COVIDVu is a prospective observational cohort study with a nationwide household probability sample of US adults using sampling methods that have been previously described [21]. A total of 39,500 US households were sampled using addresses derived from the US Postal Service Computerized Delivery Sequence File, including oversampling of households with census tracts comprising >50% Black residents and households with surnames likely to represent Hispanic ethnicity. The sampled households also include oversamples in California (16.5%) and Georgia (30.4%) to allow state-level estimation. All sampled households were shipped a study kit each. One adult resident enumerated the number of household members and each person’s age. One enumerated household member aged ≥18 years was randomly selected and offered participation in the study. Consenting participants were asked to complete an online survey, self-collect an anterior nares swab and a dried blood spot card, and return the specimens to a central laboratory via a prepaid mailer. Participants who returned the specimens were compensated at least US $40. Using procedures previously described for this study, sample and design weights were applied to estimate unbiased measures for noninstitutionalized, housed US adults in 2020 [21].

Informed consent was obtained from each participant in the study. The study was conducted in compliance with federal regulations governing protection of human subjects and was reviewed and approved by Emory University’s Institutional Review Board (protocol 00000695).

The following primary dependent measures were used in this study: mask wearing, social distancing, handwashing, and hand sanitizer use. Mask wearing was defined by the question “When you go out, do you wear a face mask?” with response options of always, often, sometimes, rarely, and never. Social distancing was defined by the question “How often are you trying to keep at least 6 feet between you and other people you don’t live with to avoid spreading illness?” with response options of always, often, sometimes, rarely, and never. Those who reported always wearing masks or maintaining social distancing were defined as consistently practicing these behaviors. Handwashing was defined by the question “In the past 24 hours, about how many times did you wash your hands with soap and water?” Hand sanitizer use was defined by the question “In the past 24 hours, about how many times did you use an alcohol-based hand sanitizer spray, gel, or wipes?” Based on a prior study examining handwashing effectiveness at preventing seasonal coronavirus [22], we categorized handwashing and hand sanitizer use as 0-5 times per day (the prior study referent group), 6-10 times per day (the prior study effective intervention group), and ≥11 times per day (the prior study intervention group that was not effective). For dichotomous analyses, those in the categories of 6-10 and ≥11 times per day were defined as frequently practicing hand hygiene.

Independent analysis measures included standard demographic characteristics of gender, race/ethnicity, age group, US Census region, ZIP code–based urbanicity, highest education, annual household income, and current health insurance. For those working, the job type was collected using 2018 US Bureau of Labor Statistics major job categories [23]. Some types of jobs, such as food service, education, health care, retail, and transportation services, may have had additional recommendations about mask wearing and hand hygiene [24]. These job types were differentiated from others in the analyses. For those who leave home for work, we assessed whether their jobs were completely indoors or outdoors/mix/other.

We developed sample weights to represent noninstitutionalized, housed adults (aged ≥18 years, US population). In brief, hierarchical hot deck imputation was performed to ensure no participants were missing data for key variables needed for weighting, such as gender, education, race, ethnicity, and marital status; each had <3% missingness [25]. Design weights, adjusted with classification and regression tree (CART) analysis for a differential nonresponse, were developed to facilitate population inference. A raking procedure aligned weighted distributions to the observed distributions from the Census along the lines including age, race-ethnicity, education, and income [26]. To address outlier weights, those at the 99th percentile of each distribution side were trimmed.

Using the sampling weights, we estimated the weighted prevalence and 95% modified Wilson score confidence limits for (1) consistent mask wearing, (2) social distancing, (3) handwashing, and (4) hand sanitizer use. Prevalence estimates were descriptively summarized by sociodemographic factors (race, sex, age, region, urbanicity, highest level of education, annual income, health insurance, job type), personal behaviors (leaving home for work), knowledge of mitigation behaviors, and month of sampling. To identify significant differences in the prevalence of mitigation behaviors by sociodemographic factors, prevalence ratios (PRs) and corresponding 95% CIs were estimated using weighted logistic regression procedures. All prevalence analyses were performed using SAS v9.4 (SAS Institute) and SUDAAN (RTI International).

People may follow all public health recommendations for mitigation practices similarly, or there may be individual variations in mitigation practices. This information may be useful for understanding how different groups respond to multicomponent prevention messages. We therefore conducted latent class analysis (LCA) with polytomous outcomes variables to classify participants based on their responses to the 4 primary dependent measures (mask wearing, social distancing, handwashing, and hand sanitizer use). Each measure was included as a single item. Considering we did not know the number of classes represented by these data, we fit several models with a different number (1-6) of classes. All models were fit using the polCA package in R v4.1.0 (R Foundation for Statistical Computing), which uses maximum likelihood parameter estimation with robust SEs. Each model was estimated with 30 different sets of starting values and allowed a maximum of 3000 iterations for convergence. To select the final model, we compared fit statistics (ie, Bayesian information criteria [BIC], Akaike information criteria [AIC], adjusted BIC) and accuracy statistics (eg, entropy) [27]. Each participant was classified into a latent class (ie, mitigation) group by the largest posterior probability for belonging to each class indicated by the final model. We estimated the weighted prevalence of each latent class group in the entire sample and by sociodemographic characteristics. Finally, as a minimal internal validity check, we examined whether class membership was associated with answering the following statements with “true”: “Consistently wearing a face mask will provide me with 95% or better protection from getting infected with the new coronavirus” and “It is not necessary for children and young adults to take measures to prevent infection by the COVID-19 virus.” PRs and corresponding 95% CIs were estimated using weighted logistic regression procedures in SUDAAN to identify any differences by sociodemographic characteristics.

A total of 4654 participants completed baseline enrollment procedures and were included in this study (Table 1). These participants represented 242,875,582 US adults in 2020. Most were female, were non-Hispanic White, aged 55 years or older, lived in the South, lived in metropolitan areas, had at least a bachelor’s degree, had an income of US $50,000-$99,000, and were privately insured. The highest monthly enrollment occurred in November 2020. Among the available response options in this study for job types, most respondents worked in health care and social services; however, most had other job types not available for selection in the study. Most of those whose jobs required leaving home worked completely indoors.

The estimated national period prevalence of consistently wearing a mask from August through December 2020 was 71.1% (95% CI 68.8-73.3; Table 1). Consistent mask wearing was significantly more prevalent among those who were female; were Hispanic or non-Hispanic Black; lived in a metropolitan area; had a graduate degree, were insured through Medicare, Medicaid, or other public health insurance; worked in educational services; or worked completely indoors (among those who were working from somewhere other than at home). Consistent mask wearing was significantly less prevalent among those who were less than 65 years old, lived in the Midwest (compared to the West), and worked in transportation or warehouse services.

The estimated national prevalence of consistently practicing social distancing was 42.9% (95% CI 40.5-45.3; Table 2). Consistent social distancing was significantly more prevalent among those who were female, were Hispanic or non-Hispanic Black, lived in the South (compared to the West), had a graduate degree or some college education, or had an annual income of less than US $25,000. Consistent social distancing was significantly less prevalent among those aged 18-34 years (compared to ≥65 years) or lived in the Midwest (compared to the West).

Among the 4090 participants who were administered the hand hygiene questions, the average number of times the participants washed hands in the past 24 hours was 8.8 (SE 0.3). The estimated national prevalence of individuals frequently washing hands was 55.0% (95% CI 52.3-57.7; Table 3). Frequent handwashing was significantly more prevalent among those who were female, were Hispanic or non-Hispanic Black, were aged 35-54 years (compared to ≥65 years), were enrolled in November 2020 (compared to August 2020), or worked in health care and social assistance services.

Among the 4090 participants who were administered the hand hygiene questions, the average number of times the participants used a hand sanitizer in the past 24 hours was 4.99 (SE 0.2). The estimated national prevalence of frequently using a hand sanitizer was 21.5% (95% CI 19.4-23.8; Table 4). Frequent use of a hand sanitizer was significantly more prevalent among those who were female, Hispanic or non-Hispanic Black, less than 65 years of age, lived in the South (compared to the West), had an annual income of less than US $25,000, were enrolled in September or November 2020 (compared to August 2020), or worked in accommodation, food services, health care, social assistance, retail trade, or transportation/warehouse services. Frequent use of a hand sanitizer was significantly less prevalent among those who had an annual income of US $100,000-$199,000 or were insured through Medicare, Medicaid, or other public health insurance (compared to those uninsured).

The final classification model identified 3 latent classes: (1) optimal mitigation, (2) optimal mitigation with additional hand hygiene, and (3) lowest mitigation. Optimal mitigation was consistent mask wearing, consistent social distancing, and handwashing or hand sanitizer use 6-10 times per day. Optimal mitigation with additional hand hygiene was consistent mask wearing, consistent social distancing, and handwashing or hand sanitizer use >11 times per day. The lowest mitigation was inconsistent mask wearing, inconsistent social distancing, and handwashing or hand sanitizer use 0-5 times per day. There were no classes that had suboptimal use of only some mitigation strategies but optimal use of others. All participants were categorized into these classes. Two-thirds (n=2656, 67%) practiced optimal mitigation by consistently wearing a mask, consistently following social distancing, and frequently washing their hands or using a hand sanitizer (Tables 5-7). Furthermore, 1 in 5 (n=771, 20.6%) practiced the lowest mitigation by inconsistently or infrequently engaging in all mitigation practices. The final class made up the remainder (n=463, 12.4%) who consistently wore masks and maintained social distance but had the highest frequency of handwashing or sanitizer use (>11 times per day). Compared to optimal mitigation practices, the likelihood of being in the lowest-mitigation class was significantly greater among those who were male, less than 65 years of age, lived in the Midwest (compared to the West), lived outside a metropolitan area, had no health insurance (compared to Medicare, Medicaid, or other public insurance), worked in transportation or warehouse services, or worked somewhere other than completely indoors (Table 8). Compared to just the optimal-mitigation class, the likelihood of being in the class with optimal mask wearing and social distancing but with additional handwashing or sanitizer use was significantly greater among those who were Hispanic or non-Hispanic Black, were less than 65 years of age, had less than a bachelor’s degree, or worked in any of the selected job types (compared to other jobs).

Compared to the optimal-mitigation class, those in the lowest-mitigation class were 20% less likely (PR 0.80, CI 0.79-0.90) to agree that masks provide 95% or better protection against COVID-19 and were twice as likely (PR 2.00, CI 1.27-3.15) to state that it was not necessary for youth to take measures to prevent COVID-19 infection (Table 8). There were no significant differences between the optimal-mitigation class and the class with additional hand hygiene for both the mask-wearing (PR 0.93, CI 0.82-1.06) and youth prevention (PR 1.70, CI 0.92-3.16) questions.

We report the first national probability survey estimates of the prevalence of COVID-19 mitigation strategies among US adults. During the 2020 peak of COVID-19 incidence, nearly three-quarters of adults consistently wore a mask when going out, about half consistently practiced social distancing or frequently washed their hands, and about a quarter frequently used a hand sanitizer. There were 3 distinct patterns of use of these mitigation practices. Two-thirds practiced optimal mitigation, with consistent and frequent use of all mitigation strategies; about 1 in 5 practiced the poorest mitigation practices, with inconsistent or infrequent use of all mitigation strategies; and about 1 in 9 consistently wore a mask and practiced social distancing and may have followed excessive hand hygiene practices. Finally, all mitigation practices and grouping of practices varied substantially among people with different demographic characteristics.

The prevalence of consistently wearing a mask in our population-based study was similar to earlier estimates from polls and convenience samples [5,9,10], but the estimate from our population-based sample was substantially lower than the 89% reported by the online convenience sampling–based COVID Impact Survey in June 2020 [12]. The difference in prevalence could be due to selection bias in the convenience sampling–based COVID Impact Survey if those who were more likely to wear masks were also more likely to respond to the survey. Our survey items were also slightly different, with our survey stipulating mask wearing when going out and asking about the frequency of mask usage, whereas the COVID Impact Survey asked a general question about mask wearing, without regard to context or frequency. It is also possible that mask wearing decreased between June and August 2020 (the beginning of our study), as public facilities reopened and mask requirements in each jurisdiction became more complex and possibly confusing.

There was a similar discrepancy between our findings and those of the COVID Impact Survey for practicing social distancing and washing hands, but for these practices, our estimates were even lower than those of the COVID Impact Survey, which reported >80% prevalence for both [12]. Although the COVID Impact Survey reported a slightly decreased prevalence of social distancing and hand hygiene in June compared to April 2020, a simple extrapolation of that decreasing trend would not explain the difference we found later in 2020. Similar selection biases and differences in survey items for these practices could explain part, but not all, of the difference between the findings of our study and the COVID Impact Survey. ConsumerStyles panel surveys that examined handwashing practices in October 2019 and June 2020 in specific contexts (eg, before eating, after sneezing, or after coughing) also found a substantially higher prevalence of handwashing than we did, providing further evidence that the survey question type (eg, making the questions conditional on situations in which handwashing is recommended even outside of COVID-19 times) can substantially affect prevalence estimates. We structured our questions based on the only published study on the effectiveness of hand hygiene for preventing seasonal coronavirus infection [22]. As we would expect during the COVID-19 pandemic with frequent communications about the importance of hand hygiene, our prevalence estimate of 57.7% who washed their hands 6-10 times in the past 24 hours was substantially higher than the 39.5% reported in the UK study conducted between 2006 and 2009. The prevalence of use of a hand sanitizer in our study was also substantially lower (21.5% vs 70.7%) compared to only 1 other previous paper, by Czeisler et al [14], that reported this as a separate behavior from handwashing. This difference in prevalence was likely due to context-specific differences in behaviors, where Czeisler et al [14] assessed hand sanitizer usage only after contact with high-touch public surfaces.

The distinct sets of mitigation practices (optimal mitigation, lowest mitigation, and optimal mitigation with additional hand hygiene) were also novel findings of our study. Those in the optimal-mitigation and optimal-mitigation-with-additional-hand-hygiene groups frequently wear masks and practice social distancing when they go out in public. Although there was a clear distinction between these groups based on the frequency of handwashing predetermined based on Beale et al’s [22] effectiveness study, there were no differences between these groups in their agreement with the mask-wearing and youth prevention questions. Our findings did indicate that the optimal-mitigation and optimal-mitigation-with-additional-hand-hygiene groups differed on multiple demographic characteristics, which supports the idea that these groups may have fundamental differences in their approaches toward COVID-19 prevention. Further study on the context of hand hygiene practices may clarify some of these issues, and we are now implementing a context-specific set of mitigation practice questions in our 3- and 6-month follow-up surveys with this cohort. The lowest-mitigation group, which was inconsistent in all mitigation practices, comprised an unfortunately large proportion of 1 in 5 US adults. The demographic differences between the optimal- and lowest-mitigation groups were even more pronounced, emphasizing the demographic disparities in COVID-19 mitigation practices.

This heterogeneity in COVID-19 mitigation practices among demographic groups in our study has also been partly reported in other published papers for individual practices [5,9,12,15,20], and those prior published findings are reasonably consistent with the demographic heterogeneity we found. Our study goes a step further to illustrate how persons from various backgrounds combine the individual mitigation recommendations in practice. The demographic heterogeneity in these empirically determined grouping of mitigation practices is even more evident than in individual practices. Compared to men and younger adults, women and older adults are much more likely to optimally use all mitigation practices. These differences in patterns of use may reflect greater risk perception, more exposure to COVID-19 prevention messages, or other contextual factors, such as leaving the home or living in group settings. US adults who were Black or Hispanic (compared to White, non-Hispanic), had no college degree, or worked in service-oriented jobs were more likely to report excessive hand hygiene, while also consistently wearing a mask and maintaining social distancing. These differences might also reflect greater risk perception and prevention message exposure but are more likely due to other contextual factors, such as hand hygiene requirements of their jobs. Finally, US adults who live outside metropolitan areas were likely to engage in all mitigation practices inconsistently. This might be due to differing risk perceptions or exposure to prevention messages in less densely populated areas [28].

This study has several limitations. First, there was a lack of contextual information for some mitigation practices that could better clarify whether people are engaging in practices in only some settings or situations but not in others. These situational assessments have been added to our follow-up surveys, which were completed by mid-2021, and will be included in subsequent analyses. Second, some demographic heterogeneity could be explained by confounding, which could be elucidated with additional modeling. Multivariable modeling is planned for follow-up survey analyses. Third, we did not assess the quality of the mitigation behaviors, such as correctly wearing masks, or the effectiveness of those behaviors on preventing COVID-19 infection. The prospective component of the study will directly examine these associations. Fourth, the enrollment and baseline surveys occurred during a 5-month period of substantial changes in the COVID-19 pandemic and response. There may be time frame heterogeneity in the mitigation behaviors during these changes, but we were unable to analyze these baseline data as cross-sectional time series due to sampling method changes and prioritization of the entire survey sample weighting for national estimates [21]. Finally, although household probability sampling methods and weighting allowed for national estimation of these essential mitigation practices, there is likely still selection bias due to nonresponse.

Although the prevalence of consistently wearing a mask was relatively high among US adults, there were still millions who were not doing so during the time of the highest COVID-19 incidence to date in the pandemic. Even greater numbers of US adults did not consistently practice social distancing outside their homes and did not frequently practice hand hygiene. These practices remained crucial to blunting the surge of COVID-19 infections, especially since we had not yet achieved sufficient vaccine coverage to stop the pandemic. Despite clear public health evidence of their importance, the implementation of these practices was further undermined by a confusing array of local jurisdiction messages about mask requirements and restrictions on public gatherings. In future infectious disease outbreak responses, monitoring mitigation practices in a context of changing mandates and messages will help us refine communication strategies to increase the adoption and persistence of effective mitigation behaviors. This monitoring will also help ensure that disparities in mitigation practices do not widen further, leading to even greater disparities in infectious disease incidence and continuation of high-level community transmission.