The risk relationships between alcohol consumption, alcohol use disorder, and alcohol use disorder mortality: a systematic review and meta-analysis

Abstract

Background and Aims:

Increasing levels of alcohol use are associated with a risk of developing an alcohol use disorder (AUD), which, in turn, is associated with considerable burden. Our aim was to estimate the risk relationships between alcohol consumption and AUD incidence and mortality.

Method:

A systematic literature search was conducted, using Medline, Embase, PsycINFO, and Web of Science, for case-control or cohort studies published between January 1, 2000, and July 8, 2022. These were required to report alcohol consumption, AUD incidence, and/or AUD mortality (including 100% alcohol-attributable deaths). The protocol was registered with PROSPERO (CRD42022343201). Dose-response and random-effects meta-analyses were used to determine the risk relationships between alcohol consumption and AUD incidence and mortality, and mortality rates in AUD patients, respectively.

Results:

Of the 5,904 reports identified, seven and three studies from high-income countries and Brazil met the inclusion criteria for quantitative and qualitative syntheses, respectively. In addition, two primary US data sources were analysed. Higher levels of alcohol consumption increased the risk of developing or dying from an AUD exponentially. At an average consumption of four standard drinks (assuming 10 g of pure alcohol/standard drink) per day, the risk of developing an AUD was increased seven-fold (relative risk (RR) 7.14, 95 % confidence interval [CI] 5.13–9.93) and the risk of dying four-fold (RR 3.94, 95 % CI 3.53–4.40) compared with current non-drinkers. The mortality rate in AUD patients was 3.13 (95% CI 1.07–9.13) per 1,000 person-years.

Conclusions:

There are exponential positive risk relationships between alcohol use and both alcohol use disorder incidence and mortality. Even at an average consumption of 20 g/day (about one large beer), the risk of developing an AUD is nearly threefold that of current non-drinkers and the risk of dying from an AUD is approximately double that of current non-drinkers.

INTRODUCTION

Alcohol use disorder (AUD) is characterised by an impaired ability to control alcohol intake and compulsive alcohol use over extended periods of time.(1,2) Affected individuals display escalating patterns of drinking that may result in serious consequences on both their physical and mental health, and within their social environments.(1,3) Thus, AUDs are highly disabling and potentially lethal.(4,5) However, despite their potential for disability or death, they are among the most undertreated mental disorders with less than one in five individuals with an AUD receiving treatment for the condition/episode at the time of onset. (1,3,6)

AUDs rank as one of the most prevalent mental disorders globally.(7,8) Previously, they have been shown to predominantly affect men, though sex differences in prevalence are narrowing over time, due to a global increase in the number of women drinking at higher risk levels since 2000.(9,10) In 2016, 8.6% of adult men and 1.7% of adult women were affected by AUDs globally.(11) Variations in prevalence of AUDs are also apparent for other factors, including country income level and geographic region.(12) Rising prevalence of AUDs contribute tremendous risks to both health and social life and are associated with considerable burden.(13) In 2016, for example, alcohol use led to a loss of 117.2 million disability-adjusted life years (DALYs) and 2.0 million premature deaths globally,(14) and AUD specifically caused health harms that accounted for 145,000 deaths.(15)

With higher levels of alcohol intake, the risk of developing or dying from an AUD increases.(16–18) This exposure-outcome relationship seems intuitive, but little is known about the shape of the underlying dose-response relationship connecting the level of alcohol use to AUD incidence or mortality. Additionally, these risk relationships are likely dependent on several factors including age, sex, race/ethnicity, and socioeconomic status (SES). Relative to men, women drink less often and consume less alcohol but, are more susceptible to specific alcohol-related problems and are less likely to receive help for these problems.(19,20) Furthermore, among women themselves, there are racial/ethnic disparities in the access to treatment and quality of care that they receive.(21) For instance, in the United States (US), both Black and Latina women have a significantly lower likelihood than White women to access specialty treatment for AUD.(21) Additionally, previous research suggests that socioeconomic inequalities are approximately two times higher for 100% alcohol-attributable deaths compared to all-cause mortality.(22) As a result, alcohol use may not relate to AUD incidence or mortality in the same way across sociodemographic subgroups.

The aim of this study was to disentangle the complex associations between alcohol use, AUD, and AUD mortality by analyzing three risk relationships, separately. We conducted a systematic review and meta-analysis, which was additionally supported by two primary data analyses (described in more depth in the “Methods” section) exploring the risk relationships of: (i) the level of alcohol use and AUD incidence [use->AUD inc.]; (ii) the level of alcohol use and AUD mortality [use->AUD mort.]; (iii) having an AUD and dying from an AUD [AUD->AUD mort.] (Figure 1). Given substantial differences in drinking patterns and AUD prevalence across sociodemographic groups, we also aimed to account for the effects of age, sex, race/ethnicity, and SES on each of these relationships, where possible.

Figure 1.

Display of the three risk relationships to be evaluated in this review.

AUD: alcohol use disorder

METHODS

Search strategy

A systematic literature search was conducted on July 8, 2022, via Medline, Embase, and PsycINFO (via OVID), and Web of Science, to find studies on the associations between alcohol use, incidence of AUD, and AUD mortality, with no language restrictions applied. The electronic databases were searched from January 1, 2000, to July 8, 2022, for original, observational studies using search terms including the study design (“case-control” or “cohort”), outcome (“incidence” or “mortality”), and exposure (“alcohol use” or “alcohol consumption”, and “alcohol dependence” or “alcohol abuse”) (see Supplementary Material eTable 3). Screening of the references was conducted first at title and abstract and then at full text by three reviewers (TC, LLF, CK). Every reference was screened by two independent reviewers and inter-rater reliability amongst the reviewers was determined using the Cohen’s kappa statistic.(23) Backward and forward citation tracking was performed for each of the included articles, in addition to a systematic grey literature search (see Supplementary Material eMethods 5).

Studies were included in the review if they met the following criteria: (a) participants were at least 18 years of age; (b) the study used a prospective or retrospective cohort design drawn from the general population (for risk relationships i [use->AUD inc.] and ii [use->AUD mort.]) or a sample of AUD patients (for risk relationship iii [AUD->AUD mort.]) or a case-control design with a sample of cases of AUD incidence or mortality and controls; (c) they reported at least one of the three risk relationships between average alcohol consumption level, AUD, and AUD mortality (Figure 1), using either odds ratios (OR), risk ratios (RR) or hazard ratios (HR) with corresponding confidence intervals (CI), or mortality rates (risk-relationship iii only), or sufficient data to calculate respective quantitative results. AUD was defined according to Roerecke and Rehm’s criteria: AUD diagnosed by a medical professional (physician or psychiatrist), participated in an AUD treatment program, drove while intoxicated or registered at a temperance board, and/or met criteria for AUD on a standardized and validated questionnaire.(4) AUD mortality was defined as dying from mental and behavioural disorders due to the use of alcohol (F10) according to the International Classification of Diseases, 10^th revision (ICD-10) criteria.(24) Studies that included causes of death that are 100% alcohol-attributable and a likely result of an AUD characterized by heavy alcohol use over time (e.g., alcoholic liver cirrhosis) in addition to AUD mortality, were also eligible for inclusion. This is because registration of the cause of death as a 100% alcohol-attributable category rather than F10 may be mainly a reflection of coding practices with an AUD being the underlying driver. Thus, the following ICD-10 codes were included: E24.4, G31.2, G62.1, G72.1, I42.6, K29.2, K70–70.4, K70.9, K85.2, K86.0, R78.0, X45-X45.9, X65, Y15-Y15.9, Y90, Y91, Y91.0-Y91.3, Y91.9 when reported grouped together with F10–10.9 (see Supplementary Material eTable 4). Studies were excluded if they employed an experimental or cross-sectional design, if they represented a specific subgroup, or if they reported results for AUD incidence and mortality combined. The review protocol was registered with PROSPERO under the ID number CRD42022343201, and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) criteria were applied (25) (see Supplementary Material eTable 1).

Data extraction

From all included articles the following details were extracted by one reviewer (TC) and checked by another independent reviewer (CP or AL): authors’ names, year of publication, country, design and duration of study, follow-up years, age, sex, number of participants and events (AUD incidence/AUD mortality), operationalization of alcohol use/AUD/cause of death, adjustments used, and OR/RR/HR with corresponding CIs or mortality rates (relationship iii). When stratified by the covariates of interest, estimates of RR and their CIs were additionally extracted. Any inconsistencies were discussed and decided by consensus between reviewers. When information required for analysis was not available, the authors were contacted to obtain the necessary results. Data was additionally extracted from two US data sources including the longitudinal National Epidemiologic Survey of Alcohol and Related Conditions (NESARC) and the National Health Interview Survey (NHIS) linked to mortality data.

Secondary data sources

We used secondary data from two US studies that allowed us to perform primary analyses. 1) Two longitudinal waves of the NESARC survey, conducted in 2001/2002 and 2004/2005 were used to calculate point estimates for the risk of developing an AUD at various levels of alcohol consumption [logistic regression model, refer to table S9 for sample size etc.]. 2) NHIS 1997–2018 linked to the 2019 National Death Index data were used to calculate HRs of dying from an AUD at different levels of alcohol use [Cox proportional hazard model]. Greater detail surrounding the methodology can be found in Supplementary Material eMethods 9 and 10.

Statistical analysis

One-stage random-effects dose-response meta-analyses were used to explore the nature of the dose-response risk relationships between alcohol use and AUD incidence (i) [use->AUD inc.] and mortality (ii) [use->AUD mort.] at various levels of alcohol use in grams per day (g/day).(26) Alcohol consumption in average grams of pure alcohol per day served as the exposure variable. When alcohol consumption was not reported in grams of pure alcohol per day, it was estimated from the reported quantity in standard drinks and frequency. If the study provided a range for quantity or frequency estimates (e.g., 3 to 5 standard drinks per occasion), the midpoint was taken, and if there was no upper bound to the highest category (e.g., 6 or more standard drinks per occasion), 75% of the width of the previous range was added to the lower bound of the highest level to represent the point estimate for this category (e.g., in our example 8.25 standard drinks). The country-specific definition of a standard drink provided in the respective study was used when converting standard drinks to grams of pure alcohol. The outcome variable was AUD incidence for (i) [use->AUD inc.] and AUD mortality for (ii) [use->AUD mort.]. Two models (linear and quadratic) were tested for each of the dose-response relationships, and optimal fit was determined based on the AIC (Akaike Information Criterion) and the BIC (Bayesian Information Criterion), with the lowest value indicating the best fit to the data.

In the risk relationship (iii) [AUD->AUD mort.], we were interested in the AUD mortality risk of people with AUD. The outcome of interest was AUD mortality rate per person-years (PY). When PY were not provided, they were estimated as follows: If provided with the total number of all-cause deaths and the all-cause mortality rate per PY within the target population (AUD patients), these values were divided to obtain the PY. For one study by Haver et al.,(27) PY were estimated based on a survival plot by extracting the survival rates at various time points and multiplying these figures by the number of people in the sample. The total number of survivors at each time point was summed together to get the total PY. Here, a random-effects meta-analysis was conducted using single mortality rates to obtain the weighted average of the mortality risk of people with an AUD.

To evaluate the between-study heterogeneity, Cochran’s Q and the I² statistic were used, such that a significant Q value was indicative of substantial heterogeneity, along with an I² value of ≥75%. I² > 25% and < 75% represented moderate heterogeneity and I² ≤ 25% represented low heterogeneity.

A qualitative summary was performed where a study met the inclusion criteria but the way in which alcohol consumption was reported did not allow for accurate conversion into grams per day. For example, one study, Zaridze et al.,(28) assessed consumption in terms of only one beverage (vodka), making it relatively incomparable to other studies, even when the grams of pure alcohol per day were obtained.

Statistical analyses were conducted in R 4.2.1, using the dosresmeta package for (i) and (ii),(29) and the meta package for (iii).(30)

Risk of bias assessment

The Newcastle-Ottawa Scale (NOS) for case-control and cohort studies was adapted to assess the risk of bias of the selected studies (see Supplementary Material eTable 7,8).(31) Risk of bias was assessed using three subcategories by two independent reviewers per study and any conflicts were resolved through deliberation amongst the reviewers. Results were reported on a scale of 0 to 4 for the selection category where the risk of bias was high if rated a score of ≤1, moderate if rated a 2 or 3, and low if rated a 4. All cohort studies that used mortality rates (iii) [AUD->AUD mort.] were scored one point lower on a scale of 0 to 3 points, with 3, 2, and ≤1 representing a low, moderate, and high risk of bias, respectively, as one item within the selection category was not applicable (no comparison to an unexposed cohort). The same scale was used for the outcome category. For the comparability category which reported results on a scale of 0 to 2, risk of bias was low if rated a 2, moderate if rated a 1, and high if rated a 0.

RESULTS

Of the initial 5,904 records screened for inclusion, a total of ten studies were included in the systematic review, seven of which were eligible for quantitative synthesis, (27,32–37) in addition to two secondary US data sources (i.e., NESARC and NHIS) (Figure 2). A substantial reviewer agreement of Kappa ranging between 0.60 and 0.69 was achieved amongst the reviewers for both the screening of titles and abstracts as well as full texts. In total, the included studies (quantitative analysis only) reported on results from Sweden (three), the US (two), Brazil (one), Finland (one), Iceland (one), and Norway (one), with 3,760 alcohol-related events (AUD incidence and mortality cases) observed overall in the quantitative analysis. For most of the studies (n=6), a longitudinal cohort design was employed, with record linkage to register data. One study was of case-control design and used matched general population controls with data from a cause of death registry (Table 1). Reports that provided sex and age-stratified estimates were limited, however, for risk relationship (ii) [use->AUD mort.], three studies contributed male-specific estimates of AUD deaths,((33,34), NHIS) and for risk relationship (iii) [AUD->AUD mort.], two studies contributed female-specific estimates of AUD deaths, for those with an AUD.(27,35) None of the included studies investigated the role of age, SES, or race/ethnicity on any of the three risk relationships, neither was it possible to explore sex-specific dose-response relationships between alcohol use and AUD incidence.

Figure 2.

PRISMA flowchart of study selection.

*One study informed two risk relationships. NESARC: National Epidemiologic Survey of Alcohol and Related Conditions, NHIS: National Health Interview Survey

Table 1.

Characteristics of studies and data sources included in quantitative analysis.

Study/data source	Country	Study years	Mean follow-up years (range)	Study design	Sample size	Sex	Mean age or range at baseline	Number of cases	Alcohol consumption/exposure	Reference category	Outcome	Adjustments	Risk of bias (selection/comparability/outcome)
Risk relationships (i) and (ii): [use->AUD inc.] and [use->AUD mort.]
Laatikainen et al., 2013 (34) (ii)	Finland	1987–1997	7.3 (5–10)	Longitudinal cohort study, with record linkage to national mortality register	5,092	100% M	42.1 (heavy drinkers), 45.5 (no heavy drinking)	32	Heavy drinking pattern	No heavy drinking occasions	100% alcohol-attributable mortality; ICD-10	N/A	4/2/3
Romelsjö et al., 2012 (33) (ii)	Sweden	1969–2004	N/A (1–35)	Longitudinal cohort study, with record linkage to mortality data	48,716	100% M	18–20	210	Alcohol consumption (g/day)	Abstainers (0g/day)	100% alcohol-attributable mortality; ICD-10	N/A	4/0/3
Thern et al., 2021 (32) (i and ii)	Sweden	2002–2007	13.3 (1–16)	Longitudinal cohort study, with record linkage to national registers	37,484	45.6% M, 54.4% W	25–70	1301 (i), 67 (ii)	Alcohol consumption (g/day)	Light drinkers (6g/day)	AUD incidence [including identical ICD-10 codes as for mortality], standardized questionnaire; ICD-10 [Swedish index of alcohol-related diagnoses] 100% alcohol-attributable mortality; specifically, AUD, alcohol-related liver cirrhosis, alcohol poisoning; ICD-10	Age, sex, country of birth	4/2/3
*NESARC (see Supplementary Material eMethods 9) (i)	United States	2001–2005	3 (N/A)	Longitudinal study	24,581	33.9% M, 66.1% W	N/A	1265	Alcohol consumption (g/day)	Lifetime abstainers	AUD incidence; AUDADIS-4 interview; DSM-IV	Age, sex, education, race/ethnicity	4/2/3
*NHIS (see Supplementary Material eMethods 10) (ii)	United States	1997–2018	10.5 1–22	Longitudinal study	562,042	43.8% M, 56.2% W	50.3	825	Alcohol consumption (g/day)	Lifetime abstainers	100% alcohol-attributable mortality; ICD-10	Sex, education, race/ethnicity, marital status, survey year, survey design	4/2/3
Risk relationship (iii): [AUD->AUD mort.]
da Roza et al., 2022 (36)	Brazil	2002–2016	11.3 (0–14)	Longitudinal study, with database linkage	803	54.8% M, 45.2% W	35	14	AUD diagnosis (F10)	N/A	100% alcohol-attributable mortality; ICD-10	Age and sex	3/0/2
Gunnarsdottir et al., 2014 (35)	Iceland	2002–2008	N/A (1–7)	Prospective cohort study, with record linkage to a national cause-of-death registry	107,237	53.3% M, 46.7% W	41.5	18	AUD diagnosis (F10)	N/A	100% alcohol-attributable mortality and alcoholic liver disease mortality; ICD 10	Age, sex, number of visits, year of entrance, mental and behavioural disorders at discharge	3/1/3
Haver et al., 2009 (27)	Sweden	1981–2007	N/A (0–25)	Case-control study between patients and matched GP controls, with data from cause of death register	357	100% W	42.5	20	AUD diagnosis	N/A	100% alcohol-attributable mortality; alcohol explicitly mentioned in diagnostic category	N/A	2/2/2
Hjemsaeter et al., 2019 (37)	Norway	1997–2016	N/A (1–19)	Prospective, longitudinal cohort study, with record linkage to cause of death registry	102	72% M, 28% W	45.8	17	AUD diagnosis	N/A	100% alcohol-attributable mortality; ICD-10	N/A	3/1/3

AUD: alcohol use disorder; AUDADIS-4: Alcohol Use Disorder and Associated Disabilities Interview Schedule, Diagnostic and Statistical Manual of Mental Disorders, Fourth version; GP: general population; N/A: not applicable; NESARC: National Epidemiologic Survey of Alcohol and Related Conditions; NHIS: National Health Interview Survey;

^*Secondary data sources analyzed by the authors

Risk of bias assessment

Four studies, including the two secondary data sources, achieved the maximum rating, indicative of a low risk of bias. All of these studies observed the relationship between alcohol use and AUD mortality, except for NESARC which informed the relationship between alcohol use and AUD incidence. Three other reports received a moderate risk of bias in the category of comparability due to the absence of additional control variables (i.e., race/ethnicity or SES). Within the remaining reports, descriptions of their cohort/case selection criteria or comparability controls were lacking thus, they achieved the lowest ratings in these categories and had the highest risk of bias. For category specific ratings, see Table 1.

Alcohol use and AUD incidence (i)

Quantitative summary

In total, two identified studies reported 2,566 AUD incident cases (limited to one per individual) out of 57,460 individuals. The dose-response relationship between alcohol use and AUD incidence was best described by a model including a quadratic term (formula and AIC/BIC values available in Supplementary Material eMethods 11 and eResults 12). The quadratic dose-response relationship between alcohol use and the risk of AUD incidence (non-log scale) is shown in Figure 1 (for log scale, see Supplementary Material eResults 13). Overall, the risk of incident AUD increases immediately and exponentially as alcohol use surpasses 0g/day (Table 2). At 20 g/day, the RR is 2.74 (95% CI 1.48–5.08) and at 40 g/day, the RR is 7.14 (95% CI 5.13–9.93), relative to 0 g/day. In other words, a person drinking 40 grams of pure alcohol per day has a 7.1-fold increased risk of developing an AUD compared to non-drinkers. Beyond this point, the RR begins to increase more drastically, and as consumption surpasses 60 g/day, the RR increases by more than 1 for every 1 g increase in the amount of pure alcohol consumed (Figure 1). However, given the low number of point estimates for high levels of consumption, the dose-response relationship at high levels of consumption is subject to large uncertainty.

Table 2.

Relative risk of incident AUD for increasing levels of alcohol use.

Grams of pure alcohol per day	Relative risk	95% Confidence interval
10	1.67	1.09–2.54
20	2.74	1.48–5.08
40	7.14	5.13–9.93
60	17.64	6.95–44.76

Qualitative summary

Two additional studies explored the relationship between alcohol consumption and the risk of AUD incidence, in which alcohol use was assessed using frequency of “binge” or “risky” drinking. Though they reported an average level of consumption over a given time, this could not accurately be converted into a grams of pure alcohol per day measurement. This made them ineligible for inclusion in the quantitative analysis (Table 3).

Table 3.

Characteristics of studies included in the qualitative analysis.

Study	Country	Study years	Follow-up years	Study design	Sample size	Sex	Mean age	Alcohol consumption/exposure	Reference category	Outcome
Tavolacci et al., 2019 (38)	France	2017	N/A	Retrospective case-control study	166	86.7% M, 13.3% W	34.6	Alcohol consumption (frequency of binge drinking before age 18, between 18–25, and between 25–45)	Non-alcohol-dependent (AUDIT score < 8) (controls)	AUD incidence
Dawson et al., 2008 (39)	United States	2001–2005	3	Prospective cohort study	22,122	61.9% M, 38.1% W	38.9	Alcohol consumption (frequency of risk drinking in year preceding Wave 1 interview)	No risky drinking	AUD incidence; AUDADIS-4 interview; DSM-IV
Zaridze et al., 2009 (28)	Russia	2001–2005	Deaths occurred from 1990–2001. Data collected by proxy from 2001–2005	Case-control study, reported by proxy	48,557	64.9% M, 35.1% W	15.74	Alcohol consumption (half-litre bottles of vodka, or equivalent, by usual weekly intake)	Usual weekly consumption always <0.5 bottles of vodka or equivalent, and maximum consumption of spirits in 1 day always <0.5 bottles	100% alcohol-attributable mortality; ICD-10 (X45, Y15, F10)

Tavolacci et al.(38) is a retrospective case-control study that investigated the association between binge drinking during the ages of 18–25, and AUD in adulthood, using data from a sample of adults who sought treatment for AUD in France (13.3% women, mean age of 34.6 years). They included 83 cases of AUD (13.3% women) and 83 matched controls (13.3% women). The study assessed frequency and prevalence of consumption and binge drinking, with binge drinking defined as consumption of four/five or more alcoholic drinks in a single day for women/men respectively, and frequent binge drinking (>twice a month). Frequent binge drinking in early adulthood was a risk factor for AUD in adulthood with three-fold increased odds compared to those who did not report binge drinking or reported occasional binge drinking in early adulthood. The frequency of binge drinking occasions between 18 and 25 were significantly higher in AUD cases relative to controls.

Dawson et al.(39) investigated the association between the frequency of risky drinking (defined as five or more alcoholic drinks in a single day for men, and four or more for women) and the incidence of adverse outcomes (including AUD) approximately three years later, using US NESARC Waves 1 and 2 data (n=22,122, 38.1% women, mean age of 38.9 years). Individuals were categorized into four groups depending on their frequency of risky drinking and the incidences of AUD were presented as OR, adjusted for several factors including age, sex, race/ethnicity, SES (marital status, education, employment), and health status. The findings suggested that baseline risky drinking led to a significantly increased risk of incident AUD; even when risky drinking occasions occurred at low frequency (<1/month), the ORs for incident “alcohol abuse” and “alcohol dependence” were 1.59 (95% CI 1.25–2.02), and 1.35 (95% CI 1.05–1.73), respectively. The adjusted ORs increased steadily with the frequency of risky drinking occasions, up to 3.93 (95% CI 2.40–6.44) for “alcohol abuse” and 7.23 (95% CI 4.75–11.00) for “alcohol dependence” at the highest frequency of daily/near daily.

Together, the studies reported an overall increased risk of incident AUD when binge drinking occasions, regardless of the number of occasions per month, occurred 3+ years prior to AUD diagnosis. This was compared to matched controls or those who never engaged in binge drinking. These results support our findings that AUD incidence risk increases with greater quantities and frequencies of alcohol consumption.

Alcohol use and AUD mortality (ii)

Quantitative summary

A total of four identified studies included 612,964 individuals who reported their alcohol consumption, 1,134 individuals later died from an AUD. For the dose-response relationship between alcohol use and AUD mortality, the linear model showed the best fit (formula and AIC/BIC values available in Supplementary Material eMethods 11 and eResults 12). Figure 2 displays the linear dose-response relationship between the level of alcohol use (average grams of pure alcohol per day) and the log-RR of dying from an AUD, for men, women, and all participants, on a non-log scale (for log scale, see Supplementary Material eResults 13). As alcohol consumption increases, the risk of AUD mortality increases in an accelerated manner. As shown in Table 4, at 20 g/day, the RR is 1.99 (95% CI 1.88–2.10), increasing to 3.94 (95% CI 3.53–4.40) at 40 g/day. Similarly, as consumption increases from 60 to 80 g/day, the RR increases from 7.82 (95% CI 6.63–9.22) to 15.52 (95% CI 12.46–19.34). Two studies included men only, and among this group, the RR of AUD mortality increases with every 20 g increase in consumption in a similar manner to the values reported above for the combined men, women, and all participants (see Supplementary Material eResults 14).

Table 4.

Relative risk of AUD mortality for all participants at increasing levels of alcohol use.

Grams of pure alcohol per day	Relative risk	95% Confidence interval
20	1.99	1.88–2.10
40	3.94	3.53–4.40
60	7.82	6.63–9.22
80	15.52	12.46–19.34
100	30.81	23.41–40.56

Qualitative summary

One study similarly observed the RR of death from AUDs and alcohol poisoning (F10, X45, Y15) at varying levels of alcohol intake per week, reported by proxy (spouse/partner, sibling, parent, adult offspring, or other adult relative) (Table 3).(28) In this report by Zaridze et al.,(28) alcohol consumption was presented in half-litre bottles of vodka or equivalent for men and women separately (n=48,557, 35.1% women, age range=15–74.) A fixed-effect model was used to combine estimates for men and women to create an “all participants” category with “reference drinkers” defined as 16g of pure alcohol per day as the reference category. Despite these efforts, data from the study could not be used because consumption patterns in Russia vary significantly from the rest of the world. For example, alcohol intake was measured at significantly higher doses (0 to approximately 300 g/day) in this study relative to the other included studies.

AUD and AUD mortality (iii)

Quantitative summary

Across all four identified studies, 60 AUD deaths were observed in 22,216 PY of follow-up. The mortality rate due to AUD for those with an AUD for each study, as well as the weighted average by sex, is displayed in Figure 3. Participants were grouped into three categories. The “all participants” category was comprised of overall estimates for all participants in the study, while the “women” and “men” categories included estimates that were stratified by sex. The overall mortality rate per 1,000 PY was 3.13 for all participants (95% CI 1.07–9.13), 2.17 (95% CI 1.09–4.35) for men, and 1.50 (95% CI 0.33–6.78) for women with an AUD, respectively. Between-study heterogeneity was substantial for the “all participants” category, in which Hjemsaeter et al.(37) produced a significantly higher mortality rate than the other included studies. For the sex-specific risk estimates, the interpretation is limited by the fact that only one and two studies provided estimates for men and women, respectively. A leave-one-out analysis was performed for each study to determine if studies contributed substantial leverage to estimated results (see Supplementary Material eResults 15).

Figure 3.

Dose-response relationship between the average level of alcohol consumption and the relative risk of developing an AUD based on a dose-response meta-analysis with two studies [(32), NESARC]. Bubble size indicate inverse variance weights. The dashed line and grey area indicate the 95% confidence interval of the dose-response relationship. All data points are based off of a sample including men and women.

AUD: alcohol use disorder; g/day: average grams of pure alcohol consumed per day.

DISCUSSION

To our knowledge, this is the first study to describe the risk relationships between alcohol use, AUD, and AUD mortality, which are imperative for understanding the harmful effects of alcohol consumption and preventing associated harms. The study identified exponential risk relationships between alcohol use and both AUD incidence and death. Moreover, even at an average consumption of 20 g/day (about one large beer), the risk of developing an AUD is nearly threefold that of current non-drinkers. At this level of consumption, the risk of dying from an AUD is approximately double that of current non-drinkers. Among men and women with diagnosed AUD, the average AUD mortality rate was found to be 3.1 per 1,000 PY, overall. This means if we would observe 500 individuals for two years, we would expect about 3 AUD deaths. Importantly, the studies were based on very young samples with an average age around 30; hence, AUD mortality rates are expected to be higher in samples with an older mean age. For each of the studied risk relationships, the effect modification by age, race/ethnicity, and SES could not be accounted for, due to a lack of available evidence. Insufficient data of this type can have significant consequences. Failure to consider variations in alcohol use and health outcomes across racial/ethnic and SES groups can further perpetuate or worsen existing health disparities and limit the accessibility or effectiveness of treatment for more vulnerable populations, in particular. Future studies should seek to address these key research gaps by investigating the risk relationships individually and how they vary by age, sex, race/ethnicity, SES, as well as by levels and patterns of alcohol use, as an important foundation for a targeted public health strategy to reduce the burden of disease related to AUD.

Our data suggest that the risk of developing and dying from AUD may differ by sex. However, the availability of sex-specific data was limited and gender-specific data was not available. There are several explanations for potential sex differences. For example, drinking patterns and beverage preferences among men may lead to increased risks of AUD incidence and mortality compared to women. Men are more likely to engage in heavy and frequent drinking, with a higher consumption of spirits, which could explain higher mortality rates among men.(20,40) Furthermore, there is a potential excess risk associated with spirits consumption due to rapid ethanol intake and intoxication which may contribute to elevated risks among men.(40) On the other hand, several factors may contribute to higher risks among women. First, at the same level of alcohol intake, women tend to have higher blood alcohol concentrations and may experience more progressive liver damages,(41) potentially contributing to higher AUD incidence and mortality rates among women. Moreover, women with AUD may face more barriers to learning about AUD treatment options (42) and to receiving treatment.(43,44)

Overall, our results are consistent with previous findings that suggest the prevalence of AUD significantly increases with the level of consumption and the frequency of heavy episodic drinking.(45) However, data on the specific levels of consumption at which the risk increases are still limited, both for AUD incidence and mortality. According to our findings, a reduction in daily average consumption from 60 g/day to 40 g/day may lower the RR from approximately 18 to 7 for developing an AUD, and from 8 to 4 for dying from an AUD. Similarly, the risks are about halved when reducing an average consumption from 40 g/day to 20 g/day.

While this is the most comprehensive review on the risk relationships between alcohol use, AUD, and AUD mortality as of now, there are some limitations to consider when interpreting its results. First, the outcome of AUD mortality was defined by both acute and chronic causes of death under the assumption that they are 100% alcohol-attributable, which reflects a broader definition of AUD mortality. As recording of AUD as an underlying cause of death is subject to coding practices that vary by country and over time, and is likely subject to stigmatization that may introduce further variation,(46) we believe that this approach provides a better and more realistic estimate of AUD mortality. However, this introduced some heterogeneity in the included causes of death across included studies. Furthermore, none of the included studies used lifetime abstainers as the reference category for exposure to alcohol use, which makes our estimates prone to the sick quitters effect and a potential underestimation of the observed risk relationships.(47) These estimates may have additionally been affected by our use of the midpoint of an alcohol exposure range; it is possible that participants consumed on average in the lower end of a given range since alcohol consumption generally follows a right-skewed distribution.(48)

The majority of these studies relied on self-reported alcohol use, which is likely to be under-reported due to drinkers’ recall bias and the use of surveys employing simple quantity-frequency indices that may not accurately reflect consumption habits.(49) In addition, the risk relationship between AUD and AUD mortality was based on a select group of individuals who attended treatment for an AUD. This may have either led to an under-estimation of the risks as the included individuals received AUD treatment or to an over-estimation based on the inclusion of a group of individuals with greater severity of AUD. We expect that our findings best reflect an over estimation of risk as previous studies have demonstrated that those treated for an AUD compared to those with an AUD in the general population have significantly greater all-cause mortality rates, with RRs of 3.38 (95% CI 2.98–3.84) for men and 1.91 (95% CI 1.51–2.42) for women, likely due to greater severity of dependence and higher rates of comorbidities.(4) We were also unable to comment on the latency period of exposure to outcome, as is true with many chronic conditions. However, given the included longitudinal studies that used HR for survival analysis and accounted for time of follow-up, the bias should have been minimized. Due to a gap in available research, the total number of studies included in the review was low. As a result, we had to combine research findings reporting on both men and women with those referring to men only, as was done for risk relationship (ii) [use->AUD mort.]. This may have resulted in RR values that better represent the (potentially higher) mortality risk of alcohol use on men, rather than just women. Finally, most studies included in this review were from high-income countries, limiting the generalisability of our findings.

In conclusion, there is a clear dose-response relationship between levels of alcohol consumption and the risk of AUD incidence and mortality. Further research is needed to validate and explore these risks, particularly regarding subgroup analyses, to gain a more comprehensive understanding of AUD incidence and mortality in relation to alcohol consumption levels and patterns. This will have important implications for public health and policy implementation. By understanding the risks associated with the level of alcohol use, individuals can make informed decisions regarding their drinking habits, seek earlier interventions, and adhere more effectively to treatment plans. At the same time, policies can focus on improving public awareness of alcohol-related risks, while providing interventions that strive to reduce alcohol-related harms, in order to improve public health outcomes.

Figure 4.

Dose-response relationship between the average level of alcohol consumption and the relative risk of death due to an AUD based on a dose-response meta-analysis, including four studies [(32–34),NHIS]. Inverse variance weights are indicated by bubble size. Men are represented by the blue bubbles, women by red, and “All participants” are indicated by green. The 95% confidence interval of the resulting dose-response relationship is indicated by dashed lines and the grey area.

AUD: alcohol use disorder; g/day: average grams of pure alcohol consumed per day.

Figure 5.

Observed mortality rate per 1000PY, due to an AUD, in individuals previously diagnosed with an AUD based on four studies [(27),(35–37)].

AUD: alcohol use disorder; CI: Confidence Interval; PY: person-years