key: cord-297513-fxap5sw4
authors: Liu, S. H.; Liu, B.; Li, Y.; Norbury, A.
title: Time courses of COVID-19 infection and local variation in socioeconomic and health disparities in England
date: 2020-05-29
journal: nan
DOI: 10.1101/2020.05.29.20116921
sha: 
doc_id: 297513
cord_uid: fxap5sw4

Objective: To identify factors associated with local variation in the time course of COVID-19 case burden in England. Methods: We analyzed laboratory-confirmed COVID-19 case data for 150 upper tier local authorities, from the period from January 30 to May 6, 2020, as reported by Public Health England. Using methods suitable for time-series data, we identified clusters of local authorities with distinct trajectories of daily cases, after adjusting for population size. We then tested for differences in sociodemographic, economic, and health disparity factors between these clusters. Results: Two clusters of local authorities were identified: a higher case trajectory that rose faster over time to reach higher peak infection levels, and a lower case trajectory cluster that emerged more slowly, and had a lower peak. The higher case trajectory cluster (79 local authorities) had higher population density (p<0.001), higher proportion of Black and Asian residents (p=0.03; p=0.02), higher multiple deprivation scores (p<0.001), a lower proportions of older adults (p=0.005), and higher preventable mortality rates (p=0.03). Local authorities with higher proportions of Black residents were more likely to belong to the high case trajectory cluster, even after adjusting for population density, deprivation, proportion of older adults and preventable mortality (p=0.04). Conclusion: Areas belonging to the trajectory with significantly higher COVID-19 case burden were more deprived, and had higher proportions of ethnic minority residents. A higher proportion of Black residents in regions belonging to the high trajectory cluster was not fully explained by differences in population density, deprivation, and other overall health disparities between the clusters.

over time to reach higher peak infection levels, and a lower case trajectory cluster that emerged 26 more slowly, and had a lower peak. The higher case trajectory cluster (79 local authorities) had 27 higher population density (p<0.001), higher proportion of Black and Asian residents (p=0.03; 28 p=0.02), higher multiple deprivation scores (p<0.001), a lower proportions of older adults 29 (p=0.005), and higher preventable mortality rates (p=0.03). Local authorities with higher 30

proportions of Black residents were more likely to belong to the high case trajectory cluster, even 31 after adjusting for population density, deprivation, proportion of older adults and preventable 32 mortality (p=0.04). 33 34 Conclusion: Areas belonging to the trajectory with significantly higher COVID-19 case burden 35 were more deprived, and had higher proportions of ethnic minority residents. A higher 36 proportion of Black residents in regions belonging to the high trajectory cluster was not fully 37 explained by differences in population density, deprivation, and other overall health disparities 38 between the clusters. 39

Introduction 40 41 England currently has amongst the highest worldwide recorded cases of the novel coronavirus 42 , with 243,303 cases as of May 17, 2020 [1] . However, increasing evidence suggests 43 this burden disproportionately affects certain groups. Provisional data from the Office for 44 National Statistics has indicated significantly increased COVID-19 mortality estimates for Black, 45

Asian, and mixed race/ethnicity individuals, even after indirect adjustment for age, broad 46 geographical region, and measures of self-reported health and disability [2, 3] . There are also 47 disparities in COVID-19 mortality across geographic regions. Based on COVID-19 deaths 48 occurring in March and early April, the age-standardized mortality rate was twice as high in the 49 most deprived areas of England, compared with the least deprived [4] . Importantly, a recent 50 study assessing hypothetical vulnerability to respiratory pathogen pandemicscalculated in 51 terms of likely demand characteristics versus supply-side differences in existing healthcare 52 provisionfound greater vulnerability in regions of England with higher economic deprivation 53 levels [5] . 54 55 Here, we sought to identify whether local-level variation in trajectories (changes over time) in 56 confirmed COVID-19 cases were related to variation in socioeconomic deprivation and self-57 reported racial/ethnic identity. We also assessed whether time courses of caseloads differed 58 according to important population health measuresincluding healthy life expectancy, 59 preventable mortality, and proportion of individuals with chronic health conditions that may 60 increase susceptibility to COVID-19 morbidity and mortality. The ability of different areas to 61 control COVID-19 infection rates over time may provide a more sensitive readout of 62 vulnerability than cumulative case or mortality data, as this time course will depend on a 63 combination of sociodemographic variables and local capacity to enact public health measures 64 that help limit disease spread [5] . Critically, we examined whether differences in demographic 65 measures between trajectories remained significant after adjusting for covariance with other 66 factors that may affect COVID-19 caseload (population density, proportion of older adults, 67 relative deprivation, and preventable mortality rates) [6] [7] [8] . 68 69 70 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020 Methods  71   72   Data on daily laboratory-confirmed COVID-19 cases for the period spanning January 30 to May  73 6, 2020 were accessed via Public Health England [9] . These case counts are reported based on 74 the date the specimen was taken from the patient, not the date the lab carried out testing and 75 reported results to Public Health England. Case data were extracted for upper tier local 76 authorities (henceforth referred to as local authorities), a level of analysis which was chosen in 77 order to balance geographical resolution against availability of up-to-date clinical and 78 sociodemographic data. Case counts were adjusted for local authority population size, using 79 population estimates provided by the Office for National Statistics (ONS), as of April 2019 [10] . 80 81 We identified clusters, or homogenous groups, of local authorities with distinct COVID-19 case 82 trajectories. Clusters were identified using a partition around mediods approach [11] , a clustering 83 analysis for time-series data. A dynamic time warping similarity metric was used to capture 84 nonlinear similarity between two time-series based on their shape. This metric was chosen 85 because it is has been shown in the literature to be optimal for clustering of time series data. 86

Unlike the Euclidean distance which is commonly used for cluster analysis of non-time series 87 data, the dynamic time warping distance is not sensitive to time shifts, which is helpful as some 88 local authorities may have had earlier cases than others. The clustering algorithm partitions local 89 authorities in order to maximize intra-cluster similarities in case trajectories, and maximize inter-90 cluster differences in case trajectories. In order to identify the optimal number of clusters, 2-, 3-, 91 4-, and 5-cluster models were compared using five established cluster validity indices 92 (Silhouette, Score function, Davies-Bouldin, modified Davies-Bouldin, and COP) [12] . 93 94 Sociodemographic, economic deprivation and public health data for each local authority were 95 compiled from Public Health England Fingertips repository using FingertipsR [13] . Specifically, 96 we extracted the most recent data on ethnicity (proportions of Black or Black British ethnic 97 group, Asian or Asian British ethnic group, Mixed/multiple ethnic group and White residents); 98 older adults (proportion of adults aged 65 or older), index of multiple deprivation (IMD, a 99 composite measure of deprivation calculated across multiple domains, including income, 100 employment, education, health, crime, barriers to housing and services, and living environment: 101 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116921 doi: medRxiv preprint [14] ); employment deprivation (proportion of working-age population involuntarily excluded 102 from the labor market, calculated from claimants of various out-of-work social security 103 allowances: [14] ), healthy life expectancy (average number of years a person would expect to 104 live in good health based on contemporary mortality rates and prevalence of self-reported good 105 health); preventable mortality (age-standardized mortality rate from causes considered 106 preventable, per 100,000 people); pre-existing chronic health conditions (proportion of registered 107 patients with a general practitioner (GP)-recorded diagnosis of coronary heart disease, diabetes, 108 hypertension or obesity); self-reported physical activity (proportion of adults completing at least 109 150 minutes of moderate-intensity physical activity per week); and nursing home admissions 110 (permanent admissions to residential and nursing care homes, per 100,000 people aged 65+). 111

Population density was calculated as number of people per land area square kilometer, according 112 to standard area measurements of local authorities available from the Office of National 113

Statistics Open Geography portal [15] . 114 115 Differences in sociodemographic, deprivation and public health measures between the clusters 116 were tested using the nonparametric Kruskal Wallis test. We presented the median and 117 interquartile range (IQR) of the measures in each cluster. In addition, separate logistic regression 118 models were used to determine whether proportions of minority ethnicity groups and deprivation 119 predicted cluster membership, after adjusting for potential confounding variables. Odds ratios 120 (OR) and 95% confidence intervals (CI) for adjusted models are presented. All statistical 121 analyses were carried out using R, version 3.6.1 (R Core Team, 2019). Birmingham, Sheffield and London), however members of this cluster also included less densely 131 populated areas (such as Cumbria, Northumberland, and Herefordshire; Figure 1b) . A second 132 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116921 doi: medRxiv preprint cluster of 71 local authorities followed a lower daily case trajectory. These areas tended to be 133 less densely populated, including the south-west region, but also included several London 134 boroughs (Hillingdon, Richmond, Westminster, Camden, Islington, City of London, Tower 135

Hamlets and Greenwich; Figure 1b, The high case trajectory cluster also had significantly greater population density (p<0.001). This 163 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. to be more prevalent [16] . 206

The geographic disparities in case trajectories were in general consistent with the trend in 208 COVID-19 mortality data, where the death toll in the most deprived areas of England and Wales 209 were approximately twice that in the least deprived areas [17] . Often, populations living in 210 deprivation also are co-living with a higher burden of chronic health conditions that elevate risk 211 for COVID-19 morbidity and mortality, producing a compounded effect [18] [19] [20] . Similar health 212 inequalities in COVID-19 outcomes have been reported elsewhere with regard to socioeconomic 213 and demographic profiles. [21] [22] [23] [24] . Our results are also aligned with estimates based on 214 individual-level data from the UK Biobank, which showed a 2 to 4-fold higher risk of COVID-215 19 infection among Black, Asian, and minority ethnic (BAME) communities than their white 216 counterparts, after adjusting for socioeconomic status, lifestyle, obesity, and comorbidities [25] . 217 218 However, it is important to note that it is difficult to separate racial, socioeconomic, and health 219 contributions when reporting COVID-19 disparitiesand, crucially, that insufficient 220 contextualization of findings may lead to misunderstandings that undermine the goal of 221 eliminating health inequalities [26] . For example, recent analysis from the Office for National 222 Statistics suggests that occupation may pay an important role in COVID-19 exposure and 223 mortalitywith workers in social care, public transport, and sales/retail all found to have 224 increased death rates [27] . The increased exposure to other individuals and reduced ability to 225 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116921 doi: medRxiv preprint maintain physical distancing associated with these roles is likely to interact with findings 226 stratified by ethnicity groups, as there are significantly greater proportions of BAME individuals 227 employed in caring and customer service professions [28] . Further, there are demonstrable long-228 term health implications of structural inequality and discrimination with respect to race and 229 ethnicity. Research has demonstrated that the cumulative effects of physiological responses to 230 external stressors, especially racial discrimination, are linked to increased risk of chronic health 231 conditions such as cardiovascular disease and diabetes (the 'weathering' hypothesis, [29, 30] ) -232 further exacerbating risk from COVID-19 [31, 32] . Finally, migrant groups in the UK may have 233 limited healthcare entitlements, and face additional barriers to healthcare access, including fear 234 of the imposition of high charges and/or sharing of their data with immigration enforcement 235 agencies [3, 33] . 236 237 Interestingly, areas with a high daily case trajectory had a significantly lower proportion of older 238 adults (aged 65 or above), which preliminarily suggests that it may not be older population that is 239 driving the higher burden of COVID-19 infection in these regions. However, as the healthy life 240 expectancy of males is significantly lower in the high trajectory areas, and the admissions into 241 nursing homes is higher in the high trajectory areas, it is possible that older adults in these areas 242 may have poorer health; however, this warrants further study. 243

The burden of COVID-19 deaths has fallen disproportionately in major urban centers -245 particularly London, Liverpool, Birmingham and Manchester [4, 17] . As expected, local 246 authorities in these cities tended to belong to the high case trajectory cluster. However, we note 247 that within London, some wealthier local authorities (e.g. Westminster, City of London) 248 belonged to the lower case trajectory cluster. People with means may be more able to enact 249 physical distancing, and have the ability to leave London, as 250,000 people were estimated to 250 have left London in mid-March prior to the lockdown based on smartphone data [34] . 251 252 This study had some limitations. There may be some local differences in availability and 253 provision of COVID-19 testing, although this is not expected to be a significant limitation, as 254 there is standardized government guidance for members of the public who suspect they have 255 COVID-19, to contact NHS 111 or visit nhs.net, rather than contacting their GP. Further, there 256 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116921 doi: medRxiv preprint may be outflow and inflow of residents to areas which were not their primary residence prior to 257 and during the lockdown period, which could slightly alter the data profile of local authorities. 258

Our ecological study design serves as an important complement to analyses of individual-level 259 patient data, to help understand how structural inequities at the population level affect COVID-260 19 burden. Here, we also adjusted for multiple potential confounders in our models to better 261 identify associations between proportions of ethnicity groups in a local authority, or deprivation 262 of the local authority, and COVID-19 case burden. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116921 doi: medRxiv preprint is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116921 doi: medRxiv preprint sickness, disability or carer responsibilities) represents the most recently available data, from 293 2015. 5 Population density is defined as number of persons per land area square kilometer. 294 6 Healthy life expectancy (average number of years a person would expect to live in good health 295

based on contemporary mortality rates and prevalence of self-reported good health: M, for males; 296 F, for females); 7 Preventable mortality represents the age-standardized mortality rate from causes 297 considered preventable per 100,000 population for 2016-2018 (M, for males; F, for females). 298 8 Chronic conditions: percentage of registered patients with a general practitioner (GP)-recorded 299 diagnosis of coronary heart disease, diabetes, hypertension or obesity; 9 permanent admissions to 300 residential and nursing care homes, per 100,000 people aged 65+. 10 Cumulative COVID-19 301

infections per 1 million (1M) population are those reported as of May 6, 2020. *p< 0.05, **p< 302 0.01; Kruskal Wallis tests. Data provided by Public Health England. 303 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020. income, employment, education, health, crime, barriers to housing and services, and living 308 environment. 4 Employment deprivation scale (proportion of the working-age population in an 309 area involuntarily excluded from the labour market, due to factors such as unemployment, 310 sickness, disability or carer responsibilities) represents the most recently available data, from 311 2015. Odds ratios and associated confidence intervals were calculated using logistic regression 312 adjusted for different covariates. Data provided by Public Health England. 313 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 29, 2020. . https://doi.org/10.1101/2020.05.29.20116921 doi: medRxiv preprint

COVID-19 situation update worldwide, as of 18

COVID-19) related deaths by ethnic group

Asian and Minority Ethnic groups in England are at increased risk of death from COVID-19: indirect standardisation of NHS mortality data

4 Deaths involving COVID-19 by local area and socioeconomic deprivation -Office for National Statistics

Population Vulnerability to Unexpected Health Shocks: Geographical Disparities in England at the Time of COVID-19

Is ethnicity linked to incidence or outcomes of covid-19?

Sharpening the global focus on ethnicity and race in the time of COVID-19

Redefining vulnerability in the era of COVID-19

COVID-19) in the UK

Population estimates -Office for National Statistics

An extensive comparative study of cluster validity indices

fingertipsR: Fingertips Data for Public Health

English indices of deprivation 2019

Standard Area Measurements (2019) for Administrative Areas in the United Kingdom

COVID-19 Attack Rate Increases with City Size

Social Science Research Network 2020

Covid-19: Deprived areas have the highest death rates in England and Wales

Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study

Chronic morbidity, deprivation and primary medical care spending in England in 2015-16: a cross-sectional spatial analysis

Health inequalities: Income deprivation and north/south divides

At the Heart of the Matter: Unmasking and Addressing COVID-19's Toll on Diverse Populations. Circulation;0

Report 22 -Equity in response to the COVID-19 pandemic: an assessment of the direct and indirect impacts on disadvantaged and vulnerable populations in low-and lower middleincome countries

The global impact of COVID-19 and strategies for mitigation and suppression

Ethnicity, comorbidity, socioeconomic status, and their associations with COVID-19 infection in England: a cohort analysis of UK Biobank data

Racial Health Disparities and Covid-19 -Caution and Context

COVID-19) related deaths by occupation

Weathering" and Age Patterns of Allostatic Load Scores Among Blacks and Whites in the United States

The weathering hypothesis as an explanation for racial disparities in health: a systematic review

A glossary for social epidemiology

Racism and discrimination in COVID-19 responses

Medecins du Monde's (MdM) 2019 Observatory Report Left Behind: The State of Universal Healthcare Coverage. MHADRI. 2019

Oxford COVID-19 Impact Monitor

Importance of collecting data on socioeconomic determinants from the early stage of the COVID-19 outbreak onwards