COVID and College: How the Pandemic Impacted Alcohol Use Disorder Status Among Students

Abstract

Objective:

Alcohol consumption patterns during the COVID-19 pandemic have varied notably.

Participants:

We examined the acute impact of the pandemic on alcohol use disorder (AUD) in a generalizable sample of college students who were surveyed pre-pandemic and re-surveyed in May 2020.

Method:

Items assessed pre-pandemic included DSM-5 AUD and mental health symptoms. A COVID-19 impacts questionnaire was administered, and alcohol and mental health items re-assessed.

Results:

AUD symptoms decreased from pre-pandemic to during the pandemic, demonstrating a change in trajectory compared to prior cohorts. Students with persistent AUD reported greater concurrent symptoms of PTSD, depression, and alcohol consumption than those with remitted AUD (ps≤.02), but not increased COVID-19 impact. Persistent AUD status was predicted by higher sensation seeking and alcohol consumption.

Conclusions:

Students with concurrent mental health problems are at continued risk for persistent AUD. Findings highlight the impact of the college environment and social context for drinking on AUD.

A wealth of research examining alcohol use and problems during the COVID-19 pandemic has demonstrated varied patterns¹. While some have demonstrated increases in alcohol consumption^2,3 and relapse (AUD⁴), others have demonstrated no change^5,6, demonstrated differences based on the type of consumption (e.g., no change in quantity/frequency, but increases in binge episodes⁷ and demonstrated decreasing rates^8,9). Although some work has identified increases in alcohol use problems², most of this extant work focuses on alcohol consumption.

The differential pattern of changes in alcohol consumption identified during the pandemic may in part be explained by population (i.e., increases in binge drinking for all age groups except 18–24¹⁰; decreases in alcohol consumption in college students^11,12). Individuals that were already problematic drinkers and those with comorbid mental health conditions tend to be at more risk for increased consumption and problems^13,14. The pandemic and associated quarantine measures and closures profoundly impacted the context in which drinking occurs¹⁵, with the closure of dormitories and bars and cancellation of large gatherings and events. Thus, despite the increased stressors associated with the pandemic and related closures, the closure of campus also provided a unique, natural experiment in which to examine whether these closures, and resultant changes to social environments in which drinking likely occurs for college students, impacted alcohol use outcomes.

College students are a particularly high-risk population for problematic drinking^16,17. The college environment itself may confer risk, with evidence that problematic drinking is more frequent in college students compared to their non-college, same-age peers^16,18 and student status is associated with high-intensity drinking, and more so for students who reside away from, compared to with, their parents¹⁹. The pandemic rendered rapid, dramatic changes to campus life; many students returned home due to the closure of residence halls and most experienced the loss of social environments where alcohol consumption frequently occurs (e.g., college parties, bars²⁰) as a result of quarantine measures. Social motives tend to be one of the most commonly associated motives for drinking in college students²¹; thus, loss of these contexts would be expected to impact drinking behavior.

In this study, we examined the acute impact of the COVID-19 pandemic on AUD symptoms and status in a sample of college students from an ongoing, longitudinal study with multiple cohorts²² through two complimentary aims. This sample with data collected at four time points (three time points prior to [freshman fall, freshman spring, sophomore spring] and one time point during [junior spring] the COVID-19 pandemic) allowed us to examine the change in AUD symptoms and status in the context of the pandemic and compare this pattern to prior cohorts and trajectories in the absence of the pandemic. For aim 1, we examined within-group change in DSM-5 AUD symptoms from before to during the pandemic in the current study cohort (referred to hereafter as the COVID-19 cohort) and compared the trajectory of AUD symptoms to prior cohorts unaffected by the pandemic who were assessed at the same developmental time-point. We hypothesized an overall decrease in AUD symptoms from pre-pandemic to during the pandemic and compared to prior years and cohorts, even during the ongoing stressor of the pandemic. For aim 2, we used a subset of the COVID-19 cohort who met DSM-5 AUD criteria pre-pandemic to examine COVID-19-related stressors and concurrent mental health symptoms as correlates of AUD status change (i.e., remitted or persistent). We also utilized prior pre-pandemic data on mental health symptoms and other well-established risk factors for alcohol use in college, specifically, familial risk for AUD^23,24, peer deviance²⁵, and impulsivity²⁶ to examine predictors of AUD status change. We hypothesized that greater COVID-19-related stressor impact and mental health symptoms during the pandemic would be associated with persistent, compared to remitted, AUD. We also hypothesized that positive family history, pre-pandemic mental health symptoms, and impulsivity would increase the likelihood of persistent AUD symptoms, while peer deviance would decrease the likelihood (presumably via decreased interaction with peers).

Materials and Methods

Sample and Procedures

Participants came from a large, ongoing longitudinal study of college students at a mid-Atlantic public university. This study was approved by the university’s review board and all participants provided informed consent. Detailed information on the broader study is available elsewhere²². Briefly, data have been collected (online through Research Electronic Data Capture [REDCap]²⁷) across five cohorts of first-year college students (N=12,385). Participants were on average 18.49 years old at baseline, 61.9% female. The sample reflected the racial and ethnic composition of the university population from which it was drawn: 47.9% White, 19.3% African-American, 16.6% Asian, 6.6% Hispanic/Latino, 9.6% other/multi-race/unknown/declined to respond. Participants completed baseline surveys during fall freshman year and follow-up assessments during each subsequent spring semester. Data for the present study comes from four timepoints: freshman fall, freshman spring, sophomore spring, and junior spring. In addition to the primary COVID-19 cohort (described below), we leveraged longitudinal data from the four prior cohorts of the study, specifically those who had AUD symptom data at these four study waves, to compare the COVID-19 cohort to prior cohorts.

The active study sample represents students who were first enrolled as part of cohort 5 in Fall 2017. Those still enrolled as students in the spring of 2020 during the onset of the pandemic (their junior year) were recruited for a COVID-19-related survey administered in May of 2020 (of 2131 in cohort 5,1899 still enrolled and invited, 897 completed) after COVID-19 was declared a pandemic. The university closed for in-person instruction, dormitories were closed for the remainder of the semester, and bars/restaurants in the surrounding city were closed for on-site dining. Those who completed the COVID-19 survey, compared to those who did not, were more likely to be female and Asian, but these differences were small (Cramer’s V of .16 and .12, respectively²⁸) and had on average fewer AUD symptoms at the prior survey, again a small effect (M=1.50 compared to M=1.92; Cohen’s d = 0.19). Details on the overarching COVID-19 survey in this cohort can be found elsewhere²⁸. In the COVID survey, 572 participants had AUD data. Among those, 421 also had AUD data in the prior survey and make up the present sample. Those who had AUD data in the COVID-19 survey, compared to those who did not, were less likely to be Black/African American and Asian and more likely to be White (Cramer’s V of .30), but there were no differences in gender. The majority of the sample was cis-gender female (80.9%) and as with the broader longitudinal survey, generally reflected the student population from which it was drawn (49.9% White; 17.5% African American, 14.5% Asian, 10.1% Hispanic/Latino, 7.9% other/multiracial).

Measures

Assessed pre-COVID and during COVID.

Alcohol use disorder (AUD) status determination was made based on the Diagnostic and Statistical Manual of Mental Disorders (DSM-5; American Psychiatric Association, 2013), using items adapted from the Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA²⁹) at each wave. The 11 DSM-5 symptoms (e.g., “Have you given up or greatly reduced important activities while drinking, for example sports, work, or associating with friends or relatives?”; “In situations where you couldn’t drink, did you have such a strong desire for it that you couldn’t think of anything else?”) on a Likert scale (never, 1–2 times, 3 or more times). Any positive endorsement (i.e., 1–2 times or 3 or more times) was marked as meeting that symptom threshold. Prior surveys assessed past-year AUD symptoms. Because the goal of the COVID-19 survey was to assess AUD criteria specifically during the pandemic, the COVID-19 survey assessed for the presence of AUD symptoms since the pandemic’s onset (i.e., past three months). AUD symptoms were summed for the first study aim. For aim two, AUD status was determined using DSM-5 criteria for the mild threshold (i.e., at least two symptoms). Individuals were classified as remitted if they met DSM-5 AUD criteria (mild severity or greater) at the survey the year prior to the pandemic and no longer met AUD criteria at the COVID-19 survey. Individuals were classified as persistent if they continued to meet any AUD criteria at the COVID-19 survey.

Alcohol consumption was measured at each wave using ordinal quantity and frequency items from the Alcohol Use Disorders Identification Test (AUDIT³⁰) converted into grams of ethanol consumed in the past month, as done in prior work³¹.

Past-month depressive and anxiety symptoms were assessed at each wave using items from the Symptom Checklist-90 Revised³². Reliabilities were high (as .89–.96).

Assessed during COVID.

Participants were administered two measures related to the pandemic adapted from the Coronavirus Health Impact Survey (CRISIS³³) and the Epidemic-Pandemic Impacts Inventory (EPII³⁴). We used five correlated factors of the COVID impact as predictors of AUD status: exposure, worry, housing/food instability, social media use and substance use. Factors were derived in the S4S COVID-19 sample using factor analyses. Details of the factor analyses and the measurement models are described in Bountress et al.²⁸

Participants were also asked to complete the PTSD Checklist (PCL-5³⁵) for past-month PTSD symptoms with regard to the pandemic as the index event.

Assessed pre-COVID.

To assess family history of alcohol problems, participants answered separate yes/no questions about drinking problems for four types of relatives. In the present study, we focused on first-degree relatives (biological parent or sibling), coded as a binary item, indicating endorsement of at least one first-degree relative with drinking problems.

Impulsivity, assessed at baseline, was measured using a subset of items from the UPPS-P³⁶, using 4-point Likert scale. As is recommended³⁷, the five subscales were examined separately (negative urgency, positive urgency, lack of premeditation, lack of perseverance, and sensation seeking; as ranged from .60–.74).

A measure of college peer group deviance was assessed during the first spring follow-up survey using a composite score of six items adapted from prior work²³ that assess on a 5-point Likert scale how many of the respondent’s friends engage in specific behaviors (e.g., got drunk, in trouble with the law) since attending college (a=.87).

Data Analytic Plan

The grams of ethanol consumed in the past month variable and AUD symptom sum were log transformed due to skewness and kurtosis, as done in prior work. All other continuous variables were within acceptable ranges for skewness and kurtosis. Analyses were conducted in SPSS (v. 26). Composite scores were prorated in the case of individual item missingness if over half of items were completed. Listwise deletion was used for missing composite variables.

To address aim one, comparing AUD symptoms during the COVID-19 pandemic to pre-pandemic time-points and prior study cohorts, a mixed-effects regression model was constructed of log-transformed AUD symptom sum as the primary outcome variable. Self-reported race, gender, time, and cohort were included as covariates, and an interaction term for time by cohort was tested because we expected AUD symptoms to vary across time differently between cohorts. The mixed model includes a random effect to account for the correlation between the repeated AUD screenings within individuals. We used contrasts to compare the difference in mean AUD symptom count between cohort 5 and the average of the means of cohorts 1–4 at each of the 4 timepoints. P-values for multiple comparisons tests via contrast statements were adjusted using Holm’s method.

For aim two, to examine correlates and predictors of AUD status, chi square tests and a series of four t-tests examined differences in mental health symptoms and alcohol consumption between the remitted and persistent groups. A logistic regression analysis incorporating the five COVID-19 stressor factors was conducted. A combined logistic regression analysis was conducted, with demographic factors of gender and race/ethnicity, pre-pandemic predictors of prior mental health symptoms, family history, impulsivity, and peer group deviance.

Results

Total sample size for the mixed model was 11,881 subjects and a total of 27,940 observations; samples sizes at each cohort and time point are reported in Supplementary Table 1. Because we determined a priori to adjust for the effects of cohort, time, sex, and self-reported race, we did not engage in traditional model building to test these in a univariate fashion for their association with AUD score sum. We did test the importance of the time-by-cohort interaction term using a likelihood ratio test and determined that the interaction was significant (likelihood ratio test χ² statistic = 225.96, DF = 12, p-value, full model statistics reported in Table 1). Main effects for cohort and time are not directly interpreted in the presence of their interaction terms, but marginal effects for Asian, Black, Latino, and Multiracial races, as compared to White (used as the reference because it was the largest category) were all significant, as was the sex effect. Model coefficients represent mean expected change in the log-transformed AUD score sum per unit change, or change from reference category, in the predictor. Back transforming, we observe that Asian, Black, Latino, and Multiracial students across all cohorts reported approximately 29, 25, 9, and 5% fewer AUD symptoms than White students, and male students reports approximately 3% more symptoms than female students. Cohort 1 and Year 1, Fall served as the reference levels for the cohort-by-time interaction terms, and 7 out of the total 12 interaction terms were significant; rather than interpreting these coefficients directly, we instead interpret the cohort by time interaction in the context of specific contrasts.

Table 1.

Multivariate regression model coefficients, with standard errors of estimates, degrees of freedom, t-statistics, and marginal effect p-values.

Term	Estimate	Standard Error	Degrees of Freedom	t Statistic	Marginal P-value
(Intercept)	0.7141	0.0168	16800.2042	42.5953	<0.0001
Cohort 2	0.0272	0.0212	18546.211	1.2823	0.1998
Cohort 3	0.0406	0.0213	18477.6228	1.9084	0.0564
Cohort 4	0.0476	0.0216	18544.8828	2.2035	0.0276
Cohort 5	0.0166	0.0215	18475.0774	0.7738	0.4391
Y1S	0.2021	0.0171	18321.1129	11.8172	<0.0001
Y2S	0.2769	0.0181	18420.0639	15.283	<0.0001
Y3S	0.3067	0.0203	18510.6004	15.0922	<0.0001
Race (American Indian)	−0.0089	0.0843	11855.501	−0.106	0.9156
Race (Asian)	−0.3451	0.0171	11266.3299	−20.2226	<0.0001
Race (Black)	−0.2895	0.0162	11414.2886	−17.8498	<0.0001
Race (Latino)	−0.099	0.025	11557.3277	−3.9564	0.0001
Race (Multiracial)	−0.0508	0.0252	11728.7419	−2.0148	0.0439
Race (Pacific Islander)	−0.0075	0.0741	11542.5366	−0.1007	0.9198
Race (Unknown)	−0.1064	0.1059	12010.0217	−1.0049	0.3150
Male	0.0257	0.0127	11773.1036	2.027	0.0427
Cohort 2*Y1S	−0.0717	0.0244	18181.9887	−2.9374	0.0033
Cohort 3*Y1S	−0.0446	0.0248	18237.1978	−1.7978	0.0722
Cohort 4*Y1S	−0.0689	0.0254	18362.9894	−2.7078	0.0068
Cohort 5*Y1S	−0.0874	0.025	18356.3299	−3.4985	0.0005
Cohort 2*Y2S	−0.0302	0.0262	18373.373	−1.1491	0.2505
Cohort 3*Y2S	1.00E-04	0.0262	18372.3274	0.0021	0.9983
Cohort 4*Y2S	−0.1971	0.027	18498.056	−7.3075	<0.0001
Cohort 5*Y2S	−0.1494	0.0272	18554.6233	−5.4995	<0.0001
Cohort 2*Y3S	0.0351	0.0289	18450.135	1.2153	0.2243
Cohort 3*Y3S	0.0196	0.0288	18445.529	0.6823	0.4950
Cohort 4*Y3S	0.0346	0.0307	18583.6833	1.125	0.2606
Cohort 5*Y3S	−0.3096	0.034	18671.6396	−9.1139	<0.0001

Means compared by contrasts are reported in Table 2 and all means are plotted by cohort and timepoint in Figure 1. Mean AUD symptom scores were not significantly different between cohort 5 (the dot-and-dash line, Figure 1) and the average of the means of cohorts 1–4 (the small dash line, Figure 1) at the first timepoint, but were different at timepoints 2 through 4. As a post-hoc test, we also compared the changes from baseline between cohort 5 and the averages of the means for cohorts 1–4 and found that all changes were significantly different.

Table 2.

Contrasts comparing (a) in rows 1–4, the mean for cohort 5 and the average of the means for cohorts 1:4, at each timepoint, and (b), in rows 5–8, the changes from baseline (Y1F) between the mean for cohort 5 and the average of the means for cohorts 1:4.

Contrast	Diff	DF	χ 2	P-value
Y1Fμ5 – Y1Fμμ1:μ4	−0.093	1	0.4960	0.4813
Y1Sμ5 – Y1Sμμ1:μ4	−0.215	1	6.3232	0.0238
Y2Sμ5 – Y2Sμμ1:μ4	−0.351	1	20.3782	<0.0001
Y3Sμ5 – Y3Sμμ1:μ4	−0.721	1	132.8151	<0.0001
(Y1Sμ5 – Y1Fμ5) – (Y1Sμμ1:μ4 – Y1Fμμ1:μ4)	−0.122	1	4.1137	0.0425
(Y2Sμ5 – Y1Fμ5) – (Y2Sμμ1:μ4 – Y1Fμμ1:μ4)	−0.258	1	17.1412	0.0001
(Y3Sμ5 – Y1Fμ5) – (Y3Sμμ1:μ4 – Y1Fμμ1:μ4)	−0.628	1	129.2196	<0.0001

Note: Diff = difference in the contrasted means, in raw AUD score sum; DF = degrees of freedom; Y1F = year 1, Fall; Y1S = year 1, Spring; Y2S = year 2, Spring; Y3S = year 3, Spring; μ₅ = mean AUD score sum for cohort 5; μ_μ1:μ4 = averages of the mean AUD score sums for cohorts 1 through 4

Figure 1.

AUD Symptom Count by Cohort Across Study Waves

AUD symptom count based on DSM-5 criteria. Cohort 5 represents those administered the COVID-19 survey. In the COVID-19 cohort, the year 3 spring represents the COVID-19 survey. Note that the full y-axis scale should be from 0–11 (for the 11 AUD symptoms) but this has been truncated for ease of viewing; thus, overall effects seen here are small. Y1F=year 1 fall; Y1S=year 1 spring; Y2S=year 2 spring; Y3S=year 3 spring

For aim two, among those in the COVID-19 cohort, 41.1% (n=173) met AUD criteria (mild or more) the year prior and 33.3% (n=140) met AUD criteria at the COVID-19 survey. Approximately 46.8% of participants with prior year AUD were classified as remitted at the COVID survey (M_AUDsymptom=3.52, SD=1.86 pre-COVID and M_AUDsymptom=0.49, SD=0.51, at the COVID survey; compared to M_AUDsymptom=4.18, SD=2.11 and M_AUDsymptom=3.90, SD=2.14 pre-COVID and the COVID survey, respectively, in the persistent group [53.2%]). Results of the t-tests are presented in Table 2. On average, individuals with persistent AUD status reported greater concurrent symptoms of alcohol consumption, PTSD symptoms, and depression symptoms, but no differences in anxiety symptoms. However, we note that depression symptoms did not remain significant following multiple testing correction (Bonferroni adjusted p = .01). In the regression analysis of COVID-19 impact factors, the only factor associated with persistent AUD status was the COVID-19 stressor factor of increases in substance use during the pandemic (OR=5.71, p=.003; Supplementary Table 2). The regression model examining relevant predictors of AUD status demonstrated that likelihood of persistent AUD status was associated with only higher levels of prior alcohol consumption (OR=2.14, p=.03) and sensation seeking (OR=1.93, p=.03; Supplementary Table 1). We note that the pattern of findings was the same after accounting for COVID-19 stressor factors in the model as well.

Discussion

Leveraging a large, ongoing longitudinal project, this study aimed to investigate the acute impact of the COVID-19 pandemic on AUD symptoms in a sample of college students in comparison to prior cohorts at the same time points across the college years and to examine predictors and correlates of AUD status.

As hypothesized, findings from the first aim demonstrated that average AUD scores at Y3S were lower in cohort 5 students, as compared to prior, pre-pandemic cohorts. The primary hypothesized comparison yielded statistically significant differences in mean scores at Y3S of approximately −0.72. Although this is a small difference, given the threshold for mild AUD of two or more symptoms, this mean difference is nearing clinical utility, although clinical interpretations should be tempered given the overall low means. Post-hoc testing similarly yielded statistically significant differences between change from baseline means for all time point comparisons, but only the difference in changes from baseline from study beginning (Y1F) to end (Y3S) of approximately −0.63 symptoms is nearing clinical utility. Findings align with another COVID-19 related survey utilizing longitudinal data in a college sample, which found decreases in alcohol consumption¹². The present study extends upon that work to demonstrate a decrease in AUD symptoms in comparison to prior cohorts at the same developmental phase whose college experiences were not interrupted by the pandemic, thus demonstrating a divergence from the expected trajectory. The different trajectories of alcohol consumption and problems demonstrated during the pandemic (i.e., increasing for some, decreasing for others) seems to be in part due to population, with students and young adults demonstrating decreasing trends. Students who continued to meet AUD criteria were those with greater psychiatric distress during the pandemic, as hypothesized, and in line with other studies finding increases in consumption in students with anxiety/depression symptoms^14,38. Contrary to our hypothesis, severity of COVID-19 impact across the COVID impact domains (exposure, worry, housing/food instability, social media use and substance use) was generally not associated with AUD status. Such findings may suggest mental health symptoms in the context of chronic stressors, not necessarily the stressors themselves, are associated with persistent AUD status.

Contrary to our a priori hypothesis, we did not find associations between most of the examined predictors with AUD status change. This is unexpected, but intriguing and adds further support to the impact of the college environment on not only alcohol consumption, but AUD symptoms. For college students, living at home is protective¹² and while we were unable to specifically examine whether students returned home to parents or stayed close to the campus area (i.e., off-campus housing), as done by White and colleagues¹², the closing of residence halls, bars/restaurants, and cancellation of large group gatherings nevertheless impacted the broader social context. This highlights the important impact of college student setting and environments in which to consume alcohol on AUD symptoms. In college students, the loss of opportunities to drink with peers, typically associated with heavy drinking³⁹, may have resulted in decreased problems, even in the context of ongoing COVID-19-related stressors. Sensation seeking was also associated with likelihood of persistent AUD status. Behavioral undercontrol, such as sensation seeking, has been associated with reduced rates of remission from AUD⁴⁰ and thus remains an important factor overall, as well as in the context of the pandemic in particular. Finally, only a subset of potential predictors was examined based on what was available in the longitudinal dataset; there are likely a myriad of other factors (e.g., place of residence, availability of alcohol) that may have decreased symptoms and impacted AUD status.

While the environmental change induced by the pandemic may have been associated with an improvement in AUD symptoms and disorder status in many students, for a subset, namely those with concurrent psychiatric distress, AUD status was sustained regardless of environmental change. Existing community-based samples suggest that individuals with psychiatric comorbidities are also at greater risk for relapse⁴¹. University programs aiming to decrease problematic alcohol use may benefit from collaborative substance use and mental health services targeting efforts on such students, as they are most likely to sustain problems in the face of stressors and regardless of environmental changes (during and beyond college) and are more likely to drop out of college⁴². It is hoped that findings from longitudinal studies during various stages of the pandemic such as this study will help shed light on risk and protective factors in the face of other large-scale stressors. Findings from this study can inform university efforts to reduce the burden of alcohol use problems in the college population, particularly as students return more fully to campus and the associated expectation that alcohol consumption/problems will increase again, or even rebound above baseline.

Study findings should be evaluated in the context of limitations. In the COVID-19 survey, AUD symptoms were measured regarding the past three months to assess the acute impacts of the pandemic, while in the prior year survey, AUD symptoms were measured regarding the past year. Although the measurement timeframe between the two waves differed by design, additional research is needed to determine whether the change in AUD symptoms persists over time. Currently, it remains unknown whether AUD symptoms will rebound, or even be exacerbated, as a function of returning to campus or the impact of the ongoing, prolonged nature of the pandemic. Future research with longitudinal data collected during the pandemic can better address these questions. It is possible that students who completed the survey were faring better than those who did not, and although the overarching survey is representative of the broader university student population, our findings may be more representative for females.

Table 3.

T-tests of COVID-19 correlates for those in the remitted and persistent any AUD groups (n = 174)

	Remitted AUD (n = 82)		Persistent AUD (n = 92)
	Mean	SD	Mean	SD	t	p
Alcohol consumption grams ethanol (log10)	1.20	0.91	1.57	1.02	−2.50	.01
PTSD sum	19.33	14.92	27.60	20.64	−3.04	.003
Anxiety sum	3.39	3.95	4.42	4.67	−1.58	.12
Depression sum	6.38	4.34	8.08	4.92	−2.40	.02