Changes in US drinking and alcohol use disorders associated with social, health, and economic impacts of COVID-19

Abstract

Aims

The COVID-19 pandemic increased alcohol consumption in the USA as a result of widespread individual changes in drinking patterns. Few studies have utilized longitudinal data allowing the prediction of increased or decreased drinking from COVID-19 economic, social, and health impacts.

Methods

Data are from 1819 respondents in the 2019–20 National Alcohol Survey and a one-year follow-up in early 2021. Changes in past-year alcohol volume, drinking days, days with 5+ drinks, and Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) alcohol use disorder (AUD) severity were measured as outcomes. Measures of COVID-19 economic, health, and social impacts were assessed for the individual and household. Economic impacts were combined into Self and Household scores. Analyses utilized multinomial logistic regression models to estimate meaningful increases or decreases in outcomes, while generalized estimating equation models estimated overall effects.

Results

Increases in alcohol use and AUD severity were larger and more prevalent than decreases, and differences between sociodemographic groups in the prevalence of meaningful increases and decreases were found. Models of meaningful changes found that higher self-economic impact scores predicted increases in 5+ days and AUD severity. Generalized estimating equation models also found that the self-economic impact score predicted increased AUD severity and additionally that being an essential worker was associated with reductions in alcohol volume and 5+ days.

Conclusions

Substantial changes in drinking and AUD severity were observed, with increases in these outcomes being more prevalent and larger than decreases. Results highlight the importance of the pandemic’s economic impacts in predicting changes in drinking and AUD severity.

Introduction

The COVID-19 pandemic created pervasive social, health, and economic impacts that spilled over to affect people’s mental health and substance use. Alcohol use has been of particular interest given its wide availability and, in some places in the USA, increased access to alcohol during the pandemic via expanded delivery and “to-go” drink options (Trangenstein et al. 2023b). Studies with varying designs have addressed changes in alcohol use during the pandemic. A recent meta-analysis by Acuff et al. (2022) found that in US studies, 29% of people who used alcohol increased their alcohol consumption in the early pandemic period, while 16% decreased. This contrasts with global findings of increases in consumption by 23% of people who used alcohol that was offset by a commensurate 23% who decreased consumption (Acuff et al. 2022). US studies have also identified important racial and ethnic differences in changes in alcohol use, with greater increases in alcohol consumption observed for Black and other people of color who used alcohol than among their White counterparts (Acuff et al. 2022).

The 2019–20 National Alcohol Survey (NAS) and the 2021 NAS COVID-19 follow-up study assessed changes in alcohol use patterns and alcohol use disorder (AUD) symptoms in individuals surveyed both prior to and during the COVID-19 pandemic. This longitudinal study found increased alcohol volume (especially greater spirits volume), increased frequency of use, and increased prevalence of moderate or severe DSM-5 AUD (Kerr et al. 2022). Particularly large increases in daily drinking and AUD were seen for Black respondents and women, and for AUD among those aged 35–49, while reduced current and heavy drinking was seen for men and White respondents. These survey data findings are consistent with changes in apparent per capita alcohol consumption (PCC) and other US population surveys (Barbosa et al. 2021; Castaldelli-Maia et al. 2021; National Institute on Alcohol Abuse and Alcoholism 2021; Patrick et al. 2022; Slater and Alpert 2023). A historically large annual increase of 3% in total PCC occurred in 2020, and PCC continued to increase in 2021 by an additional 2.8% (Slater and Alpert 2023). Spirits PCC was the main beverage type accounting for these increases, rising by 7% in 2020 and 6.3% in 2021. These increases in alcohol use were reflected in increased alcohol-related mortality rates, which rose substantially in 2020 (White et al. 2022). There were also substantial increases in liver disease mortality rates in both 2020 and 2021, even after accounting for the existing upward trend in mortality rates prior to the pandemic (Gao et al. 2023).

There are a variety of possible explanations for these increases in alcohol consumption and alcohol-related mortality during the COVID pandemic. First, while there were temporary bar and restaurant closings in most states for varying periods of time, liquor stores remained open nearly everywhere. Further, increased delivery and to-go options from bars and restaurants may have facilitated increased drinking, with studies showing that those who ordered alcohol via delivery drank more alcohol (Trangenstein et al. 2023a) and that those reporting purchasing delivery or to-go drinks endorsed more AUD symptoms than those who did not (Trangenstein et al. 2023b). Mental health impacts of COVID-19 and social restrictions during the initial phases of the pandemic may have also contributed to studies finding increased rates of depression and anxiety and drinking to forget worries and problems among women in the early pandemic months (Martinez et al. 2022). Studies also found that anxiety and depression were associated with increased alcohol consumption during the pandemic (Capasso et al. 2021; Lechner et al. 2021; Weerakoon et al. 2021).

Additional social, health, and economic impacts of COVID-19 also could have contributed to increased alcohol use and associated problems. Both pandemic-specific stress and general stress were associated with changes in drinking and alcohol problems (Grossman et al. 2020). Increases in risky drinking were also associated with racial tension, financial distress, psychological distress, and illness-related stress (Lechner et al. 2021). The Acuff et al. meta-analysis found that income loss, financial stress, job loss, working remotely, and being an essential worker were all related to increased drinking, but the level of social support was not associated with drinking changes (Acuff et al. 2022). Few of these studies utilized a longitudinal design, however, with most relying on respondents’ subjective reports of their drinking changes from prepandemic patterns. As such, there is need for additional longitudinal data describing the details of drinking pattern changes during the pandemic and evaluating the role of potential predictors, including economic, social, and illness impacts from the COVID-19 pandemic.

To address these gaps in the literature, the current study identifies predictors of meaningful changes in past-year alcohol use and AUD measures using longitudinal data comparing the pandemic period (beginning 1 April 2020) with measures from the same individuals before the pandemic. Hypothesized predictors of increased drinking and problems include economic, social, and health impacts related to COVID-19 for both individuals and their households. We utilized two types of models to provide different perspectives on these impacts. First, multinomial logistic regression models estimate relationships of economic, social, and health impacts with meaningful changes in the outcomes (including both increases and decreases). Second, generalized estimating equation (GEE) models estimate the overall effects of economic, social, and health impacts on outcomes. These complimentary perspectives provide relevant details for understanding how economic, social, and health impacts from COVID-19 changed alcohol use and AUD during the first year of the pandemic.

Materials and Methods

Data

This study utilizes data obtained from 2019–20 National Alcohol Survey (N14) and its COVID-19 follow-up survey (N14C). Data collection for N14 occurred between February 2019 and April 2020 with a representative sample of noninstitutionalized adults aged 18 years and older, covering all 50 US states and the District of Columbia (Kerr et al. 2018). The N14 survey employed two probability samples, a random digit dialed (RDD) landline and mobile sample and an address-based sample (ABS), as well as a nonprobability sample from a web panel. Data were collected through telephone interviews or web questionnaires administered in either English or Spanish. The overall N14 sample included 8493 complete interviews (1326 RDD; 5184 ABS; 1983 web panel) with a combined RDD and ABS cooperation rate of 42.2%, as measured by the American Association for Public Opinion Research COOP4 cooperation rate (The American Association for Public Opinion Research 2011). The web panel was not eligible for N14C.

Of the 6510 respondents who completed interviews through ABS or RDD, 3146 respondents (48%) were willing to participate in future studies. Data collection for N14C took place exclusively through a web-based survey between 30 January 2021 and 28 March 2021. During this period, 1819 interviews were completed, achieving a response rate of 58%. N14 respondents who lived in the South were in the CATI sample, were infrequent spirits drinkers, and had worse general health were more likely to agree to be contacted. Among those willing to participate in future studies, participants who were successfully followed were statistically more likely to be female, have higher levels of education, report being married or cohabitating, report past-year spirits drinking, were in the ABS sample, and report a higher quality of life. For a more detailed description of N14 and N14C, please refer to Kerr et al. (2022). ICF, Inc. of Fairfax, Virginia, conducted the fieldwork for both surveys.

Measures

Change in drinking and alcohol problems between N14 and N14C

‘12-month volume consumption of alcohol’ was measured with the graduated frequency (GF) series (Greenfield 2000; Greenfield et al. 2009) for all beverages combined. Following a 12-month maximum drinks in a day item (Greenfield et al. 2006), frequencies of consuming 12+, 8–11, 5–7, 3–4, and 1–2 drinks of any alcoholic beverage during the last year were coded as 14, 9.5, 6, 3.5, and 1.5 drinks, respectively. These were multiplied by associated frequencies of use, ranging from “every day or nearly every day” to “once in last 12 months” and summed to calculate the total volume. Volumes for N14 and N14C were then compared to construct the categorical measure of ‘change in volume’. Those whose absolute difference in volume between the two time points was no more than either 12 drinks (one drink per month on average) or a 30% change in average volume were coded as “relatively constant in volume.” Those whose absolute difference was greater than both 12 drinks and 30% of their average volume, with larger volume reported in N14C (or N14), were coded as “increase (decrease) in volume.”

12-month drinking frequency for both N14 and N14C was coded into the following categories: “every day or nearly every day,” “3–4 times a week,” “1–2 times a week,” “3–4 times a month,” “about once a month,” and “less than once a month.” For change in drinking frequency, those who were in the same frequency category for both periods were defined as “relatively constant,” while those who reported a higher (lower) frequency category in N14C compared to N14 were defined as “increase (decrease) in frequency.”

Heavy (5+) drinking days were derived from the GF series by summing the drinking days for 5–7, 8–11, and 12+ drinks. Change in heavy drinking was defined as “relatively constant” when the absolute difference in 5+ days between the two periods was no more than 20% of the individual’s average 5+ days, while those with changes of more (less) than 20% of the average 5+ days were classified as “increase (decrease) in heavy drinking” if the individual reported more (fewer) 5+ days in N14C.

‘AUD criteria’ for both N14 and N14C were based on the DSM-5 definition (American Psychiatric Association 2013). The AUD criteria counts include 11 criteria (hazardous use, failure to fulfill major obligations, interpersonal problems, tolerance, withdrawal, drinking more than intended, unsuccessful efforts to control use, giving up pleasures or interests to drink, spending a great deal of time in drinking, continued use despite problems, and craving). Individuals are categorized into current AUD status of “none” (0–1 criteria endorsed), “mild” (2–3), “moderate” (4–5), and “severe” (6+). ‘Change in AUD severity’ was then defined as “relatively constant” if an individual stayed in the same AUD status category between two time points and “increase (decrease) in AUD” if the individual moved up (down) to a more (less) severe category in N14C.

COVID-19 impact measures

A series of COVID-19 impact measures in N14C asked respondents how the pandemic had changed various parts of their lives since 1 April 2020. Questions asked whether the respondent or someone in the household had (i) been laid off or unemployed, (ii) pay or hours reduced at work, (iii) applied for unemployment insurance, (iv) difficulty paying rent or mortgage, or (v) lost housing. These economic impacts are analyzed separately and as combined measures of a respondent’s Self and Household economic impacts because of strong intercorrelations between indicators. The mean of the five economic indicators was calculated for the total COVID-19 economic impact score (range 0–1) for the Household (including both the respondent and others in the household), and four economic indicators were averaged for the respondent only (excluding #4, asked of the whole household). Additional COVID-19 impact measures asked whether respondents had experienced moderate or severe changes in their access to extended family and trusted friends (social impacts), whether they had been occasionally or frequently without enough food or without good quality food (food insecurity), whether the respondent or immediate family member had been diagnosed with COVID-19, and whether the respondent or someone in the household was an essential worker.

Demographics

‘Baseline demographic factors’ include gender, age (18–34, 35–49, 50+), race and ethnicity (White non-Hispanic, Black non-Hispanic, Hispanic, all other groups), education (high school graduate or less, some college, college graduate), family income (≤$20 000, $20 001–40 000, $40 001–60 000, $60 001–80 000, and >$80 000), marital status (married, never married, and separated, divorced, or widowed) and employment status (employed versus others).

Data analysis

First, we examined changes in drinking and alcohol problems from the pre- to post-COVID-19 period predicted by baseline demographic factors and N14C COVID-19 impact measures using multinomial logistic regression. Using the “relatively constant” category as the referent, these models generate separate odds ratios for the demographic and COVID-19 impact predictors on increases and decreases in drinking or alcohol problems after COVID-19. A second approach examined the effects of COVID-19 impacts using GEE models to predict drinking outcomes over time (0 = baseline, 1 = follow-up), accounting for the COVID-19 impact predictors independently and in interactions with the time indicator. The GEE models estimate the effects of COVID-19 impacts on changes in drinking outcomes using continuous measures (total volume in number of drinks, total number of 5+ days, and number of AUD criteria endorsed). The GEE models account for within-person correlation and use a negative binomial distribution with a log link function. One key difference between the two analyses is that the GEE models estimate absolute changes in levels of drinking and alcohol problems, while the multinomial logistic regression models predict both substantive increases and substantive decreases in outcomes (compared to staying relatively stable), thus providing additional information regarding COVID-19 impacts. Consistent results from both methods, examining relative and absolute changes, further enhance the robustness of the findings. All analyses adjust for sampling weights, which account for N14 weights and nonresponse during follow-up by adjusting for the above-described characteristics associated with both willingness to participate and completion of follow-up [see Kerr et al. 2022 for more details].

Results

Out of the total sample in N14C (1819 individuals), 344 respondents were nondrinkers in both N14 and N14C. They were excluded, resulting in a sample of 1475 respondents who reported drinking alcohol in at least one of the two surveys. Supplemental Table S1 shows the socio-demographic and COVID-19 impact characteristics for the total sample and the longitudinal analytic sample. Table 1 shows descriptive statistics for drinking changes on the four outcome measures (volume, frequency, 5+ days, and AUD) for respondents categorized as “increased,” “relatively constant,” and “decreased.” For the volume of alcohol consumed, 28.6% decreased volume, 30.5% increased volume, and the remaining 40.9% stayed relatively stable. For those who decreased volume, the average past-year (PY) consumption dropped from 419 drinks in N14 to 123 drinks in N14C, an average individual decrease of 296 drinks. In contrast, for those who increased volume, average PY consumption increased from 138 drinks to 521 drinks, an average individual increase of 383 drinks. Average PY consumption for those who stayed relatively stable was ~161–162 drinks in both periods. For changes in AUD, a larger proportion of the drinker sample saw their drinking problem getting more severe in N14C than getting less severe (9.0% vs. 5.8%).

Table 1.

Drinking change groups between N14 baseline and follow-up with respect to volume, drinking frequency, 5+ days, and DSM5 severity^a

Volume change	Decreased	Relatively constant	Increased
N	423	585	467
Weighted %	28.6%	40.9%	30.5%
Baseline volume 12 m (drinks)
Mean	418.5	160.8	138.2
(SE)	(42.2)	(26.8)	(12.1)
Follow-up volume 12 m (drinks)
Mean	122.9	162.0	521.0
(SE)	(15.3)	(26.0)	(39.2)
Drinking frequency change	Decreased	Relatively constant	Increased
N	386	596	491
Weighted %	27.4%	40.1%	32.5%
Baseline frequency 12 m (days)
Mean	67.3	91.0	33.2
(SE)	(5.8)	(6.7)	(2.9)
Follow-up frequency 12 m (days)
Mean	18.8	91.4	113.0
(SE)	(2.1)	(6.7)	(7.8)
5+ drinking change	Decreased	Relatively constant	Increased
N	249	1037	189
Weighted %	16.7%	69.3%	14.0%
Baseline 5+ 12 m (days)
Mean	41.7	4.6	8.6
(SE)	(6.2)	(1.7)	(1.9)
Follow-up 5+ 12 m (days)
Mean	6.0	4.7	70.1
(SE)	(1.4)	(1.7)	(8.8)
DSM5 severity change	Decreased	Relatively constant	Increased
N	87	1257	131
Weighted %	5.8%	85.2%	9.0%
Baseline symptom counts
Mean	3.37	0.23	0.58
(SE)	(0.25)	(0.03)	(0.10)
Follow-up symptom counts
Mean	1.17	0.22	3.75
(SE)	(0.21)	(0.03)	(0.25)

1819 total follow-up sample.

^aExcludes those who were nondrinkers at both baseline and follow-up (N = 344, 22.3% of the 1819 total follow-up sample)

Table 2 shows associations of sociodemographic factors with changes in alcohol outcomes over time. Significant bivariate relationships for volume were observed for age and marital and employment status, but when all demographic factors were entered together (Table 3), only age effects were statistically significant (P < .05) with more changes occurring at younger ages. Changes in drinking frequency, heavy drinking days, and AUD severity also showed significant effects of age with the 50+ group less likely to change in either direction than the 18–34 group, suggesting greater instability in drinking for younger respondents. For drinking frequency, higher incomes were associated with fewer decreases and increases, indicating greater instability for those with lower incomes. Supplemental Table S2 shows results for age in five categories breaking down the 18–34 and 50+ groups and for region of residence (not statistically significant for any measure).

Table 2.

Percentages (95% confidence interval [CI]) who increased or decreased their drinking or DSM-5 AUD severity between N14 baseline and follow-up by baseline demographics^a

	Volume		Drink days		5+ days		DSM-5 AUD severity
	Decreased	Increased	Decreased	Increased	Decreased	Increased	Decreased	Increased
Gender
Women	26.2 (22.4, 30.0)	30.1 (26.1, 34.1)	25.9 (22.1, 29.7)	35.8 (31.5, 40.0)	14.3 (11.3, 17.3)	12.3 (9.2, 15.4)	4.5 (3.0, 6.1)	8.2 (5.5, 10.9)
Men	31.3 (26.1, 36.4)	30.8 (25.7, 35.9)	29.0 (23.9, 34.2)	28.9 (23.8, 34.0)	19.3 (15.1, 23.6)	15.8 (11.7, 19.9)	7.1 (4.1, 10.1)	9.9 (6.5, 13.3)
Chi2 P-value		.189		.130		.035		.197
Age
18–34	33.6 (28.1, 39.1)	35.1 (29.5, 40.7)	34.9 (29.2, 40.6)	33.5 (28.0, 39.1)	24.6 (19.6, 29.6)	16.7 (12.3, 21.0)	11.1 (7.1, 15.1)	13.4 (9.4, 17.4)
35–49	30.5 (24.0, 37.0)	36.1 (29.6, 42.6)	24.7 (18.5, 30.9)	34.6 (28.3, 41.0)	15.3 (10.3, 20.3)	20.5 (14.7, 26.3)	4.1 (1.5, 6.7)	8.3 (4.6, 11.9)
50+	23.9 (19.2, 28.7)	23.9 (19.1, 28.6)	23.6 (18.9, 28.4)	30.6 (25.2, 35.9)	11.9 (8.4, 15.4)	8.3 (4.7, 11.8)	3.0 (1.2, 4.7)	6.3 (2.9, 9.7)
Chi2 P-value		<.001		.005		<.001		<.001
Race/ethnicity
White	30.3 (26.3, 34.4)	28.2 (24.3, 32.1)	28.8 (24.8, 32.9)	29.9 (25.9, 34.0)	16.4 (13.3, 19.5)	12.5 (9.4, 15.6)	5.2 (3.4, 7.1)	9.3 (6.4, 12.2)
Black	20.9 (13.7, 28.2)	43.6 (33.8, 53.4)	24.2 (15.8, 32.6)	38.0 (28.6, 47.3)	17.8 (9.3, 26.3)	14.5 (8.3, 20.6)	7.2 (1.4, 13.0)	13.3 (6.9, 19.7)
Hispanic	27.3 (18.5, 36.1)	30.6 (22.3, 39.0)	23.4 (15.7, 31.2)	36.4 (27.4, 45.4)	16.1 (9.9, 22.3)	19.4 (11.7, 27.0)	7.2 (2.2, 12.3)	4.5 (1.2, 7.8)
Other groups	29.3 (19.4, 39.2)	27.5 (17.7, 37.3)	28.3 (18.0, 38.5)	37.2 (25.9, 48.4)	18.0 (9.5, 26.5)	14.5 (5.5, 23.5)	5.0 (0.9, 9.1)	8.3 (2.3, 14.3)
Chi2 P-value		.102		.513		.649		.320
Education
≤HS grad	29.3 (21.9, 36.7)	28.6 (21.4, 35.7)	26.4 (19.3, 33.5)	37.2 (29.5, 45.0)	15.9 (10.6, 21.2)	16.6 (10.5, 22.8)	5.6 (2.3, 8.8)	10.4 (4.9, 15.9)
Some college	27.0 (22.2, 31.8)	32.2 (27.0, 37.5)	28.6 (23.5, 33.8)	34.4 (29.1, 39.8)	16.8 (12.4, 21.1)	13.9 (9.9, 17.9)	5.4 (2.6, 8.3)	8.3 (5.2, 11.3)
College grad	29.7 (25.4, 34.0)	29.8 (25.7, 33.9)	26.9 (22.6, 31.2)	25.6 (21.8, 29.5)	17.0 (13.5, 20.4)	11.3 (8.4, 14.2)	6.4 (4.0, 8.8)	8.7 (6.1, 11.4)
Chi2 P-value		.884		.041		.570		.904
Family income
≤20k	30.6 (22.4, 38.7)	33.8 (25.0, 42.7)	30.6 (22.0, 39.3)	42.6 (33.1, 52.0)	15.4 (9.5, 21.2)	18.9 (10.8, 27.0)	8.7 (3.7, 13.8)	13.7 (6.3, 21.1)
20 001–40k	32.4 (25.0, 39.8)	27.1 (20.2, 34.0)	34.5 (26.8, 42.1)	30.7 (23.5, 37.9)	18.6 (12.6, 24.7)	13.4 (8.1, 18.6)	6.5 (2.2, 10.7)	9.0 (4.4, 13.5)
40 001–60k	28.0 (21.2, 34.9)	29.5 (22.7, 36.4)	26.6 (19.8, 33.3)	30.8 (24.1, 37.5)	15.4 (10.6, 20.2)	11.3 (6.3, 16.4)	4.5 (1.7, 7.4)	8.6 (4.0, 13.2)
60 001–80k	23.7 (14.7, 32.6)	27.6 (18.5, 36.6)	23.1 (14.2, 32.1)	26.7 (17.4, 36.0)	16.0 (7.4, 24.6)	14.3 (6.6, 22.1)	2.7 (0.3, 5.0)	7.5 (2.0, 13.0)
>80k	27.8 (22.2, 33.4)	33.6 (27.9, 39.2)	21.4 (16.6, 26.2)	32.5 (26.7, 38.2)	17.8 (12.9, 22.6)	14.3 (10.1, 18.4)	6.3 (3.0, 9.6)	7.4 (4.5, 10.3)
Chi2 P-value		.609		.004		.807		.302
Marital status
Married	28.3 (24.1, 32.4)	30.6 (26.5, 34.6)	25.1 (21.1, 29.1)	32.7 (28.4, 36.9)	14.6 (11.5, 17.6)	13.4 (10.3, 16.5)	4.5 (2.8, 6.2)	7.7 (5.3, 10.1)
Never mar	35.8 (29.0, 42.7)	32.8 (25.9, 39.6)	35.6 (28.6, 42.5)	31.9 (25.2, 38.6)	25.6 (19.5, 31.8)	16.2 (10.6, 21.8)	12.1 (6.6, 17.5)	13.1 (8.2, 18.0)
Sep/Wid/Div	21.4 (14.6, 28.1)	27.6 (19.9, 35.3)	24.8 (17.4, 32.3)	32.8 (24.4, 41.2)	12.9 (6.8, 19.0)	13.3 (6.6, 20.0)	2.5 (0.4, 4.6)	8.3 (1.9, 14.7)
Chi2 P-value		.008		.085		.009		.001
Employment
Employed	31.8 (27.8, 35.7)	32.0 (28.0, 36.0)	29.1 (25.1, 33.1)	32.4 (28.3, 36.4)	19.6 (16.2, 23.0)	16.2 (13.0, 19.4)	7.5 (5.1, 10.0)	9.0 (6.7, 11.3)
Others	23.7 (18.6, 28.9)	28.1 (22.8, 33.3)	24.7 (19.5, 29.9)	32.8 (27.1, 38.4)	12.2 (8.5, 16.0)	10.6 (6.4, 14.8)	3.1 (1.5, 4.8)	8.9 (4.8, 13.1)
Chi2 P-value		.004		.383		.002		.049

^aExcludes those who were nondrinkers at both baseline and follow-up (N = 344, 22.3% of the 1819 total follow-up sample)

Table 3.

Relative risk ratios (RRRs) and 95% confidence intervals (CIs) from multinomial logistic regressions predicting decrease or increase in drinking (vs. staying relatively constant) by baseline demographics^a,b

	Volume		Drink days		5+ days		DSM-5 AUD severity
	Predicting decrease	Predicting increase	Predicting decrease	Predicting increase	Predicting decrease	Predicting increase	Predicting decrease	Predicting increase
Gender
Women	Ref		Ref		Ref		Ref
Men	1.32 (0.91, 1.91)	1.13 (0.79, 1.62)	1.07 (0.74, 1.54)	0.74 (0.52, 1.05)c	1.44 (0.97, 2.14)c	1.42 (0.90, 2.23)	1.55 (0.87, 2.75)	1.38 (0.82, 2.34)
Age
18–34	Ref		Ref		Ref		Ref
35–49	1.02 (0.62, 1.68)	0.95 (0.60, 1.50)	0.64 (0.39, 1.05)c	0.79 (0.51, 1.24)	0.63 (0.39, 1.02)c	1.05 (0.62, 1.78)	0.38 (0.17, 0.87)d	0.56 (0.29, 1.06)c
50+	0.61 (0.37, 1.00)d	0.44 (0.28, 0.70)e	0.58 (0.36, 0.92)d	0.71 (0.45, 1.11)	0.46 (0.27, 0.78)e	0.37 (0.20, 0.70)e	0.41 (0.18, 0.94)d	0.41 (0.21, 0.78)e
Race/ethnicity
White	Ref		Ref		Ref		Ref
Black	0.64 (0.36, 1.14)	1.66 (0.97, 2.82)c	0.67 (0.38, 1.18)	1.26 (0.73, 2.18)	0.85 (0.43, 1.67)	0.95 (0.48, 1.85)	1.10 (0.46, 2.62)	1.29 (0.65, 2.59)
Hispanic	0.70 (0.39, 1.28)	0.92 (0.55, 1.52)	0.63 (0.36, 1.10)	1.03 (0.61, 1.73)	0.87 (0.50, 1.52)	1.25 (0.68, 2.31)	0.97 (0.41, 2.25)	0.37 (0.16, 0.86)d
Other groups	0.79 (0.43, 1.43)	0.90 (0.50, 1.62)	0.97 (0.55, 1.72)	1.60 (0.92, 2.81)c	0.91 (0.47, 1.77)	1.14 (0.55, 2.38)	0.74 (0.28, 1.92)	0.84 (0.38, 1.89)
Education
≤HS grad	Ref		Ref		Ref		Ref
Some college	1.04 (0.63, 1.71)	1.24 (0.77, 2.02)	1.08 (0.64, 1.81)	0.95 (0.60, 1.52)	0.91 (0.53, 1.55)	0.75 (0.41, 1.36)	0.89 (0.38, 2.08)	0.77 (0.38, 1.56)
College grad	1.09 (0.65, 1.83)	1.06 (0.65, 1.73)	0.90 (0.54, 1.52)	0.55 (0.34, 0.90)d	0.76 (0.43, 1.35)	0.53 (0.29, 0.98)d	1.08 (0.42, 2.78)	0.90 (0.45, 1.82)
Family income
≤20k	Ref		Ref		Ref		Ref
20 001–40k	0.96 (0.51, 1.79)	0.67 (0.36, 1.25)	0.88 (0.47, 1.65)	0.53 (0.29, 0.98)d	1.32 (0.68, 2.55)	0.77 (0.36, 1.65)	0.67 (0.28, 1.64)	0.65 (0.28, 1.48)
40 001–60k	0.75 (0.38, 1.48)	0.73 (0.39, 1.38)	0.56 (0.28, 1.09)c	0.47 (0.25, 0.88)c	1.07 (0.55, 2.05)	0.60 (0.27, 1.34)	0.51 (0.20, 1.27)	0.61 (0.25, 1.51)
60 001–80k	0.53 (0.24, 1.15)	0.61 (0.29, 1.32)	0.41 (0.19, 0.88)d	0.36 (0.17, 0.76)e	1.27 (0.53, 3.04)	0.93 (0.36, 2.36)	0.29 (0.09, 0.92)d	0.57 (0.20, 1.61)
>80k	0.73 (0.36, 1.51)	0.98 (0.51, 1.87)	0.42 (0.21, 0.83)d	0.56 (0.29, 1.07)c	1.57 (0.76, 3.23)	1.08 (0.52, 2.24)	0.78 (0.29, 2.12)	0.61 (0.29, 1.29)
Marital status
Married	Ref		Ref		Ref		Ref
Never mar	1.33 (0.82, 2.15)	0.95 (0.59, 1.53)	1.30 (0.82, 2.07)	0.85 (0.53, 1.36)	1.85 (1.12, 3.06)d	1.14 (0.65, 2.02)	2.01 (0.95, 4.26)c	1.35 (0.77, 2.36)
Sep/Wid/Div	0.70 (0.42, 1.18)	0.88 (0.53, 1.46)	1.01 (0.59, 1.71)	0.94 (0.58, 1.53)	1.29 (0.66, 2.52)	1.46 (0.75, 2.84)	0.78 (0.30, 2.01)	1.27 (0.54, 2.97)
Employment
Employed	Ref		Ref		Ref		Ref
Others	0.67 (0.44, 1.03)c	0.89 (0.61, 1.31)	0.69 (0.45, 1.07)c	0.79 (0.54, 1.16)	0.65 (0.41, 1.04)c	0.69 (0.40, 1.19)	0.47 (0.22, 0.99)d	1.06 (0.56, 2.02)

^aExcludes those who were non-drinkers at both baseline and follow-up (N = 344, 22.3% of the 1819 total follow-up sample).

^bAll demographic variables are entered simultaneously.

^c P < .10,

^d P < .05,

^e P < .01.

Table 4 shows the multinomial logistic regression results predicting both increases and decreases (compared to staying relatively constant) in volume, 5+ days, and AUD severity, with each set of COVID-19 impact measures entered separately, controlling for demographic factors. None of the COVID-19 impact measures was significantly associated with volume changes. However, several measures significantly predicted increases in 5+ days and AUD severity. Respondent job loss was significantly associated with increases in AUD severity: compared to those experiencing no job loss themselves or within their household, people who had lost a job due to COVID were about twice as likely to report an increase in AUD severity (incidence rate ratio [IRR] = 1.98 [1.09, 3.59]) rather than no change. Both respondent and household member experiences of pay cuts were associated with increases in 5+ days. Those having difficulty paying their rent or mortgage were ~80% more likely to increase 5+ days (IRR = 1.81 [1.12, 2.93]), and AUD severity (IRR = 1.78 [1.00, 3.15]) rather than showing no change. Note that they were also more likely to decrease 5+ days (IRR = 1.23 [0.81, 1.88]) and AUD severity (IRR = 1.53 [0.82, 2.87]) rather than showing no change. While the effects were not statistically significant for these reductions in 5+ days and AUD severity, the overall patterns suggest greater instability in heavy and problem drinking among economically impacted groups. The household COVID-19 economic impact score was positively associated with an increase in 5+ days, while the personal COVID-19 economic impact score was positively associated with increases in both 5+ days and AUD severity. Those whose access to family or close friends was reduced during COVID-19 were also more likely to show increases in AUD severity. There were no significant effects of having COVID-19, reduced access to food, or essential worker status for either the respondent or a household member. We also examined the effects of the personal COVID-19 economic impact score separately by age group and found only one significant estimate due to compromised power for interaction analysis (see Supplemental Table S3).

Table 4.

Effect of COVID impact on change in drinking between N14 baseline and COVID follow-up: RRRs and 95% CIs from multinomial logistic regressions predicting decrease or increase in drinking related to staying relatively constant^a,b

	Volume		5+ days		DSM-5 AUD severity
	Predicting decrease	Predicting increase	Predicting decrease	Predicting increase	Predicting decrease	Predicting increase
Lost job
Self	1.04 (0.63, 1.69)	1.54 (0.98, 2.42)	1.23 (0.76, 2.00)	1.25 (0.73, 2.15)	1.18 (0.58, 2.40)	1.98 (1.09, 3.59)c
Other in family, not self	1.44 (0.84, 2.44)	0.94 (0.56, 1.57)	1.07 (0.59, 1.93)	1.10 (0.58, 2.11)	1.86 (0.81, 4.28)	0.86 (0.41, 1.83)
Pay cut
Self	1.45 (0.93, 2.27)	1.50 (0.97, 2.32)	1.33 (0.83, 2.13)	1.86 (1.12, 3.08)c	1.40 (0.74, 2.63)	1.55 (0.84, 2.86)
Other in family, not self	1.55 (0.90, 2.68)	1.37 (0.82, 2.28)	1.69 (0.96, 2.99)	2.56 (1.35, 4.86)d	1.31 (0.52, 3.28)	1.87 (0.84, 4.15)
Applied unemployment insurance
Self	1.25 (0.75, 2.08)	1.26 (0.76, 2.08)	1.44 (0.85, 2.43)	1.58 (0.89, 2.81)	1.41 (0.68, 2.89)	1.21 (0.62, 2.33)
Other in family, not self	1.30 (0.73, 2.29)	0.98 (0.58, 1.66)	1.56 (0.84, 2.90)	1.43 (0.76, 2.67)	1.37 (0.56, 3.34)	0.75 (0.33, 1.70)
Difficult pay rent/mortgage	0.86 (0.56, 1.32)	0.96 (0.65, 1.42)	1.23 (0.81, 1.88)	1.81 (1.12, 2.93)c	1.53 (0.82, 2.87)	1.78 (1.00, 3.15)c
Lost housing	1.35 (0.47, 3.88)	1.44 (0.61, 3.44)	1.07 (0.32, 3.54)	1.51 (0.61, 3.69)	1.33 (0.31, 5.73)	0.83 (0.32, 2.14)
Household COVID Eco impact score	1.44 (0.75, 2.78)	1.53 (0.81, 2.87)	1.86 (0.91, 3.80)	2.95 (1.36, 6.40)d	2.25 (0.84, 6.06)	2.34 (0.97, 5.64)
Self-COVID Economic impact score	1.14 (0.51, 2.55)	1.73 (0.83, 3.61)	1.66 (0.71, 3.90)	2.76 (1.21, 6.30)c	1.92 (0.65, 5.71)	2.85 (1.12, 7.25)c
Access to family/friends affected	0.86 (0.59, 1.26)	1.24 (0.88, 1.77)	0.70 (0.46, 1.05)	1.33 (0.85, 2.10)	1.09 (0.60, 1.96)	2.01 (1.16, 3.47)c
Access to food affected	0.87 (0.45, 1.68)	0.67 (0.38, 1.19)	0.58 (0.28, 1.18)	0.73 (0.35, 1.53)	1.10 (0.41, 2.98)	0.99 (0.47, 2.06)
Having Covid
Self	1.54 (0.85, 2.81)	1.40 (0.78, 2.51)	1.09 (0.59, 2.02)	1.79 (0.93, 3.44)	1.21 (0.50, 2.91)	1.63 (0.85, 3.12)
Other in family, not self	0.89 (0.55, 1.45)	0.89 (0.57, 1.38)	0.72 (0.44, 1.18)	1.13 (0.62, 2.03)	1.11 (0.51, 2.41)	1.47 (0.77, 2.81)
Essential worker
Self	0.86 (0.56, 1.33)	0.80 (0.53, 1.22)	0.96 (0.61, 1.53)	0.89 (0.54, 1.47)	0.88 (0.45, 1.71)	0.98 (0.56, 1.71)
Other in family, not self	1.19 (0.67, 2.13)	1.01 (0.57, 1.80)	1.17 (0.63, 2.18)	1.48 (0.74, 2.93)	0.52 (0.22, 1.22)	1.20 (0.52, 2.78)

^aExcludes those who were nondrinkers at both baseline and follow-up (N = 344, 22.3% of the 1819 total follow-up sample).

^bCOVID impact variables are entered separately; all models control for demographic factors.

^c P < .05,

^d P < .01.

In the second analysis, GEE models were fit to predict changes in continuous measures of volume, 5+ days, and the AUD criteria count by COVID-19 impact measures (Table 5, see Supplemental Table S4 for distributions of drinking outcomes). The COVID-19 impact measures (examined separately), the time indicator (1 = follow-up, 0 = baseline), and the COVID-19 impact by time interaction were entered in GEE models controlling for sociodemographic factors. Only the interaction estimates are shown in IRRs in Table 5, as these are estimates of differential effects of COVID-19 impact on change in drinking over time. Most of the significant predictors of increases in AUD severity observed using multinomial logistic regression models were also significant in the GEE models. Respondent job loss, having difficulty paying rent/mortgage, personal COVID-19 economic impact score, and reduced access to family/friends were all significantly associated with increases in AUD criteria counts over time. For example, the rate of change in AUD criteria count over time was 47% higher (IRR = 1.47 [1.04, 2.08]) for those having difficulty paying their rent or mortgage. There also were some significant predictors of changes in alcohol volume in the GEE models, with reduced volume being associated with a household member’s job loss and being an essential worker. Notably, the only impact measure significantly associated with an increase in 5+ drinking over time was the respondent having COVID.

Table 5.

IRRs and 95% CIs for the effect of COVID impact on change in drinking using GEE-negative binomial models, controlling for N14 baseline drinking and socio-demographics

	Volume	5+ days	DSM-5 AUD severity
Lost job
Self	1.02 (0.76, 1.38)	0.65 (0.36, 1.16)	1.56 (1.04, 2.33)a
Other in family, not self	0.67 (0.47, 0.96)a	0.72 (0.30, 1.75)	0.85 (0.58, 1.25)
Pay cut
Self	1.19 (0.86, 1.64)	1.01 (0.55, 1.87)	1.15 (0.79, 1.67)
Other in family, not self	0.93 (0.65, 1.33)	1.07 (0.46, 2.52)	1.61 (0.95, 2.74)
Applied unemployment insurance
Self	1.18 (0.85, 1.64)	1.41 (0.70, 2.84)	0.95 (0.64, 1.40)
Other in family, not self	0.81 (0.61, 1.08)	1.05 (0.47, 2.35)	0.82 (0.54, 1.25)
Difficult pay rent/mortgage	1.27 (0.96, 1.66)b	0.96 (0.55, 1.68)	1.47 (1.04, 2.08)a
Lost housing	0.88 (0.33, 2.34)	0.50 (0.13, 1.94)	1.24 (0.43, 3.62)
Household COVID Eco impact score	1.11 (0.77, 1.60)	0.89 (0.42, 1.89)	1.55 (0.94, 2.54)
Self-Covid Economic impact score	1.46 (0.94, 2.27)	0.96 (0.41, 2.24)	1.75 (1.03, 2.99)a
Access to family/friends affected	1.21 (0.94, 1.56)	1.28 (0.74, 2.21)	1.46 (1.05, 2.03)a
Access to food affected	1.18 (0.67, 2.07)	0.85 (0.29, 2.45)	1.47 (0.82, 2.63)
Having Covid
Self	1.26 (0.80, 1.98)	3.24 (1.18, 8.91)a	1.23 (0.73, 2.08)
Other in family, not self	0.94 (0.65, 1.36)	1.08 (0.58, 1.99)	0.91 (0.60, 1.37)
Essential worker
Self	0.70 (0.54, 0.91)c	0.58 (0.34, 0.98)a	0.94 (0.66, 1.35)
Other in family, not self	0.93 (0.63, 1.37)	1.33 (0.55, 3.23)	1.60 (0.90, 2.87)

^a P < .05,

^b P < 0.10,

^c P < .01.

Discussion

Findings are largely consistent with prior US studies, with more people reporting increases than decreases in alcohol consumption volume, frequency of drinking, and AUD criteria (Acuff et al. 2022). About 60% of drinkers reported meaningful changes in alcohol volume and frequency, indicating a high degree of variability in past-year drinking from 2019 to the 2020 COVID-19 pandemic period. There were more decreases than increases reported for 5+ days, reflecting the reduction in the prevalence of consuming 5+ drinks in the past year documented in our prior study (Kerr et al. 2022). This is likely due to reduced social activity and temporary bar closings. Our analysis of the magnitude of the mean increase and decrease in alcohol consumption indicators adds important information to these results on the prevalence of change. Notably, our findings for volume, frequency, and 5+ drinking change all indicate that the mean size of the increase was considerably larger than the mean decrease; as an example, for the group that increased drinking frequency, the follow-up mean days of drinking was 80 days higher than baseline, whereas, among those who decreased drinking frequency, the follow-up mean was 48 days fewer than baseline. The increase in mean alcohol volume was due to an increase in the frequency of alcohol use—including 5+ drinking days—among those who increased consumption relative to reductions in frequency of alcohol use among those who decreased consumption. These additional drinking days may have been facilitated by more remote work, fewer social activities, and less time working for some people (Weerakoon et al. 2021), together with sustained or even increased alcohol availability via delivery and to-go purchasing options.

Some differences between the GEE model and multinomial logistic regression models were observed. For example, the GEE model for 5+ days did not observe significant effects for COVID-19 economic impact measures such as taking a pay cut, difficulty paying rent, or mortgage and COVID-19 economic impact scores, as was seen in the logistic regression models. The GEE models also found reduced alcohol volume and 5+ days for those who were essential workers and reduced alcohol volume where a household member experienced job loss. The GEE model estimates ‘absolute’ changes in ‘levels’ of drinking and alcohol problems, while the multinomial logistic regression predicts both increases and decreases in change categories compared to staying relatively stable. Thus, some variables may predict both increases and decreases (as seen in Table 4 multinomial regression results), which may be partially or fully canceled out when considering absolute changes (as seen in Table 5 GEE results). We illustrate the potential differences between the two analytic methods using the impact measure of having difficulty paying rent or mortgage as an example in Supplemental Table S5.

The effects of pandemic period economic impacts on alcohol use patterns and AUD are relevant to the health impacts of pandemic-era restrictions and the importance of economic policies employed to counter those impacts. There were significant increases in 5+ drinking days and AUD symptoms associated with economic impacts such as respondents losing jobs, pay cuts to the respondent or others in the household, household difficulty paying rent or mortgage, and the combined economic impact scores. The net increases in 5+ drinking days and AUD severity may be due in part to economic impacts, in addition to the general disruption and anxiety that affected nearly everyone during the pandemic period. There was variation across US states in the impact of job loss and income reductions during the pandemic, with lower rates of depression and anxiety among those experiencing economic shocks in states with more supportive social policies (Donnelly and Farina 2021). Our findings are similar to earlier studies of the Great Recession, which found that severe economic impacts such as job or housing loss were associated with the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) alcohol dependence and other alcohol-related problems (Zemore et al. 2013; Mulia et al. 2014). The finding in the GEE models that job loss by a household member was associated with reduced alcohol volume suggest that the reduction in household income led to reduced drinking where the effects of own job loss on mental health and time at home led to increased 5+ and AUD severity.

Given the timing of data collection, only 10% of the sample reported having COVID-19 illness during the study period. These cases came before vaccines were available, and they tended to be relatively more serious compared to later COVID-19 variants (Florensa et al. 2022). No significant changes in outcomes were associated with COVID-19 illness in the logistic regression models; however, in the GEE models, respondents who had COVID-19 showed increases in both 5+ drinking days and AUD severity.

Social impacts from COVID-19 (reduced access to family and friends, specifically) predicted increases in AUD severity. The first year of the pandemic period saw the most significant disruption of social activity in recent history, with closures of workplaces, schools, retail stores, bars, restaurants, and event venues for varying periods of time depending on the area. Individuals also reduced contact with family and friends, which is important because greater social support has been shown to mitigate the effects of negative economic impacts on alcohol problems (Murphy et al. 2014; Marroquín et al. 2020).

Findings of increased alcohol consumption and problems during the COVID pandemic in the US contrast with European studies, where decreasing alcohol use was seen in most countries (Kilian et al. 2022). Globally, the prevalence of increasing and decreasing individuals was about the same, resulting in no significant changes in mean alcohol consumption on average (Acuff et al. 2022). However, the USA has less job and economic security than many European countries, which may contribute to more significant impacts of economic recessions on mental health and related outcomes such as suicide (Norström and Grönqvist 2015). People in the USA also may be more likely to be intrapersonally motivated (for example, reporting drinking to cope), rather than socially motivated (for example, drinking to enhance a social gathering), so reductions in social activities may have reduced drinking less than in countries where drinking is more socially motivated (Prestigiacomo et al. 2021; Kilian et al. 2022; Tucker et al. 2022).

Study limitations include the use of self-reports, which might lead to recall and social desirability biases for both alcohol measures (Greenfield and Kerr 2008) as well as the economic and social impact measures. Measures of household economic impacts may have been affected by respondents’ limited knowledge of specific economic impacts experienced by other household members. Also, these measures of event experiences do not capture respondents’ appraisals of the event’s impact, which may be relevant to impacts on drinking. The measure of social impacts was also limited to reduced access to friends and family, and the effects of other social restrictions (such as closures of workplaces) were not measured. Recall periods for the alcohol measures also differed slightly between the surveys, with outcomes assessed over just 10 or 11 months for N14C compared to a full 12 months in N14, which could slightly bias downward measures in N14C, although most were seen to increase at follow-up. Finally, the logistic regression models utilized groupings based on the size of changes (increases or decreases) in the alcohol use and AUD measures, so results could be sensitive to the operational definitions used to define change and stability. However, most key findings were replicated in the continuous GEE models.

Conclusion

Study results provided insights into changing drinking patterns and AUD severity during the first year of the COVID-19 pandemic and the effect of COVID-19 impacts on these changes. Notably, substantial changes in alcohol consumption and AUD severity were observed, and increases in these outcomes were more prevalent and considerably larger in magnitude than decreases. Importantly, results show the salience of the pandemic’s economic impacts in predicting increases in 5+ days and AUD severity, with impacts experienced directly by respondents being especially important in relation to AUD severity. Overall, our findings suggest that people who experience firsthand the economic fallout from a global health crisis may be at a heightened risk for heavy drinking and the development of more severe AUD. Efforts to identify people at heightened risk should be enacted at the onset of a crisis and readily available supports should be established to prevent and address consequent increases in alcohol use and AUD symptoms. Analyses also uncovered some of the complexity of changes in alcohol use and AUD, with many substantial individual changes in both directions with more changes occurring among younger age and lower income groups.

Author contributions

William C. Kerr (Conceptualization, Funding acquisition, Project administration, Writing—original draft [lead], Methodology, Writing—review & editing [equal]), Yu Ye (Conceptualization, Writing—original draft [supporting], Data curation, Formal analysis [lead], Methodology, Writing—review & editing [equal]), Priscilla Martinez (Funding acquisition, Methodology, Project administration [supporting], Writing—review & editing [equal]), Katherine J. Karriker-Jaffe (Methodology [supporting], Writing—review & editing [equal]), Deidre Patterson (Data curation, Methodology [supporting], Writing—review & editing [equal]), Thomas K. Greenfield (Methodology [supporting], Writing—review & editing [equal]), and Nina Mulia (Methodology [supporting], Writing—review & editing [equal]).

 

Conflict of interest: Drs Kerr and Greenfield and Ms. Patterson have received funding and travel support from the National Alcoholic Beverage Control Association for work not associated with this manuscript. Dr Kerr has been paid as an expert witness regarding cases on alcohol policy issues retained by the Attorney General’s Offices of the US states of Indiana and Illinois under arrangements where half of the cost was paid by organizations representing wine and spirits distributors in those states.

Funding

The work of all authors was supported by the U.S. National Institute on Alcohol Abuse and Alcoholism (NIAAA) at the National Institutes of Health (NIH) (grant numbers P50 AA005595 and R01 AA029921). Content and opinions are those of authors and do not reflect official positions of the U.S. National Institute on Alcohol Abuse and Alcoholism or the National Institutes of Health, which played no role in the manuscript development and decision to submit.

Data availability

Data underlying this article are available by reasonable request from the Alcohol Research Group. The N14C data will also be available in the NIAAA data archive.