key: cord-0726484-lk1w0d3v authors: Jalali, A. M.; Khoury, S. G.; See, J.; Gulsvig, A. M.; Peterson, B. M.; Gunasekera, R. S.; Buzi, G.; Wilson, J.; Galbadage, T. title: Delayed Interventions, Low Compliance, and Health Disparities Amplified the Early Spread of COVID-19 date: 2020-08-04 journal: nan DOI: 10.1101/2020.07.31.20165654 sha: 7a3768d93e8020dfdeb52bca3e74007cdf3c5506 doc_id: 726484 cord_uid: lk1w0d3v The United States (US) public health interventions were rigorous and rapid, yet failed to arrest the spread of the Coronavirus Disease 2019 (COVID-19) pandemic as infections spread throughout the US. Many factors have contributed to the spread of COVID-19, and the success of public health interventions depends on the level of community adherence to preventative measures. Public health professionals must also understand regional demographic variation in health disparities and determinants to target interventions more effectively. In this study, a systematic evaluation of three significant interventions employed in the US, and their effectiveness in slowing the early spread of COVID-19 was conducted. Next, community-level compliance with a state-level stay at home orders was assessed to determine COVID-19 spread behavior. Finally, health disparities that may have contributed to the disproportionate acceleration of early COVID-19 spread between certain counties were characterized. The contribution of these factors for the disproportionate spread of the disease was analyzed using both univariate and multivariate statistical analyses. Results of this investigation show that delayed implementation of public health interventions, a low level of compliance with the stay at home orders, in conjunction with health disparities, significantly contributed to the early spread of the COVID-19 pandemic. Amidst the unprecedented early public health response to the Coronavirus Disease 2019 pandemic, the United States (US) was unable to contain the spread of the Coronavirus. Public health officials implemented rigorous evidence-based nonpharmaceutical interventions (NPIs) originating from measures utilized during the 1918 influenza pandemic to help "flatten the curve" (Ferguson et al., 2006; Bootsma and Ferguson, 2007; Hatchett et al., 2007) . The effectiveness of NPIs implemented was dependent on four factors that included the concurrent use of NPIs, early intervention, duration, and the rigor of the preventative measure (Hatchett et al., 2007) . Seemingly rigorous and early preventative measures that were employed to slow the spread of the COVID-19 have, unfortunately, resulted in subpar and inconsistent outcomes across the US to COVID-19 fatality (Holmes et al., 2020) . Minority groups, including African Americans, Hispanics, and Native Americans, are reported to be more likely to experience socioeconomic disadvantages at some point in their life than non-minority groups (Williams et al., 2010) . In addition, previous studies have shown that low-income groups are at an increased risk for mental illness, chronic diseases, lower life expectancy, and higher mortality (Belle Doucet, 2003; Braveman et al., 2010; Mode et al., 2016) . The Center for Disease Control and Prevention (CDC) has stated that underlying health conditions and comorbid conditions are major risk factors for COVID-19, thereby subjecting individuals from minority populations to an even greater risk (CDC, 2020) . When addressing the spread of COVID-19, it is important to take into account many influencing factors that may lead to the disproportionate spread of the disease. These may include, but not be limited to, the effectiveness of the preventative measure, community-level compliance to stay at home orders, and underlying health disparities. In this study, we characterized these three aspects in detail as we evaluated the early spread of COVID-19 in the US. We compared these factors individually and collectively in the 30 most populous US counties, to identify possible associations with the disproportionate spread of the disease. The COVID-19 county-level number of cases and deaths were obtained from the data provided by the Johns Hopkins University COVID-19 data repository (JHU, 2020) . Case rates, mortality rates, and case-fatality rates were calculated from this dataset. The COVID-19 case rates were defined as the cumulative number of cases per unit county population on a given date. Mortality rates were defined as the cumulative number of deaths per unit county population on a 2), (2) stay at home orders (Supplemental Table 3 and 4), and (3) face mask requirements (Supplemental Figure 5 and 6 ). For each of the counties in this study, we characterized the quality and intensity of their public health interventions until May 10 th , 2020, focusing on the following four specific criteria. (1) The duration of the corresponding intervention was implemented. (2) The number of days the corresponding intervention was delayed before it started and a case rate of 1 per 10,000 was used as a reference start date. (3) The number of COVID-19 cases in each county the day before the start of the corresponding intervention. (4) The COVID-19 case rate in each county, the day before the start of the corresponding public health response. The public health interventions were then compared in the three COVID-19 spread groups (high, mid, and low). We used Community Mobility Reports and the Unacast Social Distancing Scoreboard to assess the publics' compliance with public health interventions enacted by their state or county Unacast, 2020) . The Google dataset provided mobility trends showing percent change in the number of visits over time by geography, across different categories of places. The baseline was six weeks of pre-COVID-19 (before March 2020) using anonymously collected google location history data. Location included (a) retail and recreation, (b) groceries and pharmacies, (c) parks, (d) transit stations, (e) workplaces, and (f) residential. Percent change in the residential category represented a change in duration while all other categories represented a change in the total number of visitors. The Unacast dataset provided mobile device location data. Devices were assigned to counties based on where a specific device was recorded for the longest time on a particular day. The pre-COVID-19 period was defined as four weeks before March 8 th , 2020. Percent changes in the movement are shown in three categories: (g) distance traveled, (h) non-essential points of interest (POIs) visitation, and (i) All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 human encounters. Total overall movement (j) includes a combined score for the distance traveled, POIs visitation, and human encounters. Change in mobility and movement was calculated as the mean of daily percent changes from the start of stay at home orders until May 10 th , 2020, or the end of the stay at home orders (whichever came first), for the corresponding county. For each of the above categories (a-j), we plotted the time series data for percent change of mobility as a 7day rolling average for the high, mid, and low groups. Demographic and social determinants of health data from the 2018 US Census with fiveyear estimates (when available) were obtained for all the counties in this current study (USCB, 2020) . For age and sex data, table S0101 "Age and Sex" were used. Age ranges studied included years 0-19, 20-44, 45-64, >65. We compared the percent population that was ≥ 65 years old in each of the counties. The land area of each county was obtained from table LND01 2011 dataset. The population density was calculated using total county population information and land area for each county in square miles. For race and ethnicity, we used table B03002 "Race and Ethnicity." We compared the percent population that was Caucasian, Hispanic, African American, and Asian. For income and poverty data, table B17002 "Income and Poverty" were used. Only a 1-year estimate was available for the B17002 dataset. This table provides data on the ratio of income to the poverty level in the past twelve months. For this study, we used ratio ranges for < 0.99 (below poverty line) and > 5 times the poverty level (higher income). For data on housing units, table B25001 "Housing Units" was used, and information regarding housing density was collected. Table B25008 , "Rent or Own Houses," was used to gather data on the percent population that rents or owns their homes. For information on household and family size table S1101, "Households and Families" was used. Table B27001 , "Health Insurance," was used to determine the percentage of All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 individuals without health insurance . To determine the level of education attainment table B15003, "Education Attainment 25 years and older," was used. We used the percent population that had at least one year of college education in this study. We used the Johns Hopkins University COVID-19 Resource Center to collect state information on the number of staffed hospital beds and intensive care unit (ICU) beds available in each of the study counties (JHU, 2020) . We compared these health determinants among the three COVID-19 spread groups (high, mid, and low) to identify any health disparities that may have contributed to the disproportionate spread of the Coronavirus in the US. Non-parametric Kruskal-Wallis tests were employed to compare the three COVID-19 spread groups (high, mid, and low) in individual categories. These analyses characterized the associations between the differential spread of COVID-19 in the US and various parameters, including public health interventions, community compliance, and health disparities. Pearson's correlation coefficients and partial correlation coefficients were calculated for all pairs of numeric variables under study. We also performed multiple regression analyses for each of the three categories, (1) public health interventions, (2) community compliance, and (3) health determinants to assess the combined contribution of each of these parameters. To show that there were differences in the early spread of the COVID-19 pandemic across the US, we examined COVID-19 case rates in the 30 most populous US counties. On May 10 th , 2020, we observed case rates of 7.7 to 292.2 per 10,000 county population, even among these most All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint populous US counties (Table 1, Supplemental Table 1 ). We grouped these counties into high, mid, and low case rate groups, with ten counties in each spread group (Table 1) . We then plotted case rates for these counties from March 1 st to May 31 st , 2020, and the mean case rates for the high, mid, and low spread groups (Figure 1 a-d) . We also looked at case rates in two-week intervals from April 12 th to May 24 th , 2020, and noted significant differences between these three groups [p < 0.0001] (Figure 1 e-h). While these were the most populous US counties and were affected very early in the pandemic, the start of the COVID-19 spread had some temporal differences. To account for the delay in the increase of cases among these counties, we used 1 case per 10,000 as a point of reference and followed the increase in case rates for the next 92 days (13 weeks) ( Figure 1 i-l). The increase in the case rates showed different patterns among the three groups. The high spread group showed an initial exponential increase followed by a plateauing of their case rates. The mid group had a somewhat constant increase in case rates, while the low group displayed a slow rise in cases initially followed by a delayed exponential rise in their case rates (Figure 1 i-k). Fourteen-day (2 weeks) intervals showed that there were consistently significant differences in the case rates among these three groups temporally. [p < 0.0001] (Figure 1 m-p) . These results showed that there were significant differences in the spread of the early pandemic in the 30 most populous US counties. Next, we evaluated the mortality rates for the counties in the high, mid, and low spread groups similar to our analyses for case rates. We found that higher spread groups had the highest mortality rates, followed by the mid spread group and the low spread groups (Supplemental Figure 2 ). To examine if there were differences in the number of deaths per case in the high, mid, and low spread groups, we calculated the case-fatality rates in each of the counties on May 10 th , 2020 (Table 1) . Case-fatality rates ranging from 2.2 to 11.9 per 100 cases were observed among these All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10. 1101 counties. To evaluate the differences in these case-fatality rates, we plotted case-fatality rates for these counties from March 1 st to May 31 st , 2020 (Figure 2 a-c). The mean case-fatality rates for the high spread group were significantly greater than the case-fatality rates in the mid and low spread groups (Figure 2 d) . To characterize these differences further, we looked at case-fatality rates in two-week intervals from April 12 th to May 24 th , 2020. We noted a significant difference among these three groups at all-time points [p = 0.0412, p = 0.0117, p = 0.0011, and p = 0.0006 respectively] (Figure 2 e-h). Among all 30 counties, the correlation between case rate and the mortality rate was 0.93, between case rates and case-fatality rates was 0.57, and between mortality rate and case-fatality rate was 0.76. Assessment of the three main interventions implemented across the US at both the state and county levels, including restrictions on mass gatherings, stay at home orders, and facemask requirements, were conducted. This included the duration of the intervention, delay in implementation, number of cases, and the case rate at the start of these actions within the high, mid, and low case rate groups ( (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint longer durations could reflect counties with high case burdens implementing more rigorous measures. The differences in delay in enacting stay at home orders, number of cases and case rate at the start of the intervention suggests that the early implementation of stay at home orders was associated with lower case rates of COVID-19 ( Figure 3 f-h). Duration and delay in the implementation of face mask requirements alone showed no significant difference between the high, mid, and low spread groups [p = 0.3395 and p = 0.2339, respectively] (Figure 3 i,j). The bimodal distribution observed in the duration and delay of the low spread group may, in part, account for this observation. There was a significant difference among the three spread groups for the number of cases and the case rate at the time of implementation of face mask requirements [p = 0.0001 and p < 0.0001 respectively] (Figure 3 k,l). This difference shows that the early introduction of face mask requirements alone was associated with significantly lower case rates. Rigorous public health interventions can only be effective to the degree to which the intended population follows them. Google community mobility reports were utilized to determine if the state and county level public health measures were effectively implemented (Figure 4 a-f). The percent change in mobility in the high, mid, and low spread counties was evaluated to determine the level of compliance with the stay at home orders implemented. We defined an 81 to 100% reduction in mobility as high compliance, a 51 to 80% reduction mobility as moderate compliance, and any present change ≤ 50% as low compliance. We observed the following median (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. To confirm our Google community mobility report results, we compared mobile device location data provided by Unacast, in our study counties (Unacast, 2020) ( data showed no significant difference in the reduction of movement among the high, mid, and low spread groups. Our analyses showed a reduction in movements of 20-to-70.0% across different parameter highlights that there was low compliance to the stay at home orders enforced and supports our conclusions from the Google mobility reports. To further confirm our observations of community-level low compliance with stay at home orders, we used time-series data from both Google mobility reports and Unacast data. We plotted the average change in movements for the high, mid, and low spread groups from March 1 st to May All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint 31 st , 2020 (Figure 5 a-j). Our results show that the categories of retail, grocery, parks, travel, visitations, human encounters, and overall had a low level of compliance almost throughout the stay at home period. This is also compounded by the day and weekend variability of mobility in each of these categories. Parks, in particular, showed an increase in mobility compared to baseline in most high spread counties (Figure 5 c) . A third factor that can contribute to the differences we observe in COVID-19 case rates among the most populous US counties are underlying health disparities. We used US census data to characterize demographic differences, including social determinants of health, to examine associations with the early spread of COVID-19 in the US. We categorized these data into location- These observations show that the early spread of COVID-19 was associated with high population density and housing densities. Given the primary mode of transmission of COVID-19, this is a plausible association . Among our study counties, there was a lower percentage of Hispanics and a higher percentage of African Americans, the Hispanic population, and a higher percentage of African American population in the high spread counties. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint We did not observe significant differences among the high, mid, and low spread groups and the Next, we were interested in determining if county level health disparities observed among the high, mid, and low spread groups had any bearing on the case rates and mortality rates of COVID-19. We used demographically stratified COVID-19 cases and deaths for the five counties making up the New York Borough, to compare to the proportion of each of the county populations in the respective demographic category. When relative ratios of the case rates were compared, we (Figure 7 a, d) . A similar trend was observed for mortality rates. Relative ratios for mortalities were, age ≥ 65 years (95% C.I. 4.71-5.83), male sex (95% C.I. 1.24-1.29), and African American race (95% C.I. 1.06-1.78) were all risk factors for the disease (Figure 7 a-e). No significant differences were observed between the five-borough counties with Hispanic ethnicity (95% C.I. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint 0.74-1.45) and White ethnicity (95% C.I. 0.68-1.32). Next, the association between the level of poverty and COVID-19 spread was evaluated, after which a stepwise increase in the case rate, severe case rate, and mortality rate with the increase in poverty (7 f-h) was observed. These results show that disparities in health determinants, including advanced age, male sex, African American race, and poverty were associated with higher case rates of COVID-19. Next, the association between the level of poverty and COVID-19 spread was evaluated, after which a stepwise increase in the case rate, severe case rate, and mortality rate with the increase in poverty (7 f-h) was observed. These results show that disparities in health determinants, including advanced age, male sex, African American race, and poverty, were associated with higher case rates of COVID-19... Next, the association between the level of poverty and COVID-19 spread was evaluated, after which a stepwise increase in the case rate, severe case rate, and mortality rate with the increase in poverty (7 f-h) was observed. These results show that disparities in health determinants, including advanced age, male sex, African American race, and poverty were associated with higher case rates of COVID-19. Since this study focused on 30 counties/data points, not all variables were analyzed together in multiple regression models. The response variable for the models shown was case rates for May 10 th , 2020. In general, remarkably similar results were obtained using the other dates, mortality rates, case-fatality rates, and the slopes of lines through the case/mortality/fatality rates. Three models are reported below, one for each group of variables studied: interventions, compliance, and health disparities. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint The three explanatory variables for the intervention multiple regression model were mass gathering order duration, stay-at-home order duration, and facemask duration. All pair-wise interactions were included in the model. There were seven coefficients, including intercept, only one with a p-value < 0.05. The model statistics were F(6, 23) = 3.83, p-value = 0.0086, RSE = 71.88, R 2 = 0.50, R 2 adj = 0.37 (Supplemental Table 7 and Supplemental Figure 3 ). This implies that, at a single point in time, the interventions have some explanatory power in the case rates, but it is relatively weak, indicating there are other more significant factors. The six explanatory variables for the compliance multiple regression model were the % change in movement from the start of the stay at home orders to their end (or May 10 th ) for retail, grocery, parks, transit, workplace, and residential. There were seven coefficients, including intercept, none one with p-value < 0.05. The model statistics were F(6, 23) = 3.53, p-value = 0.0126, RSE = 73.33, R 2 = 0.48, R 2 adj = 0.34 (Supplemental Table 8 and Supplemental Figure 4 ). This implies that, at a single point in time, the compliance has some explanatory power in the case rates, but it is relatively weak, indicating that there are more significant factors. The ten explanatory variables for the health disparities multiple regression model were population density, housing density, % age ≥ 65 years, % male sex, % white ethnicity, % Hispanic ethnicity, % African American race, % with income below the poverty line, % with at least one year of college education, and % uninsured. Among multiple interaction assessments, one of the most effective models had 20 coefficients (including the intercept), and 13 of them had p-value < 0.05. The model statistics were F(19, 10) = 40.7, p-value = 0.0000005, RSE = 17.42, R 2 = 0.99, All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint R 2 adj = 0.96 (Supplemental Table 9 and Supplemental Figure 5 ). This implies that, at a single point in time, the health disparities substantially explain the differences in case rates. When the health disparities are combined in a model with the strongest intervention and compliance variables, the intervention and compliance variables are still not statistically significant, while the health disparity variables remain significant. The strongest health disparities, which are the primary drivers of the explanatory power of the model, are population density, % Caucasian ethnicity, % African American race, % with income below the poverty line, and their interactions. In this study, we evaluated the disproportionate early spread of the COVID-19 pandemic in the US, with a particular emphasis on public health interventions implemented, level of compliance to those preventative interventions, and underlying health disparities across the most populous counties. We showed that by May 10 th , 2020, the 30 study counties had vastly different case rates ranging from 7.7 to 292.2 per 10,000 (Table 1, Figure 1 ). Our analyses on the public health interventions showed that the early implementation of stay at home orders and face mask requirements likely resulted in lower COVID-19 cases rates for regions assessed. Better yet, when public health interventions, including restrictions on mass gatherings, stay at home orders, and face mask requirements were implemented concurrently and early, the spread of COVID-19 was significantly lower than regions that did not incorporate these measures. While these interventions were implemented very early on in most US states with the likely exception of New York, by early July 2020, the US started experiencing continued propagation of positive COVID-19 cases that spread across many more states (Jalali et al., 2020) . To help explain this phenomenon and evaluate the robustness of the nationwide stay at home orders, we evaluated the level of compliance observed at a community level. Our results showed that irrespective of the All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint rate of spread of COVID-19, most counties displayed a low level of compliance to the stay at home orders implemented in their state. Categorical locations that displayed the lowest level of compliance with stay at home orders were parks, grocery stores, and pharmacies. The underlying level of non-compliance, in conjunction with relaxations of these public health interventions after one to two months of implementation, likely caused a suppressed level of viral spread followed by a resurgent wave. Regardless, it appears that early interventions reported herein were effective in delaying the inevitability of viral spread, thereby providing tangible evidence to support the concept of lowering the curve. With the differences in public health responses and the varying degree of compliance to the stay at home orders observed, outcomes presented in this article cannot fully explain the vast difference we observed in the early COVID-19 case rates. However, one particular factor that is of importance to the investigators in this article is to highlight that health disparities may likely have played a large role in the disproportionate spread of COVID-19. For example, the emergence and persistence of disparities related to health often manifest through environmental, socioeconomic, or system-level factors that are complex, and may disproportionately affect minority communities (Brown et al., 2019) . Differences observed between populations are closely tied to economic, social, and environmental disadvantages that may hinder a person's ability to achieve optimal health (CDC, 2016; HP2020, 2020). In our evaluation of various health determinants across the most populous US counties, the following list of factors were all associated with amplified case rates of COVID-19 during the early stages of the pandemic. These were: population density, housing density, African American race, education, and percent uninsured. Various social determinants of health including race and ethnicity, access to healthcare, income inequality, housing, and social support have shown to contribute to the increased COVID-All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 19 spread and related mortality (Rollston and Galea, 2020; Turner-Musa et al., 2020; Webb Hooper et al., 2020; Yaya et al., 2020) . African American individuals have also been disproportionately affected by this disease due to the presence of comorbidities associated with worse outcomes, limited COVID-19 awareness, and composing of a significant portion of essential workers in the US (Dorn et al., 2020; Holmes et al., 2020; Tai et al., 2020; Wilder, 2020) (Chowkwanyun and Reed, 2020) . While our investigation was limited to 30 US counties, the outcomes presented herein may serve to aid in the preparation of a framework for future investigations aimed at better serving populations at risk for greater health disparities through public health interventions. We have identified a few limitations with this study. For example, data from the 30 most populous counties were included in this investigation. These data, unfortunately, represent a small segment of the overall population in the US, and thus it would be inappropriate to extrapolate generalizations for all regions in the US. The regions and data utilized also served as a snapshot of the early spreading of COVID-19. The investigators of this study purposefully aimed to reduce All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 the potential introduction of a time-based bias so that public health interventions could be compared uniformly across all counties. Our results have illustrated how early COVID-19 began to spread through populous counties. Second, the mobility data we used to determine compliance to stay at home orders were derived from time-series data. We averaged the daily percent changes to give categorical data for comparison purposes. In doing so, we may likely have overlooked daily changes that may have provided a greater level of insight. However, this is beyond the scope of this study and is a potential future direction of research. Two other factors may have impacted the case rates of some of these study counties. COVID-19 testing and screening were not uniformly administered across these counties that could lead to some bias in the number of reported cases. Besides travel within a county, there is travel across counties that can likely influence the spread of COVID-19. We did not take into account this variable as it is not readily quantifiable, and is also a potential limitation of the study. In our study, we also observed that mortality rates and case-fatality rates were higher in counties with higher case rates (Figures 2 a-d, Supplemental Figure 2 a-d) . The very high case rate to mortality rate correlation (0.93) indicates a very similar disease behavior across counties. In contrast, the substantially lower-case rate to case-fatality rate correlation (0.57) indicates the presence of additional underlying factors that are causing case-fatality rates to fluctuate across counties. Whether this is due to variations in case reporting, the rapid onset of viral transmission, the overburdened medical systems, underlying health disparities, or differences in testing needs to be carefully investigated further. Our multivariate analyses showed that there are indeed other external factors that contribute to the disproportionate case-fatality rates observes across US counties. Overall, our study takes a holistic approach to characterize the early spread of COVID-19 in the US and the multiples factors that contribute to its disproportionate spread. Identifying All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint these factors and addressing them can help mobilize the appropriate resources to implement targeted preventative measures that help promote health equality and reduce the spread of COVID- The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. AJ characterized the public health response, helped analyze data, and drafted the article. SK, JS, and AG assisted in analyzing data and helped with writing. BP and RG helped with writing and editing. GB helped with computational modeling of the data. JW performed statistical analyses and helped with writing and editing. TG led the study, helped analyze data, performed statistical analyses, prepared figures, and helped with writing and editing. No external funds were used for this study. We thank Dr. Jeffrey S. Wang, Infectious Disease Specialist at Kaiser Permanente, Anaheim, California, for his clinical insights. The Supplementary Material for this article can be found online at: (Link) All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint Tables Table 1. COVID-19 Case rates and case-fatality rates of the 30 most populous counties in the United States on May 10 th , 2020. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. The cumulative number of cases per 10,000 of the respective county population on May 10 th , 2020. 5 The cumulative number of deaths per 100 cases in the respective county on May 10 th , 2020. 6 High -a case rate of > 100 per 10,000. 7 Mid -a case rate of 15 to 100 per 10,000. 8 Low -a case rate of < 15 per 10,000. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 Figure 1. COVID-19 case rates in the 30 most populous US counties. COVID-19 case rates defined as the cumulative number of cases per unit county population. The 30 most populous US counties divided into three groups (high, mid, and low) based on their COVID-19 case rates on May 10 th , 2020. The high group with case rates of > 100 per 10,000, the mid group with case rates All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 of 15 to 100 per 10,000 and low group with case rates < 15 per 10,000. The counties included in the high, mid, and low groups are listed in Table 1 (56) days post reaching a case rate of 1 per 10,000. A non-parametric Kruskal-Wallis test was performed to compare the three groups (high, mid, and low), and the p-values are indicated above each of the corresponding results. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020 . . https://doi.org/10.1101 (b, f, j) The number of days the corresponding intervention was delayed before it started. A case All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint rate of 1 per 10,000 was used as a reference start date. (c, g, k) The number of COVID-19 cases in each county the day before the start of the corresponding intervention. (d, h, l) The COVID-19 case rate in each county, the day before the start of the corresponding intervention. A nonparametric Kruskal-Wallis test was performed to compare the three groups (high, mid, and low), and the p-values are indicated above each of the corresponding results. implemented. Change in mobility and movement shown as the mean of daily percent changes from the start of stay at home orders until May 10 th , 2020, or the end of the stay at home orders (whichever came first), for the corresponding county. Median and 95% confidence interval (95% CI) are presented. The 30 most populous US counties divided into three groups (high, mid, and low) based on their COVID-19 case rates on May 10th, 2020. High group with case rates of > 100 per 10,000, mid group with case rates of 15 to 100 per 10,000, and low group with case rates < 15 per 10,000. (a-f) Google community mobility trends showing percent change in movement over All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint time by geography, across different categories of places. The baseline was six weeks of pre-COVID-19 (before March 2020) using anonymously collected google location history data . Location included (a) retail and recreation, (b) groceries and pharmacies, (c) parks, (d) transit stations, (e) workplaces, and (f) residential. Percent change in Residential was calculated by the duration, while the other categories (a-e) were by the number of visits. (g-j) Unacast mobile device location data (Unacast, 2020) . Devices were assigned to counties based on where a specific device was recorded for the longest time on a particular day. The pre-COVID-19 period was defined as four weeks before March 8 th , 2020. Percent changes in the movement are shown in four categories, (g) distance traveled, (H) non-essential points of interest (POIs) visitation, and (i) human encounters. (j) Total overall movement includes a combined score for the distance traveled, POIs visitation, and human encounters. A non-parametric Kruskal-Wallis test was performed to compare the three groups (high, mid, and low), and the p-values are indicated above each of the corresponding results. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint (Unacast, 2020) . Percent changes in the movement are shown in four categories, (g) distance traveled, (H) non-essential points of interest (POIs) visitation, and (i) human encounters. (j) Total overall movement includes a combined score for the distance traveled, POIs visitation, and human encounters. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10.1101/2020.07.31.20165654 doi: medRxiv preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10. 1101 cases and deaths in the New York Borough, stratified according poverty level at the zip code areas. Low <10% of residents living below the poverty line. Medium 10% to <20% of residents living below the poverty line. High 20% to <30% of residents living below the poverty line. Very high ≥30% of residents living below the poverty line. (f) Cases rates. (g) Severe case rates. (h) Mortality rates. affected the early dissemination of the COVID-19 pandemic in the US were public health interventions, level of compliance to stay at home orders and underlying health disparities. Early and rigorous public health interventions helped slow the spread of the Coronavirus. Low level of compliance (<50% reduction in community mobility) during the period stay at home orders were in place helped continually spread the virus. Underlying health disparities observed across the US caused disproportionate early spread of the COVID-19 pandemic. All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted August 4, 2020. . https://doi.org/10. 1101 Disparities In Outcomes Among COVID-19 Patients In A Large Health Care System In California Poverty, Inequality, And Discrimination As Sources Of Depression Among U.S. Women The effect of public health measures on the 1918 influenza pandemic in U.S. cities Socioeconomic disparities in health in the United States: what the patterns tell us Structural Interventions to Reduce and Eliminate Health Disparities Strategies for Reducing Health Disparities People with Certain Medical Conditions Racial Health Disparities and Covid-19 -Caution and Context COVID-19 exacerbating inequalities in the US Strategies for mitigating an influenza pandemic Does COVID-19 Spread Through Droplets Alone? Front Public Health 8 See how your community is moving around differently due to COVID-19 and Case Fatality in the United States: Translational Epidemiologic Perspective and Challenges Department of Health and Human Services, Office of Disease Prevention and Health Promotion Early COVID-19 Interventions Failed to Replicate St. Louis vs. Philadelphia Outcomes in the United States. medRxiv, 2020 COVID-19 Data Repository by the Race, Neighborhood Economic Status, Income Inequality and Mortality COVID-19: Data Hospitalization and Mortality among Black Patients and White Patients with Covid-19 COVID-19 and the Social Determinants of Health Public Health Response to the Initiation and Spread of Pandemic COVID-19 in the United States The Disproportionate Impact of COVID-19 on Racial and Ethnic Minorities in the United States Examining Social Determinants of Health, Stigma, and COVID-19 Disparities Unacast Social Distancing Scoreboard