Influenza epidemics observed in primary care from 1984 to 2017 in France: A decrease in epidemic size over time

Background Epidemiological analysis of past influenza epidemics remains essential to understand the evolution of the disease and optimize control and prevention strategies. Here, we aimed to use data collected by a primary care surveillance system over the last three decades to study trends in influenza epidemics and describe epidemic profiles according to circulating influenza viruses. Methods Influenza‐like illness (ILI) weekly incidences were estimated using cases reported by general practitioners participating in the French Sentinelles network, between 1984 and 2017. Influenza epidemics were detected by applying a periodic regression to this time series. Epidemic (co‐)dominant influenza virus (sub)types were determined using French virology data. Results During the study period, 297 607 ILI cases were reported allowing the detection of 33 influenza epidemics. On average, seasonal epidemics lasted 9 weeks and affected 4.1% of the population (95% CI 3.5; 4.7). Mean age of cases was 29 years. Epidemic size decreased over time by ‐66 cases per 100 000 population per season on average (95% CI −132; −0.2, P value = 0.049) and epidemic height decreased by ‐15 cases per 100 000 (95% CI −28; −2, P value = 0.022). Epidemic duration appeared stable over time. Epidemics were mostly dominated by A(H3N2) (n = 17, 52%), associated with larger epidemic size, higher epidemic peak and older age of cases. Conclusions The declining trend in influenza epidemic size and height over the last 33 years might be related to several factors like increased vaccine coverage, hygiene improvements or changing in influenza viruses. However, further researches are needed to assess the impact of potential contributing factors to adapt influenza plans.


| INTRODUC TI ON
Influenza is a common respiratory infectious disease, manifested typically by influenza-like illness (ILI) usually defined by sudden onset of fever, myalgia and respiratory signs. 1 Although morbidity and mortality attributed to influenza epidemics vary depending on circulating strains, 2 influenza epidemics cause each year between 3 and 5 million of severe cases and between 250 and 500 thousands of deaths around the world. 3 In France, the yearly impact of winter acute respiratory illness was estimated between 22 and 29% of the population over the 2012-2017 period. 4,5 The excess respiratory deaths due to influenza seasonal epidemics were estimated to 2.9 per 100 000 population each year. 6 The socio-economic impact was estimated at 2.9 (±2.5) days of work lost per person and per flu episode, 7 for a total cost of $2.6 billion, of which $2.3 billion indirect costs (mainly due to loss of productivity) and $0.3 billion direct costs spent for medical care. 8 The health and socio-economic impact of influenza epidemics, along with the need to assess the impact of influenza vaccine in the field, justify its surveillance. Real-time monitoring of influenza epidemics is used by authorities and healthcare professionals to implement or adjust interventions. However, retrospective analysis of influenza epidemics over a long period of time seems essential to better understand disease seasonality, variation of long-term trend and potentially forecast future trends. This allows providing information to adapt overall strategies for prevention and control and to plan management of healthcare facilities.
Influenza surveillance in France is coordinated by French national public health agency (ie, Santé publique France) including hospitals, laboratories and primary care networks. 9 The French primary care surveillance network-called Sentinelles, participate in monitoring influenza epidemics since 1984 through a sample of general practitioners (GPs) reporting on a weekly basis all ILI cases seen in consultation. 10 ILI definition used had a high predictive value for influenza although other respiratory viruses could cause ILI. 11 This definition has remained unchanged over the period, allowing study of trends in ILI. There were no steady influenza virological database available in primary care since 1984, as this laboratory surveillance has evolved, first coordinated by the GROG-Groupes Régionaux d'Observation de la Grippe, before 2014 12 and then by the Sentinelles network. 13,14 The work presented here aims to study evolution of the dynamics of influenza epidemics observed in primary care in France between winters 1984/85 and 2016/17 and to describe epidemic profiles according to circulating influenza viruses.

| Study population: influenza-like illness cases
The French Sentinelles network (http://www.sentiweb.fr) is a realtime epidemiologic surveillance system in primary care created in 1984. 10 15 SGPs were similar to other GPs for age and practice of a complementary medicine, but they differed in a number of ways: they were more frequently males, their number of consultation by week were slightly higher and were not equally spread over the territory.
Since late 1984, SGPs reported throughout the year weekly numbers of ILI cases seen in consultation and described these cases. 16 An ILI case is defined as a sudden onset of fever over 39°C with myalgia and respiratory symptoms. The description form for ILI cases has changed over years adding specific questions linked to public health issues. They concern: age, sex, influenza seasonal vaccination status (since September 1986) and whether the delay between vaccination and onset of symptoms is more than decreased by -15 cases per 100 000 (95% CI −28; −2, P value = 0.022). Epidemic duration appeared stable over time. Epidemics were mostly dominated by A(H3N2) (n = 17, 52%), associated with larger epidemic size, higher epidemic peak and older age of cases.

Conclusions:
The declining trend in influenza epidemic size and height over the last 33 years might be related to several factors like increased vaccine coverage, hygiene improvements or changing in influenza viruses. However, further researches are needed to assess the impact of potential contributing factors to adapt influenza plans.

K E Y W O R D S
epidemics, influenza, influenza-like illness, primary care, surveillance The study was performed using data collected from 1984 week 40 to 2017 week 25.
The protocol was conducted in agreement with the Helsinki declaration. We obtained authorization from the French Data Protection Agency (CNIL, registration number #471393).

| Incidence estimation
In France, the size of the population covered by GPs is unknown because there is no mandatory practice register. 15 Thus, the total number of ILI cases by week (weekly incidence) is estimated by multiplying the mean number of reported cases per participating SGP for a given week by the total number of practicing GPs. To take into account the regional disparities of SGPs density, weekly incidence was first estimated at the regional level (Nomenclature of territorial units for statistics-NUTS 2 level 17 ), and then summed to obtain the national estimates. Incidence rates (per 100 000 population) were obtained by dividing incidences by yearly population size (census data). Age-specific incidence rates were estimated for the following age groups: 0-4 years old, 5-14 years, 15-64 years, 65 years and older. Confidence intervals (CI) were estimated assuming that cases reported by SGPs follow a Poisson distribution. 18

| Determination of epidemic period
Influenza epidemic detection method was based on a regression model which fits non-epidemic data to predict a non-epidemic baseline. ILI non-epidemic baseline were estimated by applying a periodic regression model including a linear trend, annual and semi-annual periodic terms on weekly ILI incidence rates below a cut-off value (defined at 279 cases per 100 000 population). 19 Epidemic thresholds were defined as the estimated baseline's upper 90% prediction bound. The epidemic is declared when at least two consecutive weekly incidence rates exceed the threshold. This epidemic detection method was selected based on its performance, evaluated in a recent study. 20 Each epidemic was named by its influenza season (ie, 2011/12 epidemic, refers to the epidemic period having occurred between September 2011 and August 2012).

| Description of epidemic profiles
In the analyses, the 2009 pandemic has been considered aside from the seasonal epidemics.
The epidemic peak is defined as the highest weekly ILI incidence during the epidemic period. Epidemic size refers to weekly cumulated incidence rates during the epidemic period. Change in epidemic size and height over time was studied using linear regression where cumulated ILI incidence rates by epidemic were predicted by a time variable (eg, year).
Epidemics were classified into three groups according to the time of the influenza season started-called "start period." Breaks were defined using the first and the third quartiles of the start of the 32 seasonal epidemics. Associations between start period and epidemic size, height and duration were studied using variance analysis.
For each epidemic, the age-specific burden of illness was assessed with the relative illness ratio (RIR). 21 This ratio is defined as the contribution of the age group i to ILI cases divide by its contribution to the general population: where C i is the number of ILI cases in age group i and N i the total population of age group i.
This ratio allows assessing the under-or over-representation of an age group among ILI cases: a ratio above one indicates an excess risk. Besides, being standardized on epidemic size, it can be compared across epidemics. Confidence intervals (CI) were estimated with the exact Poisson method. 22  Table 1.

| Influenza viruses circulating
Using these collected data, virological dominance and codominance by influenza epidemic was established using the following decision rule adapted from literature. [23][24][25] The main circulating virus type or subtype was considered as "dominant" if: (a) it accounted for 70% or more of all isolates during the season or (b) it accounted between 40% and <70% of all isolates and the second most common virus type or subtype accounted for <30%.
Two types or subtypes are considered as "co-dominant" if the main circulating virus accounted between 40% and <70% of all isolates and the second most common accounted for 30% or more of all isolates.

| Epidemic profiles
Data collected by the Sentinelles network allowed the detection of 32 influenza seasonal epidemics and one pandemic during the 33 monitored seasons ( Figure 1 and Table 1).

| Influenza viruses circulating
None of the 10 epidemics dominated or co-dominated by type B viruses started early in the season. Among the "late" epidemics, 6/9 (67%) were dominated or co-dominated by type B. Moreover, among the "early" epidemics, 6/8 (75%) were dominated by A(H3N2) viruses ( Figure 3).  (Figure 4). The observed requested hospitalization was higher than for seasonal epidemics (1% of ILI cases). During the study period, type A viruses, and specially A(H3N2) subtype, were more often dominant in France, as it was already reported for Europe and USA. 38 We highlighted that epidemics dominated by subtype A(H3N2) were mostly larger with older average age of cases and higher severity in elderly. This is consistent with greater virulence of type A, 39 particularly for subtype A(H3N2), subject to more frequent changes in the major surface antigens, and responsible for epidemics with a greater morbidity and mortality especially among elderly. 40,41 This could be explained by acquired immunity against subtypes of influenza A(H1N1) in adults and elderly encountered through past exposure to epidemics, 40,42 while the A(H3N2) viruses evolve faster 43 where a lesser acquired immunity. Moreover, the greater severity of symptoms often associated with A(H3N2) infection 40 could increase adults and elderly's propensity to consult a GP. Type B epidemics were associated with lower epidemic size and occurred late, consistent with previous reports. 12,44 Only the 2012/13 epidemic dominated by B viruses was over 3500 cases per 100 000 population, where A(H1N1) and A(H3N2) were also circulating (25% and 20%, respectively). 12 Our study has some limitations. We implemented a single influenza epidemic detection method among the various used worldwide. 45 Based on its performances 20 and to ensure consistency with all data and scientific papers on the French Sentinelles network, we opted for the periodic regression used since 1991. 19 A recruitment bias was possible as ILI cases reported by SGPs were not virologically confirmed and could be caused by other viruses than influenza. However, we used a very specific definition 11 and we included ILI cases only during the epidemic period-where influenza positivity rates of cases were the highest, allows to reduce this bias. 14 Our study based on ILI consultations with a febrile ILI definition could underestimate true influenza burden in the community, especially among elderly who usually had lower symptoms (eg, low fever). Moreover, we did not include cases observed in nursing homes, where elderly could have worse health than those consulting in primary care. Conversely, influenza burden in children could be overestimated relatively to adults' estimates as children may be more likely to consult a GP than adults. 36 However, consistency of data collection since 1984, 10 using the same ILI definition and epidemic detection method, allowed comparisons over time, as potential bias, if occurred, would be constant in time.

| D ISCUSS I ON
The main strength of our study is to rely on data collected by a long-term surveillance system in primary care. 46 For more than

ACK N OWLED G EM ENTS
We thank all general practitioners and paediatricians participating in the French Sentinelles network.

CO N FLI C T O F I NTE R E S T
None declared.