key: cord-0720188-8l2ed3vq
authors: Dahal, Sushma; Mizumoto, Kenji; Rothenberg, Richard; Chowell, Gerardo
title: Investigating spatial variability in COVID-19 pandemic severity across 19 geographic areas, Spain, 2020
date: 2020-04-17
journal: nan
DOI: 10.1101/2020.04.14.20065524
sha: 17c28053e15154edd92cfee8b349e40005a5452b
doc_id: 720188
cord_uid: 8l2ed3vq

Background: Spain has been disproportionately affected by the COVID-19 pandemic with the second highest death toll in the world after Italy. Here we analyzed estimates of pandemic severity and investigated how different factors shaped the severity of the COVID-19 pandemic. Methods: We retrieved the daily cumulative numbers of laboratory-confirmed COVID-19 cases and deaths in Spain from February 20, 2020 to April 2, 2020. We used statistical methods to estimate the time-delay adjusted case fatality ratio (CFR) for 17 autonomous areas and 2 autonomous cities of Spain. We then assessed how transmission and sociodemographic variables were associated with the CFR across areas in Spain using multivariate regression analysis. Results: We estimated the highest time-delay-adjusted CFR for Madrid (38.4%) and the average adjusted CFR in Spain at 23.9%. Our multivariate regression analysis revealed a statistically significant three predictor variables: infant mortality rate, poverty risk rate and the cumulative morbidity rate. Conclusions: Our estimates of the time-delay adjusted CFR for 12 autonomous areas/cities in Spain are significantly higher than those previously estimated for other geographic regions including China and Korea. Our results call for urgent public health interventions focusing on low socioeconomic groups to ameliorate the burden of the COVID-19 pandemic in Spain.

The case fatality ratio is a useful metric to assess pandemic severity, which is typically estimated as the proportion of deaths among the total number of cases attributed to the disease [2] . However, during the course of outbreak of an infectious disease outbreak such as COVID-19, real-time estimates of CFR need to be derived carefully since it is prone to ascertainment bias and right censoring [2, 3] . In particular, the disease spectrum for COVID-19 ranges from asymptomatic and mild infections to severe cases that require hospitalization and specialized supportive care. This may lead to overestimation of the CFR among ascertained cases. On the other hand, there is a delay from illness onset to death for severe cases [4] , which could lead to an underestimation of the CFR [3, 5] . Therefore, statistical methods that help mitigate inherent biases in estimates of the CFR should be employed to accurately plan for medical resources such as ICU units and ventilators, which are essential resources to save the lives of critically ill patients [6] [7] [8] .

Several studies have reported CFR estimates for COVID-19 [9] [10] [11] . Overall, these estimates have varied substantially across geographic regions even within the same country. For example, a recent study estimated the time-delay-adjusted CFR at 12.2% for the ground zero of the COVID-19 pandemic: the city of Wuhan [4] , whereas for the most affected region in Italy (Northwest), the delay-adjusted CFR reached 23.0% [12] . The drivers behind the geographical variations in the severity of the COVID-19 pandemic are yet to be investigated, but could . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint provide critical information to mitigate the morbidity and mortality impact of this and future pandemics [13] .

In this study we aim to estimate the severity of COVID-19 pandemic across 19 geographic areas in Spain and aim to explain how these estimates varied geographically as a function of underlying factors. For the real-time estimation of severity, we adjust for right censoring using established methods [14, 15] and report the estimates of the time-delay adjusted CFR of COVID-19 for 17 autonomous areas and 2 autonomous cities of Spain as well as for the entire Spain. We then assessed the association between different transmission and socio-demographic factors and the estimated CFRs across areas using multivariate regression analyses.

. 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 April 17, 2020. The Ministry of Health of Spain releases daily report on COVID-19 cases and deaths [19] .

From these reports we retrieved the daily cumulative numbers of reported laboratory-confirmed COVID-19 cases and deaths from February 20, 2020 to April 2, 2020.

We then stratified the data into 20 groups that included 17 CCAA, 2 African autonomous cities and for the entire Spain.

For each CCAA we obtained data on total population size, proportion of the population older than 60 years, proportion of population at risk of poverty, and infant mortality rates from the . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint Statistics National Institute (Instituto Nacional de Estadística) [20] . We also obtained the data on the total area of each CCAA [21] , and percentage of total consolidated expenditure on hospital and specialized services in the CCAA from the annual report issued by the national health system, 2018 [22] . Finally, we also included two transmission-related metrics: the COVID-19 initial growth rate during the 15 days of local transmission and the cumulative morbidity rate given by the cumulative number cases divided by the local population size.

Additionally we obtained the shapefiles of the autonomous areas of Spain from the national geographic information system of Spain [23].

The crude CFR is defined as the number of cumulative deaths divided by the number of cumulative cases at a specific point in time. For the estimation of CFR in real time, we employed the delay from hospitalization to death, h s , which is assumed to be given by h s = H(s) -H(s-1) for s>0 where H(s) is a cumulative density function of the delay from hospitalization to death and follows a gamma distribution with mean 10.1 days and SD 5.4 days, obtained from the previously published paper [4] . Let π a,ti be the time-delay adjusted case fatality ratio on reported day t i in area a, the likelihood function of the estimate π a,ti is . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint ୀ ଵ

where c a,t represents the number of new cases with reported day t in area a, and D a,ti is the cumulative number of deaths until reported day t i in area a [14, 15] . . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint

We also explored the association between time-delay-adjusted CFR with population size, population density, proportion of population aged more than 60 years, infant mortality rate, population at risk of poverty as measured by poverty risk rate, consolidated public health expenditure on hospital and specialized services as well as with two transmission-related metrics: the cumulative morbidity rate of COVID-19 and the initial incidence growth rate across CCAAs. For this analysis, we built a multivariate linear regression model with all predictor variables to identify simplified models with significant factors linked to the variation in CFR estimates across geographic areas in Spain. We used stepwise regression method to build a final model that contained significant predictors.

All statistical analyses were conducted in R version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria).

As of April 2, a total of 117,710 cases and 10,935 deaths due to COVID-19 have been reported in Spain. Moreover, the Madrid region has reported the highest number of cases at 34,188 (29%) and deaths at 4,483 (41%) followed by Catalunya with 23460 cases (19.9%) and 2335 deaths (21.3%).

. 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint adjusted CFR varies considerably for different areas. For instance, as the epidemic progresses, the adjusted CFR increases slightly in CN and NC while it shows the upward trend for first two weeks and then declines very slowly as in MD. Similarly, for AR the graph shows the downward trend, probably due to misdiagnosis of cases until the first reports of deaths.

Likewise, for CM the graph shows upward trend for first two weeks, followed by a relative decline and then again moves upwards and then stays stable.

A summary of the time delay adjusted case fatality ratio, range of median estimates and crude CFR of COVID-19 across different areas of Spain are presented in figure 3 ).

Autonomous areas with higher proportion of population at risk of poverty, areas with a higher infant mortality rate and areas with higher cumulative morbidity rate experienced higher CFRs.

These three significant factors explained 62.2% (62.2 is multiple R-squared and adjusted R-squared is 53.5%) of variance in the pandemic severity across CCAAs (P < 0.05, . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint Table 2 ). Figure 4 demonstrates the model-adjusted CFR, infant mortality rate, cumulative morbidity rate and poverty risk rate within the map of Spain for 17 CCAAs.

In this paper, we have Our findings suggest the need for additional control efforts and medical resources particularly for lower socio-economic areas which have been particularly hit hard by the COVID-19 pandemic.

The adjusted CFR estimates in Spain is higher than the estimates for Wuhan (12.2%) [4] , Korea

(1.4%) [26] and slightly less than the estimates for Northwest Italy (31.4%) [12] . However when we compare the estimates for the most affected areas across different countries, the rate . 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 April 17, 2020. We found a significant positive association between CFR and the infant mortality rate, the poverty risk rate and the COVID-19 cumulative morbidity rate across areas in Spain. In fact, these three variables explained more than 50% of the geographic variation in CFR. Infant mortality rate is an important indicator of an overall health of society while poverty risk rate reflects the socio-economic status of an area. In any pandemic situation like COVID-19, the poorer tend to exhibit the highest morbidity and mortality rates. For instance, lower . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint socioeconomic groups were also disproportionately affected by the 1918 influenza pandemic [34, 35] .

Those with the poor economic status have higher odds of having pre-existing conditions such as cardiovascular diseases, obesity, diabetes, and cancer [36] [37] [38] . According to WHO-China-joint mission on COVID-19, the patients with no comorbid conditions had a crude CFR of 1.4% compared to very higher rates among those with preexisting conditions. For example, 13.2% for those with cardiovascular disease, 9.2% for diabetes, 8.4% for hypertension, 8.0% for chronic respiratory disease, and 7.6% for cancer [10] . Moreover, preliminary COVID-19 mortality data from the US also indicates a 2-fold age-adjusted death rate among Hispanic/Latino and 1.9-fold among Black/African American compared to Whites [39] .

In our study we saw considerable variations in CFR trend across areas. For instance, as the epidemic progressed, the adjusted CFR showed a slightly upward trend in CN and NC, a rapid upward trend followed by the slow decline in MD, and a downward trend in AR. Likewise, for CM the graph showed an upward trend followed by a relative decline and then again an upward trend before staying stable. The CFR trend for the 19 autonomous areas can be helpful in the planning and implementation of health care services and prevention measures separately for . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint each of them. The downward trend in CFR as seen in some of the areas in our study suggest the improvement in epidemiologic surveillance leading to the increased capture of mild or asymptomatic cases. A higher number of mild and asymptomatic cases also indicate an increase in human-to-human transmission leading to a prolonged epidemic which can be controlled through effective social distancing measures until an effective vaccine or treatment becomes available [4] .

The upward trend in CFR indicates that the temporal disease burden exceeded the capacity of healthcare facilities and the surveillance system probably missed many cases during the early phase of the epidemic [4] , particularly due to a significant presence of mild and asymptomatic cases. It has been found that about 18% of the COVID-19 infections in Diamond Princess Cruise ship were asymptomatic [40] . The increasing trend in CFR could further be explained the nosocomial transmission affecting the health care workers, inpatients and their visitors [4] .

In China, of 44672 confirmed COVID-19 cases, 3.8% was among the health care personnel [41] . Similarly, Wang et al. in their study suspected 41% of the patients to have human-to-human hospital associated transmission of COVID-19 [42] .

Our study has some limitations. The preferential ascertainment of severe cases bias in COVID-19 may have spuriously increased our estimate of CFR [3] , which is a frequent caveat . 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 April 17, 2020. . https://doi.org/10.1101/2020.04. 14.20065524 doi: medRxiv preprint in this type of studies [43, 44] . Similarly, given the long infection-death time for COVID-19 which ranges between 2 to 8 weeks [29] , our estimate may have been affected by delayed reporting bias [3, 5] . Similarly, in our data, the date of report reflects the date of reporting and not the date of onset of illness. Finally, we assumed infant mortality and poverty risk rate as a proxy for areas with low socio-economic groups.

The risk of death due to COVID-19 in Spain was estimated at 23.9%, but estimates varied substantially across 19 geographic areas. The CFR was as high as 38% in Madrid (38.4%), and Castilla-La Mancha areas and as low as 4% in Melilla and 8% in Galicia and Murcia. Of the 19 autonomous areas/cities, 16 had a time-delay-adjusted CFR greater than 10% reflecting a disproportionate severity burden of COVID-19 in Spain. Importantly, our estimate of CFR for the most affected Madrid region is higher than previous estimates for the most affected areas within China, Korea, and Italy. Our findings suggest a significant association of factors such as infant mortality rate and poverty risk rate with the increased risk of death due to COVID-19.

Further studies with patient level data on mortality, and risk factors could provide a more detailed understanding of the factors shaping the risk of death related to COVID-19.

. 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 April 17, 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 April 17, 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint . 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 April 17, 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint . 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 April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint

Estimating the risk of Middle East respiratory syndrome (MERS) death during the course of the outbreak in the Republic of Korea

Potential biases in estimating absolute and relative case-fatality risks during outbreaks. PLoS neglected tropical diseases

Estimating Risk for Death from 2019 Novel Coronavirus Disease, China

Case fatality rate for Ebola virus disease in west Africa. The Lancet

Defining the epidemiology of (COVID-19)

Clinical, laboratory and imaging features of COVID-19: A systematic review and meta-analysis

Spatial variability in the risk of death from COVID-19 in Italy

Spatial-temporal excess mortality patterns of the 1918-1919 influenza pandemic in Spain. BMC infectious diseases

Early epidemiological assessment of the virulence of emerging infectious diseases: a case study of an influenza pandemic

Dynamics of the pneumonic plague epidemic in Madagascar

First case of coronavirus confirmed in Spain

Epidemiologic evidence for the relation between socioeconomic status and depression, obesity, and diabetes

Socioeconomic status and cardiovascular disease: risks and implications for care

Impact of socioeconomic status on cancer incidence and stage at diagnosis: selected findings from the surveillance, epidemiology, and end results: National Longitudinal Mortality Study. Cancer causes & control

Age adjusted fatal lab confirmed COVID-19 cases per 100,000 by race/ethnicity group

Estimating the asymptomatic proportion of coronavirus disease 2019 (COVID-19) cases on board the Diamond Princess cruise ship

Characteristics of and important lessons from the coronavirus disease 2019 (COVID-19) outbreak in China: summary of a report of 72 314 cases from the Chinese Center for Disease Control and Prevention

Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan

6% (95%CrI: 9

CL) 27.9%

0% (95%Crl: 6.9-7.2%)

GA)

Melilla(ME)

9% (95%CrI: 1.7-7.1%)

PV) 16.9% (95%CrI: 16.4-17.5%)

8%

95% credibility intervals (CrI), ¶ 95%CI: 95% confidence interval

Covid-19-studies needed. New England Journal of Medicine, 2020.

Reich, N.G., et al. . 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 April 17, 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 April 17, 2020. 

Worldometer, Spain population. 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. (which was not certified by peer review)The copyright holder for this preprint this version posted April 17, 2020. . https://doi.org/10.1101/2020.04.14.20065524 doi: medRxiv preprint