Modeling Community COVID-19 Transmission Risk Associated with U.S. Universities

doi:10.21203/rs.3.rs-1899505/v1

The ongoing COVID-19 pandemic is among the worst in recent history, resulting in excess of 520,000,000 cases and 6,200,000 deaths worldwide. The United States has recently surpassed 1,000,000 deaths. Individuals who are elderly and/or immunocompromised are the most susceptible to serious sequelae. Rising sentiment often implicates younger, less-vulnerable populations as primary introducers of COVID-19 to communities, particularly around colleges and universities. This research seeks to evaluate the potential role of college students, measured as total university enrollment, in overall community-level age-adjusted COVID-19 case and mortality rate within each American county. From January 1, 2020 through March 31, 2021, a total of 22,385,335 cases and 374,130 deaths were reported to the CDC. Adjusting for more than 32 key socio-demographic, economic, and epidemiologic variables, we implemented regression and factor analytic methods to determine the overall infection and mortality susceptibility for each county and key factors associated with case and mortality rates. Overall, counties with increasing numbers of university enrollment showed significantly lower case rates and marginal decreases in mortality rates. Contrary to common sentiment, our findings demonstrate that counties with high university enrollments may be more adherent to public safety measures and vaccinations, likely contributing to safer communities.

SARS-CoV-2

Community Transmission

Universities

Age-adjusted rates

Public Health

Pandemic

In the United States (U.S.), early assessments measuring the impact of SARS-CoV-2 (COVID-19) transmission on college campuses and their surrounding communities were challenging due to heterogeneous mitigation and testing strategies. Many public health experts warned during the summer of 2020 that reopening campuses before vaccines were widely available was risky, and subsequent outbreaks among students, staff, and faculty within weeks of arriving on campus for the Fall semester quickly led to shutdowns^1–3.

Observations of COVID-19 responses in the U.S. and other countries, and models of potential outcomes for the Fall semester demonstrated that a safe re-opening of university campuses depended on rigid adherence to containment (e.g., testing, isolation, contact tracing, and quarantine) and mitigation strategies (e.g., masking, ventilation, and social distancing). A key component of this approach was frequent testing of students, faculty, and staff for COVID-19^1,4,5. A stochastic model by Gressman and Peck⁶ evaluating mitigation found that, in addition to high COVID-19 test specificity, remote instruction and a high capacity for student quarantine were critical to successfully managing outbreaks. Findings from an agent-based model also suggested that reducing direct contact in residence halls and classrooms through class size caps or hybrid options, combined with masking and social distancing, could keep the campus viral reproductive rate at or below 1.0⁷. Additionally, evidence suggested that frequent testing and strict social distancing measures may prevent university shutdowns and the economic penalties associated with such a disruption, potentially outweighing the costs of implementing such surveillance⁸. However, given that providing masks, implementing remote learning, and conducting regular testing requires up front and ongoing funding that not all colleges and universities have, modification of these mitigation strategies is often needed ⁹.

Due to the wide variation in the mitigation and containment strategies implemented during the Fall 2020 semester on college campuses across the U.S.³, questions remain about which strategies were most effective in lowering incidence of COVID-19 on campuses, and how campuses impacted the non-university communities they belonged to. The overall sentiment, both in public opinion and some modeling studies, has been that college campuses fuel outbreaks and spread into the surrounding communities^10,11. One study found that re-opening campuses to in-person instruction contributed to increased human mobility and was strongly associated with increased COVID-19 incidence in neighboring communities¹². However, this study only analyzed teaching modality (e.g., in-person vs. hybrid vs. fully online courses) and did not examine other mitigation and containment approaches that were implemented. Stratifying the analysis by collective COVID-response plan may reveal more information about the impact of these measures. In a study of a large (~40000 individuals), northeastern urban university with ample mitigation strategies (e.g. rapid delivery of SARS-CoV-2 polymerase chain reaction (PCR) test results, contact tracing, face mask use, reduced class size and density, and enhancement of all building air systems) more than half of cases were traced to a non-campus source, and no cases were attributed to classroom exposure¹³. These contradictions spawn a twofold question: are universities, particularly those with large (>15000) student populations, impacting community spread of COVID-19, and how?

The purpose of this investigation was to: 1) estimate U.S. county-level COVID-19 case and mortality rates and compare those of counties with and without college campuses, and 2) identify the impact of mitigation efforts on case and mortality rates of COVID-19 in a sample of U.S. counties with college campuses. We hypothesized that colleges and universities with robust COVID-19 response plans (e.g. those that have increasing adherence to mask wearing, social distancing, vaccination rates, frequent testing, education campaigns, etc.) would have a protective effect both on campus members and their surrounding communities.

Overall Impact of Universities on County-level COVID-19 Cases and Deaths

Between January 1, 2020 and March 30, 2021, increasing county university enrollment was associated with significant reductions in COVID-19 case rates, but only slight differences in mortality rates (Figure 1, Table 1). Compared to counties with no universities, this equates to a 1% reduction in cases among counties with small university enrollments, an 8% reduction among counties with medium university enrollments, and a 16% reduction among counties with large university enrollments. A comparison of standardized case and mortality rates by age group shows similar patterns across all enrollment types: 20–59-year-olds accounted for most cases, with the highest case rate in the 20–29-year-old age group (Figure 2a). Despite having a lower case rate than adults under 50, adults over 50 experienced the highest mortality rates regardless of county university enrollment type, and this rate increased with age (Figure 2b). On average, those >80 had nearly twice the risk of death from COVID-19 as 70–79-year-olds, 2.5-times the risk as 60-69-year-olds, and 5-times the risk as 50-59-year-olds.

Across the three wave periods there is a significant reduction in both case and mortality rates with increasing county UE, compared to counties with no university enrollment (Figures 3 & 4). Case and mortality rates remained below 105 and 2.94 (per 100,000), respectively, for all county types through waves 1 and 2, but markedly increased in wave 3 (Figure 4a-b). These rapid increases in both case and mortality rates were experienced across all county types but were less severe as university enrollment increased. The Fall 2020 semester coincided with the beginning of wave 3, and case and mortality rates followed similar trends across all county university types (Figure 4c-d).

The effects of each included covariate were similar across county university enrollment types (Table 2). Each of the 18 covariates evaluated in this analysis was included in at least one final best-fit model; variables included in 75% of all models were 2019 population, CCVI, % receiving at least one vaccination dose, and % of time in which childcare facilities, nursing home visitations, and restaurants were closed at the county level. Statistically weighted averages of logWorth values show that enforcing the closure of nursing home visitations and childcare facilities were the two most important covariates for age-adjusted case rates, while 2019 county population & CCVI were the two most important covariates for age-adjusted mortality rates.

Factor and Cluster Analyses

A total of 1,568 colleges and universities across 740 counties were included in the sub-analysis of college reopening plans for the Fall of 2020 based on availability of reported COVID-19 mitigation plans for the Fall 2020 semester (Table S1). Of these, 235 counties had large enrollments, 196 counties had medium, 302 had small enrollment totals, and seven counties had no reported higher education enrollments.

The multivariate factor analysis revealed that the first three dimensions (for which eigenvalues were above one) only explained 20% of the cumulative variance in population adjusted COVID-19 cases during the Fall of 2020. However, within those dimensions, county-level descriptions (i.e. whether there was a state mask mandate, self-reported mask wearing, and county-level enrollment size). contributed 15% more than expected (~40% of overall contribution) to the first dimension. The second dimension’s contribution to the variance in population-adjusted COVID-19 cases was strongly defined by institutional factors (~50%) (i.e. institutional category, size category, land grant status, and school location description), and campus COVID-19 mitigation strategies (~35%) (i.e. MOI category and testing strategy). The third dimension was almost entirely defined by campus COVID-19 mitigation strategies (~65% of total contribution). Grouped variables contributing more than expected to these three dimensions (Table S1) were retained for both the cluster analysis and modeling.

From these factors, we constructed a hierarchical cluster analysis that identified three distinct descriptive school clusters based on their relation to population-adjusted county-level COVID cases (Table 3). All factors with a positive value test score above 1.96 (p-value < 0.05) were retained. Clusters were most heavily defined by the school location setting (e.g. rural vs urban), the county enrollment size, and the size category of the institution. Cluster 1 represented suburban and city-based small private institutions in counties with large student enrollments that did not report COVID measures for the fall 2020 semester. Cluster 2 represented larger, city-based land grant public institutions that remained online or in hybrid instruction mode for the fall 2020 semester, and that offered multiple testing options for students. Schools in clusters 1 and 2 were in counties under mask mandates and that reported high mask usage. The third cluster was defined by small schools in counties with either no or small student enrollments, that held in-person instruction during the fall of 2020, offered voluntary or no testing options for students, and whose counties reported low to moderate mask usage and had no mask mandates in place at the state level.

Modeling COVID-19 Mitigation Plans Impact on County-Level Cases

Best fit models that examined the association of campus COVID-19 mitigation strategies (MOI, testing strategies, school-specific variables) and economic/county-level covariates (mask usage, mask mandates, median household income) with population-adjusted cases in the Fall of 2020 showed that the most strongly contributive factors to increased county population-adjusted case numbers were low and moderate mask usage, lack of state mask mandates, and median household income (Table 4). For the school-related covariates alone, the overall best fit model predicted that larger schools, schools located in rural areas, and non-land-grant institutions were associated with more county-level cases, (i.e. schools that were part of Cluster 2 were more likely to be located in counties with lower population-adjusted COVID-19 case rates, while those in Clusters 1 and 3 were associated with higher county-level cases). Overall, MOI and testing strategies were not significantly predictive of county-level case rates during the Fall 2020 semester, except that schools that employed mandatory testing of students were associated with counties with lower population adjusted case numbers.

SARS-CoV-2 is a highly infectious respiratory pathogen that has rapidly spread throughout the world since early 2020. It has been speculated that counties with large university enrollments were at higher risk for COVID outbreaks in the U.S. In this analysis, we evaluated a total of 22,385,335 cases reported to the CDC, representing 3,047 U.S. counties from January 1, 2020 through March 30, 2021. After all cases and deaths were aggregated to the county level and categorized by their total university enrollment sizes, this analysis found that the presence of large university enrollments was associated with lower county COVID-19 case rates.

A retrospective analysis evaluating 15 months of cases and deaths caused by COVID-19 reveals small, but significant reductions in cases among counties with increasing university enrollments. However, little to no change was noted with respect to mortality rates. Further analyses focused on differences between age groups also reveal little to no variation in the case and mortality rates of each group across university enrollment size. However, notable age-related trends were discovered. First, the highest case rates were among young to middle-aged adults (20-59), with 20–29-year-olds experiencing the highest case rates than any other age group – a finding also found by Monod et al.¹⁴ Second, although the 0-9 and 80+ year old age groups experienced the lowest case rates, the 80+ year old age group’s risk of death was substantially higher than all other age groups. These trends corroborate the widely observed findings pertaining to COVID-19 fatality^15,16.

The time series plots of daily aggregated county averages of case and mortality rates (Figure 3) show observable differences in the magnitude of peaks with increasing county university enrollment. However, when evaluating COVID transmission by wave, critical differences become more apparent. Three COVID-19 waves occurred prior to May 2021, with each wave more severe than the wave before. However, the third wave saw a ~3-6-fold increase in cases and deaths as compared to waves 1 and 2. Although all counties experienced this drastic increase in COVID-19, those with more university enrollment had significantly lower case rates (5.3%, 10.6%, and 27.2% for counties with small, medium, and large university enrollments, respectively) compared to counties with no university enrollment. Counties with medium or large university enrollments experienced significantly lower COVID-19 death rates (averaging 12.8% and 29.8% lower, respectively) compared to counties with low or no university enrollments.

The second epidemiologic period of interest was among counties before and after the Fall 2020 semester. As seen in the wave analysis, all counties experienced a rapid increase in both COVID-19 cases and deaths after the start of the Fall 2020 semester. However, counties with increasing university enrollment experienced significantly lower case rates (3.7%, 10.8%, and 27.1% lower for counties with small, medium, and large university enrollments, respectively) compared to counties with no university enrollment. Counties with medium and large university enrollments also resulted in significantly lower death rates (averaging 13.2% and 30.2% lower, respectively) compared to counties with low or no university enrollment.

Our sub-group analysis of COVID-19 mitigation strategies for the Fall 2020 semester provided further evidence that colleges and universities were not associated with increases in county-level cases – even before cohesive containment plans were established. The cluster analysis identified that schools identified in Cluster 2 tended to be larger, land grant universities that maintained entirely online or hybrid course instruction during the Fall 2020 semester and were likely to be in counties with lower population-adjusted COVID cases. These schools were more likely to have mandatory student testing that was significantly associated with overall lower county cases numbers. During the Fall 2020 semester, county and state-level factors (e.g., mask usage, mask mandates, and median household income) were far more significantly predictive of overall county-level cases, which held true during the larger analysis period. Limitations to this sub-group analysis include an incomplete picture of mitigation strategies for some of the universities (e.g. Cluster 1 was defined by schools that did not fully report their COVID-19 plans). Counties without higher-education enrollments were also severely limited in this dataset, which could have impacted the analysis.

Overall, the COVID-19 pandemic through March 30, 2021 rapidly spread through all U.S. counties with similar patterns in the timing and intensity of cases and deaths, regardless of the size of university enrollment. However, the magnitude by which cases and deaths affected counties is strongly associated with university enrollment. Although there were minimal differences among death rates by university enrollment, large enrollment universities were most affected by COVID-19 in the early stages of the pandemic - a suspected driving force behind the lack of significant differences in overall mortality rates. However, as the pandemic progressed in intensity (case and deaths per day), counties with increasing university enrollments experienced decreased risks in acquiring, and dying from, COVID-19.

Despite having larger populations, counties with large university enrollments fared better than counties with little or no university enrollments, especially as COVID-19 cases surged through the winter of 2020-2021 (wave 3). In comparison to counties with little or no university enrollment, large university enrollment counties contained higher household incomes, less unemployment, and had higher vaccination rates (% with at least 1 dose). These counties also tended to enforce statewide mandates more frequently and for longer throughout the pandemic compared to counties with little to no university enrollments (Tables S2 & S3).

In addition to the differences noted above, public health decisions were dependent on several political, economic, and social factors. It is apparent that this pandemic has fueled divisions along political lines, which influenced both public health decisions and compliance. From adherence to social distancing and mask wearing to vaccination rates, political associations appear to be a strong influencing force¹⁷. Using the 2020 U.S. Presidential election as a proxy for determining a county’s political affiliation, the associations of COVID-19 cases and deaths are moderately correlated (Figures S1 & S2). Counties with higher university enrollments are also correlated with increased overall education rates, which have been found to be associated with use of pandemic control^18–20.

Counties with large university enrollments appear to have several key protective factors in place to mitigate COVID-19. However, these counties also contain significantly larger populations of black, indigenous, and people of color (BIPOC) that suffer from disproportionate racial and health inequities. Consequently, social vulnerabilities are significantly higher in counties with large university enrollments. Preliminary analyses of this dataset, in respect to COVID-19 outcomes by race and ethnicity, reveal evidence of these inequities (data not shown). However, this analysis is beyond the scope of the current study.

This study is limited in its analysis by aggregating non-lineage-specific individual-level COVID-19 case data to county of residence. By doing so, generalizations are made across varying population sizes and COVID-19 strains and cannot capture subtle differences. Numerous reporting agencies submit varying levels of completeness to the CDC, leading to a high potential for ascertainment bias. Additionally, due to the rapid onset of cases, particularly through peak wave periods, significant delays in reporting also occurred. This analysis attempted to establish etiologic-specific onset dates given the data made available. In some cases, differences of 1+ week from reported onset date to actual, true onset date likely exist and are unavoidable. However, the data account for a significant portion of the U.S. population, thus reducing the errors in generalization and representing the best source for such a study.

An increase in a county’s university enrollment is associated with a lower rate of COVID-19 cases and deaths compared to counties with no university enrollments.
Counties with the largest universities tend to have greater population sizes, higher household income, lower unemployment, a higher percentage of people partially vaccinated against COVID-19, and a higher proportion of the population voting for Biden in the 2020 presidential election, but also a higher social vulnerability, than those with no university enrollment.
Counties with university enrollments tend to have higher enforcement of statewide closing mandates than those without.
Despite having high population densities and high social vulnerabilities, individuals in counties with large university enrollments had reduced COVID-19 case and mortality rates than those without university enrollments, especially in the third wave of the pandemic.
Increasing education levels, enforcement of closing mandates, adherence to public health recommendations, and political affiliation were all associated with lower COVID-19 case and mortality rates.
There was no significant impact on community cases from universities who returned for the Fall 2020 semester; county-level factors were the leading predictors of cases.
These results offer evidence against the presumption that universities increase risk of COVID-19 community transmission.

Data Sources

Cases and Mortalities

SARS-CoV-2 laboratory-confirmed case and mortality data were acquired from the Centers for Disease Control and Prevention (CDC) data archive²¹. The time period of January 1, 2020 to March 30, 2021 was selected for three reasons: 1) the Spring (approximately January-May) and Fall (approximately September-December) periods capture the entirety of a typical university calendar year, providing a control period to adequately measure COVID-19 transmission in local communities, 2) The commencement of the fall semester (September 1, 2020) is a critical point as it is approximately the median time point for this analysis, and 3) The U.S. experienced three distinctive COVID-19 “waves” over this period, creating a large representative dataset for comparing counties with campuses across the country.

By March 30, 2021 a total of 22,385,335 cases and 374,130 deaths were reported to the CDC. Each case in this dataset contained 32 elements, including demographics (e.g., age, race and ethnicity, and sex) and county/state of residence. Numerous reporting agencies provide this information to the CDC and as a result, lineage-specific COVID-19 designation, symptom onset, and/or test positive dates were often incomplete, inconsistent, and/or provisional. All cases were aggregated with no sequencing-specific lineage designation. If multiple dates were listed, a single case-positive date was selected from the following (the same order of preference used by the CDC) data: symptom onset date, test positive date, and CDC report date. If no date or county of residence information was provided, the case was excluded. All personally identifiable information (names, addresses, etc.) were removed by the CDC prior to public release. Geographic locations for each case were aggregated to county-level for additional privacy protections.

University Enrollment

University enrollment (UE) data were acquired from the Integrated Postsecondary Education Data System (IPEDS)²². Total enrollment by postsecondary institution for the Fall semester of 2018-2019 academic year was aggregated by county and divided into four categories. Counties with total enrollment x 15000, 15000 > x > 5000, 5000 x > 0, or no enrollment, were labeled large (n = 253), medium (n = 361), small (n = 792), and absent (n = 1641), respectively.

U.S. Population and Other Key Covariates

County population was acquired from the American Community Survey (ACS) 2019 5-year estimate²³. These data provide estimates of population by age from 60 months of sampling between January 1, 2015 and December 31, 2019. For purposes of this study, age groups have been categorized by 10-year intervals as 0-9, 10-19, 20-29, 30-39, 40-49, 50-59, 60-69, 70-79, and 80+.

Other factors of interest in evaluating COVID-19 among populations at the county level included median household income²³, unemployment²⁴, COVID-19 community vulnerability index (CCVI)²⁵, percentage of population that was vaccinated as of March 30, 2021 (at least 1 dose)²⁶, and percentage vote by party candidate in the 2020 U.S. Presidential Election²⁷.

University COVID-19 Mitigation and Containment Policies

Highly incongruous COVID-19 policies were implemented across U.S. universities during of the Fall 2020 semester reopening. We therefore selected aggregated variables that could be compared widely across institutions of various size and location by the College Crisis Initiative at Davidson College²⁸. Only data from four-year universities were included in the analysis, and schools were organized by the aforementioned enrollment size categories. Reopening plan variables included the “Mode of Instruction” (MOI) as of September 1, 2020 and proposed on-campus COVID-19 testing strategy (Table S1). Other educational institution factors (e.g. land grant university status and degree of campus urbanization)²² were included in the analysis due to potential impact of funding mechanisms and population density structures. Additionally, we evaluated county-aggregate factors such as self-reported masking adherence, state-instituted mask mandates²⁹, median household income²³ and unemployment rates²⁴.

Statistical Analysis

Cases and deaths were age-adjusted using the 2019 ACS 5-year estimates as the reference population. Each county’s COVID-19-confirmed cases and deaths were organized by age group for the time period between January 1, 2020 and March 30, 2021. These were then aggregated as a total for each outcome and age-adjusted.

Age-adjusted COVID-19 case and death rates were evaluated across three time scenarios: 1) January 1, 2020 to March 30, 2021; 2) By each “wave” period, determined by natural breaks between peaks among national cases (Wave 1: January 1, 2020 – June 7, 2020, Wave 2: June 8, 2020 – September 6, 2020, Wave 3: September 7, 2020 – March 30, 2021); and 3) Before and during the Fall 2020 academic semester (January 1, 2020 - August 31, 2020 and September 1, 2020 - March 30, 2021)^30–32.

All statistical comparisons of groups were assessed using JMP® (Version 16.0)³³. Group means were compared using Student’s t-test or Dunnett’s Test (control group = counties without university enrollment). Comparisons were significant at p < 0.05. Age-adjusted case and mortality rates were modeled separately as dependent variables. Models were stratified by county university enrollment size, totaling eight standard least squares regression models. Final model selections were made using backwards Bayesian Information Criterion (BIC)-based stepwise regression. Overall significance of an independent variable was assessed by its frequency of inclusion in each of the four county university enrollment types, as well as averaging the logWorth (calculated as -log₁₀(p-value)) of that variable across all four university county types.

Since the college reopening dataset contained both quantitative and qualitative variables, a factor analysis of mixed data using the FactomineR package³⁴ in R (version 4.1.2) was conducted to identify important variables related to university reopening plans. Once the contributions of these variables to the overall variance of the dataset were assessed, a hierarchical cluster analysis was performed (using Bartlett test of sphericity, p < 0.05) to identify similar school mitigation strategy clusters associated with population-adjusted COVID-19 cases at the county level.

Finally, due to overdispersion in the case data, we fit negative binomial models using the MASS package³⁵ to identify university COVID-19 mitigation strategies and other county-level variables that significantly predicted population-adjusted county COVID cases in the Fall 2020 semester. Model fits were evaluated with the performance package³⁶ which gives a performance score based on the model’s BIC, Nagelkerke’s R², and root-mean square error.

Data availability statement

The U.S. COVID data used in this study are available at https://data.cdc.gov/Case-Surveillance/COVID-19-Case-Surveillance-Restricted-Access-Detai/mbd7-r32t after completing the required data access request. The raw U.S. university enrollment data are available from IPEDS at https://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2021100REV. All U.S. census data are available from the U.S. Census Bureau at https://data.census.gov/cedsci/. University COVID-19 Mitigation and Containment Policies are available from Davidson College’s College Crisis Initiative at https://collegecrisis.org/.

Code availability

The college reopening analysis was coded in R (version 4.1.2) and is available at https://github.com/hkopsco/COVIDonCampus.git.

Author contributions

J.A.U. and H.K. conceived the study. J.A.U. designed the analysis, developed and ran the university COVID model and wrote the first draft of the paper. H.K. developed and ran the college reopening and mitigation model. J.A.U., H.K., J.M., W.M.B., and R.L.S. interpreted the results and provided edits to the paper. J.A.U. and H.K. wrote and revised the paper.

Competing Interests

The authors declare no competing interests. All personal identifiable information was removed by the CDC prior to authorization for use in this research analysis. No human or animal subjects were researched in this study.

Yamey, G. & Walensky, R. P. Covid-19: Re-opening universities is high risk. The BMJ vol. 370 Preprint at https://doi.org/10.1136/bmj.m3365 (2020).
Wilson, E. et al. Multiple COVID-19 Clusters on a University Campus — North Carolina, August 2020. MMWR. Morbidity and Mortality Weekly Report 69, (2020).
Walke, H. T., Honein, M. A. & Redfield, R. R. Preventing and Responding to COVID-19 on College Campuses. JAMA - Journal of the American Medical Association vol. 324 Preprint at https://doi.org/10.1001/jama.2020.20027 (2020).
Paltiel, A. D., Zheng, A. & Walensky, R. P. Assessment of SARS-CoV-2 Screening Strategies to Permit the Safe Reopening of College Campuses in the United States. JAMA Network Open 3, (2020).
Cheng, S. Y., Jason Wang, C., Chiung-Tao Shen, A. & Chang, S. C. How to safely reopen colleges and universities during covid-19: Experiences from taiwan. Annals of Internal Medicine 173, (2020).
Gressman, P. T. & Peck, J. R. Simulating COVID-19 in a university environment. Mathematical Biosciences 328, (2020).
Goyal, R., Hotchkiss, J., Schooley, R. T., de Gruttola, V. & Martin, N. K. Evaluation of severe acute respiratory syndrome coronavirus 2 transmission mitigation strategies on a university campus using an agent-based network model. Clinical Infectious Diseases 73, (2021).
Losina, E. et al. College campuses and COVID-19 mitigation: Clinical and economic value. Annals of Internal Medicine 174, (2021).
Pavlov, O. v. & Katsamakas, E. COVID-19 and financial sustainability of academic institutions. Sustainability (Switzerland) 13, (2021).
Quintana, C. & Stucka, M. “Astonishingly risky”: COVID-19 cases at colleges are fueling the nation’s hottest oubreaks. USA Today (2020).
Lu, H. et al. Are college campuses superspreaders? A data-driven modeling study. Computer Methods in Biomechanics and Biomedical Engineering 24, (2021).
Andersen, M. S., Bento, A. I., Basu, A., Marsicano, C. R. & Simon, K. College Openings in the United States Increased Mobility and COVID-19 Incidence. medRxiv (2021).
Hamer, D. H. et al. Assessment of a COVID-19 Control Plan on an Urban University Campus during a Second Wave of the Pandemic. JAMA Network Open 4, (2021).
Monod, M. et al. Age groups that sustain resurging COVID-19 epidemics in the United States. Science (1979) 371, (2021).
Centers for Disease Control and Prevention. COVID-19 Case Surveillance Public Use Data with Geography. (2022).
Davies, N. G. et al. Age-dependent effects in the transmission and control of COVID-19 epidemics. Nature Medicine 26, (2020).
Painter, M. & Qiu, T. Political beliefs affect compliance with government mandates. Journal of Economic Behavior and Organization 185, (2021).
Rattay, P. et al. Differences in risk perception, knowledge and protective behaviour regarding COVID-19 by education level among women and men in Germany. Results from the COVID-19 Snapshot Monitoring (COSMO) study. PLoS ONE 16, (2021).
Brouard, S., Vasilopoulos, P. & Becher, M. Sociodemographic and psychological correlates of compliance with the Covid-19 public health measures in France. Canadian Journal of Political Science 53, (2020).
Bish, A. & Michie, S. Demographic and attitudinal determinants of protective behaviours during a pandemic: A review. British Journal of Health Psychology 15, (2010).
Centers for Disease Control. COVID-19 Case Surveillance Restricted Access Detailed Data. (2022).
National Center for Education Statistics. Integrated Postsecondary Education Data System. (2022).
United States Census Bureau. American Community Survey (ACS). (2022).
USA Facts. 2019 Unemployment Rate . (2022).
Precision for COVID. The U.S. COVID Community Vulnerability Index (CCVI): Vulnerability — how well a community handles the repercussions of a COVID-19 outbreak — matters. (2021).
Centers for Disease Control and Prevention. Rates of COVID-19 Cases or Deaths by Age Group and Vaccination Status. (2022).
McGovern, T., Larson, S., Morris, B. & Hodges, M. tonmcg/US_County_Level_Election_Results_08-16: US County-level Presidential Election Results. (2020) doi:10.5281/ZENODO.3975765.
Davidson College. The College Crisis Initiative. (2022).
Avila, Y., Harvey, B., Lee, J. C. & Shaver, J. W. See Mask Mandates and Guidance in Each State. (2022).
Sen, P., Yamana, T. K., Kandula, S., Galanti, M. & Shaman, J. Burden and characteristics of COVID-19 in the United States during 2020. Nature 598, (2021).
Feehan, D. M. & Mahmud, A. S. Quantifying population contact patterns in the United States during the COVID-19 pandemic. Nature Communications 12, (2021).
Hassan, E. M. & Mahmoud, H. N. Impact of multiple waves of COVID-19 on healthcare networks in the United States. PLoS ONE 16, (2021).
SAS Institute Inc., Cary, NC, U. Sas® 9.4. SAS Institute Inc., Cary, NC, USA (2017).
Lê, S., Josse, J. & Husson, F. FactoMineR: An R package for multivariate analysis. Journal of Statistical Software 25, (2008).
Venables, W. N. & Ripley, B. D. Modern Applied Statistics with S Fourth edition by. World vol. 53 (2002).
Lüdecke, D., Ben-Shachar, M., Patil, I., Waggoner, P. & Makowski, D. performance: An R Package for Assessment, Comparison and Testing of Statistical Models. Journal of Open Source Software6, (2021).

Table 1. Summary of United States population (U.S. Census 2010) and key socioeconomic, demographic, and political variables by race & ethnicity and county total university enrollment (none, small, medium, and large). Cases and deaths labeled as unknown/missing denote individuals where gender was not provided.

Variable		County University Enrollment
		None (n = 1641, ref)	Small (n = 792)	Medium (n = 361)	Large (n = 253)
Population		33,260,133	49,051,160	56,626,128	184,344,451
Population		1	1.47**	1.70***	5.54***
Socio-demographic Descriptors
	Average Household Income	$53,582	$55,475	$60,190	$67,690
	Average Household Income	1	1.04**	1.12***	1.26***
	Social Vulnerability	0.45	0.54	0.56	0.60
	Social Vulnerability	1	1.20***	1.24***	1.33***
	% Unemployed	4	4.10	3.89	3.59
	% Unemployed	1	1.03	0.97	0.90***
2020 Presidential Election
	% Biden	0.28	0.32	0.38	0.51
	% Biden	1	1.14***	1.36***	1.82***
	% Trump	0.70	0.63	0.56	0.41
	% Trump	1	0.90***	0.80***	0.59***
Cases
	Female	1,053,896	1,548,981	1,896,893	5,654,215
	Female	1	1.47*	1.80***	5.37***
	Male	957,526	1,391,310	1,701,010	5,125,302
	Male	1	1.45*	1.78***	5.35***
	Unknown/missing^c	61,518	50,065	46,541	103,104
	Unknown/missing^c	1	0.81	0.76	1.68
Deaths
	Female	15,512	25,328	30,388	96,144
	Female	1	1.63	1.96***	6.20***
	Male	17,839	29,099	33,319	115,240
	Male	1	1.63	1.87***	6.46***
	Unknown/missing^c	1,110	677	427	535
	Unknown/missing^c	1	0.61	0.38	0.48
Average Age-Adjusted Case Rate (x 100,000 people)		748.91	741.87	685.80	631.23
Average Age-Adjusted Case Rate (x 100,000 people)		1	0.99	0.92***	0.84***
Average Age-Adjusted Mortality Rate (x 100,000 people)		12.61	13.80	12.36	12.82
Average Age-Adjusted Mortality Rate (x 100,000 people)		1	1.09**	0.98	1.02

Vaccination
	% Completed Series	14.20	14.76	14.43	14.17
	% Completed Series	1	1.04***	1.02	1.00
	% Received 1 Dose	21.96	23.44	23.62	24.51
	% Received 1 Dose	1	1.07***	1.08***	1.12***
*Data describing trends from January 1, 2020 - March 30, 2021
^aComparison of means analyzed using Dunnett's Test (race & ethnicity control = white population; county university enrollment control = none)
^bThe following correspond to the following statistical values: * = p <0.05, ** = p<0.01, *p<0.001
^cUnable to statistically analyze

Table 2. Matrix displaying covariates (n=18) included in each final model assessing age-adjusted case (A) and mortality (B) rates by university enrollment and race & ethnicity. Overall significance for each included covariate is indicated by gray shade.

Parameter		Average Age-Adjusted Case Rates				Frequency	Average logWorth Value	Influence on Dependent Variable	Average Age-Adjusted Mortality Rates				Frequency	Average logWorth Value	Influence on Dependent Variable
Parameter		None	Small	Medium	Large	Frequency	Average logWorth Value	Influence on Dependent Variable	None	Small	Medium	Large	Frequency	Average logWorth Value	Influence on Dependent Variable
2019 population		0.017 (+)		0.0045 (+)	0.00050 (+)	3	13.878	Increasing (all)	0.0016 (-)	0.00068 (-)	0.00047 (-)	2.20e^-5 (-)	4	36.311	Protective (all)
2019 population		11.154		17.588	12.892	3	13.878	Increasing (all)	46.656	39.504	50.2399	8.846	4	36.311	Protective (all)
% State-wide Closing Mandate Enforced	K-12 Schools	2856.67 (+)	784.43 (+)		444.66 (+)	3	7.553	Increasing (all)	70.63 (+)		14.33 (+)		2	3.076	Increasing (both)
	K-12 Schools	13.046	5.077		4.537	3	7.553	Increasing (all)			3.076		2	3.076	Increasing (both)
	Day/Child Cares	1126.58 (-)	1286.64 (-)	1354.53 (-)		3	17.804	Protective (all)	37.28 (+)		35.27 (+)	47.00 (+)	3	3.524	Increasing (all)
	Day/Child Cares	17.184	23.46	12.767		3	17.804	Protective (all)	3.921		2.936	3.715	3	3.524	Increasing (all)
	Nursing Home Visitations	209.48 (-)	243.99 (-)	224.95 (-)	161.85 (-)	4	36.350	Protective (all)	2.64 (-)				1	3.797	Protective
	Nursing Home Visitations	69.214	43.23	24.794	8.163	4	36.350	Protective (all)	3.797				1	3.797	Protective
	Non-essential businesses	263.13 (-)	286.87 (-)		261.71 (-)	3	11.283	Protective (all)	28.48 (-)				1	4.106	Protective
	Non-essential businesses	22.94	8.362		2.547	3	11.283	Protective (all)	4.106				1	4.106	Protective
	Restaurants	179.36 (-)	229.97 (-)	358.30 (-)	315.20 (-)	4	5.566	Protective (all)	6.53 (+)				1	2.172	Increasing
	Restaurants	3.271	3.672	6.037	9.282	4	5.566	Protective (all)	2.172				1	2.172	Increasing
	Gyms	346.45 (-)				1	6.336	Protective	17.42 (-)	3.45 (-)			2	3.587	Protective (both)
	Gyms	6.336				1	6.336	Protective	6.604	0.569			2	3.587	Protective (both)
	Movie Theaters	248.99 (+)	66.89 (-)			2	3.318	Increasing (x1), Protective (x1)	16.80 (+)		9.71 (-)		2	5.639	Increasing (x1), Protective (x1)
	Movie Theaters	6.343	0.292			2	3.318	Increasing (x1), Protective (x1)	8.101		3.176		2	5.639	Increasing (x1), Protective (x1)
	Bars			131.96 (-)		1	2.058	Protective	9.83 (-)	6.84 (-)		4.30 (+)	3	3.338	Increasing (x1), Protective (x2)
	Bars			2.058		1	2.058	Protective	5.178	3.058		1.779	3	3.338	Increasing (x1), Protective (x2)
	Casinos			247.25 (+)	239.87 (+)	2	2.808	Increasing (both)	15.18 (+)	13.13 (+)	8.76 (+)		3	6.105	Increasing (all)
	Casinos			2.044	3.571	2	2.808	Increasing (both)	8.314	6.895	3.105		3	6.105	Increasing (all)
	Overnight Businesses		94.10 (-)		176.62 (+)	2	2.938	Increasing (x1), Protective (x1)	3.96 (-)	9.26 (-)	9.25 (-)		3	5.725	Protective (all)
	Overnight Businesses		2.599		3.277	2	2.938	Increasing (x1), Protective (x1)	2.964	9.008	5.203		3	5.725	Protective (all)
CCVI			268.71 (+)	175.62 (+)		2	6.828	Increasing (both)	15.49 (+)	11.50 (+)	20.25 (+)	10.87 (+)	4	14.722	Increasing (all)
CCVI			10.623	3.032		2	6.828	Increasing (both)	29.061	9.723	16.109	3.993	4	14.722	Increasing (all)
2019 Average Household Income		0.0046 (-)				1	9.432	Protective	0.00014 (+)	0.00022 (+)	0.00047 (+)		3	9.670	Increasing (all)
2019 Average Household Income		9.432				1	9.432	Protective	3.155	7.588	18.268		3	9.670	Increasing (all)
2019 Unemployment Rate		43.32 (-)		20.62 (+)		2	8.485	Increasing (x1), Protective (x1)		2.18 (+)	2.25 (+)		2	6.776	Increasing (both)
2019 Unemployment Rate		15.813		1.157		2	8.485	Increasing (x1), Protective (x1)		6.339	7.212		2	6.776	Increasing (both)
Vaccination Rates	Completed Series		13.69 (-)	27.82 (-)	17.24 (-)	3	4.610	Protective (all)	0.017 (+)	0.25 (+)			2	0.236	Increasing (both)
	Completed Series		4.092	7.226	2.512	3	4.610	Protective (all)	0.114	0.358			2	0.236	Increasing (both)
	Received 1 Dose	6.51 (+)	20.73 (+)	32.57 (+)	23.36 (+)	4	13.741	Increasing (all)			0.32 (+)		1	1.854	Increasing
	Received 1 Dose	11.499	15.699	19.365	8.4	4	13.741	Increasing (all)			1.854		1	1.854	Increasing
Voting % Trump			317.60 (+)	527.47 (+)	405.33 (+)	3	12.610	Increasing (all)	6.88 (+)				1	2.854	Increasing
Voting % Trump			11.3	18.68	7.85	3	12.610	Increasing (all)	2.854				1	2.854	Increasing
Model Performance	BIC	171892.5	90884.68	43677.5	31405.1				80421.54	44980.72	21473.29	16868.98
Model Performance	r²_adj	0.0993	0.1465	0.233	0.1761				0.056	0.06883	0.1244	0.0404
		p < 0.001	p < 0.01	p < 0.05					p < 0.001	p < 0.01	p < 0.05

Table 3. Within-cluster (Mod.Cla; percentages of individuals who fit this description are in this cluster) and across-cluster (Cla.Mod; this percentage of individuals in this cluster fit this description) distributions of variables. V-test values greater than 1.96 correspond to a p-values less than 0.05; and indicate an over-represented variable for the category. Only variables with a p-value < 0.05 are included as this shows that one category is significantly linked to the other categories.

		Cluster 1				Cluster 2				Cluster 3
Variable	Category	Within cluster % (Cla/Mod)	Across cluster % (Mod/Cla)	p-value	v-test score	Within cluster % (Cla/Mod)	Across cluster % (Mod/Cla)	p-value	v-test score	Within cluster % (Cla/Mod)	Across cluster % (Mod/Cla)	p-value	v-test score
Institution type	Private not-for-profit	55.33	89.62	5.39e-71	18.27
	Private for-profit
	Public					57.51	70.90	1.80e-74	18.25
Institution size	Under 1,000	62.26	38.57	4.59e-72	10.35
	1,000 - 4,999									43.35	64.27	2.15e-22	9.73
	5,000 - 9,999					58.76	24.56	1.39e-19	9.05
	10,000 - 19,999					73.71	24.78	2.22e-33	12.03
	20,000 and above					94.01	23.70	1.91e-53	15.38
Land grant status	Land grant institution					77.14	11.63	5.39e-17	8.37
Land grant status	Not a land grant institution	41.25	100.00	5.76e-69	8.20
School location	Suburb	69.34	45.05	1.66e-103	14.33
	Town									79.19	56.26	2.39e-101	21.37
	Rural									70.63	18.27	1.69e-21	9.52
	City	42.55	48.13	2.04e-2	2.31	43.69	65.73	5.27e-28	10.97
County Enrollment	Large	54.86	79.09	2.44e-49	14.50	40.00	77.15	1.77e-26	10.64
	Medium									45.84	28.33	1.84e-09	6.01
	Small									81.56	59.95	2.38e-117	23.02
	Absent									57.14	1.64	4.82e-02	1.97
On campus testing strategy	No testing mentioned	56.82	70.82	5.54e-89	14.04
	Encouraged, not provided, or prearrival only									43.47	10.26	3.00e-03	2.90
	Only symptomatic testing offered					50.58	37.50	9.44e-21	9.34
	Voluntary testing									50.72	7.18	5.40e-04	3.45
	Mandatory sample testing of some students					51.91	20.47	1.62e-11	6.73
	Mandatory testing of all students					52.27	9.91	5.01e-06	4.56
MOI	Online					47.37	48.70	1.62e-23	9.99
	Hybrid					39.44	27.80	1.70e-05	4.30
	In person									57.49	52.77	2.57e-44	13.96
	Unknown	59.03	29.49	1.28e-53	7.69
Mask usage	High	49.40	94.32	1.28e-53	14.93	34.23	87.50	1.33e-13	7.40
	Moderate									71.10	44.96	3.92e-60	16.35
	Low									95.83	14.16	6.11e-33	11.95
State mask mandate status	Masks mandated	43.05	79.57	7.63e-17	4.14	32.35	80.81	4.06e-05	4.10
State mask mandate status	Masks not mandated									47.43	39.83	3.91e-16	8.14

Table 4. Best overall fit model that predicts population-adjusted county-level COVID-19 cases during the fall of 2020 (September 1st-December 31st) including college/university characteristics and COVID-19 mitigation strategies.

Variable	Coefficient	Standard Error	z-value	p-value
Intercept/Reference	1.589e+00	3.977e-01	3.996	6.44e-05 ***
School Size	1.261e-01	4.998e-02	2.522	0.01167 *
10,000 - 19,999	1.261e-01	4.998e-02	2.522	0.01167 *
20,000 and above	3.958e-01	6.031e-02	6.564	5.25e-11 ***
5,000 - 9,999	9.977e-02	4.486e-02	2.224	0.02616 *
Under 1,000	5.367e-02	3.393e-02	1.582	0.11365
School Category	-6.186e-01	3.560e-01	-1.737	0.08231
Private, not-for-profit	-6.186e-01	3.560e-01	-1.737	0.08231
Public	-7.045e-01	3.571e-01	-1.973	0.04850 *
Land Grant Status	1.560e-0	6.632e-02	2.352	0.01868 *
Not a Land Grant Institution	1.560e-0	6.632e-02	2.352	0.01868 *
School Location	1.155e-01	5.464e-02	-2.114	0.03449 *
Rural	1.155e-01	5.464e-02	-2.114	0.03449 *
Suburb	1.158e-02	3.390e-02	0.342	0.73264
Town	-8.366e-02	4.142e-02	-2.020	0.04342 *
MOI	-5.563e-02	3.709e-02	-1.500	0.13367
Online	-5.563e-02	3.709e-02	-1.500	0.13367
None (School Closed)	-1.160e-02	1.936e-01	-0.599	0.54894
In person	2.263e-02	3.748e-02	0.604	0.54597
Unknown	2.531e-02	4.028e-02	0.628	0.52975
On-campus Testing	-1.501e-01	7.265e-02	-2.067	0.03876 *
Mandatory regular testing of all students	-1.501e-01	7.265e-02	-2.067	0.03876 *
Mandatory sample testing of some students	-1.472e-01	6.090e-02	-2.418	0.01561 *
No testing mentioned	-4.546e-02	5.139e-02	-0.885	0.37641
Only symptomatic testing offered	6.112e-03	5.466e-02	0.112	0.91097
Voluntary testing	7.706e-03	7.738e-02	0.100	0.92067
State Mask Mandate	3.615e-01	3.317e-02	10.900	< 2e-16 ***
Masks not mandated	3.615e-01	3.317e-02	10.900	< 2e-16 ***
Mask Usage	3.716e-01	6.524e-02	5.697	1.22e-08 ***
Low	3.716e-01	6.524e-02	5.697	1.22e-08 ***
Moderate	1.938e-01	3.655e-02	5.304	1.13e-07 ***
County Enrollment	-1.567e-01	1.398e-01	-1.121	0.26230
Small	-1.567e-01	1.398e-01	-1.121	0.26230
Medium	-5.075e-02	1.403e-01	-0.362	0.71749
Large	2.807e-02	1.392e-01	0.202	0.84020
Median Household Income	1.074e-05	1.543e-06	6.960	3.40e-12 ***

No competing interests reported.

Modeling Community COVID-19 Transmission Risk Associated with U.S. Universities

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Results

Discussion

Conclusions

Methods

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1