The World Health Organization risk drinking levels measure of alcohol consumption: prevalence and health correlates in US adult nationally representative surveys, 2001-2002 and 2012-2013

Abstract

Objective:

Little is known about change over time in the prevalence of World Health Organization (WHO) risk drinking levels (very high, high, moderate, low) and their association with health conditions, overall, and by gender.

Methods:

Data on current drinkers from the 2001–2002 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC; N=26,655) and the 2012–2013 NESARC-III (N=25,659) were analyzed using logistic regression. Prevalence differences between surveys were estimated for each drinking level overall and by gender. Within each survey, prevalence differences by WHO risk drinking level were estimated for health conditions: alcohol use disorders [AUD], drug use disorders, functional impairment, liver disease, depressive/anxiety disorders.

Results:

In 2012–2013, the prevalences of moderate, high and very high risk drinking were 5.9%, 3.2% and 3.5%, respectively, representing significant increases from 2001–2002 of +0.9% in very high risk drinking, +0.6% in high risk drinking and +1.0% in moderate drinking. The increase for very high risk drinking in men (+0.5%) was smaller than the increase in women (+1.4%). Within both surveys, compared to low risk, health conditions were significantly associated with very high risk (range of prevalence differences: +2.2%-+57.8%), high risk (+2.6%-+41.3%), and moderate risk (+0.6%-+29.8%) drinking. Associations were similar by gender, except there were stronger effects for AUD in men and for functional impairment and depressive/anxiety disorders in women.

Conclusions:

The increase in potentially problematic drinking levels among US adults emphasizes the need for better prevention and treatment strategies. Results support the validity of the WHO risk drinking levels, which show clinical utility as non-abstinent drinking reduction treatment goals. Such goals could engage more people in treatment, improving public health by decreasing personal and societal consequences of risk drinking.

INTRODUCTION

Heavy drinking and alcohol use disorders (AUD) contribute substantially to morbidity and mortality worldwide^1,2, mainly through liver disease, injury, cancer, cardiovascular disease and impaired psychosocial functioning^3,4. However, few individuals receive treatment for problematic drinking^5–8, often because they are not interested in abstinence, the goal most commonly offered in treatment settings^9–11. Recently, non-abstinent drinking reduction treatment goals, which may be more attainable and engage more people in treatment, have gained attention^12,13. Psychopharmacologic treatments successfully reduce drinking to non-abstinent levels^14–16, reductions are maintained over time¹⁷, and are associated with decreased mortality¹⁴, improved health, and reduced negative consequences of drinking in clinical and general population samples^17–23. In these studies, drinking reduction was measured using the World Health Organization (WHO) risk drinking levels, a gender-specific metric indicating the level of risk associated with the average daily amount of alcohol consumed: very high risk, high risk, moderate risk, or low risk²⁴.

Recent studies and a meta-analysis²⁵ of time trends in alcohol consumption among US adults show increases in any alcohol use^25–27 and heavy use (binge drinking)^25,26, specifically among women^25,26,28,29. However, none of these studies measured consumption using the WHO risk drinking level definitions. What is known about the prevalence of the WHO levels was estimated in older data (2001–2002), with 2.5% of current drinkers at very high risk, 2.5% at high risk, 4.8% at moderate risk, and 90.2% at low risk¹⁸. However, more recent prevalence data are lacking, and whether the prevalence of WHO risk drinking levels has changed over time, and if there are differences between men and women remain unknown. Further, relationships between the WHO risk drinking levels and clinically important drinking consequences, e.g., alcohol dependence¹⁸; drug dependence²³; reduced quality of life^19,21; mental health functional impairment¹⁸; impaired liver function^17,21 and disease²⁰; and anxiety and depressive disorders²² used data collected over 15 years ago. Additionally, the relationships between alcohol use and consequences differ in men and women³⁰. Given the many changes in US society and the prevalence of alcohol-related conditions^26,31–41, updated information is needed on the associations of health conditions with the WHO risk drinking levels, overall and by gender.

To examine these issues, we used data on US adults from two nationally representative surveys, the 2001–2002 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC)⁴² and the 2012–2013 NESARC-III⁸. First, we determined whether prevalence of WHO risk drinking levels changed between 2001–2002 and 2012–2013, and whether changes varied by gender. Second, we examined whether health conditions related to alcohol (alcohol dependence, AUD, drug dependence, drug use disorders, functional impairment, liver disease, and depressive/anxiety disorders) were associated with WHO risk drinking level within each survey, and whether these associations differed by gender.

METHODS

Sample and procedures

NESARC⁴² and NESARC-III⁸ are nationally representative surveys of civilian adults age≥18, sampled from households and group quarters using multistage probability sampling designs. Sample weights adjusted the data for nonresponse and selection probabilities to represent the US civilian population based on the 2000 Census for NESARC⁴³ and the 2012 American Community Survey for NESARC-III⁴⁴. Both surveys utilized similar rigorous field procedures, e.g., structured interviewer training, on-going supervision, and quality control assurance^8,42–45. The methodological similarities of the surveys have enabled their use to examine change over time in health outcomes^{26,31–34,46–48}. The National Institute on Alcohol Abuse and Alcoholism (NIAAA) sponsored the surveys, with the field work carried out by large survey organizations (NESARC: US Census Bureau; NESARC-III: Westat). Institutional Review Boards from the US Census Bureau and Office of Management and Budget (NESARC), NIAAA (both surveys), and Westat (NESARC-III) approved the protocol and consent procedures. All respondents gave informed consent after getting a complete description of the study. Interviews were conducted from 2001–2002 for NESARC and 2012–2013 for NESARC-III, with overall response rates of 81.2% and 60.1%, respectively. The total analyzed sample (N=52,314) included current drinkers with information about drinks per day, from NESARC (N=26,655) and NESARC-III (N=25,659). From all current drinkers (N=52,724), 410 (0.78%) were missing daily drinking information and excluded from the analysis.

Measures

Both surveys used the Alcohol Use Disorder and Associated Disabilities Interview Schedule (AUDADIS) to assess drinking, AUD, drug use disorders, sociodemographic variables, and other health conditions. The AUDADIS is a fully-structured computer-assisted diagnostic interview. AUDADIS-IV⁴⁹ was used for NESARC and AUDADIS-5⁵⁰ for NESARC-III.

WHO risk drinking levels

In both surveys, identical questions assessed alcohol use²⁶, which were used to estimate average ethanol consumed per day in the past year. AUDADIS measures of past year average daily ethanol showed substantial to excellent reliability (intraclass correlation coefficients=0.68–0.92)^51–53. Average daily ethanol was used to categorize respondents into the WHO very high risk, high risk, moderate risk and low risk categories (Supplemental Table 1), as in recent studies^18,20,22,23. For men, very high risk was defined as >100 g/d or >7.1 U.S. standard drinks; high risk as >60 to 100 g/d or >4.3 to 7.1 drinks; moderate risk as >40 to 60 g/d or >2.9 to 4.3 drinks; and low risk as 1 to 40 g/d or 1 to 2.9 drinks. For women, very high risk was defined as >60 g/d or >4.3 standard drinks; high risk as >40 to 60 g/d or >2.9 to 4.3 drinks; moderate risk as >20 to 40 g/d or >1.4 to 2.9 drinks; and low risk as 1 to 20 g/d or 1 to 1.4 drinks.

The WHO risk level variable was the outcome for the first aim (change between surveys in risk level prevalence) and a predictor for the second aim (association with health conditions).

Health conditions

Alcohol use disorders (AUD).

For consistency across surveys, DSM-IV criteria were used, and past-year AUD diagnoses were positive if respondents had alcohol dependence or abuse in the past year. Dependence required ≥3 of 7 DSM-IV dependence criteria, while abuse required ≥1 of 4 DSM-IV abuse criteria. Because extensive evidence indicates that all 11 criteria reflect a single continuum⁵⁴, dependence and abuse were combined into one variable (AUD). Most of the symptom items used to assess criteria in both surveys were identical, and the few trivial differences could not account for the changes in AUD prevalence across the two surveys²⁶. Alcohol dependence was included as a separate outcome because WHO drinking risk levels were previously associated with alcohol dependence¹⁸. The reliability and validity of AUDADIS DSM-IV AUD and alcohol dependence diagnoses were substantial to excellent (kappa>.60) in national and international studies, in general and clinical populations^{51–53,55–61}.

Drug use disorders.

Similarly, past-year drug use disorders (dependence or abuse) were diagnosed using the DSM-IV criteria for marijuana, cocaine, heroin, prescription opioids, sedatives/tranquilizers, hallucinogens, stimulants, inhalants, and club drugs. Respondents positive for any drug use disorder were considered positive for the drug use disorder variable. As with alcohol, the small differences in operationalization between the surveys had little effect on prevalence for marijuana³¹, heroin³², prescription opioids³⁵, and cocaine use disorders³⁴. Any drug dependence was included as a separate outcome, because WHO drinking risk levels were previously associated with drug dependence²³. In multiple studies, reliability and validity of any AUDADIS DSM-IV drug use disorder/drug dependence as well as drug-specific disorders was generally substantial to excellent^{52,53,55–60}.

Functional impairment.

Both surveys included the Medical Outcomes Study 12-Item Short Form Health Survey, Version 2 (SF-12)⁶², a valid measure of general functioning used in clinical⁶³ and general population surveys⁶⁴. The SF-12 was used to calculate a standardized mental component summary score (mean=50; standard deviation=10), shown to be related to AUD^5,8. Functional impairment was defined as ≥1 standard deviation below the mean, i.e., scores of ≤40, as done previously¹⁸.

Liver disease.

Both surveys asked identical questions about whether respondents had cirrhosis of the liver or another form of liver disease in the past year. As done previously, liver disease was considered positive if a doctor or health professional confirmed to the respondent that they had cirrhosis or other liver disease²⁰.

Depressive/anxiety disorders.

AUDADIS-IV provided diagnoses of past year DSM-IV anxiety (generalized anxiety, panic, agoraphobia, social or specific phobia) and depressive (major depression, dysthymia) disorders. Reliability was moderate for anxiety disorders (kappa=0.40–0.52)⁵¹ and moderate to substantial for depressive disorders (kappa=0.50–0.73)^51,55. AUDADIS-5 provided DSM-5 diagnoses of the depressive/anxiety disorders, which showed fair to moderate reliability (kappa=0.39–0.51)⁶⁵ and validity (kappa=0.32, 0.40, for any anxiety disorder and any depressive disorder, respectively)⁶⁶, but not DSM-IV diagnoses. Because depressive/anxiety disorders show high comorbidity^67–69 and cluster together on the internalizing dimension of the transdiagnostic model⁷⁰, they were combined into one variable, any depressive/anxiety disorder, as was done previously²². An additional variable was defined as positive for respondents with any depressive/anxiety disorder not due to substances or illness.

Sociodemographic variables

Covariates were measured identically in both surveys: gender (men; women); age (18–29; 30–44; 45–64; ≥65); education (less than high school, high school, some college, graduated college or more); race/ethnicity (Hispanic; non-Hispanic: White, Black, American Indian/Alaska Native, Asian/Native Hawaiian/Pacific Islander); current smoking (yes, no); and health insurance (any, none). These covariates were used in previous studies of WHO risk levels in NESARC data^18,20,22,23.

Statistical analysis

As in previous studies of substance-related trends^{26,31–34,47,48}, the two datasets were concatenated, and a survey variable (2001–2002 or 2012–2013) was added. SUDAAN 11.0.1 was used for analysis to incorporate survey weights and adjust for complex sampling⁷¹. Weighted prevalence was evaluated for WHO risk drinking levels, health conditions, and sociodemographic covariates, by survey.

To test for change in the prevalence of WHO risk levels over time (between surveys), we used multinomial logistic regression to model the risk level variable (outcome) as a function of time (survey), adjusting for covariates (gender, age, race/ethnicity, education, smoking status, insurance). In each survey, weighted model-predicted marginal prevalence estimates (back transformed from log odds) and standard errors were generated⁷² for each risk level. For each risk level, the prevalence difference between 2012–2013 and 2001–2002 indicated time trends. A prevalence difference significantly greater than 0 indicated that prevalence was higher in 2012–2013 than 2001–2002, i.e., it increased over time; a prevalence difference significantly lower than 0 indicated a decrease. To determine whether prevalence difference differed by gender, an interaction term of survey x gender was included in the model. Prevalence differences (trends) were estimated in men and women, and the difference in trends in men versus women (difference-in-prevalence differences) was evaluated. A difference-in-differences significantly different from 0 indicated differential trends in men and women. To adjust for potentially different covariate effects in men and women, this model also included gender x covariate interaction terms.

To evaluate association of WHO risk level and health conditions within each survey, we used logistic regression to model each health condition (outcome) as a function of WHO risk level, survey, and risk level x survey interaction, adjusting for covariates. To adjust for potentially different covariate effects in the different risk levels, risk level x covariate interactions were also included. (Both surveys were included in the same model so that adjustments would be consistent across surveys.) For each outcome, weighted model-predicted marginal prevalence estimates and standard errors were generated within each survey and each risk level. Within each survey, association is indicated by the prevalence difference between each risk level and the reference (low risk level). To determine whether association differed by gender, a three-way interaction term of risk level x survey x gender was included in the model (as well as survey x gender). In each survey, association was estimated in men and women, and contrasts were used to determine whether these differed.

All tests were 2-tailed, with significance set at p<.05, as indicated by 95% CI not including 0.

RESULTS

Sample characteristics (Table 1)

Table 1:

Sample descriptives

	2001–2002a			2012–2013a
	n	Prevalenceb		n	Prevalenceb
		%	SE		%	SE
Sociodemographics
Gender
Men	12,886	52.5	.43	11,935	50.7	.37
Women	13,769	47.5	.43	13,724	49.3	.37
Age
18–29	6,076	24.3	.41	6,384	23.8	.45
30–44	9,373	34.0	.38	7,839	28.2	.38
45–64	7,858	30.5	.36	8,451	34.7	.39
≥65	3,348	11.2	.29	2,985	13.3	.42
Race/ethnicity
White	16,562	75.3	1.44	14,236	68.5	.73
Black	4,129	9.0	.59	5,102	10.7	.58
Hispanic	4,899	10.6	1.13	4,868	14.3	.63
Asian/Native Hawaiian/Pacific Islander	660	3.2	.41	1,092	4.9	.42
American Indian/Alaska Native	405	1.9	.15	361	1.6	.13
Health insurance
Any	21,677	81.8	.55	20,466	82.8	.48
None	4,978	18.2	.55	5,193	17.2	.48
Current smoker
Yes	8,436	33.0	.61	8,248	30.9	.53
No	18,219	67.0	.61	17,411	69.1	.53
Education
Less than high school	1,097	3.4	.24	858	2.8	.17
High school	9,499	34.9	.61	8,604	31.2	.65
Some college	6,057	22.9	.39	6,035	22.7	.36
College degree	10,002	38.8	.66	10,162	43.3	.78
WHO risk drinking levels
Low risk	23,984	89.7	.29	22,202	87.5	.32
Moderate risk	1,314	4.9	.17	1,561	5.9	.19
High risk	686	2.7	.15	865	3.1	.16
Very high risk	671	2.7	.15	1,031	3.4	.17
Health Conditions
Alcohol dependence	1,457	5.8	.21	2,493	9.0	.27
Alcohol use disorderc	3,283	12.9	.35	4,571	17.4	.40
Any drug dependenced	218	0.9	.08	577	2.1	.13
Any drug use disorderc,d	689	2.8	.15	1,393	5.1	.19
Liver disease	154	0.6	.06	268	1.1	.09
Functional impairment	2,759	9.5	.23	3,957	13.7	.31
Depressive/anxiety disorderse, any	4,771	17.3	.40	5,989	22.7	.44
Depressive/anxiety disorderse, not substance or illness induced	4,595	16.7	.39	5,692	21.6	.42

^a2001–2002 data from NESARC survey (N=26,655); 2012–2013 data from NESARC-III survey (N=25,659)

^bweighted prevalence

^cuse disorder includes abuse or dependence

^dfor marijuana, cocaine, heroin, painkillers (prescription opioids), sedative/tranquilizers, hallucinogens, stimulants, inhalants, and club drugs

^eincludes anxiety (panic disorder, agoraphobia, social phobia/social anxiety disorder, specific phobia, generalized anxiety disorder) and depression (dysthymia/persistent depressive disorder, major depression), DSM-IV for 2001–2002, DSM-5 for 2012–2013

Both surveys had ~50% men, ~80% with any insurance, ~30% current smokers, and about two-thirds with at least some college education. The 2012–2013 sample was older (~50% ages 18–44) than the 2001–2002 sample (~60% ages 18–44), and had a higher prevalence of minorities (~30%) than 2001–2002 (~25%) and of health conditions (range of 1.1% for liver disease to 22.5% for depressive/anxiety disorders) than 2001–2002 (range of 0.6% for liver disease to 17.3% for depressive/anxiety disorders).

Prevalence of WHO risk drinking levels over time

Overall, compared to 2001–2002 prevalence, significant increases were observed in the 2012–2013 prevalence of moderate risk (+1.0%), high risk (+0.6%), and very high risk drinking (+0.9%), with a concomitant decrease in prevalence of low risk drinking (−2.5%) (Table 2). Trends over time differed by gender. In women, there were significant increases in moderate (+1.1%), high (+0.5%), and very high (+1.4%) levels. In contrast, in men, a significant increase was seen in moderate risk (+0.9%). In the very high risk group, the increase in men (+0.5%) was significantly smaller (−1.0%) than the increase in women (+1.4%; Table 3).

Table 2:

Change over time (trend) in WHO risk drinking levels, 2001–2002 (NESARC) to 2012–2013 (NESARC-III), overall

	2001–2002a Prevalenceb		2012–2013a Prevalenceb
					Prevalence differencec
WHO drinking risk level	%	SE	%	SE	%	95% CI
Very high risk	2.6	.14	3.5	.17	+0.9	0.49, 1.31
High risk	2.6	.15	3.2	.16	+0.6	0.17, 1.03
Moderate risk	4.9	.16	5.9	.19	+1.0	0.51, 1.49
Low risk	89.8	.28	87.4	.32	−2.5	−3.34, −1.66

CI = confidence interval

^a2001–2002 data from NESARC survey (N=26,655); 2012–2013 data from NESARC-III survey (N=25,659)

^badjusted for sample weights and sociodemographic covariates (gender, age, education, race/ethnicity, health insurance, and current smoking)

^cPrevalence at 2012–2013 minus prevalence at 2001–2002 indicates the trend. Prevalence differences whose 95% CI do not include 0 are statistically significant at p<.05 and are bolded.

Note that the prevalence and prevalence differences are rounded, such that subtracting the values may not yield the exact difference reported.

Table 3:

Trends in WHO risk drinking levels, 2001–2002 (NESARC) to 2012–2013 (NESARC-III), by gender

	Men (N=24,821)						Women (N=27,493)						Trend differences, men vs. womend
	2001–2002a		2012–2013a		Prevalence differencec		2001–2002a		2012–2013a		Prevalence differencec
WHO risk level	Prevalenceb		Prevalenceb				Prevalenceb		Prevalenceb
%	SE	%	SE	%	95%CI	%	SE	%	SE	%	95%CI	%	95%CI
	%	SE	%	SE	%	95% CI	%	SE	%	SE	%	95% CI	%	95% CI
Very high	3.3	.19	3.7	.24	+0.5	−0.09, 1.09	1.8	.18	3.3	.17	+1.4	0.93, 1.87	−1.0	−1.69, −0.31
High	3.4	.21	4.0	.24	+0.6	−0.03, 1.23	1.7	.15	2.2	.17	+0.5	0.05, 0.95	+0.1	−0.59, 0.79
Moderate	4.1	.21	5.0	.25	+0.9	0.27, 1.53	5.8	.25	6.8	.29	+1.1	0.36, 1.84	−0.2	−1.16, 0.76
Low	89.2	.38	87.2	.38	−2.0	−3.02, −0.98	90.7	.35	87.7	.42	−3.0	−4.08, −1.92	+1.0	−0.27, 2.27

CI = confidence interval

^a2001–2002 data from NESARC survey (n: men = 12,886; women = 13,769); 2012–2013 data from NESARC-III survey (n: men = 11,935; women = 13,724)

^badjusted for sample weights and sociodemographic covariates (gender, age, education, race/ethnicity, health insurance, and current smoking)

^cPrevalence at 2012–2013 minus prevalence at 2001–2002 indicates the trend. Prevalence differences whose 95% CI do not include 0 are statistically significant at p<.05 and are bolded.

^dFor each WHO risk drinking level, the trend (prevalence difference) in men minus the prevalence difference in women indicated whether change over time differs by gender. Trend differences whose 95% CI do not include 0 are statistically significant at p<.05 and are bolded.

Note that the prevalence, prevalence differences, and trends differences are rounded, such that subtracting the values may not yield the exact difference reported.

Association of WHO risk levels with health conditions

DSM-IV alcohol use disorders

In 2001–2002, the prevalence of alcohol dependence was significantly greater in very high risk (+40.4%), high risk (+21.4%), and moderate risk (+12.3%), drinkers than in low risk drinkers (Table 4). Similarly, in 2012–2013, the prevalence was significantly greater in very high risk (+49.7%), high risk (+32.0%), and moderate risk (+20.1%) drinkers than in low risk drinkers. Significant associations were observed in men and women, with stronger association for moderate risk in men than women (Table 5). Similar results were observed for AUD (Tables 4, 5).

Table 4:

Association of WHO risk drinking level with health conditions, 2001–2002 (NESARC) and 2012–2013 (NESARC-III)

	2001–2002a				2012–2013a
	Prevalenceb of condition		Prevalence differencec		Prevalenceb of condition		Prevalence differencec
	%	SE	%	95% CI	%	SE	%	95% CI
Alcohol dependence
Very high risk	43.4	2.49	40.4	35.56, 45.24	54.6	2.45	49.7	44.90, 54.50
High risk	24.4	2.11	21.4	17.24, 25.56	36.9	2.35	32.0	27.40, 36.61
Moderate risk	15.3	1.07	12.3	10.21, 14.40	25.0	1.43	20.1	17.30, 22.90
Low risk	3.0	0.14	reference		4.9	0.19	reference
Alcohol use disorderd
Very high risk	56.6	2.77	47.7	42.29, 53.11	70.4	2.32	57.8	53.25, 62.35
High risk	40.9	2.24	32.0	27.55, 36.45	53.9	2.38	41.3	36.56, 46.04
Moderate risk	33.2	1.52	24.3	21.40, 27.20	42.3	1.41	29.8	26.92, 32.68
Low risk	8.9	0.25	reference		12.6	0.34	reference
Any drug dependencee
Very high risk	5.6	0.99	5.0	3.06, 6.94	7.7	0.96	6.1	4.24, 7.96
High risk	1.1	0.39	0.6	−0.18, 1.38	4.2	0.79	2.6	1.05, 4.15
Moderate risk	1.2	0.30	0.6	0.01, 1.19	2.9	0.38	1.2	0.42, 1.98
Low risk	0.6	0.07	reference		1.7	0.13	reference
Any drug use disorderd,e
Very high risk	13.0	1.48	11.0	8.08, 13.92	17.4	1.62	13.1	9.96, 16.24
High risk	5.7	0.98	3.7	1.84, 5.56	11.1	1.23	6.8	4.37, 9.23
Moderate risk	4.7	0.74	2.8	1.39, 4.21	7.9	0.70	3.6	2.23, 4.97
Low risk	2.0	0.12	reference		4.3	0.19	reference
SF-12 functional impairment
Very high risk	21.9	2.24	13.0	8.57, 17.43	25.2	1.75	11.9	8.37, 15.43
High risk	12.0	1.59	3.1	0.01, 6.20	16.5	1.52	3.1	0.02, 6.18
Moderate risk	11.4	1.05	2.6	0.54, 4.66	13.8	0.99	0.4	−1.62, 2.42
Low risk	8.9	0.23	reference		13.3	0.33	reference
Liver disease
Very high risk	3.4	0.98	2.9	0.98, 4.82	3.1	0.85	2.2	0.51, 3.89
High risk	0.9	0.37	0.4	−0.33, 1.13	2.3	0.80	1.4	−0.17, 2.97
Moderate risk	0.6	0.28	0.2	−0.35, 0.75	1.1	0.30	0.2	−0.43, 0.83
Low risk	0.5	0.06	reference		0.9	0.09	reference
Any depressive/anxiety disorderf
Very high risk	25.8	2.16	9.2	4.95, 13.45	29.8	1.87	7.1	3.47, 10.73
High risk	19.5	1.92	2.9	−0.84, 6.64	24.7	1.78	2.0	−1.57, 5.57
Moderate risk	17.6	1.25	1.0	−1.45, 3.45	22.4	1.22	−0.3	−2.81, 2.21
Low risk	16.6	0.36	reference		22.7	0.47	reference
Any depressive/anxiety disorderf, no substance or illness induced
Very high risk	24.2	2.13	8.2	4.03, 12.37	28.8	1.80	7.1	3.47, 10.73
High risk	18.6	1.85	2.7	−0.91, 6.31	23.6	1.74	1.9	1.61, 5.41
Moderate risk	17.2	1.25	1.2	−1.25, 3.65	21.3	1.20	−0.4	2.85, 2.05
Low risk	16.0	0.36	reference		21.7	0.45	reference

CI = confidence interval

^a2001–2002 data from NESARC survey (N=26,655); 2012–2013 data from NESARC-III survey (N=25,659)

^badjusted for sample weights and sociodemographic covariates (gender, age, education, race/ethnicity, health insurance, and current smoking)

^cPrevalence in each risk level minus prevalence in the reference risk level (low) indicates the association at each survey. Prevalence differences whose 95% CI do not include 0 are statistically significant at p<.05 and are bolded.

^dabuse or dependence

^efor marijuana, cocaine, heroin, painkillers (prescription opioids), sedative/tranquilizers, hallucinogens, stimulants, inhalants, and club drugs

^fincludes anxiety (panic disorder, agoraphobia, social phobia/social anxiety disorder, specific phobia, generalized anxiety disorder) and depression (dysthymia/persistent depressive disorder, major depression), DSM-IV for 2001–2002, DSM-5 for 2012–2013

Note that the prevalence and prevalence differences are rounded, such that subtracting the values may not yield the exact difference reported.

Table 5:

Association of WHO risk drinking level with health conditions, by gender

2001–2002 (NESARC)
	Men (n=12,886)				Women (n=13,769)				Difference in prevalence differences, men vs. womenc
	Prevalence a of condition		Prevalence differenceb		Prevalence a of condition		Prevalence differenceb
	%	SE	%	95% CI	%	SE	%	95% CI	%	95% CI
Alcohol dependence
Very high risk	45.7	2.91	42.0	36.32, 47.68	39.4	3.66	37.3	30.15, 44.45	4.7	−3.79, 13.29
High risk	27.6	2.64	23.9	18.73, 29.07	20.6	3.64	18.5	11.37, 25.63	5.4	−3.38, 14.18
Moderate risk	16.1	1.68	12.4	9.09, 15.71	14.1	1.37	12.0	9.31, 14.69	0.4	−3.97, 4.77
Low risk	3.7	0.20	reference		2.1	0.17	reference
Alcohol use disorderd
Very high risk	60.7	3.09	48.8	42.72, 54.88	52.2	4.54	46.7	37.80, 55.60	2.1	−8.15, 12.35
High risk	48.0	2.91	36.1	30.36, 41.84	31.9	3.76	26.4	18.97, 33.83	9.8	0.24, 19.36
Moderate risk	39.1	2.33	27.2	22.63, 31.77	26.9	1.70	21.3	18.05, 24.55	5.9	0.53, 11.27
Low risk	11.9	0.38	reference		5.6	0.27	reference
Any drug dependencee
Very high risk	5.5	1.18	4.8	2.49, 7.11	5.7	1.47	5.3	2.42, 8.18	−0.5	−3.99, 2.99
High risk	1.7	0.61	1.0	−0.22, 2.22	0.3	0.19	−0.2	0.61, 0.21	1.1	−0.19, 2.39
Moderate risk	1.4	0.46	0.8	−0.10, 1.70	0.9	0.34	0.5	−0.21, 1.21	0.3	−0.84, 1.44
Low risk	0.7	0.11	reference		0.4	0.07	reference
Any drug use disorderd,e
Very high risk	12.1	1.68	9.7	6.39, 13.01	13.7	2.39	12.3	7.60, 17.00	−2.6	−8.24, 3.04
High risk	7.2	1.43	4.8	2.04, 7.56	3.3	1.20	1.9	−0.45, 4.25	2.9	−0.88, 6.68
Moderate risk	5.1	0.95	2.7	0.84, 4.56	4.2	0.98	2.8	0.86, 4.74	−0.2	−2.65, 2.25
Low risk	2.4	0.18	reference		1.4	0.13	reference
SF-12 functional impairment
Very high risk	15.8	2.22	9.5	5.11, 13.89	25.9	3.38	14.1	7.38, 20.82	−4.6	−12.05, 2.85
High risk	8.5	1.44	2.2	−0.60, 5.00	16.7	3.21	4.9	−1.45, 11.25	−2.7	−9.64, 4.24
Moderate risk	8.0	1.17	1.6	−0.79, 3.99	15.3	1.73	3.5	0.13, 6.87	−1.9	−6.00, 2.20
Low risk	6.4	0.25	reference		11.8	0.37	reference
Any depressive/anxiety disorderf
Very high risk	18.4	2.07	7.3	3.20, 11.40	30.5	3.71	7.8	0.49, 15.11	−0.5	−8.77, 7.77
High risk	13.6	1.79	2.5	−1.03, 6.03	27.0	3.82	4.3	−3.21, 11.81	−1.8	−10.07, 6.47
Moderate risk	11.2	1.39	0.0	−2.70, 2.70	24.6	1.97	1.9	−2.02, 5.82	−1.9	−6.49, 2.69
Low risk	11.1	0.37	reference		22.7	0.56	reference
Any depressive/anxiety disorderf, no substance or illness induced
Very high risk	16.8	2.03	6.1	2.10, 10.10	29.5	3.64	7.6	0.43, 14.77	−1.5	−9.61, 6.61
High risk	13.0	1.73	2.3	−1.11, 5.71	26.0	3.74	4.1	−3.25, 11.45	−1.8	−9.97, 6.37
Moderate risk	11.0	1.38	0.4	−2.29, 3.09	23.8	1.96	2.0	−1.88, 5.88	−1.6	−6.09, 2.89
Low risk	10.7	0.37	reference		21.9	0.55	reference
2012–2013 (NESARC-III)
	Men (n=11,936)				Women (n=13,724)				Difference in prevalence differences, men vs. womenc
	Prevalence a of condition		Prevalence differenceb		Prevalence a of condition		Prevalence differenceb
	%	SE	%	95% CI	%	SE	%	95% CI	%	95% CI
Alcohol dependence
Very high risk	55.1	3.17	49.7	43.47, 55.93	54.3	3.18	49.9	43.73, 56.07	−0.2	−8.16, 7.76
High risk	40.7	2.78	35.4	29.95, 40.85	32.8	3.39	28.4	21.76, 35.04	7.0	−0.72, 14.72
Moderate risk	30.2	2.28	24.9	20.41, 29.39	19.8	1.57	15.4	12.30, 18.50	9.5	4.21, 14.79
Low risk	5.3	0.27	reference		4.4	0.25	reference
Alcohol use disorderd
Very high risk	73.7	2.86	58.2	52.57, 63.83	66.9	3.16	57.6	51.33, 63.87	0.6	−7.19, 8.30
High risk	59.6	2.63	44.0	38.65, 49.35	48.7	3.70	39.4	32.03, 46.77	4.7	−3.65, 13.05
Moderate risk	49.1	2.28	33.5	28.84, 38.16	35.2	1.65	25.9	22.47, 29.33	7.6	1.82, 13.38
Low risk	15.5	0.38	reference		9.3	0.35	reference
Any drug dependencee
Very high risk	7.7	1.12	6.0	3.86, 8.14	7.8	1.35	6.2	3.57, 8.83	−0.3	−3.22, 2.62
High risk	5.2	1.21	3.4	1.01, 5.79	3.3	0.91	1.8	−0.00, 3.60	1.7	−1.20, 4.60
Moderate risk	3.4	0.65	1.7	0.35, 3.05	2.2	0.49	0.6	−0.36, 1.56	1.0	−0.80, 2.80
Low risk	1.7	0.18	reference		1.5	0.16	reference
Any drug use disorderd,e
Very high risk	16.0	2.08	10.8	6.74, 14.86	19.0	2.13	15.8	11.64, 19.96	−5.0	−10.29, 0.29
High risk	12.9	1.78	7.7	4.13, 11.27	9.4	1.65	6.2	3.01, 9.39	1.5	−3.24, 6.24
Moderate risk	9.3	1.15	4.1	1.75, 6.45	6.3	0.80	3.2	1.75, 4.65	0.9	−1.90, 3.70
Low risk	5.2	0.27	reference		3.2	0.21	reference
SF-12 functional impairment
Very high risk	15.7	1.78	5.6	1.99, 9.21	36.3	2.91	19.3	13.42, 25.18	−13.6	−20.26, −6.94
High risk	13.1	1.58	3.0	−0.19, 6.19	19.8	2.83	2.8	−2.88, 8.48	0.2	−6.39, 6.79
Moderate risk	10.5	1.38	0.4	−2.44, 3.24	17.5	1.50	0.5	−2.54, 3.54	−0.1	−4.37, 4.17
Low risk	10.1	0.43	reference		17.0	0.48	reference
Any depressive/anxiety disorderf
Very high risk	18.4	2.07	2.4	−1.28, 6.08	43.3	3.15	13.0	6.57, 19.43	−10.6	−17.81, −3.39
High risk	18.9	1.69	3.1	−0.41, 6.61	30.5	3.12	0.2	−5.92, 6.32	2.9	−3.80, 9.60
Moderate risk	17.5	1.71	1.7	1.79, 5.19	28.1	1.80	−2.2	−5.87, 1.47	3.9	−1.20, 9.00
Low risk	15.8	0.50	reference		30.3	0.66	reference
Any depressive/anxiety disorderf, no substance or illness induced
Very high risk	17.6	1.77	2.7	−0.83, 6.23	41.7	3.14	12.6	6.41, 18.79	−9.9	−17.09, −2.71
High risk	18.0	1.66	3.2	−0.19, 6.59	29.1	3.09	0.1	−6.05, 6.25	3.1	−3.64, 9.84
Moderate risk	16.3	1.66	1.5	1.85, 4.85	27.0	1.77	−2.1	−5.71, 1.51	3.5	−1.42, 8.42
Low risk	14.9	0.46	reference		29.1	0.67	reference

^aadjusted for sample weights and sociodemographic covariates (gender, age, education, race/ethnicity, health insurance, and current smoking)

^bPrevalence in each risk level minus prevalence in the reference risk level (low) indicates the association. Prevalence differences whose 95% CI do not include 0 are statistically significant at p<.05 and are bolded.

^cPrevalence difference in men minus prevalence difference in women indicates the differential association by gender. Differences whose 95% CI do not include 0 are statistically significant at p<.05 and are bolded.

^dabuse or dependence

^efor marijuana, cocaine, heroin, painkillers (prescription opioids), sedative/tranquilizers, hallucinogens, stimulants, inhalants, and club drugs

^fincludes anxiety (panic disorder, agoraphobia, social phobia/social anxiety disorder, specific phobia, generalized anxiety disorder) and depression (dysthymia/persistent depressive disorder, major depression), DSM-IV for 2001–2002, DSM-5 for 2012–2013

Liver disease excluded from this table due to low prevalence in women

Note that the prevalence, prevalence differences, and trends differences are rounded, such that subtracting the values may not yield the exact difference reported.

DSM-IV drug use disorders

In 2001–2002, the prevalence of any drug dependence was significantly greater in very high (+5.0%) and moderate risk (+0.6%) drinkers than in low risk drinkers (Table 4). In 2012–2013, the prevalence was significantly greater in very high (+6.1%), high (+2.6%), and moderate (+1.2%) than low risk drinkers. Similar results were observed for any drug use disorder, except that prevalence was also significantly greater in high risk (+3.7%) than low risk in 2001–2002. Similar associations were observed in men and women (Table 5).

Functional impairment

In 2001–2002, the prevalence of functional impairment was significantly greater in very high (+13.9%), high (+3.1%), and moderate (+2.6%) than low risk drinkers (Table 4). In 2012–2013, the prevalence was significantly greater in very high (+11.9%) and high (+3.1%) than low risk drinkers. In men and women, significant associations were observed with very high risk, with a significantly stronger association in women than men (Table 5).

Liver disease

In both 2001–2002 and 2012–2013, the prevalence of liver disease was significantly greater in very high risk drinkers (+2.9%; +2.2%; respectively) than in low risk drinkers (Table 4). Differences by gender were not estimated due to low prevalence of liver disease among women in NESARC.

Any depressive/anxiety disorder

In both 2001–2002 and 2012–2013, the prevalence of any depressive/anxiety disorder was significantly greater in very high risk drinkers (+9.2%; +7.1%; respectively) than low risk drinkers (Table 4). Significant associations with very high risk were observed for women at both surveys but for men at 2001–2002 only (Table 5). Similar results were observed for any depressive/anxiety disorder, excluding those occurring only during periods of substance use or illness (Tables 4, 5).

DISCUSSION

In previous studies, the WHO risk drinking levels (very high, high, moderate, and low) were associated with physical, mental, and social functioning and reduction in the WHO risk drinking categories predicted improvement in these conditions^17–23. Thus, the WHO risk levels showed potential clinical utility as treatment outcome measures^{14–17,19,21}. However, important epidemiological information was lacking, i.e., whether prevalence of the WHO risk levels changed over time, association of these levels with clinical correlates of heavy drinking in newer data, and whether results differed by gender. In adult general population current drinkers, the prevalence of moderate, high, and very high risk levels was significantly greater in 2012–2013 than in 2001–2002, with a greater increase in prevalence of the very high risk drinking level in women than in men. Health conditions (AUD; drug use disorders; functional impairment; liver disease; anxiety/depressive disorders) were associated with risk levels within each survey, in men and women.

Increases over time in moderate, high, and very high risk drinking are similar to results from other US national studies, which show increases in alcohol consumption, specifically binge drinking (any or weekly), particularly among women^25,26,28,29. Increases in heavy drinking in women are concerning, as women are less likely to receive treatment⁷³, yet may be more likely to develop health consequences than men at comparable consumption levels³⁰. Additional studies should identify the drivers of these patterns⁴¹. Inconsistent with previous studies, the current study showed increase in moderate drinking among men. This study differs from the others in two key ways: an important and widely-recognized consumption measure and analysis conducted among current drinkers. One previous study of current drinkers showed no increase in binge drinking in men or women²⁸, suggesting that binge drinking and the WHO risk drinking levels measure alcohol consumption differently. As a metric of alcohol consumption, the WHO levels are particularly useful, as they categorize drinkers based on intensity and frequency of drinking and identify which drinkers are at greatest risk for alcohol-related consequences²⁴.

In both surveys, higher WHO risk drinking levels were associated with clinically important health conditions (alcohol dependence/AUD, drug dependence/drug use disorders, functional impairment, liver disease, and depressive/anxiety disorders), similar to previous studies^18,20,22,23, suggesting that they are a valid characterization of alcohol consumption. Alcohol dependence, AUD, drug dependence, and drug use disorders were associated with all three risk levels (moderate, high, very high versus low). Functional impairment was associated with very high and high risk, the categories of greatest clinical concern¹⁸. Liver disease and depressive/anxiety disorders were associated with very high risk drinking. Generally, prevalence of these health conditions was greater in the very high/high risk levels, indicating that increased drinking shows increased risk, and suggesting that reducing drinking to moderate or low risk levels could reduce such conditions.

Associations were generally similar for women and men with some differences, mainly in 2012–2013. Women showed stronger relationships of very high risk drinking to functional impairment and depressive/anxiety disorders than men, similar to previous studies in AUD samples^74,75,76. Men showed a stronger relationship of moderate risk drinking to AUD than women. These differences may reflect that generally, men show higher prevalence of AUD and women show higher prevalence of depression/anxiety, emphasizing the need for further studies among women examining the relationship between drinking and functional impairment, depression, and anxiety.

While causality cannot be determined in these cross-sectional datasets, modeling alcohol consumption as preceding the outcomes (health conditions) is supported by the following. By definition, drinking precedes AUD. Drinking impacts liver functioning, causes liver disease, and exacerbates liver disease due to other causes³. Heavy drinking/AUD lead to functional impairment due to mental health issues⁷⁷. Drug use disorders and depressive/anxiety disorders are highly comorbid with alcohol use/AUD^5,8,42,67,78, with some (but not all) studies showing alcohol use/AUD preceding the comorbid disorders⁷⁹. Longitudinal studies showed that reduction in drinking was associated with reduced likelihood of these outcomes^18,20,22,23, justifying the inference about directionality modeled here. Further studies are warranted to better understand the complex and possibly reciprocal relationships between drinking and these conditions.

Study limitations are noted. While the direction of effect modeled was well supported, cross-sectional data cannot determine causality. Data were based on self-report, leading to the possibility that response bias could contribute to the findings. A higher response rate for NESARC-III would be preferred, since survey respondents may be healthier than non-respondents⁸⁰, and thus the prevalence of risky drinking and health conditions may be underestimated. Diagnoses were not made by clinicians, because clinician-administered interviews are not feasible in large-scale epidemiologic surveys. Future studies of health conditions could incorporate direct examinations or medical record variables. Participants were not asked whether alcohol was the cause of their liver disease, but even in those with liver disease from other causes, alcohol use leads to further damage and a worse prognosis³. Liver disease had low prevalence, especially in women. For depressive/anxiety disorders, the diagnostic systems could not be perfectly aligned, because DSM-IV diagnoses were used in 2001–2002 and DSM-5 diagnoses in 2012–2013. However, the effect of the DSM-IV/−5 differences should be small for a combined depressive/anxiety disorder variable, because some diagnoses would be made in both systems⁸¹.

Several strengths are noted: nationally representative data were used, with a sample large enough to include all the WHO risk drinking levels; there was representation of participants by gender, age, race/ethnicity, and socioeconomic status; the assessment of alcohol consumption and health conditions was detailed, rigorous and consistent; and diagnoses were reliable and valid.

Conclusions

This study provides important information about the WHO risk drinking levels. The prevalence of moderate, high and very high risk drinking increased over time, pointing to the increasing public health burden of individuals with potentially problematic drinking. Association of the WHO risk drinking levels with health conditions in both surveys, among men and women, shows their relevance as valid measures of drinking, since increased drinking was associated with increased risk. Thus, from the public health perspective, this metric of alcohol consumption is useful, and can be adopted internationally, by translating amounts of alcohol into country-specific standardized drinks. This metric also has clinical utility, since non-abstinent drinking reduction, i.e., reducing consumption by one or two WHO risk drinking levels, leads to significant physical, psychological, and emotional improvement^14,17–23. If clinicians and the general public became more aware that non-abstinent drinking reduction is feasible, sustainable and beneficial to health, more individuals could be engaged in treatment, which is of great public health importance. The WHO risk drinking levels can be leveraged in prevention and intervention strategies, for the public health goal of decreasing the personal and societal toll of risky alcohol use.