Estimating the burden of influenza‐related and associated hospitalizations and deaths in France: An eight‐season data study, 2010–2018

Abstract Background In France, each year, influenza viruses are responsible for seasonal epidemics leading to 2–6 million cases. Influenza can cause severe disease that may lead to hospitalization or death. As severe disease may be due to the virus itself or to disease complications, estimating the burden of severe influenza is complex. The present study aimed at estimating the epidemiological and economic burden of severe influenza in France during eight consecutive influenza seasons (2010–2018). Methods Influenza‐related hospitalization and mortality data and patient characteristics were taken from the French hospital information database, PMSI. An ecological approach using cyclic regression models integrating the incidence of influenza syndrome from the Sentinelles network supplemented the PMSI data analysis in estimating excess hospitalization and mortality (CépiDc—2010–2015) and medical costs. Results Each season, the average number of influenza‐related hospitalizations was 18,979 (range: 8627–44,024), with an average length of stay of 8 days. The average number of respiratory hospitalizations indirectly related with influenza (i.e., influenza associated) was 31,490 (95% confidence interval [CI]: 24,542–39,012), with an average cost of €141 million (range: 54–217); 70% of these hospitalizations and 77% of their costs concerned individuals ≥65 years of age (65+). More than 90% of excess mortality was in 65+ subjects. Conclusions The combination of two complementary approaches allowed estimation of both influenza‐related and associated hospitalizations and deaths and their burden in France, showing the substantial impact of complications. The present study highlighted the major public health burden of influenza and its severe complications, especially in 65+ subjects.

decrease transmission are also implemented during epidemics, such as handwashing or masks. Estimating the global influenza burden in terms of excess hospitalization and death and the economic consequences is therefore important for decision-makers, to optimize prevention policies. A recent study in the United States estimated that the average annual total economic burden of influenza was as high as $11.2 billion. 6 In the present study, the epidemiological and economic burden of hospitalization and in-hospital mortality directly due to influenza (i.e., influenza related) as calculated from the French healthcare databases 7,10 was supplemented by the burden calculated from excess hospitalization and mortality indirectly due to influenza (influenza associated) and related costs.

| METHODS
To evaluate burden of disease, we used two approaches. The first was the evaluation of the burden of influenza in the hospital, in terms of the number of influenza-related hospitalizations and in-hospital deaths. The second was the evaluation of the burden of influenza in the hospital, in terms of influenza-associated excess hospitalization and mortality.

| Data and data sources
The study was conducted using the French hospital databases named PMSI recording medical information about all the hospitalizations performed annually and covering reimbursement information for each public or private hospital since 2007. PMSI is part of the SNDS national health data system database, and the exhaustive CépiDc For mortality data, in the direct approach, all deaths occurring during a hospital stay for influenza (ICD-10 codes: J09-J11) were included. In the excess burden approach, weekly age-specific underlying cause-of-death data for the whole French population were obtained from death certificates collected by Inserm-CépiDc from June 2010 through June 2015 (last year with available data). 11 We compiled deaths from P&I (ICD-10 codes: J09-J18), respiratory diseases (ICD-10 codes: J00-J99), cardiovascular diseases (ICD-10 codes: CIM-10 I00-I99), and all causes based on the ICD-10.
For demographic data, we used yearly population data obtained from the National Institute of Statistics and Economic Studies (INSEE) to calculate age-standardized hospitalization or death rates, 12 using the June 2017 French population structure as a reference.
Influenza-like-illness (ILI) incidence in conjunction with documented cocirculation of influenza viruses has been shown to be a good proxy for influenza incidence in France and elsewhere 13 and an appropriate covariate in mortality burden models. 14 Weekly ILI incidence was obtained from the French Sentinelles surveillance network from 2011 to 2018. This surveillance network relies on volunteer general practitioners (1% of all general practitioners), who have been reporting medical consultations for ILI and other infections since 1984. 5 The ILI case definition consisted of a combination of fever >39 C, myalgia, and respiratory symptoms as per the French Sentinelles definition. 15  For the epidemiological burden, in the direct burden approach, the incidence and the number and proportion of patients hospitalized for influenza (DP-DR coded with ICD-10 J09-J11 or DAS coded with ICD-10 J09-J11) 16 were estimated. To estimate the mortality burden of influenza, in-hospital deaths for all causes were counted for each July-June respiratory season and expressed as a hospital mortality rate. In the excess burden approach, several indirect approaches have traditionally been used to estimate excess mortality due either to influenza or to its complications, but the preferred model was to explicitly model weekly mortality data against weekly indicators of influenza activity. 15 This ecological approach, 15 commonly used to estimate mortality impact, was applied here to both mortality and hospitalization data. [17][18][19] Precisely, different Poisson cyclic models (time series) were built, where age-and cause-specific hospitalization as well as mortality data were explained by ILI incidence, 17,19-21 time trends, and seasonal terms, using a log link. 8 (Table 1). Incidence varied between 13 and 63 per 100,000 persons.

| Excess burden approach or burden of influenza-associated hospitalization and death
Regarding performance of the model and model fitting, the hospitalization models showed good performance, with high correlations between observed and predicted values and low mean absolute percentage error (MAPE) for P&I hospital stays and respiratory causes regardless of age group (Table S1 and Figure S1). The fit between the observed and predicted values for cardiac and all causes ( Figure S1) was lower. This was particularly true among younger subjects.
The mortality models showed high correlation between observed and predicted values and MAPE for the older age groups for P&I, respiratory, cardiac, and all causes. For younger subjects (0-4 and 5-19 age groups), correlations and MAPE were weaker (correlations between 0.48 and 0.76 and MAPE between 44% and 54%), compared with older age groups (Table S1).
Regarding estimated excess influenza-associated hospitalization, the standardized P&I and respiratory excess hospitalization rates were always higher in 65+ compared with all ages taken together.  (Table 3).

| Direct burden approach
Costs varied substantially depending on the epidemic season, from  (Table 4). During the severe epidemic seasons, total cost was 4-fold to 5.5-fold higher in 65+ than 65À (€171.8 million vs. €35.0 million for respiratory causes) ( Table 4).
In 65+, for the severe epidemic seasons, total excess costs attributable to influenza accounted for 34% and 23% of the elderly P&I and respiratory total costs, respectively.

| DISCUSSION
To the best of our knowledge, this was the first national study to inte- are well-known worldwide. [22][23][24][25] Like in other studies, influenza epidemics due to H1N1 or H1N1/B affected particularly young people, whereas H3N2 strains affected particularly elderly people, where they were associated with high mortality. 26 In addition, neither median hospital stays nor the proportion of mortality directly related to the influenza hospitalization varied with epidemic severity. However, influenza can lead to complications in vulnerable people, including the elderly, sometimes some weeks after the actual influenza episode, and in this case, the death is not officially associated with influenza in the death certificate. 27 This was the reason for our indirect analysis (cyclic regression), commonly used to estimate the impact of influenza on mortality in the elderly. 17,20,[28][29][30][31] This provided substantial complementary information and refined the results of the direct approach, by showing a consistent relationship between the excess mortality attributable to influenza and the severity of the epidemic. Less than 20% of deaths were identified by the direct approach in the 65+, whereas they accounted for 90% of the estimated all-cause mortality attributable to influenza, showing the importance of the indirect burden in this vulnerable population. Results confirmed findings from other studies in high-and middle-income countries, which suggested a moderate contribution of influenza to total winter mortality among seniors, estimated at around 5% (range 2.2-16%) in the present study. 25,32,33 Finally, findings suggested a substantial contribution to respiratory winter mortality (around 14%).
We applied those models to estimate the excess hospitalization associated with influenza. The most specific cause of influenzaassociated hospitalization (P&I) showed a very good fit between the direct (hospital stays directly attributed to influenza) and indirect approach (rate of 12 vs. 13 and 62 vs. 63 per 100,000 for 2013-2014 and 2017-2018, respectively). This is a strong argument in favor of our model's validity, at least for P&I and respiratory causes. On the other hand, the lower precision of the estimates of the cardiac and allcause models suggests that these models were not specific enough to provide robust estimates of the excess influenza-associated hospitalization for cardiac causes and all causes. Such models have been found to be relevant to predict the cardiac hospitalization excess, but mostly in 80+. 28 Overall, the direct burden corresponded to 55% excess hospitalization attributable to influenza for all ages, but only 33% for the 65+.
The study further assessed the economic counterpart of both the epidemiological burden and excess hospitalization directly and indirectly associated with influenza. Consistently, associated costs differed with epidemic severity, with a cost 3-4 times higher on average in severe than low or moderate severity seasons (all ages). In agreement with the epidemiological burden, this trend was even more marked in the elderly population, which accounted for 83% of costs during severe epidemic seasons. In addition, the direct cost of  Abbreviation: P&I, pneumonia and influenza.