·500–1·00) standard drinks per day among females, whereas the NDE was estimated to be 3·19 (1·50–5·20) standard drinks per day among males and 3·51 (1·70–5·50) standard drinks per day among females. The differences in the TMREL and the NDE between males and females aged 65 years and older were not significant.

24The distribution of the causes of disease burden for a given age group varied substantially across regions, resulting in regional variations in TMRELs and NDEs, particularly in individuals aged 40 years and older. For example, among individuals aged 55–59 years in north Africa and the Middle East, 30·7% (95% UI 27·3–34·6) of alcohol-related DALYs were due to cardiovascular disease, 12·6% (11·0–14·3) were due to cancers, and 0·37% (0·27–0·55) were due to tuberculosis. By contrast, in this same age group in central sub-Saharan Africa, 20·1% (17·2–23·8) of alcohol-related DALYs were due to cardiovascular disease, 9·80% (8·31–11·7) were due to cancers, and 10·1% (6·03–14·1) were due to tuberculosis. As a result, the TMRELs for this age group were 0·876 (0·500–2·00) standard drinks per day in north Africa and the Middle East and 0·596 (0·300–2·00) standard drinks per day in central sub-Saharan Africa. The NDEs also varied, with an NDE of 3·89 (1·50–5·90) standard drinks per day in north Africa and the Middle East and 1·53 (0·600–4·70) standard drinks per day in central sub-Saharan Africa. The TMRELs and NDEs for each region by age and sex for 1990, 2000, 2010, and 2020 are shown in appendix 2 (pp 3–31).

25The distribution of the major causes of DALYs varied slightly between sexes, with injuries making up a larger share of distributions for males than for females. This resulted in mean TMRELs and NDEs that were larger among females compared to males of the same region, age, and year. When taking uncertainty into account, these differences were not significant. However, a larger proportion of males compared to females consume alcohol, and their average level of consumption is also significantly higher. As a result, young males stood out as the group with the highest level of harmful alcohol consumption (figure 4).

26Globally, 1·03 billion (95% UI 0·851–1·19) males (35·1% [29·1–40·7] of the male population aged ≥15 years) and 312 million (199–432) females (10·5% [6·72–14·6] of the female population aged ≥15 years) consumed alcohol in amounts exceeding the NDE in 2020; the number and proportion of people consuming alcohol in excess of the NDE, along with the percentage change since 1990 in the proportion of people consuming alcohol in excess of the NDE, by age group, sex, and location is reported in table 1. Since 1990, the global proportion of drinkers consuming alcohol in excess of the NDE has not changed significantly. Although the proportion of the population consuming harmful amounts of alcohol stayed at the same level over the past three decades, the number of people consuming harmful amounts of alcohol increased from 983 million (718–1190) in 1990 to 1·34 billion (1·06–1·62) in 2020, driven by population growth. Overall, among individuals consuming harmful amounts of alcohol in 2020, 76·9% (73·0–81·3) were male.

27Harmful consumption was predominantly concentrated among individuals aged 15–39 years, who had the lowest TMRELs and NDEs (figure 4). Of the 1·34 billion (95% UI 1·06–1·62) individuals consuming alcohol in excess of the NDE in 2020, 59·1% (54·3–65·4) were aged 15–39 years. Of these, 75·5% (70·3–80·7) were male (595 million [489–658]). Australasia (83·2% [71·1–86·9]), western Europe (79·3% [67·1–84·5]), and central Europe (78·3% [68·1–83·5]) had the highest percentages of males aged 15–39 years consuming harmful amounts of alcohol (table 1). Among females in the same age group, Australasia (77·7% [65·3–82·0]), western Europe (64·3% [44·3–73·6]), and southern Latin America (59·0% [42·0–67·8]) had the highest rates of harmful alcohol consumption. By contrast, only 6·55% (4·79–8·43) of individuals consuming alcohol in excess of the NDE globally were older than 65 years (appendix 2 pp 54–83). Between 1990 and 2019, 14 countries had significant increases in the prevalence of harmful alcohol consumption, 55 countries had significant decreases, and in 135 countries the prevalence did not change significantly (table 1).

28Importantly, these results were not sensitive to our approach to constructing the weighted attributable-cause relative risk curves that are used to calculate the TMREL and NDE. The TMREL and NDE for each of the sensitivity scenarios are shown in appendix 2 (pp 39–41). By including additional risk to health from alcohol use disorders and alcoholic cardiomyopathy, which are 100% attributable to alcohol use, the TMREL decreased by an average of 0·058 (95% UI 0·00–0·30) standard drinks per day in scenario A (for alcohol use disorder and alcoholic cardiomyopathy, relative risks linearly increase with consumption to a maximum relative risk of three at four standard drinks per day, at which point they plateau through ten standard drinks), 0·092 (0·00–0·50) standard drinks per day in scenario B (relative risks linearly increase with consumption to a maximum of five, at four standard drinks per day for alcohol use disorder and alcoholic cardiomyopathy), and 0·146 (0·00–0·60) standard drinks per day in scenario C (relative risks linearly increase with consumption to a maximum of 10, at four standard drinks per day for alcohol use disorder and alcoholic cardiomyopathy), whereas the NDE decreased by an average of 0·188 (0·00–0·90) standard drinks per day in scenario A, 0·310 (0·00–1·30) standard drinks per day in scenario B, and 0·508 (0·00–2·10) standard drinks per day in scenario C. The greatest differences were observed in males in eastern Europe, where alcohol use disorder is especially prevalent, as well as in individuals aged 30–54 years, among whom DALY rates from alcohol use disorder are the greatest (appendix 2 pp 39–41; Bryazka D, unpublished).

29Our results were not sensitive to the changes in relative risk estimates for the five updated outcomes, compared to the entire set of relative risk estimates published in GBD 2016 (appendix 2 pp 43–53). The global, age-standardised, both-sexes TMREL based on the full set of GBD 2016 relative risk estimates and GBD 2020 DALY rates was 0·534 (95% UI 0·00–1·00) standard drinks per day, compared to 0·511 (0·400–0·700) standard drinks per day based on the GBD 2020 relative risk estimates and GBD 2020 DALY rates, whereas the previously published global, age-standardised, both-sexes TMREL was 0·00 (0·00–0·80) standard drinks per day in 2016 based on GBD 2016 relative risk estimates and GBD 2016 DALY rates (table 2).

30We show that the estimation of the health effects associated with alcohol use requires consideration of both the relationship between alcohol consumption and disease outcomes, and the observed disease rates in each population. We found that the population-level health risks associated with low levels of alcohol consumption varied across regions and were greater for younger populations than for older populations. Although we did not find significant differences in the risks of ill health by sex or by year, we did find that males made up 76·9% (95% UI 73·0–81·3) of the population consuming harmful amounts of alcohol in 2020. Notably, 1·03 billion (0·851–1·19) males and 312 million (199–432) females drank harmful amounts of alcohol in excess of the NDE in 2020. Harmful use of alcohol was particularly concentrated in males aged 15–39 years, primarily in Australasia, western Europe, and central Europe. These findings highlight the need for tailored guidelines that discourage alcohol consumption among young people, as well as alcohol control policies and interventions that are targeted especially towards young males.

31Understanding the variation in the level of alcohol consumption that minimises the risk of ill health for populations can aid in setting effective consumption guidelines, supporting alcohol control policies, monitoring progress in reducing harmful alcohol use, and designing public health risk messaging.,,,,,,,, Most alcohol consumption guidelines for the general population combine recommendations to avoid alcohol use with the definition of lower-risk alcohol consumption thresholds, which tend to vary between 8 g and 42 g of alcohol per day for females, and between 10 g and 52 g of alcohol per day for males. Generally, thresholds are one standard drink greater for men than for women, and some lower-risk thresholds are framed in units of weekly consumption that come with a recommendation to avoid alcohol entirely for several days of the week.

32In our analysis, the population-specific TMRELs ranged between 0 (95% UI 0–0) and 0·603 (0·400–1·00) standard drinks per day among individuals aged 15–39 years across world regions, and the NDEs ranged between 0 (0–0) and 1·75 (0·698–4·30) standard drinks per day among individuals aged 15–39 years across world regions in 2020. Even if a conservative approach is taken and the lower bound of the uncertainty interval is used to set policy recommendations rather than the mean, this implies that the recommended level of alcohol consumption in existing low consumption recommendations is too high for younger populations. Our estimates, based on currently available evidence, also do not support low consumption guidelines that differ by sex. Given the known difficulties associated with translating evidence into changes in consumer behaviour, clear messaging around updates to drinking guidelines will be crucial to ensure the full improvements are realised.

33One key distinction between this study and existing recommendations on alcohol consumption is that our estimates focus on minimising health loss across all alcohol-attributable outcomes in a population. Thresholds exist for different purposes; in terms of injury prevention, several countries have moved to a zero-tolerance threshold that is consistent with evidence of the entirely harmful effect of alcohol consumption on injuries. Furthermore, individual-level as opposed to population-level risk minimisation will depend on individual-level factors, including comorbid conditions and the use of pharmaceuticals, which are more prevalent among older populations. Our results for older adults should be interpreted in the context of their additional uncertainty., Approaches to minimising individual-level risk are beyond the scope of this study and need to take into consideration not only alcohol use and specific health outcomes, but also interactions between environmental, genetic, and behavioural factors, as well as the societal and health system context of individuals.

34Broadly, this analysis highlights the need to consider the existing prevalence of diseases and injuries for specific populations when determining the total harms posed by a risk factor. Although the biological effects of alcohol are unlikely to change across populations, except in the case of specific genetic interactions such as variants in alcohol dehydrogenase, disease rates vary substantially across regions, age, sex, and time. For example, alcohol use poses a greater risk to population health in areas with a high prevalence of tuberculosis than in areas with low prevalence. Although this consideration is perhaps most important for setting effective policy recommendations for risks with both harmful and protective relationships with disease, such as alcohol use and red meat consumption, it has implications for all risk factors. As countries navigate the epidemiological transition and their background rates of disease evolve from infectious diseases and injuries to non-communicable diseases, policy recommendations will need to evolve as well.

35It is important to consider our findings in the context of those published by the GBD 2016 Alcohol Collaborators in 2018. Compared to that report, the analysis presented here includes three major changes: we updated five of the relative risk curves; we weighted the relative risk curves using DALY rates estimated as part of GBD 2020 rather than GBD 2016; and we estimated the TMREL separately for each region, age, sex, and year. Although the GBD 2016 Alcohol Collaborators found that the global, age-standardised, both-sexes TMREL was zero standard drinks per day, computing the global TMREL with the first two of these updates, we found that the global TMREL was still quite low, at 0·511 (95% UI 0·400–0·700) standard drinks per day. Re-estimating the TMREL with updated 2020 DALY weights but the former relative risk curves suggests a global TMREL of 0·534 (0–1·00) standard drinks per day. Region-specific, age-specific, and sex-specific differences between these approaches are summarised in appendix 2 (pp 45–55). Importantly, the differences across TMREL by region and age hold even with the relative risk curves estimated in 2016. The more nuanced analysis in the present study, where we explored the risks to ill health by age and region, represents a major step forward in our understanding of how to minimise health loss due to alcohol consumption across the world.

36One challenge associated with using observational studies to measure the causal effect of alcohol consumption on health is the potential for the introduction of various forms of bias, including reverse causation, selection bias, and residual confounding. Mendelian randomisation is a method that attempts to mitigate bias by using genetic variation as a proxy for risk exposure., Although a small number of Mendelian randomisation studies have been done on alcohol use to date, a recent meta-analysis reported that those done on cardiovascular disease and diabetes had varied in their findings, with 67% of studies on cardiovascular disease and 75% of studies on diabetes reporting a null association with alcohol. However, only five of 24 studies examined whether alcohol had a non-linear relationship with these health outcomes. As additional Mendelian randomisation studies from diverse populations are increasingly published, they have the potential to improve the evidence base, and estimates should be regularly revised to reflect new evidence.

37This study had various limitations that should be taken into account when interpreting the findings. First, we did not incorporate patterns of drinking, and therefore did not distinguish between individuals who infrequently engage in heavy episodic drinking and those who consume the same amount of alcohol over several days. Manthey and colleagues estimated that in 2018, 20% of adults engaged in heavy episodic drinking—the consumption of 60 g or more of alcohol on a single occasion—over the past month. Second, due to a paucity of studies reporting a dose–response relationship between the risk of alcohol use and incidence of and mortality from alcohol use disorders, the burden of alcohol use disorders was not included in the TMREL calculation. As shown by the sensitivity analyses, which used conservative hypothetical relative risk curves for alcohol use disorder and alcoholic cardiomyopathy, inclusion of these diseases results in slightly lower estimates of TMREL and NDE, particularly among males in eastern Europe and in individuals aged 30–54 years globally. The decreases in the TMREL and NDE in the sensitivity scenarios were found to be quite small, since the risk of these two conditions is likely to be concentrated at higher levels of consumption and in younger adults, resulting in minimal impact on estimates of the TMREL and NDE. Third, although we attempted to adjust for the impacts of confounding and bias in our meta-regressions, it is possible that relative risk estimates did not account and adjust for all sources of bias, including measurement bias and selection bias, as well as the potential impacts of reverse causality. Fourth, studies reporting the relative risks of alcohol use were based on self-reported alcohol consumption, which is subject to social desirability and recall biases. Fifth, we did not consider differences in risk by type or quality of alcohol. Sixth, the weights used within the weighted alcohol-attributable relative risk curve used DALY estimates that could be due to alcohol use. However, this limitation would only have had a marginal effect on estimates of the TMREL and NDE. Seventh, our estimates of the proportion of the population consuming alcohol in excess of the NDE were derived from alcohol consumption data collected through 2019. Because of delays in routine data collection on risk factors caused by the COVID-19 pandemic, we forecasted our estimates to obtain a time series through 2020. As a result, the estimates do not reflect changes in consumption patterns associated with the pandemic. Last, our results did not include health conditions with burgeoning evidence indicating a relationship with alcohol use, such as major depressive disorder, generalised anxiety disorder, or dementia, given the current scarcity of sufficient evidence to support a meta-analysis and the potential for reverse causality. Inclusions of these outcomes would possibly reduce estimates of the TMREL and NDE.

38In conclusion, the relationship between moderate alcohol use and health is complex and has raised a great deal of controversy in the scientific literature. Given that the available evidence suggests that low levels of alcohol consumption are associated with a lower risk of some disease outcomes and an increased risk of others, alcohol consumption recommendations should take into account the full epidemiological profile that includes the background rates of disease within populations. The findings of this study support the development of tailored guidelines and recommendations on alcohol consumption by age and across regions and highlight that existing low consumption thresholds are too high for younger populations in all regions. Additionally, our results suggest that guidelines should not incorporate sex-specific recommendations, given the absence of variation in TMREL and NDE by sex across geographies and locations. Finally, recognising that the majority of the world's population consuming harmful amounts of alcohol are young adults and predominantly young males, in order to minimise health loss due to alcohol consumption it is important to prioritise interventions targeted at these demographic groups.

39For access to the data used for this specific analysis before the full publication of GBD 2020, please contact Emmanuela Gakidou at gakidou@uw.edu.

40O M Adebayo reports grants or contracts from Merck Foundation and Servier Nigeria; payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing, or educational events from Merck Foundation; support for attending meetings or travel from Servier Nigeria; and a leadership or fiduciary role in a board, society, committee or advocacy group, paid or unpaid, with the Nigerian Association of Resident Doctors; all outside the submitted work. S Afzal reports honorary participation on the Institutional Review Board of King Edward Medical University (Lahore, Pakistan), the Quality Enhancement Cell at Fatima Jinnah Medical University (Lahore, Pakistan), and the Corona Expert Advisory Group (Pakistan); an unpaid leadership or fiduciary role in board, society, committee or advocacy group, with the Pakistan Society of Community Medicine & Public Health, Pakistan Association of Medical Editors, and Pakistan Society of Medical Infectious Diseases; all outside the submitted work. R Ancuceanu reports consulting fees from AbbVie; payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing or educational events from AbbVie, B. Braun Medical, Sandoz, and Laropharm; all outside the submitted work. P Atorkey reports support for the present manuscript via funding from the School of Medicine and Public Health, The University of Newcastle (Callaghan, NSW, Australia), and infrastructure support from Hunter New England-Population Health and Hunger Medical Research Institute, Australia. M Ausloos reports a research grant from the Romanian National Authority for Scientific Research and Innovation, CNDS-UEFISCDI (project number: PN-III-P4-ID-PCCF-2016-0084, title: “Understanding and modelling time-space patterns of psychology-related inequalities and polarization”) outside the submitted work. T Bärnighausen reports grants from the European Union (Horizon 2020 and EIT Health), German Research Foundation (DFG), US National Institutes of Health, German Ministry of Education and Research, Alexander von Humboldt Foundation, Else-Kröner-Fresenius-Foundation, Wellcome Trust, Bill & Melinda Gates Foundation, KfW, Joint United Nations Programme on HIV/AIDS (UNAIDS), World Health Organization; consulting fees from KfW for the OSCAR initiative in Vietnam; participation on a Data Safety Monitoring Board or Advisory Board with National Institutes of Health (US)-funded study “Healthy Options” (PIs: Smith Fawzi, Kaaya) as Chair of the Data Safety and Monitoring Board, with the German National Committee on the “Future of Public Health Research and Education”, as chair of the scientific advisory board to the EDCTP Evaluation, as a member of the UNAIDS Evaluation Expert Advisory Committee, as a National Institutes of Health (US) Study Section Member on Population and Public Health Approaches to HIV/AIDS (PPAH), with the US National Academies of Sciences, Engineering, and Medicine's Committee for the “Evaluation of Human Resources for Health in the Republic of Rwanda under the President's Emergency Plan for AIDS Relief (PEPFAR)”, with the University of Pennsylvania Population Aging Research Center (PARC) as an External Advisory Board Member; leadership or fiduciary role in a board, society, committee or advocacy group, paid or unpaid, with the Global Health Hub Germany (which was initiated by the German Ministry of Health) as a co-chair; all outside the submitted work. S M M Bhaskar reports grants or contracts from the NSW Ministry of Health, Australia; a leadership or fiduciary role in a board, society, committee or advocacy group, paid or unpaid, with the Rotary Club of Sydney as board director, with the International Rotary Fellowship of Healthcare Professionals (UK) as board director, with Global Health & Migration Hub Community, Global Health Hub Germany, Berlin as a chair or manager; all outside the submitted work. J M Castaldelli-Maia reports grants or contracts from Pfizer (Independent Grants for Learning and Change) and the French National Institute for Cancer (INCa); consulting fees from L'Oreal Mental Health Wellness International Board; all outside the submitted work. S Costanzo reports a research grant from the European Foundation for Alcohol Research (ERAB) (ID EA1767; 2018-2020); payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing or educational events from The Dutch Beer Institute Foundation - The Brewers of Europe as a member of the Organizing Committee and speaker for the 9th European Beer and Health Symposium (Bruxelles 2019), and for giving a lecture at the 13th European Nutrition Conference FENS 2019 (Dublin), sponsored by the Beer and Health Initiative (The Dutch Beer Institute foundation - The Brewers of Europe); all outside the submitted work. I Filip reports financial or non-financial support from the Avicenna Medical and Clinical Research Institute (California, USA). R C Franklin reports leadership or fiduciary role in board, society, committee or advocacy group, paid or unpaid with Kidsafe, Farmsafe, Royal Life Saving Society – Australia, and PHAA – Injury Prevention SIG outside the submitted work. C Herteliu reports research grants from the Romanian Ministry of Research Innovation and Digitalization, MCID (ID-585-CTR-42-PFE-2021, Jan 2022-Jun 2023, “Enhancing institutional performance through development of infrastructure and transdisciplinary research ecosystem within socio-economic domain – PERFECTIS”), the Romanian National Authority for Scientific Research and Innovation, CNDS-UEFISCDI (PN-III-P4-ID-PCCF-2016-0084, October, 2018, to September, 2022, “Understanding and modelling time-space patterns of psychology-related inequalities and polarization”; PN-III-P2-2.1-SOL-2020-2-0351, June, 2020, to October, 2020, “Approaches within public health management in the context of COVID-19 pandemic”), and the Ministry of Labour and Social Justice, Romania (30/PSCD/2018, September, 2018, to June, 2019, “Agenda for skills Romania 2020-2025”), all outside the submitted work. J J Jozwiak reports payment or honoraria for lectures, presentations, speakers bureaus, manuscript writing or educational events from Teva Pharmaceuticals, Amgen, Synexus, Boehringer Ingelheim, ALAB Laboratories, and Zentiva, outside the submitted work. M Kivimäki reports support for the present manuscript from the Wellcome Trust (221854/Z/20/Z), the UK Medical Research Council (MR/S011676/1), the US National Institute on Aging (R01AG056477), and the Academy of Finland (350426) in the form of research grants to their institution. K Krishan reports non-financial support from the UGC Centre of Advanced Study (Phase II), Department of Anthropology, Panjab University, Chandigarh, India, outside the submitted work. S Lorkowski reports grants or contracts paid to his institution from Akcea Therapeutics Germany; consulting fees from Danone, Novartis Pharma, Swedish Orphan Biovitrum (SOBI), and Upfield; payment or honoraria for lectures, presentations, speaker's bureaus, manuscript writing, or educational events from Akcea Therapeutics Germany, Amarin Germany, Amedes Holding, Amgen, Berlin-Chemie, Boehringer Ingelheim, Daiichi Sankyo Deutschland, Danone, Hubert Burda Media Holding, Janssen-Cilag, Lilly Deutschland, Novartis, Novo Nordisk, F Hoffmann-La Roche (Roche), Sanofi-Aventis, SYNLAB Holding Deutschland, and SYNLAB Akademie; support for attending meetings or travel from Amgen; and participation on a data safety monitoring board or advisory board for Akcea Therapeutics Germany, Amgen, Daiichi Sankyo Deutschland, Novartis, and Sanofi-Aventis; all outside the submitted work. A M Madureira-Carvalho reports grants or contracts from Instituto Universitário de Ciências da Saúde (Gandra, Portugal); consulting fees from Albert Labs and Eurox Pharma paid to her and her institution; a leadership or fiduciary role in a board, society, committee or advocacy group, paid or unpaid, with the Portuguese Association of Forensic Sciences (APCF); all outside the submitted work. A-F A Mentis reports grants or contracts from “MilkSafe: A novel pipeline to enrich formula milk using omics technologies”, a research co-financed by the European Regional Development Fund of the European Union and Greek national funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH - CREATE - INNOVATE (project code: T2EDK-02222), as well as from ELIDEK (Hellenic Foundation for Research and Innovation, MIMS-860); stock or stock options in a family winery; and support from BGI Group as a scientific officer; all outside the submitted work. C D H Parry reports grants or contracts from the South African Medical Research Council paid to their institution; consulting fees paid to them from the World Health Organization (WHO) and the University of Cape Town (Cape Town, South Africa); payment or honoraria for a lecture on alcohol & NCDs in 2020 from the University of Cape Town; support for attending meetings or travel from UCT African Union 2022, UN Office on Drugs & Crime 2019, and WHO meeting on alcohol in Uganda 2021; participation on a Data Safety Monitoring Board or Advisory Board with the UK SPECTRUM Project (multi university) Global Advisory Board, and the UN Office on Drugs & Crime WDR Scientific Advisory Board, both unpaid; all outside the submitted work. G A Roth reports support for the present manuscript from the Bill & Melinda Gates Foundation via a research grant to their institution. G A Roth reports grants or contracts from the American Heart Association, the American College of Cardiology, and the National Heart, Lung, and Blood Institute paid to their institution, outside the submitted work V Shivarov reports financial support from ICON plc. J A Singh reports consulting fees from Crealta/Horizon, Medisys, Fidia, PK Med, Two Labs, Adept Field Solutions, Clinical Care Options, Clearview Healthcare Partners, Putnam Associates, Focus Forward, Navigant Consulting, Spherix, MedIQ, Jupiter Life Science, UBM, Trio Health, Medscape, WebMD, and Practice Point Communications, the National Institutes of Health, and the American College of Rheumatology; payment or honoraria for lectures, presentations, speakers' bureaus, manuscript writing, or educational events from Simply Speaking; support for attending meetings or travel from the steering committee of OMERACT; participation on a Data Safety Monitoring Board or Advisory Board with the US Food and Drug Administration Arthritis Advisory Committee; leadership or fiduciary role in board, society, committee or advocacy group, paid or unpaid, with OMERACT as a steering committee member, with the Veterans Affairs Rheumatology Field Advisory Committee as Chair (unpaid), and with the UAB Cochrane Musculoskeletal Group Satellite Center on Network Meta-analysis and editor and director (unpaid); stock or stock options in TPT Global Tech, Vaxart Pharmaceuticals, Aytu BioPharma, Adaptimmune Therapeutics, GeoVax Labs, Pieris Pharmaceuticals, Enzolytics, Seres Therapeutics, Tonix Pharmaceuticals and Charlotte's Web Holdings, and previously owned stock options in Amarin, Viking, and Moderna Pharmaceuticals; all outside the submitted work.