Racial and ethnic differentials in COVID-19-related job exposures by occupational standing in the US

Researchers and journalists have argued that work-related factors may be partly responsible for disproportionate COVID-19 infection and death rates among vulnerable groups. We evaluate these claims by examining racial and ethnic differences in the likelihood of work-related exposure to COVID-19. We extend previous studies by considering 12 racial and ethnic groups and five types of potential occupational exposure to the virus: exposure to infection, physical proximity to others, face-to-face discussions, interactions with external customers and the public, and working indoors. Most importantly, we stratify our results by occupational standing, defined as the proportion of workers within each occupation with at least some college education. This measure serves as a proxy for whether workplaces and workers employ significant COVID-19-related risk reduction strategies. We use the 2018 American Community Survey to identify recent workers by occupation, and link 409 occupations to information on work context from the Occupational Information Network to identify potential COVID-related risk factors. We then examine the racial/ethnic distribution of all frontline workers and frontline workers at highest potential risk of COVID-19, by occupational standing and by sex. The results indicate that, contrary to expectation, White frontline workers are often overrepresented in high-risk jobs while Black and Latino frontline workers are generally underrepresented in these jobs. However, disaggregation of the results by occupational standing shows that, in contrast to Whites and several Asian groups, Latino and Black frontline workers are overrepresented in lower status occupations overall and in lower status occupations associated with high risk, and are thus less likely to have adequate COVID-19 protections. Our findings suggest that greater work exposures likely contribute to a higher prevalence of COVID-19 among Latino and Black adults and underscore the need for measures to reduce potential exposure for workers in low status occupations and for the development of programs outside the workplace.


Introduction
In the United States, Black and Latino adults have experienced substantially higher rates of COVID-19 infection and mortality during 2020 than Whites and Asians [1][2][3][4]. Researchers and journalists argue that the difference is due, at least in part, to two work-related factors: (1) Blacks and Latinos are more likely to hold jobs which have to be done at their workplace rather than remotely, and (2) Latino and Black workers face greater risks of exposure to COVID-19 in their jobs than others [5][6][7][8][9][10][11]. In this paper, we investigate these claims by examining differences in the likelihood of work-related exposure to COVID-19 by race and ethnicity.
It is difficult to quantify the importance of occupational vs. other exposures to the coronavirus. Recent studies have used two approaches to argue that occupational risks to COVID-19 are higher for marginalized racial and ethnic groups than for Whites. We use a combination of these approaches. The most common strategy has been to look at racial/ethnic differences in recent (generally pre-pandemic) employment in industries or occupations that are considered essential or frontline during the pandemic [5,12,13]. As we discuss later, these earlier studies have been based on various definitions of essential and frontline. A second approach, sometimes used in conjunction with the first, has been to estimate relative risk of different groups during the pandemic from survey data about whether jobs entail high exposure to disease or infection and/or require close proximity to others [6].
We extend this work in several ways. First, we consider 12 racial/ethnic groups. Previous studies on occupations and COVID-19 have been limited to four broad racial/ethnic categories (Whites, Blacks, Latinos and Asians), although workers in other racial/ethnic groups or in Asian and Latino subgroups may be disproportionately exposed to COVID-19 [4]. Second, we examine five distinct job characteristics that could expose individuals to a high risk of contracting COVID-19: exposure to disease/infection; proximity to others; face-to-face discussions; interactions with external customers or the public; and working mostly indoors. Third, we examine results separately by occupational status or ranking, defined here as the proportion of workers within each occupation with some college education. Occupational status is closely linked to the type of activities performed, the physical work environment, control over working conditions, and attitudes toward workplace rules [14][15][16]. Despite the high risks often faced by physicians, nurses and other health personnel, especially early in the pandemic, workers in lower status occupations are generally more likely to be exposed to COVID-19 in the workplace than those in higher status occupations. In contrast, higher status workers have better access to risk mitigation measures such as personal protective equipment (PPE), frequent sanitation, enforced distancing, partitions, better ventilation, and new filtration systems. Examining results by occupational status leads to a clearer picture of racial/ethnic differentials in potential exposure to COVID-19 transmission. We also present separate results by sex because of vastly different occupational profiles for men and women.

Data
We use data from two sources. Information on employment status, type of job, years of education, sex and race/ethnicity come from the 2018 American Community Survey (ACS), which is the most recent large, nationally representative survey fielded prior to the pandemic.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020. 11.13.20231431 doi: medRxiv preprint Access is provided by the Integrated Public Use Microdata Series (IPUMS) [17]. Variables related to work context are drawn from the Occupational Information Network (O*NET), a database of work characteristics by detailed occupation collected by the US Department of Labor [18]. Using continuous data collection from random samples of incumbent workers, O*NET obtains information about job characteristics -such as work-related activities, work environment, and skills required to do the job -for almost 1000 detailed occupations. We use six variables from O*NET version 24.3, which contains information collected through 2019 and was released in May 2020 [18]. These variables capture potential exposure to SARS-Cov2 -a virus that appears to be transmitted primarily through respiratory droplets and aerosols when an infected person coughs, sneezes, talks or breathes near others [19]. The six workplace variables reflect five types of risk: exposure to infection, physical proximity to others, face-to-face discussions, interactions with external customers and the public, and working indoors.

Variables
Race/ethnicity Individuals were classified according to the following 12 race/ethnicity categories as reported by the head of the household in the ACS: White, Black, Latino, Chinese, Filipino, Vietnamese, Korean, other Asian, Native American (American Indian/Alaskan Native), Pacific Islander (Native Hawaiian/Pacific Islanders), mixed race, and other race/ethnicity. The three largest groups in the other Asian category are Indian, Japanese and Pakistani. In our analysis, all racial/ethnic categories except Latino exclude those who reported themselves as Latino/Hispanic in the ACS question on ethnicity.

Occupational Status
In order to rank occupations, we use a well-established sociological indicator of occupational ranking, known as "occupational education" [20]. For each occupation, we calculate occupational education from the ACS as the percent of persons in the occupation who completed at least one year of college education (note that this variable classifies individuals by the educational attainment of all people in their occupation rather than by their own educational attainment). We use this measure, which we refer to as "occupational status" or "occupational ranking," to divide the 409 occupations reported by ACS respondents in our analysis into quartiles, i.e., each group comprising about one-quarter of the occupations (not one-quarter of the workers).

Frontline status
During the pandemic, many workers have been sheltered from occupational exposure to infection by working remotely, e.g., from home. To distinguish between workers who are more vs. less able to work remotely, studies have used varying criteria to define "essential" and/or "frontline occupations." In this analysis, we include only occupations classified as "frontline" based on the definition offered by Dingel and Neiman [21] and used by Blau et al. [13]: occupations in which one-third or fewer workers can feasibly work from home, ascertained from responses to 15 questions in O*NET. A total of 249 out of 409 occupations in our analysis are considered frontline according to these criteria.
We do not consider whether occupations are classified as essential or subject to lockdown because there has been enormous variation across states, localities, and time period in definitions . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint and their application during the pandemic. For example, janitors, maids, bus drivers, retail sales workers, and personal care workers would not be classified as essential according to industry guidelines issued by the Department of Homeland Security (DHS) Cybersecurity and Infrastructure Security Agency (CISA) [13], yet many individuals in these occupations have likely been working away from home much or most of the time since March, 2020.

Defining High Risk Occupations
The five types of risk we consider from O*NET data -exposure to infection, physical proximity to others, face-to-face discussions, interactions with external customers and the public, and working indoors -are based on the following O*NET questions (in their original wording): 1. How often does your current job require that you be exposed to diseases or infection? 2. How physically close to other people are you when you perform your current job? 3. How often does your current job require face-to-face discussions with individuals and within teams? 4. In your current job, how important are interactions that require you to deal with external customers (as in retail sales) or the public in general (as in police work)? 5. How often does your current job require you to work outdoors, exposed to all weather conditions? [And:] How often does your current job require you to work outdoors, under cover (like in an open shed)?
Each of these questions has five possible categorical responses; the responses reflect frequency of exposure for questions 1,3, and 5; degrees of closeness for question 2, and importance of interactions for question 4.
For each of the first four risk indicators, we define high risk as working in an occupation for which the mean response in the O*NET data falls in the highest quartile (25%) of the full set of 409 occupations in the analysis. For the fifth risk indicator we use two questions reflecting the frequency with which employees work outdoors; the responses for these two questions are highly correlated. Because indoor work is higher risk than outdoor work ceteris paribus, we take the maximum of the O*NET values for these two variables and define high risk as the quartile of occupations with the lowest value, thereby identifying jobs with the lowest frequency of working outdoors as high risk.

Analytic Strategy
In order to match occupations reported in the ACS with characteristics in O*NET, we converted the occupation code (Standard Occupational Classification or SOC) in O*NET into the four-digit 2010 census occupational code recorded in the 2018 ACS using a crosswalk provided at https://www.bls.gov/emp/documentation/crosswalks.htm. In cases where one census occupational code corresponded to multiple O*NET occupations, we took unweighted averages of the characteristics of the O*NET occupations and assigned these to the census occupational code. We linked occupations between the ACS and O*NET for all employed persons who reported holding a job in ACS with the exception of those in the military (because O*NET data were not available for military occupations); this linkage yielded about 1.9 million individuals in the ACS in 409 occupations. Approximately 88 percent of these individuals reported on the job they held during the past year with the remaining respondents reporting on the job held between . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint one and five years prior to the survey. It is important to note that unemployment and labor force participation rates vary considerably by race, ethnicity, and sex. For example, in the third quarter of 2020, the unemployment rate for Black men 16+ years old was 13.8 percent compared to 7.4 percent for White men and 9.6 percent for Asian men [22]. These rates have also fluctuated throughout the period of the pandemic. Our analysis does not account for differential unemployment and labor force participation rates.
As noted above, the analysis includes only individuals who worked in the five years prior to the ACS, the vast majority of whom worked in the year prior to the ACS; we refer to these individuals as either "recent workers" or simply as "workers." We make four types of comparisons by racial/ethnic composition for female and male workers: (1) all workers vs. frontline workers; (2) frontline workers by quartiles of occupational status; (3) all frontline workers vs. high risk frontline workers for each our five indicators of risk; and (4) high-risk frontline workers by occupational status, for each of our five risk indicators. All distributions are calculated using weights provided in the ACS, but we also report unweighted sample sizes.
The racial/ethnic compositions of the different occupational categories described above are shown in two sets of bar graphs: the first considers 12 racial/ethnic groups whereas the second excludes Whites, Blacks and Latinos in order to increase visibility of the smaller groups. Each set of graphs comprises seven panels for each sex. The percentages corresponding to each of these graphs are presented in S1 Appendix. Table 1 shows the percentages of workers in frontline jobs by sex and race/ethnicity. Employment in frontline jobs varies considerably across groups. For both men and women, Latinos are the most likely workers to hold frontline jobs whereas Chinese and Korean workers are the least likely to do so. Latino, Black, Native American, and Pacific Islander men are the most likely to have frontline jobs-more than 70% of male workers in each of these groups are classified as frontline. In contrast, Vietnamese, Latino, and Filipino women are the most likely female workers to hold frontline jobs. The comparisons between all workers and frontline workers in Figs 1 and 2 show that, for both men and women, Black and Latino workers are proportionately more likely, and Whites proportionately less likely, to be frontline workers -although the differences in the distributions between all workers and frontline workers are relatively modest. However, panel 2 underscores the huge differences in the racial/ethnic composition of frontline workers by occupational . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint

Results
The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint ranking: moving from the highest (4 th ) to lowest (1 st ) quartile of occupational status, the percentage of White workers declines substantially, while the percentages of Black and Latino workers modest generally increase -dramatically in the case of Latinos. The most common frontline jobs in the highest quartile are primarily in the healthcare sector, most notably registered nurses, physicians, and surgeons. In contrast, the most common frontline jobs in the lowest quartiles include cashiers, drivers, janitors, laborers, stock clerks, and housekeeping cleaners.
A comparison of the racial/ethnic distribution of all frontline jobs (top bar in panel 2) with the distribution for the subset of frontline jobs that are high-risk by each of our indicators (top bar in each of panels 3-7) highlights the complexity of identifying the racial/ethnic groups with the largest occupational exposure to the virus. Contrary to expectation, White frontline workers are often overrepresented in high risk jobs -particularly indoor jobs and those requiring external interactions or face-to-face discussions -while Black and Latino frontline workers are generally underrepresented in these types of jobs. Perhaps as surprising is the finding that the racial/ethnic distribution of frontline jobs with a high risk of infection (top bar in panel 3) -a risk likely to be greatest in the healthcare industry -is similar to the distribution of all frontline workers (top bar in panel 2).
These findings, by themselves, would suggest that Black and Latino workers do not face higher occupational risks overall than Whites, contrary to expectation. However, this conclusion is likely to be incorrect. The problem is that the O*NET measures that we (and other researchers) are using provide no information on variations among occupations in COVID-19 risk reduction measures deployed in the workplace or by individual workers. For example, food service workers (low occupational status and disproportionately Latino) and physicians (high occupational status and disproportionately White) are both in frontline occupations at high risk for physical proximity to other people, but physicians are far more likely to have access to and use PPE, work in frequently sanitized environments, and have the knowledge and some ability to reduce potential exposure at work. The problem is compounded by the fact that O*NET provides measures of only average job characteristics, which offer no insight into the variability in the risks among workers within occupations by worker characteristics. To get a clearer picture of COVID-19 transmission-related risks faced by workers in different occupations and different racial/ethnic groups, we use occupational status as a proxy measure for the actual risk in the work environment during the COVID-19 pandemic. For each of the five risk indicators, S2 Appendix lists the most frequently held high-risk frontline occupations within each occupational status quartile (separately by sex); only occupations that . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint include at least five percent of workers in the quartile are shown. Across the five risk indicators, the highest status occupations are primarily in the healthcare sector -e.g., physicians and surgeons; registered nurses; pharmacists; dentists; technicians; and physical therapists. Some healthcare occupations also appear in the lowest two quartiles, most notably personal care aides and nursing, psychiatric and home health aides, jobs that are particularly common among women. However, there is a wide range of low status occupations outside of the healthcare sector that are high-risk on at least one of the five indicators: e.g., janitors, maids, and other cleaners; maintenance and repair workers; taxi and bus drivers; food preparation workers and servers; machine operators; barbers and hairdressers; cashiers; and correctional officers and jailers.
Figs 3 and 4 present the same information as Figs 1 and 2, but these graphs are limited to numerically smaller ethnic groups. The total length of the bars denotes the proportion of workers summed across all smaller ethnic groups in the particular occupational category; these proportions vary across rows and panels but are consistently below 20% (in contrast to 100% for Figs 1 and 2, which include all workers).
As shown in panel 1 and the top bar in panel 2, male workers in these ethnic groups are equally represented or underrepresented in frontline occupations compared with their representation in all occupations, except for Native American men who have a higher proportion in frontline than in all occupations. The case for females is more mixed, with Filipino, Vietnamese, and Native American workers overrepresented in frontline jobs compared to the population of all female workers.
However, in contrast to Native American frontline workers, who disproportionately occupy the lowest occupational status group, Asian frontline workers are generally overrepresented in the highest occupation status group. This is especially true for Filipino and "other Asian" male and female workers and for Chinese and Korean male workers. The majority of Filipino men and women, and Chinese and Korean women, who are in high status frontline occupations are registered nurses. The other frequently held occupations for these groups are also in the healthcare sector. Chinese and Korean men with the highest status frontline occupations are most likely to be physicians and surgeons, but are also employed as dentists, pharmacists and registered nurses. The high proportions of Asians in the healthcare sector largely accounts for their disproportionate presence in high-risk occupations, particularly those involving exposure to infection, close proximity to others, and indoor locations.

Discussion
In this study, we extended earlier work on racial/ethnic differences in potential work-related exposure to COVID-19 by examining 12 racial/ethnic groups rather than the four used in previous studies and by considering five indicators of risk exposure. Our central innovation is to use occupational status as a proxy for whether workplaces and workers employ significant COVID-19-related risk reduction strategies (PPE, distancing, sanitation, improved ventilation, etc.) -something not measured in the data we use nor in any other large, nationally representative data sets, to our knowledge. We argue that higher occupational status workers, even in high risk occupations, are more likely to have access to (and to use) workplace COVID-. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review) preprint
The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint 19 risk mitigation measures than low status workers because they are more likely: (a) to demand risk reduction equipment and other measures and have the bargaining power to obtain it, (b) to be perceived as valuable and hard to replace by their employers, and (c) to understand (or learn about) COVID-19 transmission routes and to comply with risk reduction strategies or implement their own.
Once we disaggregate our results by occupational status, we see large differences in the racial and ethnic distribution of frontline occupations. In contrast to Whites, Latino and Black frontline workers are overrepresented in lower status occupations overall, as well as lower status occupations associated with high risk, and are thus less likely than Whites to have adequate COVID-19 protections.
In contrast to Native American frontline workers, who are overrepresented in lower status occupations, Asians show more complex patterns, reflecting factors such as immigration history, labor market segregation, and ethnic economic niches. For example, female Filipino workers are overrepresented among frontline workers, and they are disproportionately at risk for all but one of the five risk categories. However, they are much more likely to be in higher status occupations than other Asian groups, in part, because of a history of immigration of Filipino nurses and others health professionals to the US [23].
Our conclusions contrast with those of earlier studies by Selden and Berdahl and Hawkins [6,12]. Based on data from the 2014-2017 Medical Expenditure Panel Survey and the 2017-2018 American Time Use Survey, Selden and Berdahl find that Blacks, Latinos, and Asians are only marginally more likely than whites to work in essential jobs, i.e., at businesses that have been allowed to remain open during COVID-19-related shutdowns [12]. Hawkins uses data from the 2019 Current Population Survey linked with two measures of risk from O*NET (infection and close proximity) and finds an elevated risk for Black workers but virtually none for Latinos for jobs in essential industries [6]. Unlike these studies, we focus on frontline occupations rather than essential industries, because of the large variation among states and local jurisdictions and over time in the definition of essential industries versus the relative stability of the frontline occupation classification. However, our main contribution is that we distinguish between lower and higher status occupations, a strategy that uncovers important differences among racial and ethnic groups that were not apparent in the previous studies.
The racial/ethnic differences by occupational status highlighted in this paper are consistent with a large literature unrelated to the COVID-19 pandemic. Social science and public health research shows that occupations in the US are highly segregated by race and ethnicity, with Latino and Black workers holding many of the lowest status and least secure jobs [24][25][26][27]. There is also ample evidence that workers in low status occupations were already at higher risk of accidents, injury, infection, and other health problems prior to the pandemic and that these workers also have less control and decision-making ability over how their workplaces are run [28]. Workers in low status jobs may also be more likely to face employer resistance to implementing stricter safety measures that may reduce productivity or profits [29][30][31][32]. Thus, a realistic depiction of COVID-19 risks for Latino and Black workers requires consideration of occupational status.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint Despite the insights provided by this study, our analysis has several important limitations. First, the 2018 ACS data on recent jobs used here may not provide an accurate picture of employment or occupational distributions during the pandemic. Unemployment rates rose markedly in many areas of the country during 2020. At the time of this writing, the peak rate was 14.7% in April 2020 compared to 3.5% in February 2020 for the US population age 16+ [33]. Unemployment rates were even higher for some racial/ethnic groups, including Latinos (18.9% in April) and Blacks (16.8% in May) [34,35]. Business closures, layoffs, and unemployment rates have varied a great deal from state to state and over time. Thus, it is impossible to determine whether or not workers were employed during 2020 at a particular time and place. Other consequences of the pandemic, such as virtual schooling, have led many parents, especially mothers, to leave the labor force entirely, at least until the pandemic is over [36]. Despite these caveats, we believe that the ACS data provide the best picture currently available of the occupational distribution of the US population at a time close to the onset of the pandemic. A related limitation is that, because our analyses provide no evidence about exposure for people who are unemployed or outside the labor force, our conclusions are necessarily restricted to the employed population.
Additional concerns pertain to the use of O*NET data for measures of risk. These measures were not designed with a pandemic in mind and do not include key information on viral transmission risk. A critical omission for analyses of COVID-19 is data on the implementation of, and worker compliance with, COVID-19 transmission workplace mitigation measures. This lack of information underscores the need to disaggregate workers by occupational status, as we have done here. There are other limitations to the O*NET data when used in studies of infection-related risks. For example, in the case of contact with others, it would be useful to know the average duration of close contacts, the nature of the contacts and the characteristics of the people with whom the worker interacts. For all of our measures of risk, our analysis would benefit if O*NET provided information on how characteristics of each individual occupation vary by demographic factors (e.g., race/ethnicity, sex, and age) of the workers.
Despite these limitations, our results strongly suggest that higher work exposures to COVID-19 likely contribute to a higher prevalence of the virus among Latino and Black, compared to White, adults in the US. Until there is a highly effective and widely available prevention or treatment method for COVID-19, the focus needs to be on risk reduction, particularly in places where transmission is high. Reducing potential exposure for workers in low status occupations would require multiple coordinated measures. At the employer level, measures might include incentivizing employers (through tax breaks and rebates, grants, or low interest loans) to improve work environments (e.g., frequent sanitizing, upgrading air filtration and ventilation systems, installing partitions, and providing more space so that workers can be at least six feet apart) and potentially penalizing employers who fail to implement these measures. Another strategy would require employers to test all workers regularly and to close for two weeks and sanitize their facilities if an employee tests positive for the virus. For businesses employing workers in lower status occupations, policies such as this are more likely to be effective (and more equitable) if they come with free or subsidized health care for workers who become ill and financial incentives to businesses to pay workers and keep the business afloat during shutdowns. Finally, a whistle blower system in which workers can confidentially report potentially risky COVID-19related conditions and propose improvements to reduce risk could be highly effective if . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint adequately funded by local, state, or national governments -but only if claims could be quickly adjudicated and lead to rapid changes, where appropriate.
Equally important, however, are programs outside of the workplace, including: (1) effective community-based and free or low-cost testing, tracing, and treatment programs particularly centered on the uninsured and underinsured population, (2) a unified national, evidence-based health communications campaign to provide clear messaging about the pandemic and prevention measures in multiple languages and venues, and (3) COVID-19 health communication and education programs tailored for Black and Latino populations which address distrust and experiences of racism in public health and medical care.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted November 16, 2020. ; https://doi.org/10.1101/2020.11.13.20231431 doi: medRxiv preprint