Excess all-cause mortality during the first wave of the COVID-19 epidemic in France, March to May 2020

Through a weekly all-cause mortality surveillance system, we observed in France a major all-cause excess mortality from March to May 2020, concomitant with the coronavirus disease (COVID-19) epidemic. The excess mortality was 25,030 deaths, mainly among elderly people. Five metropolitan regions were the most affected, particularly Île-de-France and the Grand-Est regions. Assessing the excess mortality related to COVID-19 is complex because of the potential protective effect of the lockdown period on other causes of mortality.


A reactive all-cause mortality surveillance
Since 2005, the French public health agency (Santé Publique France) has been following up the all-cause mortality routinely, based on the administrative data transmitted by the French National Institute for Statistic and Economic Studies (INSEE). The mortality surveillance is based on ca 3,000 civil status offices of city halls dispatched all over the territory (including overseas) [1]. For each death, administrative information is collected (age, sex, date and place of death, date of registration). The medical causes of death are not recorded. This sample from 3,000 cities represents 77% of the total French mortality, varying from 42% to 98% in the 100 districts [1]. In order to analyse mortality at the level of all cities, the number of deaths recorded from the sample in each district was divided by the coverage rate of the system at the district level. Numbers of deaths estimated at district level were aggregated to obtain estimated numbers of deaths at regional and national levels.
We obtained the excess all-cause mortality from the difference between the estimated number of deaths and the expected number of deaths (baseline). The weekly expected number of deaths was provided by a statistical model developed by the European monitoring of excess mortality for public health action network (EuroMOMO, www.euromomo.eu) [2]. This common model is used weekly by 24 European countries or regions and uses a time-series Poisson regression to predict the baseline, adjusted for a linear or nonlinear trend and seasonal variation.
The excess deaths are expressed in proportion to the baseline (excess mortality × 100, divided by the baseline). A standardised indicator around the baseline (z-score) is also calculated, enabling standardised comparison of mortality variations at different geographical levels. The analysis is stratified by age groups (< 15, 15-44, 45-64, 65-84 and ≥ 85 years) and sex at national and regional levels.
The use of mortality data in the frame of public health surveillance and epidemiological studies has been authorised by the French National Commission for Data Protection and the Liberties (CNIL).

Estimates of excess all-cause mortality
From week 1 to week 10 in 2020, the estimated number of deaths varied around the expected value within 2 z-scores. The study period covered 2 March to 31 May 2020 (calendar weeks 10-22). Mortality increased sharply in week 11 in 2020 (crossing 2 z-scores) until week 14 in 2020. In week 14, the estimated number of deaths was 7,100 deaths (+60%) above the expected number of deaths (n = 11,800 deaths) ( Figure 1). The mortality then decreased until week 18. The estimated number of deaths returned to the expected range from week 18 to week 22 (Figure 1). Overall for the study period, the estimated number of deaths was 16.6% above the baseline for the whole of France (n = 25,030 excess deaths) (Table). Elderly people (≥ 65 years) were the most affected ( Figure 2). From week 10 to week 22 in 2020, 23,400 excess deaths were estimated in this age group (+18.2%), while persons aged 15-64 years experienced a moderate excess of deaths (+1,510 deaths, +7% compared with baseline) (Table). In this age group, the most affected were also the oldest (45-64 years) and more specifically men (+12% vs +5% in women compared with the baseline) (Figures 2 and 3). The estimated number of deaths in persons aged 15-44 years was comparable to the expected number, for both men and women. In elderly people, the proportion of allcause excess deaths for men (19.1%) was greater than for women (16.7%). The estimated number of deaths in children (< 15 years-old) was lower than the expected value (−14%, n = −170 deaths), particularly for boys (−20%).
In comparison with the five full previous seasons, the cumulative all-cause excess mortality was close to three other seasons (2014/15, 2016/17 and 2017/18) for adults older than 85 years and to season 2017/18 for adults aged 45-64 years, but was much greater for 65-84 year-old ( Figure 2).
Five other regions experienced a moderate to elevated excess all-cause mortality with z-scores varying from 2 to 5.4. Eight regions had no observed excess mortality. Among these last regions, the estimated number of deaths was significantly lower than the baseline (ca −4%) in two regions (Nouvelle-Aquitaine and Brittany).
Spatial disparities were noticed within the regions, with 12 districts experiencing an exceptional excess mortality, including all Île-de-France districts ( Figure  4). Excess mortality among adults aged 15-64 years was predominantly observed in Île-de-France.

Discussion
In the absence of medical information in civil status data, it cannot be estimated which part of the all-cause excess mortality is attributable to COVID-19. During this unusual pandemic, two complementary surveillance systems were implemented to report COVID-19 deaths in hospitals and long-term care facilities in France [3], but no system was able to report the COVID-19 mortality at home. A total of 29,200 COVID-19 deaths were reported by these two systems during the study period   [4]. This number of COVID-19-related deaths may be overestimated, since virological testing (PCR) was not systematically performed for all deaths reported by long-term care facilities.
The difference between our estimate of the all-cause excess mortality and the reported number of COVID-19 deaths in hospitals and long-term care facilities reflects the complexity to precisely estimate the epidemic impact since COVID-19 induced: • an increase in deaths directly associated to severe acute respiratory disease coronavirus 2 (SARS-CoV-2), • an increase in deaths due to other severe health conditions left untreated because of the impact of the epidemic on the health system and the lockdown, • a decrease in mortality due to the suspension of business, personal and tourism trips and activities during the lockdown period, • a decrease in mortality following an earlier death of the most frail persons who could have died from COVID-19 several days or weeks before they would have died without the epidemic.
The baseline corresponds to a mortality situation in absence of events that influence mortality. In the epidemic context, the lockdown period may have a protective effect on mortality and the number of deaths during the epidemic may be lower than the calculated baseline. The protective effect may concern external causes of deaths (such as traffic or occupational accidents), deaths related to air pollution or to other communicable diseases. This hypothesis is consistent with the situation observed in two regions with a lower mortality than expected that started 3 weeks after the beginning of the lockdown and remained until the end of the lockdown period and even longer in one of the two regions. Such results were also reported in Portugal, for persons younger than 55 years [5]. Further analysis using medical causes of deaths will allow disentangling the impact of lockdown on severe illnesses.   Different effects causing increases and decreases in mortality occur simultaneously during the same period. Providing an estimate for each effect is a great challenge. All-cause mortality data are insufficient to separate COVID-19-related deaths (directly or indirectly associated with the epidemic) from deaths related to other causes, but do provide rapid and valid information on excess mortality impact.
Spatial disparities in the excess mortality were mainly related to the spatial spread of the epidemic (data not shown) [4]. Further studies will be required to examine other factors that could explain disparities, in particular in Île-de-France where an excess mortality in 15-64-year-old adults was observed. Those multifactorial effects are mainly related to health status and socio-demographic and economic characteristics of the population residing in these territories, as well as medical and to do with health care geography.
Our results by age group were consistent with previous studies that have reported a major impact on the elderly [5][6][7][8], but they differ for children. For children, our study showed a significantly smaller number of deaths than the baseline, whereas at the European level and in England, the number of deaths during the epidemic remained comparable to the baseline [6,7]. The estimate of excess mortality by sex is rarely explored. In Italy, the excess among men was also higher than among women [8].
Overall, excess all-cause mortality is an essential indicator of disease impact, particularly during unusual events such as the current pandemic. All-cause mortality monitoring contributes to guide public health decisions and actions in order to reduce its impact at national and local level. The French surveillance system also contributes to the excess mortality surveillance at the European level through the EuroMOMO network that is particularly of interest in case of international threats [7]. High excess (5 ≤ Z-score <7) Very high excess (7 ≤ Z-score <12) Exceptional excess (Z-score ≥ 12)