Skip to main content
NCBI home page
PLOS One logoLink to PLOS One
. 2020 Oct 14;15(10):e0240151. doi: 10.1371/journal.pone.0240151

Social determinants of COVID-19 mortality at the county level

Rebecca K Fielding-Miller 1,*, Maria E Sundaram 2, Kimberly Brouwer 1
Editor: Nickolas D Zaller3
PMCID: PMC7556498  PMID: 33052932

Abstract

As of August 2020, the United States is the global epicenter of the COVID-19 pandemic. Emerging data suggests that “essential” workers, who are disproportionately more likely to be racial/ethnic minorities and immigrants, bear a disproportionate degree of risk. We used publicly available data to build a series of spatial autoregressive models assessing county level associations between COVID-19 mortality and (1) percentage of individuals engaged in farm work, (2) percentage of households without a fluent, adult English-speaker, (3) percentage of uninsured individuals under the age of 65, and (4) percentage of individuals living at or below the federal poverty line. We further adjusted these models for total population, population density, and number of days since the first reported case in a given county. We found that across all counties that had reported a case of COVID-19 as of July 12, 2020 (n = 3024), a higher percentage of farmworkers, a higher percentage of residents living in poverty, higher density, higher population, and a higher percentage of residents over the age of 65 were all independently and significantly associated with a higher number of deaths in a county. In urban counties (n = 115), a higher percentage of farmworkers, higher density, and larger population were all associated with a higher number of deaths, while lower rates of insurance coverage in a county was independently associated with fewer deaths. In non-urban counties (n = 2909), these same patterns held true, with higher percentages of residents living in poverty and senior residents also significantly associated with more deaths. Taken together, our findings suggest that farm workers may face unique risks of contracting and dying from COVID-19, and that these risks are independent of poverty, insurance, or linguistic accessibility of COVID-19 health campaigns.

Introduction

A novel coronavirus, SARS-CoV-2, is causing a global pandemic of COVID-19 respiratory disease. This pandemic has resulted in nearly 22 million cases and over 800,000 deaths since early January [1]. As of August 17 2020, the United States has more cases than any other nation in the world, with just over 5.4 million cases and 170,000 deaths [1]. Preliminary data indicates that existing health inequities in the United States are likely linked to COVID-19 morbidity and mortality [2].

Both infectious and non-communicable disease impact marginalized populations at disproportionate rates. While individual-level data is not currently available at the national level, data from county and state level entities suggest that COVID-19 may follow similar patterns. In the state of California, Latinos make up approximately 39% of the total population but represent just over 53% of total cases [3]. In New York City, Black/African American and Hispanic residents have significantly higher rates of COVID-19 illness and mortality than white residents, with a nearly doubled risk of mortality for Black/African American residents compared to white residents [4]. Journalistic reportings and early analyses suggest that unsafe working conditions among essential workers—who are more likely to be immigrants and/or racial/ethnic minorities [5]—and concerns about immigration status may be responsible for underlining existing health disparities among racial and ethnic immigrants across the United States, as well as potential language barriers, poverty, and a lack of insurance [6, 7].

We sought to assess the associations between COVID-19 mortality and farm worker, immigrant, and uninsured populations at the county level. We hypothesized that counties with a higher percentage of farm workers would report more deaths due to COVID-19, adjusting for poverty, insurance rates, population, age, and density.

Methods

We built a series of spatial autoregressive models to assess county-level associations between the number of reported COVID-19 deaths in a county and the percentage of individuals engaged in hired farm work [8] in the county as of 2018. To account for potential confounders, we adjusted our analyses for the percentage of Non-English speaking households (defined as households in which no one 14 years or older reports speaking English at least “very well”), the percentage of uninsured individuals under the age of 65, percentage of individuals living at or below the poverty line, percentage of residents age 65 or older, county population, and county density, measured as the number of residents per square mile.

COVID-19 mortality data was sourced from county public health agencies, aggregated and made publicly available by the New York Times (NYT) [9]. For these analyses we used the NYT’s “historical” dataset, with counts updated once per day with the final count of deaths as of that day [10]. Per the NYT:

The data is the product of dozens of journalists working across several time zones to monitor news conferences, analyze data releases and seek clarification from public officials on how they categorize cases…At times, cases have disappeared from a local government database, or officials have moved a patient first identified in one state or county to another, often with no explanation. In those instances, which have become more common as the number of cases has grown, our team has made every effort to update the data to reflect the most current, accurate information while ensuring that every known case is counted.

Further details on the NYT’s methodology as well as the full dataset is available at https://github.com/nytimes/covid-19-data. The proportion of households with limited English speaking ability was drawn from the American Community Survey’s (ACS) 2014 5-year estimate [11]; percentages of individuals living below poverty and percentage of residents over the age of 65 were from 2017 ACS data [12, 13]. The percentage of farmworkers was taken from the US Bureau of Economic Analysis [14]. Percent uninsured was based on the US Census Small Area Health Insurance Estimates (SAHIE) program’s 2018 estimates. Density was measured as the number of individuals per square mile, based on US census data. In addition to hypothesized predictors and potential confounders, we adjusted our models to account for the stage of the local epidemic by including a variable for the number of days since the first case of COVID-19 was reported in a given county.

Counties with 1000 residents or more per square mile were coded as urban; counties with less than 1000 residents per square mile were coded as non-urban. While there are many ways to classify counties, we chose to use 1000 people per square mile for two reasons. First, the US census uses this threshold to designate census blocks as urban vs. non-urban [15]. Second, we felt that doing so allowed us to more clearly delineate major metropolitan areas and their associated resources and public health infrastructures from neighboring suburban or exurban counties.

All analyses were completed in Stata 16.0 (College Station, TX). We first built a series of simple linear regression models to assess the bivariate association between number of deaths within a county and our hypothesized predictors. We then constructed a spatial contiguity matrix and checked the assumption that residuals were distributed spatially using a Moran’s I test.

We next built three separate spatial autoregressive models to assess the association between number of deaths and our hypothesized social determinants, adjusting for potential confounders, and fit the model with a spatial lag of the dependent variable based on our contiguity matrix. Our first model assessed relationships across all counties. We then stratified our analyses to measure the association between mortality and our hypothesized predictors in urban and non-urban counties. Our spatial autoregressive model used the generalized spatial two-stage least squares estimator, specifically Stata’s spregress command with the “g2sl” option. G2sl is a generalized method of moment (GMM) modeling approach. The GMM makes no assumptions about variable distribution and is an appropriate statistical approach to address data that are highly skewed or have unknown distributions [16].

After assessing nation-wide trends, we conducted series of sub-analyses by US Census region within the contiguous 48 states, again stratifying by urban and non-urban counties. We first assessed the association between absolute number of deaths and the same predictors assessed in the nation-wide models. We then fit an additional series of models with the number of deaths per 100,000 residents as our primary outcome of interest, rather than the number of deaths in a county.

Results

This analysis encompassed 3024 counties from all 50 states. As of July 12, 2020, the number of deaths reported in the New York Times’ aggregated dataset ranged from 0 to 22,755 per county, with a median of 2 (IQR: 0–11) (Fig 1) Within this dataset, the 5 boroughs/counties of New York are treated as a single entity. We have done the same in these analyses, assigning all 5 counties the values associated with New York County. We classified 115 counties as urban and 2909 counties as non-urban. Deaths in urban counties ranged from 4–22,755, with a median of 272 and an IQR of 90–774. Deaths in non-urban counties ranged from 0–1,133 with a median of 2 and an IQR of 0–9 (Table 1). The number of deaths per 100,000 residents ranged from 0 to 350, with an overall median of 6 (IQR: 0–11) (Fig 2). The number of deaths per 100,000 was substantially higher in urban than non-urban counties. In urban counties the median was 39.1 per 100,000 (IQR: 15.0–107.3) while in non-urban counties it was 5.6 (IQR: 0–19.7). The geographic distributions of our primary predictors of interest are shown in Fig 3.

Fig 1.

Fig 1

Open in a new tab

Table 1. Primary predictor and covariates of interest across all counties and stratified by urban and non-urban.

All counties (n = 3024) Non-urban counties (n = 2909) Urban counties (n = 115)
median IQR median IQR median IQR
Number of Deaths 2 0–11 2 0–9 272 90.0–774.0
Deaths per 100,000 6 0–22 6 0–20 39 15–107
% Farm workers 2.3 0.9–4.9 2.4 1.1–5.0 0.1 0.0–0.1
% Non-English speakers 4.9 2.8.– 10.1 4.7 2.8–9.4 19.0 11.4–29.8
% Residents uninsured 10.6 7.4–14.6 10.7 7.5–14.6 8.3 5.8–12.6
% Residents in poverty 15.1 11.4–19.4 15.1 11.4–19.6 13.0 8.9–16.7
% Residents Over 65 16.4 13.9–19.0 16.6 14.2–19.1 12.5 11.0–14.5
Residents per square mile 45.5 17.9–111.9 42.7 17.0–98.6 1754.9 1313.4–2715.3
Population (thousands) 26.5 11.7–68.7 25.0 11.2–60.1 735.3 492.3–999.0
Open in a new tab

Fig 2.

Fig 2

Open in a new tab

Fig 3. Percentage of farmworkers, non-English speaking households, uninsured residents over 65, and residents living in poverty, by county.

Fig 3

Open in a new tab

In our fully adjusted model with all 3024 US counties that had reported at least one case of COVID-19 as of July 12, 2020, the percentage of farm workers in a county, percentage of residents living at or below the federal poverty line, number of residents per square mile (ie, population density), and the percentage of residents over the age of 65 were all significantly associated with a higher number of reported COVID-19 deaths (Table 2). Each additional percentage point of farmworkers in a county was associated with 5.79 more deaths (5.51 directly, 0.28 via indirect ‘spillover’ to the next county, p <0.001), while each additional percentage point of individuals living in poverty was associated with 4.41 additional deaths (4.20 directly, 0.22 indirect, p <0.001). The percentage of residents over 65, number of residents per square mile, and county population were all also significantly associated with more deaths.

Table 2. Full spatial regression models: Absolute number of deaths associated with each covariate for all counties and stratified by urban/rural.

All counties (n = 3024) Non-urban counties (n = 2909) Urban counties (n = 115)
b-direct* b-indirect b-total p-value b-direct b-indirect b-total p-value b-direct b-indirect b-total p-value
% Farm workers 5.51 0.28 5.79 0.001 0.64 0.06 0.70 0.005 2280.53 258.98 2539.51 0.03
% Non-English speakers -2.77 -0.14 -2.92 0.83 0.09 0.01 0.10 0.24 -13.18 -1.50 -14.67 0.24
% Residents uninsured 0.16 0.01 0.17 0.25 -0.41 -0.04 -0.45 0.15 -66.27 -7.53 -73.80 0.029
% Residents in poverty 4.20 0.22 4.41 <0.001 0.44 0.04 0.49 0.04 4.47 0.51 4.98 0.86
% Residents Over 65 4.36 0.22 4.58 0.002 0.67 0.07 0.73 <0.001 0.30 0.03 0.34 <0.001
Residents per square mile 0.24 0.01 0.25 <0.001 0.08 0.01 0.08 <0.001 -15.76 -1.79 -17.55 0.75
Population (thousands) 0.62 0.03 0.65 <0.001 0.23 0.02 0.25 <0.001 1.22 0.14 1.36 <0.001
Open in a new tab

*b-direct can be interpreted as the number of deaths associated with a given coviariate within a given county, b-indirect accounts for the spillover spatial effects on neighboring counties, and b-total can be interpreted as the total number of deaths associated with a covariate in a given county plus neighboring counties.

In urban counties (n = 115), the percentage of farmworkers and uninsured individuals were both significantly associated with more deaths, as was the percentage of residents over 65 and county population. Contrary to our initial hypotheses, in urban counties the percentage of uninsured individuals was associated with lower reported COVID-19 mortality. In these 115 counties, each percentage point decrease in the number of uninsured individuals was associated with 73.8 fewer reported COVID-19 deaths (p = 0.03). In rural areas, each increase in the percentage of uninsured individuals was associated with a direct effect of 0.69 fewer deaths within the county (p <0.001) and 0.19 fewer deaths in neighboring counties (p = 0.01). In non-urban counties (n = 2909), each percentage point increase in the number of farmworkers in a county was associated with 0.70 additional deaths (0.6 directly, 0.06 indirectly, p = 0.01). While neither the percentage of non-English speaking households nor the percentage of uninsured residents was associated with a higher number of reported deaths, each percentage point increase in poverty was associated with 0.49 additional deaths (0.4 directly, 0.04 indirectly, p = 0.04).

In sub-analyses by US Census region, distinct spatial patterns emerged (Fig 4). The percentage of farmworkers in a county was not significantly associated with more deaths in West South Central, Mountain, and Pacific states. A higher percentage of non-English speaking households was associated with a higher number of deaths in the Mid-Atlantic and West North Central regions. However, the same variable had a negative association with the number of COVID-19 deaths reported across all counties in the East North Central region. A higher percentage of uninsured individuals over the age of 65 is associated with a higher number of deaths in the East North Central, and Pacific region; however, the relationship is not significant in non-urban counties in the Pacific region, and the relationship is reversed in New England, the West North Central, and West South Central states, where a higher percentage of uninsured individuals is associated with fewer reported COVID-19 deaths.

Fig 4. Absolute number of deaths associated with covariates by US census region in all counties and non-urban counties.

Fig 4

Open in a new tab

When we assessed the rates of death per 100,000 individuals across the 9 census regions, we found similar risk patterns with one main exception: the percentage of farmworkers in a county was not significantly associated with number of deaths per 100,000 individuals in any region (Fig 5). Instead, the percentage of non-English speaking households in a county was significantly associated with higher rates of death across all counties in New England (total b = 2.9, p = 0.02), the Mid-Atlantic (total b = 4.0, p < 0.001), and West North Central states (total b = 1.5, p < 0.001) and in non-urban counties in Mountain states (b = 1.20, p = 0.02). The percentage of uninsured individuals was associated with fewer reported COVID-19 deaths per 100,000 residents across all counties in New England (b = -3.6, p = 0.04) and in non-urban New England counties (b = -3.1, p = 0.02), but with higher rates across all counties in the Mid-Atlantic (b = 6.8, p = 0.01). Poverty was associated with 4.2 fewer reported deaths per 100,000 residents across all Mid-Atlantic counties (p <0.001) and 3.7 fewer reported COVID-19 deaths per 100,000 residents in non-urban Mid-Atlantic counties (p = 0.002), but with 4 more reported deaths per 100,000 residents in all counties in the East South Central region (p = 0.02) and 3.7 more reported deaths per 100,000 residents in non-urban counties in the same region (p = 0.03).

Fig 5. Number of increased deaths per 100,000 residents associated with covariates in urban and non-urban counties by census region.

Fig 5

Open in a new tab

Discussion

We used spatial autoregression models to assess the role of select social determinants of health as risk factors and drivers of the COVID-19 pandemic across the United States and by region. Our findings highlight the fact that the US COVID-19 epidemic is better conceptualized as multiple simultaneous outbreaks across geographic regions rather than a single homogenous outbreak. The wide variety of responses and policy environments across states and regions has lead to corresponding variation in the role of socioeconomic factors such as insurance rates, poverty, and immigration status, as measured by the percentage of non-English speaking households.

The negative associations we found between poverty and mortality rates in the Mid-Atlantic and uninsured residents and mortality in New England are concerning. The CDC has noted higher than expected numbers of death across the United States in recent months, suggesting that COVID-19 mortality is potentially higher than what has thus far been captured by state and county level surveillance [17]. It is possible that these associations represent gaps in testing and linkage to care among immigrants and the uninsured, and/or a gap in ascertaining deaths due to COVID-19 among these same individuals.

The percentage of farmworkers in a county appears to be independently associated with a higher number of deaths, with the specific relationship varying by region. Across all 9 census regions, there was no association between the percentage of farmworkers and mortality when we used the number of deaths per 100,000 as our primary outcome, rather than the absolute number of deaths. Taken together, these two findings suggest that farmworkers may face unique risks of COVID-19 beyond issues of language, insurance, or economics, and that while their unique risk can be seen in higher absolute numbers of deaths, more farm workers in a county is not a major contributor to epidemic spread. Our finding that a higher percentage of non-English speaking households in a county is associated with a higher rate of deaths per 100,000 individuals does suggest that individuals who do not speak English may be at particularly high risk.

Farm labor is considered essential work, but there are reports of inadequate protections for this group of people, including inadequate personal protective equipment and inadequate social distancing guidelines, as well as a lack of enforcement [6, 18]. In addition to the risk to individual farmworkers, it is important to note that widespread outbreaks of COVID-19 among farmworkers also has the potential to impact food systems across the US. Although we cannot draw conclusions about individual risk profiles, our findings do suggest that farm work may create unique risk factors and that farmworkers may require additional protections, such as personal protective equipment and/or targeted outreach. Immigrants provide approximately 75% of all farm labor in the United States [8]. Among those engaged in crop work specifically, nearly three quarters are migrant workers, meaning that they travel from farm to farm during different growing seasons; and approximately half have undocumented citizenship status [8]. Undocumented status may impede an individual’s willingness or ability to seek healthcare, or their ability to request additional protections from an employer if they worry doing so could result in their own deportation or that of a family member [19].

Conclusion

COVID-19 mortality appears to be statistically significantly associated with social determinants of health at the county level, and these relationships may be more pronounced in non-urban counties. Individuals who do not speak English, individuals engaged in farm work, and individuals living in poverty may be at heightened risk for COVID-19 mortality in non-urban counties.

Supporting information

S1 File

(DOCX)

Click here for additional data file. (17KB, docx)

Data Availability

Analyses used publicly available data which are cited appropriately.

Funding Statement

RFM was provided support by the National Institute of Mental Health, K01MH112436 and a National Institute of Minority Health and Health Disparities Loan Repayment Contract. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

  • 1.Dong E, Du H, Gardner L. An interactive web-based dashboard to track COVID-19 in real time. The Lancet Infectious Diseases. 2020;20(5):533–4. 10.1016/S1473-3099(20)30120-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Dorn AV, Cooney RE, Sabin ML. COVID-19 exacerbating inequalities in the US. The Lancet. 2020;395(10232):1243–4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Who is getting infected with COVID-10: State of California; 2020 [updated May 17, 2020. https://update.covid19.ca.gov/.
  • 4.Rates of cases, hospitalizations, and deaths by race/ethnicity group. New York City Health Department; 2020 May 14, 2020.
  • 5.Bier D. Immigrants are about 1/3 of California’s “essential workers”. The Cato Institute; 2020 March 30, 2020
  • 6.Borunda A. Farmworkers risk coronavirus infection to keep the U.S. fed. National Geographic. 2020 April 10, 2020.
  • 7.Data show COVID-19 is hitting essential workers and people of color hardest [press release]. ACLU of Massachusetts, April 8, 2020 2020.
  • 8.Farm Labor: USDA; 2020 [updated April 22, 2020. https://www.ers.usda.gov/topics/farm-economy/farm-labor/-size.
  • 9.Data from The New York Times, based on reports from state and local health agencies. The New York Times. New York, NY2020.
  • 10.US Counties Historical Data. New York Times: github; 2020.
  • 11.Bureau USC. Estimate—PERCENT IMPUTED—Ability To Speak English—No One Age 14 And Over Speaks English Only Or Speaks English "Very Well". In: Estimate A-Y, editor. American Community Survey2014.
  • 12.Bureau USC. Population 65 years and over in the United States. In: Table A-YES, editor. American Community Survey: U.S. Census Bureah; 2017.
  • 13.Bureau USC. Poverty Status in the Past 12 months. American Community Survey 1-year estimate2017.
  • 14.Analysis BoE. Employment by County, Metro, and Other Areas Washington, D.C.2020 [https://www.bea.gov/data/employment/employment-county-metro-and-other-areas.
  • 15.Ratcliffe M, Burd, C., Holder, K., Fields, A. Defining Rural at the U.S. Census Bureau. In: Bureau USC, editor. Washington, DC2016.
  • 16.Chaussé Pierre. “Computing generalized method of moments and generalized empirical likelihood with R.” Journal of Statistical Software 3411 (2010): 1–35. [Google Scholar]
  • 17.Excess Deaths Associated with COVID-19: Centers for Disease Control; 2020 [updated April 29, 2020. https://www.cdc.gov/nchs/nvss/vsrr/covid19/excess_deaths.htm.
  • 18.Altan D, Cediel A. COVID’s Hidden Toll. Frontline. United States: PBS; 2020. p. 54:26. [Google Scholar]
  • 19.Philbin MM, Flake M, Hatzenbuehler ML, Hirsch JS. State-level immigration and immigrant-focused policies as drivers of Latino health disparities in the United States. Social Science & Medicine. 2018;199:29–38. [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision Letter 0

Nickolas D Zaller

30 Jun 2020

PONE-D-20-18322

Social determinants of COVID-19 mortality at the county level

PLOS ONE

Dear Dr. Fielding-Miller,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by July 20, 2020. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

  • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

  • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

  • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Nickolas D. Zaller

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. To meet our reproducibility criteria, please ensure that the Methods section contains more information on data extraction; and that the data used is presented in supplementary tables.

3. We note you have included a table to which you do not refer in the text of your manuscript. Please ensure that you refer to Table 2 in your text; if accepted, production will need this reference to link the reader to the Table.

4. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see http://journals.plos.org/plosone/s/data-availability.

Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories. Any potentially identifying patient information must be fully anonymized.

Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access.

We will update your Data Availability statement to reflect the information you provide in your cover letter.

Additional Editor Comments (if provided):

As per reviewer #1, please consider conducting a sensitivity analysis using county population size to determine rates of mortality.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This study provides an assessment of the associations between county-level factors and COVID-19 mortality, with an emphasis on the impact of higher concentrations of non-English speaking residents and residents who are farm workers. The study evaluates associations overall and by urban/non-urban status. The study provides valuable information for identifying which populations may be at high risk; however, the study would potentially benefit by addressing the comments below, with particular emphasis on considering evaluating a rate (e.g., per 1,000 residents) of COVID-19 mortality rather than a raw count of deaths.

Major

1. There is inconsistency in the lists of independent variables and associated findings, which causes confusion. For example, the abstract describes a significant finding with respect to “higher density” and urban status; however, this variable is not listed in the previous sentence as variables that were evaluated. It may be helpful to additionally clarify that other non-psychosocial variables were evaluated.

2. The word “more” is not necessarily accurate in the abstract (e.g., more farmworkers), as “more” relates to an absolute number rather than a relative number. The phrasing “a higher percentage of” may be more appropriate.

3. While I do not disagree with the content, I am not sure that the final sentence of the abstract is the best concluding sentence given other findings from the study.

4. Why were some counties (and/or county equivalents) excluded from the study? There are 3,142 counties and county equivalents in the 50 states, excluding DC. Based on the number of counties in the study, there are more counties excluded than if the difference were only based on excluding county equivalents. If this data were simply not available, that is all that needs to be stated (e.g., the data from XXX counties were unavailable).

5. I am curious as to why the analysis was not calculated based on rates of death (e.g., per 100,000 residents), rather than raw numbers. Given the range of number of individuals who live in counties, a value of 1 additional death may be relatively very different across counties, based on population size. While this is somewhat mitigated by separating based on urban and non-urban, the population sizes within these stratifications still vary substantially. For example, the population in LA county is greater than 10,000,000 versus the population in several urban counties in Virginia of only 5,000 residents. Note that some VA counties may have been excluded because of their county equivalency status, but even if so, the lowest non-VA county population is 69,000 residents (in Colorado). I think at least a sensitivity analysis using county population size to determine rates of mortality may be important additional information.

6. My understanding is that coefficients associated with the primary analyses indicate the association of a 1 unit (e.g., a 1 percentage point increase in percent uninsured) in a given variable “holding all else constant,” and the “constant” is at the means of each other variable. As such, additionally having averages displayed in Table 1 may be helpful for some readers.

7. It is my understanding that if a given model has an input variable that is on a scale from 0 to 100, the interpretation of the coefficient would be “a 1 percentage point increase in variable XX,” rather than a “percentage” increase (e.g., line 89).

8. I think it would be most helpful for the readers to have more descriptive column headings, associated with Table 2. For example, what exactly is “b direct.”

9. The first paragraph of the discussion is meaningful and should not be removed from the manuscript; however, the discussion would benefit from a first paragraph that first outlines the findings from the study.

Minor

10. “currently” (in first sentence of abstract and in Line 4 of the introduction) may be most appropriately replaced with the current date (e.g., “As of June 2020, the United States…”), to clarify for readers. For example, if an individual reads this in 2021 and there were to be another outbreak in a different country, this may be confusing for the reader. Relatedly, a more specific time frame should be associated with the first sentence of the introduction (e.g., “between January 2020 and June 2020”).

11. In the abstract when numbering the evaluated variables, the number “3” is used twice.

12. Add (“SIP”) for “shelter in place” in line 52.

13. Suggest “fitted the model” in line 71 be “fit the models”

14. IQR has a typo (either values switched or one number wrong) on line 81.

Reviewer #2: Incredibly well-written and thoughtful

Lines 17-22: a lot in that sentence. Could you break it down more, maybe 2 sentences. There is something about "language barrier" that seems judgy. What about "lack of multi-linguial public health communication" or something like that?

Line 39-43: I found this sentence describing the ACS survey confusing. Why did you use 2014 for one and 2017 for another?

Line 39: can you give 1-2 lines about how the NYT calculated this. Limitation: place of death may not be where they lived. Does the NYT report the deaths based on where they died or lived? I am guessing in rural counties, this is very important. There also has been some data about people going back and adjusting mortality with presumed COVID-19. Do you know if I tried to pull the same data today, it would be updated with different numbers?

Line 49: The calculation for "high risk days" is confusing to me. There is a high risk for contracting and then high risk for diagnosis. I agree, the risk of the virus contraction went down with the shelter order, but the risk of diagnosis did not go down until probably 2 weeks and then mortality 3-4 weeks after that. The mortality lags. I am not sure how important this variable is to your analysis, or if it will change it, but it is a definite weak point that should be addressed.

Lines 55-61: is there a reference that could support why you did this?

Line 65: would just write out Shelter in place order.

I found the table hard to read. why is deaths not capitalized? Could you put the non percent (residents per square mile) first or last to keep all the of % ones together? The title was also hard to understand. Is there a way you could make the table more readable? I am guessing a lot of the rural towns were very small, and that the data is reported by where they lived rather than the hospital based on the numbers

Line 91-103 and 110-120 so powerful. Well done.

Discussion

You start your discussion off apologizing, being humble. Why not just say Our findings suggest that farm work..." You have already discussed your methods, and individual risk profiles were not the goal. The first para of discussion 2 long. Can you shorten so that the reader will read 3-4 sentences that are really powerful, then go on to the next para

Line 142: "point of concern" is sort of wishy washy. It is not shocking. It is not new. Th negative association.... supports other literature that people who are poor experience barriers in accessing care."

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Alysse Wurcel MD MS

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2020 Oct 14;15(10):e0240151. doi: 10.1371/journal.pone.0240151.r002

Author response to Decision Letter 0

25 Aug 2020

Reviewer #1: This study provides an assessment of the associations between county-level factors and COVID-19 mortality, with an emphasis on the impact of higher concentrations of non-English speaking residents and residents who are farm workers. The study evaluates associations overall and by urban/non-urban status. The study provides valuable information for identifying which populations may be at high risk; however, the study would potentially benefit by addressing the comments below, with particular emphasis on considering evaluating a rate (e.g., per 1,000 residents) of COVID-19 mortality rather than a raw count of deaths.

Major

1. There is inconsistency in the lists of independent variables and associated findings, which causes confusion. For example, the abstract describes a significant finding with respect to “higher density” and urban status; however, this variable is not listed in the previous sentence as variables that were evaluated. It may be helpful to additionally clarify that other non-psychosocial variables were evaluated.

We have added the following language to the abstract and believe our methods are now clearer:

We further adjusted these models for total population, population density, and number of days since the first reported case in a given county.

2. The word “more” is not necessarily accurate in the abstract (e.g., more farmworkers), as “more” relates to an absolute number rather than a relative number. The phrasing “a higher percentage of” may be more appropriate.

This is a good point, and we appreciate the reviewer’s observation. We have made the suggested revision.

3. While I do not disagree with the content, I am not sure that the final sentence of the abstract is the best concluding sentence given other findings from the study.

We have revised the abstract extensively.

---

4. Why were some counties (and/or county equivalents) excluded from the study? There are 3,142 counties and county equivalents in the 50 states, excluding DC. Based on the number of counties in the study, there are more counties excluded than if the difference were only based on excluding county equivalents. If this data were simply not available, that is all that needs to be stated (e.g., the data from XXX counties were unavailable).

We were not sufficiently clear in describing our data source. The data reflect ever county in the United States in which at least one COVID-19 case has been reported as of the date the analyses were begun. The “missing” counties were those that had not yet reported a case of COVID-19. We have clarified this in the manuscript text. We have also updated our analyses with more recent data (as of July 13), and as a result our analyses contain a higher number of counties (n=3024)

5. I am curious as to why the analysis was not calculated based on rates of death (e.g., per 100,000 residents), rather than raw numbers. Given the range of number of individuals who live in counties, a value of 1 additional death may be relatively very different across counties, based on population size. While this is somewhat mitigated by separating based on urban and non-urban, the population sizes within these stratifications still vary substantially. For example, the population in LA county is greater than 10,000,000 versus the population in several urban counties in Virginia of only 5,000 residents. Note that some VA counties may have been excluded because of their county equivalency status, but even if so, the lowest non-VA county population is 69,000 residents (in Colorado). I think at least a sensitivity analysis using county population size to determine rates of mortality may be important additional information.

We thank the reviewer for this comment, and agree with their point. We reflected on this recommendation at length. Because the COVID-19 epidemic in the United States is perhaps better understood as a series of sub-epidemics with a great deal of spatial heterogeneity, we conducted a series of 9 sub-analyses looking at the rate of deaths per 100,000 by census region. We believe comparing these sub-analyses with our original models (which we have elected to retain, while adding county population as an additional independent variable) significantly enhances the richness of our study, and we thank are grateful for the suggestion.

6. My understanding is that coefficients associated with the primary analyses indicate the association of a 1 unit (e.g., a 1 percentage point increase in percent uninsured) in a given variable “holding all else constant,” and the “constant” is at the means of each other variable. As such, additionally having averages displayed in Table 1 may be helpful for some readers.

Thank you, we have added the median of each variable to Table 1 to assist the reader’s interpretation of our findings.

7. It is my understanding that if a given model has an input variable that is on a scale from 0 to 100, the interpretation of the coefficient would be “a 1 percentage point increase in variable XX,” rather than a “percentage” increase (e.g., line 89).

The reviewer is correct. We have modified the language accordingly throughout.

8. I think it would be most helpful for the readers to have more descriptive column headings, associated with Table 2. For example, what exactly is “b direct.”

We appreciate this point. We have added additional language to the body of the text to explain the meaning of b-direct and b-indirect. We understand that this analytic approach is not currently widely used, and that additional interpretation is necessary. In the interest of readability, we have decided to leave the explanation in the manuscript text and the headings as is, however at the editors discretion we are happy to modify.

9. The first paragraph of the discussion is meaningful and should not be removed from the manuscript; however, the discussion would benefit from a first paragraph that first outlines the findings from the study.

After reflecting at length on this comment and others, we have rewritten the discussion entirely. We believe it is much stronger now and appreciate the reviewer’s critique.

Minor

10. “currently” (in first sentence of abstract and in Line 4 of the introduction) may be most appropriately replaced with the current date (e.g., “As of June 2020, the United States…”), to clarify for readers. For example, if an individual reads this in 2021 and there were to be another outbreak in a different country, this may be confusing for the reader. Relatedly, a more specific time frame should be associated with the first sentence of the introduction (e.g., “between January 2020 and June 2020”).

The reviewer makes an excellent point, and we appreciate their confidence in future reader’s interest in this paper! We have modified the language as suggested and updated to the most recent numbers.

11. In the abstract when numbering the evaluated variables, the number “3” is used twice.

We have extensively revised the abstract and addressed this typo.

12. Add (“SIP”) for “shelter in place” in line 52.

At the suggestion of reviewer 2 we have removed this variable from our models. (see below)

13. Suggest “fitted the model” in line 71 be “fit the models”

We have made the suggested change.

14. IQR has a typo (either values switched or one number wrong) on line 81.

The numbers have been updated with more recent data and the typo addressed.

 

Reviewer #2: Incredibly well-written and thoughtful

Lines 17-22: a lot in that sentence. Could you break it down more, maybe 2 sentences. There is something about "language barrier" that seems judgy. What about "lack of multi-linguial public health communication" or something like that?

This is a fair point, and we appreciate the reviewer’s point that the barrier is in the lack of accessible health communications rather than vulnerable communities preferred language of communication. We have changed the line as follows:

“…and that these risks are independent of poverty, insurance, or linguistic accessibility of COVID-19 health campaigns.

Line 39-43: I found this sentence describing the ACS survey confusing. Why did you use 2014 for one and 2017 for another?

We have revised this sentence for clarity. The 2014 5-year estimate was the most recent county-level data available from the ACS for the percentage of households in which “no one age 14 and over speaks English Only or Speaks English “very well”. More recent county-level data were available for the population age 65 and over in the United States. We have included citations for all tables in an attempt to clarify.

Line 39: can you give 1-2 lines about how the NYT calculated this. Limitation: place of death may not be where they lived. Does the NYT report the deaths based on where they died or lived? I am guessing in rural counties, this is very important. There also has been some data about people going back and adjusting mortality with presumed COVID-19. Do you know if I tried to pull the same data today, it would be updated with different numbers?

This is such an important point, and we appreciate the reviewer’s comment. We have updated this section with further details on the NYT’s methodology, and with a more obvious link to the data for interested readers. The “patchwork nature” of COVID-19 reporting across the United States makes it difficult to offer a definitive answer to the reviewer’s question. An analysis conducted by Headwater Economics suggests that this issue may also differ geographically, with more hospital beds per capita in the western United States vs. Eastern. To address this and other regional differences we have conducted a second set of analyses by US region. (see response to reviewer 1, item 5). (https://headwaterseconomics.org/equity/hospital-access-seniors/)

Line 49: The calculation for "high risk days" is confusing to me. There is a high risk for contracting and then high risk for diagnosis. I agree, the risk of the virus contraction went down with the shelter order, but the risk of diagnosis did not go down until probably 2 weeks and then mortality 3-4 weeks after that. The mortality lags. I am not sure how important this variable is to your analysis, or if it will change it, but it is a definite weak point that should be addressed.

We agree, particularly given the different definitions of sheltering in place across counties and states. This variable has been removed, with no dramatic changes in model outcomes.

Lines 55-61: is there a reference that could support why you did this?

We have inserted the appropriate citation.

Line 65: would just write out Shelter in place order.

This variable has been removed from our analyses.

I found the table hard to read. why is deaths not capitalized? Could you put the non percent (residents per square mile) first or last to keep all the of % ones together? The title was also hard to understand. Is there a way you could make the table more readable? I am guessing a lot of the rural towns were very small, and that the data is reported by where they lived rather than the hospital based on the numbers

We have revised the tables somewhat for clarity. We have opted to leave the residents per square mile and population at the bottom of the table, as we wanted to showcase outcomes and predictors in descending order of theoretical importance to our models. We have also added additional figures showcasing some of the

Line 91-103 and 110-120 so powerful. Well done.

Thank you

Discussion

You start your discussion off apologizing, being humble. Why not just say Our findings suggest that farm work..." You have already discussed your methods, and individual risk profiles were not the goal. The first para of discussion 2 long. Can you shorten so that the reader will read 3-4 sentences that are really powerful, then go on to the next para

We have extensively revised the discussion and hope it now addresses botht his point and the one below.

Line 142: "point of concern" is sort of wishy washy. It is not shocking. It is not new. Th negative association.... supports other literature that people who are poor experience barriers in accessing care."

We have changed this language.

Attachment

Submitted filename: response to reviewers.docx

Click here for additional data file. (263.2KB, docx)

Decision Letter 1

Nickolas D Zaller

4 Sep 2020

PONE-D-20-18322R1

Social determinants of COVID-19 mortality at the county level

PLOS ONE

Dear Dr. Fielding-Miller,

Thank you for submitting your manuscript to PLOS ONE. While you have thoughtfully addressed many of the previous reviewer comments, some concerns remain. Therefore, we feel that your manuscript does not fully meet PLOS ONE’s publication criteria as it currently stands. We invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

One of the more important concerns relates to the modeling approach and that there is a lack of detail regarding how the authors addressed (or did not address) the fact that a quarter of counties had a mortality equal to 0.  There are multiple statistical approaches that could be used to address potential modal instability due to a relatively large number of 0s.  At the very least, this should be further discussed in the limitations section of the manuscript. 

Please submit your revised manuscript by October 1, 2020. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

  • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

  • A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

  • An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

Nickolas D. Zaller

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: This paper evaluates the association of COVID-19 mortality with social determinants at the county level. The authors have taken great efforts to address the comments by the reviewers, and the paper remains important in that it highlights aspects related to the COVID-19 pandemic. A few questions regarding method choices should be addressed as outlined below.

Major comments

1. The authors have taken the reviewer’s suggestion to evaluate rate of mortality per 100,000 residents stratified by census region. I suggest that the findings from this analysis be provided as well. Relatedly, I’m unsure if the information in lines 189-192 means overall or in the regional analysis. This could be a reasonable place to note the overall analysis per 100,000 in an appendix.

2. The authors should clearly address limitations of their study and what implications the limitations may have. For example, 25% of the counties included in the analysis have 0 deaths. With a highly skewed outcome (as evident by the range versus the IQR and median) it is important to highlight what this may mean in terms of the analysis using spatial autoregressive models. Second, there should be a discussion about the range of the primary variable of focus (percent of farmers) among urban counties, which has an IQR of 0.0 to 0.1. This may be related to the finding of an increase of 2,500 deaths for each 1 percentage point increase in urban counties. It is likely that this may not be a stable model.

3. Figures 4 and 5 may be a bit challenging to interpret without more context. Potentially change the words “percentage” to “a one percentage point change in.”

4. I am a bit confused on the map legends in Figure 3. There are multiple categories that show 0.0-0.0, which likely are extra categories or simple mislabeling/typos. However, I am unsure how the percent of residents in poverty are all in the hundreds or over 1,000.

Minor comments

1. There remains two more uses of “more” rather than “percent of” in the abstract and in the last sentence of the introduction. I suggest modifying these for consistency and accuracy

2. Median death count on line 103 does not match what is in table 1. Text median = 2, and table median = 6.

3. The population (thousands) variable is flipped between non-urban and urban counties in table 1.

4. In table 2, please clarify that these outcomes are related to the death counts analysis (rather than per 100,000). Potentially include a footer to explain what b-direct/indirect/total mean for the reader.

5. The highest end of the largest level in the legend in Figure 2 (1615.6) does not match the maximum given in the text 350).

Reviewer #2: Thank you for taking time to make a powerful analysis even sharper.

One thing:

Something off with line 186: "Even when adjusting for the percentage of non-186 English speaking households, percentage of people living in poverty, percentage of people without insurance, and the percentage of farmworkers in a county appears to be independently associated with a higher number of deaths, and that the relationship varies by region."

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Alysse G. Wurcel MD MS

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

PLoS One. 2020 Oct 14;15(10):e0240151. doi: 10.1371/journal.pone.0240151.r004

Author response to Decision Letter 1

14 Sep 2020

PONE-D-20-18322R1

Social determinants of COVID-19 mortality at the county level

PLOS ONE

Dear Dr. Fielding-Miller,

Thank you for submitting your manuscript to PLOS ONE. While you have thoughtfully addressed many of the previous reviewer comments, some concerns remain. Therefore, we feel that your manuscript does not fully meet PLOS ONE’s publication criteria as it currently stands. We invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

One of the more important concerns relates to the modeling approach and that there is a lack of detail regarding how the authors addressed (or did not address) the fact that a quarter of counties had a mortality equal to 0. There are multiple statistical approaches that could be used to address potential modal instability due to a relatively large number of 0s. At the very least, this should be further discussed in the limitations section of the manuscript.

We have addressed this potential shortcoming under Reviewer 1, Comment 2.

Comments to the Author

Reviewer #1: This paper evaluates the association of COVID-19 mortality with social determinants at the county level. The authors have taken great efforts to address the comments by the reviewers, and the paper remains important in that it highlights aspects related to the COVID-19 pandemic. A few questions regarding method choices should be addressed as outlined below.

Major comments

1. The authors have taken the reviewer’s suggestion to evaluate rate of mortality per 100,000 residents stratified by census region. I suggest that the findings from this analysis be provided as well.

These findings are presented in lines 154 – 170. We present these findings by region rather than nationally because the national level findings would mask the significant variation by region.

Relatedly, I’m unsure if the information in lines 189-192 means overall or in the regional analysis. This could be a reasonable place to note the overall analysis per 100,000 in an appendix.

We have added language to this section to clarify that we are discussing sub-analyses. The line now reads as follows:

Across all 9 census regions, there was no association between the percentage of farmworkers and mortality…

2. The authors should clearly address limitations of their study and what implications the limitations may have. For example, 25% of the counties included in the analysis have 0 deaths. With a highly skewed outcome (as evident by the range versus the IQR and median) it is important to highlight what this may mean in terms of the analysis using spatial autoregressive models. Second, there should be a discussion about the range of the primary variable of focus (percent of farmers) among urban counties, which has an IQR of 0.0 to 0.1. This may be related to the finding of an increase of 2,500 deaths for each 1 percentage point increase in urban counties. It is likely that this may not be a stable model.

We appreciate the reviewer’s thoughtfulness. We agree, these data are highly non-normal and careful attention to model building is necessary. We have added the following language to the text regarding our approach:

Our spatial autoregressive model used the generalized spatial two-stage least squares estimator, specifically Stata’s spregress command with the “g2sl” option. G2sl is a generalized method of moment (GMM) modeling approach. The GMM makes no assumptions about variable distribution and is an appropriate statistical approach to address data that are highly skewed or have unknown distributions (15).

3. Figures 4 and 5 may be a bit challenging to interpret without more context. Potentially change the words “percentage” to “a one percentage point change in.”

We have revised the figures and hope they are now more clear.

4. I am a bit confused on the map legends in Figure 3. There are multiple categories that show 0.0-0.0, which likely are extra categories or simple mislabeling/typos. However, I am unsure how the percent of residents in poverty are all in the hundreds or over 1,000.

Thank you, some of these were errors from the software, others were typos. We have revised figure 3 extensively for clarity.

Minor comments

1. There remains two more uses of “more” rather than “percent of” in the abstract and in the last sentence of the introduction. I suggest modifying these for consistency and accuracy

2. Median death count on line 103 does not match what is in table 1. Text median = 2, and table median = 6.

Thank you for catching this oversight. We have corrected the text.

3. The population (thousands) variable is flipped between non-urban and urban counties in table 1.

Thank you! We appreciate the reviewer’s attention to detail. This has been corrected

4. In table 2, please clarify that these outcomes are related to the death counts analysis (rather than per 100,000). Potentially include a footer to explain what b-direct/indirect/total mean for the reader.

We appreciate this suggestion and agree that it will enhance the paper’s clarity.

5. The highest end of the largest level in the legend in Figure 2 (1615.6) does not match the maximum given in the text 350).

We have modified all figures to address the reviewer’s comments.

Reviewer #2: Thank you for taking time to make a powerful analysis even sharper.

One thing:

Something off with line 186: "Even when adjusting for the percentage of non-186 English speaking households, percentage of people living in poverty, percentage of people without insurance, and the percentage of farmworkers in a county appears to be independently associated with a higher number of deaths, and that the relationship varies by region.".

We have modified this sentence and believe it now reads more clearly.

Decision Letter 2

Nickolas D Zaller

22 Sep 2020

Social determinants of COVID-19 mortality at the county level

PONE-D-20-18322R2

Dear Dr. Fielding-Miller,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Nickolas D. Zaller

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have addressed all of the comments, and the manuscript provides important findings regarding the pandemic.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Acceptance letter

Nickolas D Zaller

28 Sep 2020

PONE-D-20-18322R2

Social determinants of COVID-19 mortality at the county level

Dear Dr. Fielding-Miller:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Nickolas D. Zaller

Academic Editor

PLOS ONE

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    S1 File

    (DOCX)

    Click here for additional data file. (17KB, docx)
    Attachment

    Submitted filename: response to reviewers.docx

    Click here for additional data file. (263.2KB, docx)

    Data Availability Statement

    Analyses used publicly available data which are cited appropriately.

    Articles from PLoS ONE are provided here courtesy of PLOS

    RESOURCES