Determinants of COVID-19 Incidence and Mortality in the US: Spatial Analysis

The US continues to account for the highest proportion of the global coronavirus disease-2019 (COVID-19) cases and deaths. Amid the second wave, it is important to contextualize the spread and success of mitigation efforts. The objective of this study was to assess the ecological determinants (policy, health behaviors, socio-economic, physical environment and clinical care) of COVID-19 incidence and mortality. Data from the New York Times COVID-19 repository (01/21/2020-10/27/2020), 2020 County Health Rankings, 2016 County Presidential Election Returns, and 2018-2019 Area Health Resource File were used. County-level logged incidence and mortality rate/million were modelled using Spatial Autoregressive Combined model and spatial. Counties with higher proportions of republican voters, and racial minorities (African Americans, Native Americans and Hispanics), those not proficient in English, and higher population density, pollution-particulate matter, residential segregation between non-Whites & Whites were associated with high incidence rates. Subsequently, counties with higher republican voters, excessive drinkers, children in single-parent households, uninsured adults, racial minorities, females, and high population density, pollution-particulate matter, and residential segregation between non-Whites & Whites was associated with high COVID-19 mortality rates. The study spatial models identified length of order, population density, income, and uninsurance rate and race/ethnicity as some important determinants of the geographic disparities.


INTRODUCTION
regression. The final list of county-level covariates included in the model is described in duration, and month (January/February/March) of the first reported case at the state-level were included 1 2 5 as covariates. Finally, the county-level composition of the Republican voters was also added to 1 2 6 determine the extent to which county's community health was associated with voting preference for the 1 2 7 2016 presidential elections. Descriptive univariate statistics of the weighted county-level characteristics. The presence of 1 3 0 spatial correlation was confirmed by performing Moran's I test for correlation of ordinary least square 1 3 1 regression residuals. Due to spatial correlation in the data, the current study employed spatial regression 1 3 2 analysis approach. Two island counties were excluded from the spatial regression analysis. Prior to evaluate the association between period prevalence of COVID-19 and county-level characteristics 1 3 5 [14]. The SAC model was also adopted for modeling cumulative incidence and deaths if both the spatial 1 3 6 lag parameter (rho) and spatial error parameter (lambda) were statistically significant. However, model 1 3 7 simplification was attempted when one of the parameters was not significant. A first order queen spatial 1 3 8 weight matrix was employed for all spatial models. The queen matrix defines neighbor relationships if 1 3 9 the counties either share a border or a vertex. All analysis was performed in SAS Studio University

4 4
The final analysis included data from 3,101 counties from the mainland US. Between January 20 1 4 5 to October 27, 2020, the population-weighted cumulative incidence and mortality rates for the mainland 1 4 6 US were 26,576 and 632 per million, respectively. Republican (50% of the votes in the county were for the Republican presidential candidate in 2016). The 1 5 1 weighted proportion of adult smokers, adults with obesity, adults with physical inactivity, and Medicare 1 5 2 enrollees that were administered influenza vaccines and adults with some college education were 15%, 1 5 3 29%, 23%, 19%, and 65%, respectively. Premature (<75 years) age-adjusted mortality was 344 per 1 5 4 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted December 4, 2020. ; 100,000. Among socio-economic factors, unemployment was 4%, 12% of adults were uninsured, the 1 5 5 mean income inequality ratio was 5. About 12% were African Americans, 18% Hispanics, 1% Native 1 5 6 Americans and 4% population not proficient in English. Mean percentage of the population older than 1 5 7 65 years and less than 18 years were 16% and 23%, respectively. On average, 19% counties were rural, 1 5 8 homeownership rate was 65%, 33% of the children lived in single-parent households and 18% of 1 5 9 households had severe housing problems. At the county-level, the rate of primary care physicians 1 6 0 (logged) and preventable hospitalization (logged) was 4 and 8, respectively. Importantly, there was no 1 6 1 multicollinearity-related issue in our analysis as VIF for the selected variables was less than 7. Spokane areas had clusters of high incidence rates. In the Midwestern region, Tribal lands and the Great had clusters with high incidence rates. In the South, in the Delta region along the Mississippi river incidence rates. The Corpus Christie area and along the US-Mexico border in Texas had clusters of high 1 7 0 incidence rates. Additionally, the high incidence rate clusters were found in the Tallahassee, clusters of high mortality rates were found from the greater Philadelphia area to Boston in 1 7 5 Massachusetts, including the Tristate area. Contrarily, the mortality rate clusters in Wisconsin's Superior Upland area were smaller.  Table 2 presents the results from spatial regression models that assessed the impact of ecological 1 7 8 determinants on COVID-19 incidence and mortality rates (logged). The rho and lambda model 1 7 9 parameters both were significant for the incidence rate SAC regression. However, only the rho 1 8 0 parameter was significant for the mortality rate regression model. Therefore, the SAC model was used 1 8 1 for incidence rate analysis, while the spatial lag model was used for mortality rate analysis. The analysis 1 8 2 further tested the spatial correlation for the residuals of both models using Moran's I statistic. The Moran's I for the model residuals was not significant for incidence rate (SAC model) and mortality rate 1 8 4 (spatial lag model) regression, which indicated no significant spatial autocorrelation of residuals. The 1 8 5 length of lockdown order, a policy-related factor, was significantly associated with incidence rate.
CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10. 1101/2020 Compared to the counties that had no orders implemented, counties with an order length of 1-27 days, 1 8 7 36-58 days and 59 or more days had a 0.290, 0.228 and 0.335 unit decrease in logged COVID-19-related increase in logged incidence rates, respectively. Conversely, 1% increase in 65 years or older and those 1 9 8 less than 18 years of age was associated with 0.020 and 0.011 decrease in logged incidence rates. While 1 9 9 a unit increase in black and white residential segregation , and a percentage point increase in rurality was 2 0 0 associated with a 0.002 and 0.001 unit decrease in logged incidence rates, on the other hand a unit 2 0 1 increase in white and non-white residential segregation and air pollution-particulate matter were 2 0 2 associated with an 0.003 and 0.062-unit increase in these rates. in logged COVID-19-related mortality rates, respectively. Compared to counties from states that first 2 0 6 reported COVID-19 cases in January, those with the first reported case in February were associated with 2 0 7 0.628-unit decrease in logged mortality rates. Among health behaviors, 1% increase in excessive logged mortality rates, respectively. However, 1% increase in adults with some college education was is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20242685 doi: medRxiv preprint logged mortality rates. Whereas residential segregation-African Americans/White, and rurality were 2 1 9 negatively associated with mortality rates by 0.010 and 0.014 units. To the best of our knowledge, this is the first spatial analysis study that captured and assessed the 2 2 2 cumulative incidence and deaths during the majority of the year 2020 (January 21 to October 28, 2020) Republican voters, social association rates, racial minorities (African Americans, Native Americans and 2 2 5 Hispanics), those not proficient in English and counties with higher residential segregation between non-2 2 6 Whites & Whites, population density, pollution-particulate matter were associated with higher incidence 2 2 7 rates. However, counties with longer length of stay-at-home orders, higher proportion of adults with 2 2 8 some college education, high-income, elderly, children, rurality and higher segregation between African 2 2 9 Americans & Whites had lower incidence rates. Correspondingly, counties with higher Republican 2 3 0 voters, excessive drinkers, children in single-parent households, uninsured adults, racial minorities 2 3 1 (African Americans, Native Americans and Hispanics), females, and higher population density, 2 3 2 pollution-particulate matter, and residential segregation between non-Whites & Whites had higher 2 3 3 COVID-19 mortality rate. population density, uninsured rate to be associated with increased susceptibility to COVID-19 outcomes 2 3 7 [12]. Khazanchi et al. and Nayak et al. also reported a similar association between higher county-level 2 3 8 susceptibility score and higher incidence of COVID-19 incidence and deaths [11,22]. Similar to our Americans, Hispanics and African Americans) [11,14,24]. Moreover, with increase in population 2 4 5 density, incidence and mortality rates increased, which has been reported by Sun et al. Unlike this study, Mollalo et al. spatial analysis found that increase in household income was associated with increased 2 4 7 incidence rate [25]. However, that same study reported that increase in providers (example: nurse al. also have pointed out 'partisanship' as a risk factor for non-adherence to preventive guidelines and 2 5 0 . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted December 4, 2020. ; The study received no funding. The authors declare no competing interests. All authors have consented for this manuscript to be published.

7.
Pfizer's covid vaccine is more than 90 percent effective in first analysis, company reports -The file.html). Accessed 27 October 2020. Journal. 2020. Report No.: 2020-116. . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted December 4, 2020. ; https://doi.org/10.1101/2020.12.02.20242685 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted December 4, 2020.  is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted December 4, 2020.

(which was not certified by peer review)
The copyright holder for this preprint this version posted December 4, 2020.
; https://doi.org/10.1101/2020.12.02.20242685 doi: medRxiv preprint . CC-BY-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted December 4, 2020. ;