Introduction
Though advocated by the World Health Organization (WHO) and the European Centre for Disease Prevention and Control (ECDC), surveillance of severe acute respiratory infections (SARI) that require hospital admission is implemented in only few Western European countries [
1‐
4]. In contrast, surveillance of acute respiratory infections (ARI) or influenza-like illness (ILI) in primary care is well-established as part of the European Influenza Surveillance Network (EISN), coordinated by the ECDC [
5,
6]. Such primary care surveillance covers the community dwelling population and is thus focused on patients with milder illness. The number of patients developing serious complications and who require hospitalization is not available through this system and therefore, our understanding of the burden of respiratory infections is incomplete. The WHO recommends development and application of additional SARI measures that can be used to assess the severity of every influenza epidemic, whether seasonal or pandemic, and creating a historical SARI baseline [
7]. The proposed measures include: SARI incidence, SARI peak levels, SARI mortality, and the SARI/ILI ratio. A SARI is a so-called syndrome group, i.e. based on rapidly available (initial) diagnoses that can be monitored as an indicator of infectious disease trends, outbreaks, and burden [
8].
The most important complication of influenza virus infection is pneumonia (primary viral or secondary bacterial pneumonia) [
9]. While costs can be high, the precise burden remains a blind spot [
10‐
15]. Such burden information is crucial for prevention and response considering that vaccination, the main control measure against influenza infection, is aimed at preventing complications. A severe influenza season may also lead to hospital capacity problems, especially in ICUs. In this study we analyze comprehensive retrospective ICU data to fill the current knowledge gap.
Discussion
This study helps to fill the knowledge gap on the burden of severe respiratory illness. It provides robust numbers on occurrence of ICU SARI and how these vary between the different influenza epidemics in the Netherlands. It shows how different parameters, proposed by the WHO for monitoring influenza severity [
7], can vary greatly by season and that they complement each other: incidence, peak levels, ICU SARI mortality, mean APACHE IV score, and the ICU SARI/ILI ratio. They provide insight into different aspects of the severity of separate influenza seasons.
The ratio of ILI to ICU SARI admissions varied from influenza epidemic to influenza epidemic confirming that the intensive care burden of influenza is not predictable from ILI trends [
25]. Depending on the epidemic, 8–17 SARI admissions occurred per 1000 cases of ILI in primary care (or a ratio of 1.3–2.7 ICU SARI admissions per each actual observed ILI in the 0.7% national ILI sentinel coverage).
Weekly medical ICU admissions were higher when SARI admissions were higher (Fig.
1) indicating that pressure on ICUs is expected to be higher in respiratory seasons. During the eight influenza epidemics under study, ICU SARI peaked at (min, max) 101 up to 188 ICU admissions in one week in the Netherlands. This is an almost twofold difference in peak-incidence between some seasons. Preparedness for severe or unusual seasons or outbreaks includes ICU capacity preparedness. The current situation in the Netherlands relies on crisis management when hospitals are to be overwhelmed. Whether and how often this occurs is not known due to lack of SARI surveillance. A SARI surveillance system might provide better insight into the location of potential bottlenecks and provide potential time gain allowing capacity planning instead of on-the-spur crisis management.
For influenza severity assessment, monitoring a SARI syndrome instead of specific diagnoses would be of use as distinguishing between bacterial and viral causes of pneumonia is difficult and bacterial pneumonia is often secondary to viral infection [
26‐
28]. Only 4.7% of the diagnoses are registered as viral pneumonia, which may be an underestimation: laboratory data are not yet available in the NICE registry, and the specific etiology of SARI cases could not be determined. Furthermore, laboratory testing for influenza virus or other respiratory viruses is not standard practice in hospitalized patients with SARI; less than half of patients admitted to the ICU with suspected pneumonia are tested [
29]. Notably in our data, in the 2015/2016 season a large proportion of ICU SARI was registered as viral. Whether this is a true finding, a registration artifact, or whether more laboratory testing occurred in that year is not clear. Further, inherent to a syndromic surveillance approach based on rapidly available provisional data, a SARI syndrome may contain some patients with an incorrect initial diagnosis at admission. Another limitation is that no data were available on influenza vaccination status or use of oseltamivir.
In terms of the highest extremes, notable influenza epidemics are those of 2007/2008, 2012/2013, and 2014/2015 for different parameters. The 2014/2015 epidemic was the longest recorded (20 weeks) and corresponded with the highest ICU SARI and ILI incidence but not with highest values of other ICU SARI characteristics. The highest peak of ICU SARI (188 in one week, standardized) was in the 2012/13 epidemic (16-week-long epidemic). In contrast, the 2007/2008 epidemic was much shorter (8 weeks) but had the highest ICU SARI/ILI ratio as was also reported in the USA [
30], and the largest proportion of deaths. These results may, however, potentially have been affected by the low ICU coverage in that first study year (around 50%), although we do not know in which direction this may have impacted results.
In terms of the lowest extremes the 2009/2010 season was the pandemic season (A(H1N1)pdm09 circulation), known to have been milder in older people [
31], showing the lowest ICU SARI/ILI ratio indicating a lower than average ICU burden per influenza infection in that year. The dominant virus in the following epidemic (2010/2011) was also influenza virus A(H1N1)pdm09, showing no drift to the previous season and which coincided with the smallest proportion of in -ICU SARI deaths. The 2013/2014 season was a relatively short epidemic (6 weeks) with the lowest SARI peak and with the lowest incidence of both SARI and ILI.
As the most severe outcome of any disease, mortality data, if available, could provide an important indication of flu season severity. Mortality can be high in pneumonia hospitalizations, but to date the burden of influenza mortality remains uncertain and is estimated mostly from modeling studies [
13,
32]. Fatal outcome may be more prevalent in influenza-positive SARI [
3]. In our study, in-ICU mortality in patients with SARI during influenza epidemics varied between 13% and 20%. To what extent this reflects differing severities of influenza seasons is not entirely clear because SARI can be caused by different respiratory pathogens, patients may die after their ICU or total hospital stay [
33], and shifts in afflicted age groups can impact mortality numbers. In a German study, mortality in hospitalized patients with SARI (thus not only in the ICU) was lower (5-year-average 11%), but also included non-flu weeks [
2]. Future exploration of a more comprehensive mortality measure than in-ICU deaths (i.e. 30-day mortality or survival to hospital discharge) would improve the estimate of influenza impact.
An important strength of our study is that the coverage increased over time to include almost all adult ICUs in the country, thus providing a clear estimate of the total number and incidence of ICU SARI in the Netherlands. Comparing these with other Western European countries is difficult. Most countries do not have SARI surveillance and in those that do or those that have analyzed retrospective data, the SARI case definition varies due to different coding systems and the type of hospitalizations included (ICU, other wards) [
1,
2]. Additionally, criteria for ICU admission vary between countries, with Dutch ICUs being relatively restrictive. In Germany, retrospectively, ICU SARI admissions during influenza epidemics peaked at 17% [
2], in the range of the weekly peak of 15–25% that we observed in the different epidemics. Of hospitalized SARI in Belgium, influenza was laboratory-diagnosed in 46%, and approximately 10% of SARI were in ICUs (sentinel surveillance of 2016/2017 epidemic) [
1], similar to the 8–11% in ICUs in a Dutch pilot study in two hospitals [
12] (Marbus S, et al: Acute respiratory infections in secondary care versus influenza-like illness in primary care in the Netherlands: hospital incidence peaks first, submitted). Although such estimates are not available from the national registry data (NICE) they give a rough indication that total SARI hospitalizations (when including non-ICU SARI) in the Netherlands could be around 10-fold higher than SARI in ICUs and that influenza can potentially play a large role in overall SARI admissions. Some other countries record laboratory-confirmed influenza infections in ICUs, but without tracking a denominator or testing practices, thus not providing estimates on total SARI burden [
34‐
39].
Acknowledgements
We thank Eric van der Zwan for preparing the aggregated NICE dataset and for support with data-management, and Jeroen Alblas for support with data management.