Associations of Alcohol, Marijuana and Polysubstance Use with Non-Adherence to Covid-19 Public Health Guidelines in a US Sample

Abstract

Background:

We sought to understand the association between heavy alcohol and frequent drug use and non-adherence to recommended social distancing and personal hygiene guidelines for preventing the spread of Covid-19 early in the US pandemic.

Methods:

A survey was offered on the crowdsourcing platform, Amazon Mechanical Turk (MTurk) during April 2020 (the early days of strict, social distancing restrictions). The study included 1,521 adults ages 18 years and older who resided in the US and were enrolled as MTurk workers, i.e., workers who are qualified by Amazon to complete a range of human interaction tasks, including surveys through the MTurk worker platform. Main predictors included measures of heavy drinking, marijuana, and polysubstance use. The dependent measures were measures of social distancing and personal hygiene, based on guidelines recommended at the time of the survey by the US Centers for Disease Control to prevent the spread of Covid-19.

Results:

We found consistent negative associations between heavy drinking and drug use and adherence to social distancing and personal hygiene. Additionally, two control variables, age and race/ethnicity, were significant correlates of adherence to these measures.

Conclusions:

The findings here are consistent with previous research exploring links between substance use and other adverse health behaviors. Further, the negative association between heavy drinking (five or more drinks in one sitting) and adherence underscore the public health risks entailed with unrestricted reopening of public drinking establishments.

Introduction

The emergence of the Covid-19 pandemic in early 2020, and concomitant enforcement of “shelter-in-place” regulations and other public health measures, caused major disruptions to the physical health and financial well-being of the worldwide population. Research is just beginning to document the impact of these events on the psychosocial well-being of citizens in China^1,2 and the US.³ Many states in the US and many Asian and European nations implemented rigorous social distancing measures as well as clearly prescribed and proscribed public health measures designed to reduce viral transmission in the community.⁴ While the closing of many commercial establishments during the pandemic restricted opportunities to gather and thus stray from public health recommendations, adherence to many of these guidelines involved personal decision-making.

Heavy consumers of substances such as alcohol, marijuana, and other drugs may be impaired in their ability to make rational health behavior decisions.^5,6 Relatedly, associations between frequent use of alcohol and drugs and other health risk behaviors have been well-established. For example, in the field of HIV risk research, myriad studies have demonstrated that those who engage in heavy alcohol,^7,8 marijuana,⁹ injection drug¹⁰, and polysubstance use⁹ are substantially more likely to engage in sexual activity that increases their risk for infecting themselves and others. Additional research suggests that those who use substances and who have comorbid health conditions (such as HIV) are less likely to adhere to prescribed treatment regimens when they are diagnosed (e.g., with tuberculosis).¹¹

In the context of this established body of research, our paper investigates associations between substance use (including alcohol, marijuana, and polysubstance use) and adherence to guidelines promoted by the US Centers for Disease Control⁴ (CDC) to prevent the spread of Covid-19. The guidelines span at least two broad categories – behaviors associated with social distancing measures and behaviors associated with personal hygiene measures. Accordingly, we employed two corresponding dependent measures based on CDC guideline content. Hypothesizing that more frequent drug use is associated with greater risk behavior, our independent variables are measures of frequent or heavy alcohol and/or drug use. This study is based on a sample of adults surveyed during April 2020, as the virus was rapidly spreading across the continental US. We hypothesized that participants who were heavy alcohol users and frequent users of marijuana and other substances would be at greater risk for non-adherence to disease prevention guidelines.

Methods

Data Collection and Sample Participants

The University of Connecticut IRB (X20-0057) approved all study materials as an exempt protocol. Mechanical Turk (MTurk) is a website marketplace run by Amazon where those who seek to have research tasks completed can enlist participation from people who are eligible to complete those tasks; participants are compensated for their efforts. Participants, who were part of the MTurk online worker pool, agreed to participate in the study as advertised on the MTurk website and provided informed consent prior to completing baseline questionnaires. The project was advertised as an anonymous, longitudinal study of the impact of Covid-19 on daily life, providing participants with $2 in Amazon.com credit for completing the baseline survey and $3 for each follow-up survey completed. Data presented here are drawn from the initial baseline survey and were collected over a three-week period beginning April 7, 2020. Thus, data collection began approximately three weeks after widespread shelter-in-place recommendations were first issued in the US. The survey, which was formatted in Qualtrics and designed to take no more than 30 minutes to complete, contained questions about adherence to CDC guidelines for preventing Covid-19⁴ in addition to questions about recent alcohol and drug use.

Eligible participants were adults aged 18 years or older (we over-recruited young adults ages 18-25 years into the study because of our specific interest in understanding the impact of the pandemic on emerging adults), residing in the US and able to read English. MTurk workers are more diverse than typical student or online forum samples and fairly representative of the larger US population.^12-14 Evaluations of the reliability and validity of MTurk have found the data to be high-quality, replicable, and valid across comparisons with frequently used academic platforms and student and professional samples.^12,15-17 Additionally, a recent review suggests that health and medical-related findings from MTurk samples are comparable to findings from samples obtained from other sources.¹⁸ Of particular relevance for our purposes is a recent review of this platform that found MTurk to be a useful source for recruiting engaged convenience samples for addictions research, especially if researchers adhered to specific guidelines for screening and retaining participant surveys.¹⁹ Accordingly, this study followed best-practice guidelines for excluding surveys collected from this platform. To verify the inclusion of unique individual human respondent cases (as opposed to computerized bot responses) we screened the dataset for duplicate cases. We eliminated surveys determined to be subpar due to quick responses (i.e., when respondent completion time was less than ten minutes; see details in Park et al.).³

Measures

Independent Variables – Alcohol and Drug Use.

For the present analyses, we focused on three measures of substance use, including one measure of heavy drinking and two measures of drug use. The heavy drinking measure assessed whether or not respondents reported drinking five or more drinks in one sitting during the past two weeks, based on the Core Drug and Alcohol Survey.²⁰ Our drug use measures, also adapted from the Core Survey, were based on responses to questions about 30-day drug use. Frequent marijuana use evaluated whether the respondent reported marijuana use on 3-5 or more days during the past 30 days. We also created a composite measure of recent polysubstance use by summing three substance use indicators (five or more drinks in one sitting in the past two weeks, any marijuana use in the past 30 days, and any other drug use in the past 30 days) so that a respondent could receive scores ranging from 0 to 3. Those who received a score of 2 or higher (i.e., they reported use of two or more of these substances) were classified as recent polysubstance users. Note that “other drug use” was defined as reporting use of any of nine different drugs including cocaine, amphetamines, sedatives, hallucinogens, opioids, inhalants, designer drugs, steroids, and “other illegal drugs.” Supplemental analyses evaluated three additional variables: hazardous drinking; any past 30-day marijuana use; and any past 30-day “other” drug use. Based on their established importance as epidemiological correlates of substance use disorder (Grant et al., 2016) and considering their potential importance for adherence, additional demographic variables (see “Results” for details) included in the analyses as control variables were gender, age group, race/ethnicity, and region of the country where respondent resided.

Dependent Variables – Adherence to Public Health Measures:

Using a Qualtrics slider bar (where respondents move a slider across a 0-100% scale), the survey asked respondents to indicate the percentage of time that they adhered to 12 different public health measures related to the spread of Covid-19 as outlined on the CDC’s website for preventing the spread of the disease.⁴ The first six items focused on adherence to social distance. Sample social distance items included “avoid in person gatherings” and “avoid discretionary travel.” The second six items focused on adherence to personal hygiene guidelines. Sample personal hygiene items included “avoid touching your eyes, nose and mouth with unwashed hands” and “cover your mouth and nose with a tissue when you cough or sneeze (or use the inside of your elbow).” An average social distancing adherence score was calculated differently for participants from the following states that had and had not issued “shelter-in-place” guidelines on the first day of data collection on April 7, 2020: North Dakota; South Dakota; Nebraska; Iowa; and, Arkansas.²¹ For residents from states where such guidelines were issued by April 7, average adherence was calculated across nine of the 12 items, excluding items that assessed visiting of bars/restaurants/food courts; nursing homes/retirement homes/long-term care facilities; or plans for taking time off from work/school if ill. For residents of North Dakota, South Dakota, Nebraska, Iowa, and Arkansas, where those activities were still possible, average adherence was calculated across all 12 main administered items. A similar mean score approach was taken for the six personal hygiene items to create the adherence to personal hygiene measure. A supplemental table (Table S1) lists the items and inclusion criteria (by state) for each of the two measures.

Statistical Analysis

Using Stata Version 16,²² we first investigated the prevalence of each substance use item as well as the bivariate associations between the four demographic control variables (i.e., age group, gender, race, and geographic region) and these measures (using Chi Square tests of association). Subsequent analyses involved regressing the two adherence measures, social distance and personal hygiene, on the three substance use indicators. Ordinary least squares (OLS) regression models included each of the four sociodemographic control variables. R², the percentage of explained variance in a multiple regression model, was used to estimate the overall strength of the different regression models. White tests for heteroscedasticity²³ and visual plots of residuals were examined for each regression model. Consequently, we computed all regression models using Stata²² robust variance estimators, which are appropriate under conditions of heteroscedasticity.²⁴ We tested separate models because of the strong multicollinearity between indicators of substance use. Separate models also allowed us to uniquely compare and contrast the relative importance of the three different substance use measures to better inform the direction of prevention and education strategies. We looked at the significance and magnitude of the regression coefficients (betas) to assess the relative importance of each substance use indicator, net of the covariates in each model.

Results

Sample Description:

We screened out respondents who provided subpar responses due to quick completion times (n=115), duplicate observations (n=77), and missing values on the age variable (n=29). After dropping those who were non-binary on gender (n=29), we were left with a total sample size of 1,521 adults. With respect to gender, 44.0% (n=669) of participants were male and 56.0 % (n=852) were female. With respect to age, 30.8% (n=469) of participants were 18 to 25 years, 39.3% (n=597) were 26-40 years, and 29.9% (n=455) were 41 years or older. Based on bivariate analyses (see below), our regression included two dummy variables contrasting those who were aged 26 to 40 years and those who were aged 41 years or older with those in the youngest group (18-25 years old). The racial/ethnic breakdown (which was derived from responses to questions about both race and Hispanic ethnicity) was as follows: White non-Hispanic respondents were 68.1% (n=1,036) of the sample; Black non-Hispanic respondents were 5.9% (n=90) of the sample; other non-Hispanic respondents (which included 90 Asian/Pacific Islanders, five Native Americans, 146 respondents who endorsed more than one race, and six respondents who endorsed another race or no race at all) were 16.2% (n=247) of the sample; Hispanic respondents were 9.7% (n=148) of the sample. For the purposes of the regression analysis, we included three dummy variables for race/ethnicity, using non-Hispanic White respondents as the reference category. Finally, with respect to region of the country, we divided the sample into four regions per the US Census.²⁵ Regression models included three dummy variables for region, leaving respondents from the South region as the reference group.

Adherence Measures:

The adherence to social distance measure ranged from 0 to 100 with a mean of 90.0, a median of 96.3, and a standard deviation of 16.1. The adherence to personal hygiene measure also ranged from 0 to 100; this variable had a mean of 84.6, a median of 90.0, and a standard deviation of 17.3. Both measures were negatively skewed.

Substance Use Prevalence:

Table 1 summarizes the overall prevalence of alcohol and drug use measures in the sample. Close to one in five respondents reported recent heavy drinking (22.3%) or frequent marijuana use (18.7%); over one in ten respondents (12.0%) reported recent polysubstance use. Our supplemental findings underscore the high prevalence of heavy drinking in the pandemic, with over 40% of the sample meeting the criteria for hazardous drinking (see Table S2 for prevalence and descriptions of demographic associations with supplemental measures). Our sample included 12.0% (n=183) polysubstance users in the past 30 days; specifically, 8.4% (n=128) of all study respondents reported using two substances in the composite measure; 3.6% (n=55) reported using all three substances in the composite measure. For the purposes of our analysis, polysubstance use was analyzed in the regression by creating a dummy contrast between those scoring 2 or higher on the measure with those scoring 0 or 1.

Table 1:

Substance use measures by sociodemographic variables (n=1521)

Subst. Use Measure	Variable	Yes (%)	X 2	P
5+ drinks/P2 WKS		22.3	--	--
	Gender		17.70	<.001
	Male	27.4
	Female	18.3
	Race/Ethnicity		20.47	<.001
	White NH	21.8
	Black NH	14.4
	Other NH	19.0
	Hispanic	35.8
	Age Group		22.06	<.001
	18-25 years	26.7
	26-40 years	24.6
	>40 years	14.7
	Region		1.26	0.738
	Northeast	23.3
	Midwest	22.3
	South	20.9
	West	23.7
>3-5 Days MJ Use P30 2		18.2	--	--
	Gender		0.18	0.672
	Male	18.7
	Female	17.8
	Race/Ethnicity		3.37	0.338
	White NH	18.2
	Black NH	16.7
	Other NH	15.8
	Hispanic	23.0
	Age Group		29.66	<.001
	18-25 years	24.5
	26-40 years	18.9
	>40 years	10.8
	Region		4.47	0.215
	Northeast	20.4
	Midwest	15.7
	South	16.9
	West	20.8
Polysubstance Use 3		12.0	--	--
	Gender		8.64	0.003
	Male	14.8
	Female	9.9
	Race/Ethnicity		19.15	<.001
	White NH	11.1
	Black NH	5.6
	Other NH	12.2
	Hispanic	22.3
	Age Group		26.67	<.001
	18-25 years	16.8
	26-40 years	12.9
	>40 years	5.9
	Region		1.84	0.606
	Northeast	12.9
	Midwest	11.3
	South	11.0
	West	13.7

¹NH=Non Hispanic

²Other NH includes Asian/Pacific Islanders, Native Americans, respondents who endorsed more than one race, and respondents who endorsed another race or no race at all

³Respondent had two or more of the following substance use indicators: 5 or more drinks in one sitting, any marijuana use, any other drug use; other drug use includes: cocaine, amphetamines, sedatives, hallucinogens, opioids, inhalants, designer drugs, steroids, and “other” illegal drugs.

Demographic Associations with Substance Use:

Age group was consistently associated with alcohol and drug use, with those in the 18-25 year old group reporting more hazardous drinking and higher rates of drinking five or more drinks in one sitting in the past two weeks. The rates reported in the youngest age group were higher than rates reported in each of the two older age groups for past 30-day marijuana use, past 30-day frequent marijuana use, past 30-day other drug use, and polysubstance use. Race/ethnicity was associated with elevated levels of both drinking measures, with both White non-Hispanic and Hispanic respondents showing elevated rates compared to other groups. Hispanic respondents also reported elevated rates of any past 30-day use of marijuana, past 30-day frequent use of marijuana, and past 30-day polysubstance use compared to other race/ethnicity groups. Gender was significantly associated with reports of five or more drinks in one sitting in the past two weeks, with men reporting higher rates. Men were also significantly more likely to report past 30-day polysubstance use.

Regression of Adherence Measures on Substance Use Indicators and Control Variables:

Evaluating R² values (which were relatively low but significant in every model), models explaining adherence to social distance (Table 2) performed slightly better than models explaining adherence to personal hygiene (Table 3). R² for social distance ranged from .047 to .082; R² for personal hygiene ranged from .030 to .041. Noteworthy in these models is the finding that age group and gender were consistently associated with measures of social distancing adherence. The two dummy variables for age group were both significant and positive in all of the social distancing regression models, suggesting that 26-40 year old respondents and those 41 years and older had significantly higher levels of social distancing adherence compared to those in the 18-25 year old group. Similarly, the two dummy variables for age group were both significant and positive in all of the personal hygiene regression models, suggesting that these two older age groups had higher levels of personal hygiene adherence compared to those in the 18-25 year old group. Across the six models, the betas for the 26-40 year old age group ranged from .072 to .094 and the betas for the 41+ age group ranged from .092 to .142, suggesting slightly larger associations for comparisons between the youngest and oldest age groups. Additionally, across all six models, men had significantly lower levels of adherence compared to women (beta values ranging from −.069 to −.104). Race/ethnicity dummy variables showed negative associations with social distancing adherence. Models 1-3 suggest that social distancing was lower among Hispanic participants compared to White non-Hispanic participants, (with betas ranging from −.072 to −.086), and lower among those in the other non-Hispanic group compared to the White non-Hispanic group (with betas ranging from −.072 to −.079).

Table 2.

Social Distance Adherence Regression Models

	Alcohol Model			Marijuana Model			Polysubstance Model
Variable	b	β	S.E.	b	β	S.E.	b	β	S.E.
Age
26-40 (vs. 18-25)	3.10b	0.094	1.03	3.09b	0.094	1.026	2.69b	0.081	1.017
41+ (vs. 18-25)	4.00c	0.114	1.030	4.24c	0.120	1.032	3.22b	0.092	1.028
Gender (Male vs. Female)	−2.28b	−0.070	0.834	−2.76b	−0.085	0.835	−2.24b	−0.069	0.829
Race/Ethnicity
(Black NHd vs. White)	−3.01	−0.044	2.031	−2.64	−0.039	2.070	−3.27	−0.048	2.056
(Othere NH vs. White)	−3.35a	−0.077	1.326	−3.23a	−0.074	1.335	−3.45b	−0.079	1.307
(Hispanic vs. White)	−4.07a	−0.075	1.789	−4.67a	−0.086	1.810	−3.94a	−0.072	1.751
Region
(Northeast vs. South)	3.45b	0.086	1.072	3.46b	0.086	1.072	3.47b	0.087	1.056
(Midwest vs. South)	0.72	0.018	1.153	0.59	0.015	1.160	0.72	0.018	1.153
(West vs. South)	0.42	0.011	1.118	0.46	0.012	1.140	0.62	0.016	1.104
Alcohol/drug Measure
5+ Drinks Past 2 Wks	−5.93c	−0.153	1.120
>3-5 Days MJf P30g				−3.30b	−0.079	1.159
Polysubstance h							−4.18c	−0.207	0.617
Model R2 (d.f.=10,1510)	.063c			.047c			.082c

^ap < .05

^bp <.01

^cp<001

^dNH: Non Hispanic

^eOther: includes Asian/Pacific Islanders, Native Americans, respondents who endorsed more than one race, and respondents who endorsed another race or no race at all

^fMJ: Marijuana

^gP30: Past 30 Days

^hPolysubstance users had two or more of the following substance use indicators: 5 or more drinks in one sitting, any marijuana use, any other drug use.

Table 3.

Personal Hygiene Adherence Regression Models

	Alcohol Model			Marijuana Model			Polysubstance Model
Variable	b	β	S.E.	b	β	S.E.	b	β	S.E.
Age
26-40 (vs. 18-25)	2.76a	0.078	1.079	2.85b	0.080	1.073	2.54a	0.072	1.079
41+ (vs. 18-25)	5.00c	0.132	1.130	5.38c	0.142	1.126	4.61c	0.122	1.132
Gender (Male vs. Female)	−3.32c	−0.095	0.913	−3.63c	−0.104	0.907	−3.32c	−0.095	0.913
Race/Ethnicity
(Black NH d vs. White)	−1.05	−0.014	2.174	−0.78	−0.011	2.178	−1.17	−0.016	2.188
(Othere NH vs. White)	–2.12	–0.045	1.329	–1.95	–0.042	1.339	–2.16	–0.046	1.325
(Hispanic vs. White)	–1.10	–0.019	1.805	–1.48	–0.025	1.819	–1.06	–0.018	1.799
Region
(Northeast vs. South)	2.09	0.049	1.175	2.06	0.048	1.179	2.10	0.049	1.168
(Midwest vs. South)	–0.14	–0.003	1.246	–0.19	–0.004	1.245	–0.15	–0.003	1.247
(West vs. South)	0.30	0.007	1.199	0.25	0.006	1.211	0.40	0.010	1.198
Alcohol/drug Measure
5+ Drinks Past 2 Wks	−3.67b	−0.088	1.122
>3-5 Days MJf P30g				−0.43	−0.010	1.124
Polysubstance h							−2.33c	−0.107	0.621
Model R2 (d.f.=10,1510)	.038c			.030c			.041c

^ap < .05

^bp <.01

^cp<001

^dNH: Non Hispanic

^eOther: includes Asian/Pacific Islanders, Native Americans, respondents who endorsed more than one race, and respondents who endorsed another race or no race at all

^fMJ: Marijuana

^gP30: Past 30 Days

^hPolysubstance users had two or more of the following substance use indicators: 5 or more drinks in one sitting, any marijuana use, any other drug use.

Alcohol and drug use measures were consistently significant negative predictors of adherence to social distance (Models 1-3). The significant betas for the adherence to social distancing models ranged from a low of −.079 (for frequent marijuana use) to a high of −.207 (for polysubstance use). Similar patterns were obtained for the three measures shown in the supplemental regression models (hazardous drinking, any marijuana use, and any other drug use; see Table S2). For the adherence to personal hygiene models, two of the three alcohol and drug use measures were significant negative predictors, with the marijuana variable failing to reach statistical significance. The significant betas for the adherence to personal hygiene models were −.088 for heavy drinking and −.107 for polysubstance use. Again, similar patterns were obtained for the alcohol, marijuana, and other drug use measures shown in the supplemental regression models (See Tables S3 & S4).

Discussion

In general, our results confirmed and amplified our expectations. Consistent with findings from other areas of health risk behavior, heavy drinking and drug use were consistently negatively associated with adherence to social distancing and personal hygiene measures intended to slow the spread of Covid-19 during the early weeks of the pandemic in the US. It is important to underscore that these associations emerged for both alcohol and drug use indicators and that these associations were net of other important sociodemographic correlates. Heavy drinking and polysubstance use were negatively associated with both outcomes; the latter measure was particularly problematic with respect to social distancing adherence. Significant marijuana associations were limited to social distancing. The associations demonstrated between our control variables and adherence outcomes suggest that age group, gender, and race/ethnicity are all critical components of the pandemic response that need to be considered in research assessing the implementation and impact of public health responses.

The findings regarding the association between heavy drinking (five or more drinks in one sitting) and non-adherence to both public health measures are particularly problematic in light of the importance that many in the US and elsewhere have placed on the reopening of public drinking establishments (bars) while the virus is still spreading. Journalists²⁶ have pointed out that since bars have been the source of Covid-19 outbreaks throughout the world, they are literally “plaguing” reopening efforts. Bars are likely to facilitate binge drinking, thereby undermining both social distancing and personal hygiene guidelines and potentially promoting the spread of Covid-19. These policy-relevant findings underscore the point that opening bars in the absence of enforcement of considerable restrictions will undermine public health.

As in any study focused on substance use measures, potential caveats about the validity of self-reported substance use should be noted.^27,28 Nevertheless, relevant research suggests substance use underreporting is not particularly problematic for alcohol²⁹ or marijuana.²⁷ Indeed, our elevated rates of reported substance use in this study suggest that underreporting was unlikely to have impacted participant responses. Further limitations relate to the MTurk sample itself, which is a convenience sample and not necessarily representative of the US population. Additionally, this survey (and analyses) omitted socio-economic indicators such as income, education, and employment; these factors could have potentially mediated the relationship between the alcohol and drug use measures and the two public health outcomes.

We also caution readers that these data are cross sectional. Our analyses are correlational in nature and we cannot conclusively attribute substance use as a cause of non-adherence. Accordingly, the language in this manuscript has referred to “associations” and not causes. That said, it is hard to conceptualize how higher levels of non-adherence to Covid-19 public health measures would increase substance use. Nevertheless, we do note that recent research suggests that constraints of enforced social distancing (and consequent isolation from social supports) may increase the risk for drinking (and potentially other substance abuse) as a coping mechanism for some vulnerable people.³⁰ This suggests a potentially positive association between the experience of social distancing and frequent drug or alcohol use. While these findings are not consistent with our study, they do underscore the complexity of the social distancing construct. Social distancing may be construed as both a public health “guideline” to be adhered to by individuals as well as a social norm that potentially constrains social interactions (and supports) in a community.

Finally, it is noteworthy in this study that our participants reported an unusually elevated prevalence of heavy drinking (See Table S1) compared to other community samples in the US.³¹ These findings must be interpreted with caution since the MTurk sample is not representative of the US population. Further, since our hazardous drinking measure was based on reports of behavior in a typical year, these elevated rates are not necessarily a consequence of the stresses of the COVID-19 pandemic. Nevertheless, considering the elevated reports of five or more drinks in one sitting in the past two weeks (22.3%) – along with an extensive literature suggesting that elevated substance use can be a response to community-wide stressors^32,33 and recent literature suggesting elevated drinking may be a particularly problematic consequence of the Covid -19 pandemic³⁴ – it is important to monitor this response as the pandemic and related behavior guidelines unfold in the coming months.