key: cord-257325-pvf0uon3
authors: Zeitoun, Jean-David; Faron, Matthieu; Lefèvre, Jérémie H.
title: Impact of Local Care Environment and Social Characteristics on Aggregated Hospital-Fatality Rate from COVID-19 in France: Nationwide Observational Study
date: 2020-10-10
journal: Public Health
DOI: 10.1016/j.puhe.2020.09.015
sha: 
doc_id: 257325
cord_uid: pvf0uon3

Objectives We aimed to investigate possible differences in aggregated hospital-fatality rate from COVID-19 in France at the early phase of the outbreak, and to determine whether factors related to population or healthcare supply before the pandemic could be associated with outcome differences. Study design Nationwide observational study including all French hospitals from January 24, 2020 to April 11, 2020. Methods We analysed aggregated hospital-fatality rate. A Poisson regression was performed to investigate associations between characteristics pertaining to populational health, socioeconomic context and local healthcare supply at baseline, and the chosen outcome. Results On April 11, 2020, a total number of 30 960 patients were hospitalized among the 3 046 French healthcare facilities, including 6 832 patients in intensive care unit (ICU). A total of 8 581 deaths due to Covid-19 had been recorded, with a median mortality rate per 10 000 people per department of 0.53 (IQR: 0.29-1.90). There were significant variations between the 95 French departments even after adjusting on outbreak inception (p<0.001). After multivariable analysis, four factors were independently associated with a significantly higher aggregated hospital-fatality rate: a higher ICU capacity at baseline (estimate=1.47; p=0.00791), a lower density of general practitioners (estimate=0.95; p=0.0205), a higher fraction of activity from the for-profit private sector (estimate=0.99; p<0.001), and the ratio of people over 75 (estimate=0.91; p=0.0023). Conclusions Aggregated hospital-fatality rate from COVID-19 in France seems to vary among geographic areas, with some factors pertaining to local healthcare supply being associated with outcome.

First cases of coronavirus disease 19 , the viral pneumonia related to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), were officially identified in December 2019 in China and were notified to the World Health Organization (WHO) on December 31, 2020. 1 Since then, the epidemic has expanded well beyond China and the pandemic has officially been declared by the WHO on March 11, 2020. 2 While Italy has been the earliest disease cluster in Europe 3 , France has rapidly followed. On February 23, 2020, the French Ministry of Health issued the phase I of the national epidemic. Phases II and III were respectively announced on February 29, 2020 and March 14, 2020. 4 Fatality rate, defined as the number of deaths of patients in whom COVID-19 was confirmed, divided by the total number of COVID-19 cases, seems to vary among countries. Italian reports have shown a casefatality rate ranging from approximately 7% to 10% 5 , while other countries such as South Korea have observed much lower figures. 6 Even if there is uncertainty due to variations in case recording, we lack definitive explanations for possible differences in case-fatality rates between countries. The number of tests that could be made to screen and insulate patients has been raised as a possible factor contributing to differences. Also, it is not known whether this outcome varies within a country.

Several factors can likely explain differences such as affected population profile, healthcare environment and quality of care. There has been concern in France regarding critical care capacity with respect to the probable high number of simultaneous severe cases during the outbreak peak. 7 It has been estimated by the French Ministry of Health that there were approximately 5,000 intensive care unit (ICU) beds in France yet with differences between regions. Estimates forecasted that this capacity would be exceeded. 7 J o u r n a l P r e -p r o o f Therefore, we sought to measure aggregated hospital-fatality rate from COVID-19 in France, and to examine the association between populational and local healthcare supply characteristics, and this outcome.

We used official and publicly available sources to retrieve and gather the needed data: We also retrieved the number of hospital beds per 10 000 people, including surgery beds, medicine beds, obstetrical beds, physical medicine beds, psychiatry beds and those in long-term care facilities (2017) according to a 2019 report from the French Ministry of Health, 9 and the total number of adult intensive care beds in each department at baseline, i.e. before the outbreak (2020). Last, the fraction of hospital care activity as measured by hospital-days, performed by the for-profit private sector was collected (2017).

For each department, the following health indicators were retrieved: overall mortality 

Aggregated hospital-fatality rate was chosen as study outcome (i.e. for each day of the study period, the number of hospital deaths divided by the number of admitted patients). We chose not to analyze case-fatality rate since it would be unreliable in the French case. Indeed, France has not performed systematic or large SARS-CoV-2 testing, and the number of recorded cases has repeatedly been recognized as being orders of magnitude below actual frequency. Conversely, all serious cases of suspected COVID-19 were required to be tested for confirmation. Hospitalized cases, whether in regular wards or intensive care units (ICUs), therefore represent a reliable denominator for calculation. For each day of study period and in each of the 95 French departments, the number of hospitalized COVID-19 patients and the number of COVID-19 patients in ICUs were collected. Also, for each day of study sample, the J o u r n a l P r e -p r o o f cumulative number of COVID-19-related in-hospital deaths over study period was collected.

To account for gaps in outbreak start between areas, the time origin for each department was set to the first day where at least 10 deaths due to Covid-19 had been recorded in total. To investigate the relationship between our covariates and the selected outcome, a mixed-effects Poisson generalized linear regression was used.

Models were adjusted for the number of people living in the department and the corrected day since the beginning coded as a third order polynomial as fixed effects.

To account for the hierarchical structure of our data, the department (grouping variable) was used as a random effect. Both a random intercept and random slope (for the corrected days since the beginning) were used. Any variable achieving a pvalue < 0.2 in the univariable analysis was proposed in the multivariable model. In 

There were a total number of 3046 healthcare facilities (including public hospitals, Table 1 . The median area of the 95 departments was 5 880 km 2 (IQR: 4 977-6 817 km 2 ).

The study included data from January 24, 2020 (first French case) to April 11, 2020. 

The details of univariate and multivariable analyses are given in Table 1 . Following univariate analysis, eleven factors were included in the multivariable analysis. Apart from the population, four factors were independently associated with a significantly higher aggregated hospital-fatality rate from Covid-19: a higher ICU capacity at baseline (estimate=1.47; p=0.00791), a lower density of general practitioners (estimate=0.95; p=0.0205), a higher fraction of activity from the for-profit private sector (estimate=0.99; p<0.001) and the ratio of people over 75 (estimate=0.91; p=0.0023). No health indicator was associated with our outcome in the multivariable analysis.

In this nationwide observational study regarding COVID-19 in France, we found significant differences between areas in terms of aggregated hospital-fatality rate.

Four factors were associated with our study outcome: a higher density of ICU beds at baseline, a lower fraction of hospital care activity from the for-profit private sector, a J o u r n a l P r e -p r o o f lower density of general practitioners, and a greater proportion of people over 75 were all predictors of higher aggregated hospital-fatality rate in the current model.

Our study has several strengths. First, it is a nationwide analysis gathering exhaustive data from reliable sources. For most of covariates, year of availability was very recent, thereby limiting timeliness issues. In addition, the variables of interest are unlikely to significantly change across a relatively short period of time. Second, we collected a very diverse set of data regarding demographics, populational health, wealth, and also characteristics of care supply and local healthcare ecosystems. Populational health data were in particular critical to incorporate in the model since they are factors likely to influence disease outcome. We had very fine health data beyond age, namely prevalence of chronic conditions that have already been recognized as risk factors for COVID-19 outcome. 3, 11, 12 Third, we used a robust statistical model to analyse the data, namely a Poisson linear model as the variables were daily counts and a mixed model as the observed data were not independent (repeated measures within a department), which allows separate intercept and slopes for each department. Also, time-adjustment was made so as to align all departments on a similar basis and take into account timeliness issues.

Our findings have implications. Critical care capacity has been a matter of concern regarding COVID-19 outbreak. It has been predicted that France did not have enough ICU beds to absorb all of the patients in need along several days or weeks. Yet we found no evidence that less ICU beds at baseline in a given area were associated with a worst outcome. Conversely, we found that areas with an initial higher density of ICU beds were associated with a higher aggregated hospital-fatality rate. We do not have any certain explanation for those unexpected findings. It may be that critically ill patients were more often transferred from rural areas or smaller facilities to more J o u r n a l P r e -p r o o f comprehensive facilities. It also should be underlined that hospitals have anticipated the outbreak progression by resetting their organization and creating new ICU capacity in other wards. We could not measure actual ICU beds at a given time since those data were not consistently reported. This will need further investigation. We also found that areas in which the density of general practitioners was higher were associated with a better outcome. Even though this should be interpreted with caution, one may hypothesize that general practitioners played a critical role in the epidemic, through adequate orientation of COVID-19 patients to hospitals while maintaining others at home. Last, it is remarkable that social and wealth factors were not associated with the chosen outcome. The relationship between wealth and health has been consistently documented by a huge body of literature. Again, we cannot certainly explain why herein departments with more deprivation were not associated with a higher aggregated hospital-fatality rate yet it should be recalled that France has a very protective social system with a great safety net. Perhaps it helped to attenuate the social risk in the case of the epidemic.

This study has limitations. Firstly, as an observational study, it cannot establish definitive causality. We cannot exclude the possibility that our results might be confounded by factors that were not measured. In particular, we cannot rule out that criteria for admitting patients were different among areas and that some hospitals had more serious cases than others, whether in regular wards or ICUs. Also, we did not have access to age-and gender-structure of hospitalized patients. Last, we did not take into account control measures implemented in the different departments even though those measures were thought to be very similar. Secondly, the follow-up was intentionally limited. However, given the high urgency that many healthcare systems are currently facing worldwide, we aimed at rapidly providing a first evaluation of J o u r n a l P r e -p r o o f hospital-fatality rates from COVID-19 in a markedly affected country. Subsequent work over the outbreak course will say whether local differences and their associated factors persist. Thirdly, we did not have access to hospital data or patient data. Thus, we could not calculate individual hospital-fatality rate and had to deal with aggregate measures which have been updated on a daily basis at the department level over the study period. Fourth, we intentionally excluded nursing home since the related data were not available across the whole study period. This represents a selection bias.

Last, as of March 28, 2020, the French government decided to implement targeted transfers of seriously ill patients by medicalized trains or helicopters in order to improve resource allocation within the whole territory. Those transfers may have interfered with our results even though we believe it is unlikely. Indeed, reported counts of those transfers showed it involved very few patients as compared to the magnitude of the epidemic. It seems implausible that it significantly influenced the findings from the regression analysis, which were otherwise consistent over time.

In conclusion, we found significant differences in aggregated hospital-fatality rate across French areas over the early period of the COVID-19 outbreak. Several factors pertaining to local healthcare supply were associated with a worst outcome, such as a higher ICU capacity at baseline and a lower involvement from the private sector as well as a lower density of general practitioners. Those findings clearly deserve further investigation with hospital-or patient-level data and over a longer follow-up. Those departments have been chosen to illustrate the heterogeneity of situations across the whole French territory (see Figure 1 ).

World Health Organization. WHO Director-General's opening remarks at the media briefing on COVID-19 -11

Critical Care Utilization for the COVID-19

Early Experience and Forecast During an Emergency Response

Arrêté du 14 mars 2020 portant diverses mesures relatives à la lutte contre la propagation du virus covid-19 2020

Case-Fatality Rate and Characteristics of Patients Dying in Relation to COVID-19 in Italy

Transmission potential and severity of COVID-19 in South Korea

Coronavirus : les simulations alarmantes des épidémiologistes pour la France

Health as an independent predictor of the 2017 French presidential voting behaviour: a crosssectional analysis

Les établissements de santé -édition

The Association Between Income and Life Expectancy in the United States

Clinical Characteristics of Coronavirus Disease 2019 in China

Characteristics of and Important Lessons From the

COVID-19) Outbreak in China: Summary of a Report of

Cases From the Chinese Center for Disease Control and Prevention