The long and intense influenza epidemic in winter 2017/2018 in a number of European countries, including the Netherlands, led to substantial morbidity and increased mortality, especially because of pneumonia as a complication of influenza virus infection [1]. A sudden increase in the number of patients requiring hospitalization for (complications of) acute respiratory infections may pose a significant burden for hospitals in managing bed and staff capacity. It may also severely limit the possibilities to isolate patients suspected of influenza [2,3]. Surveillance of severe acute respiratory infections (SARI), defined as an acute respiratory infection requiring hospitalization, is considered a priority by the European Centre for Disease Prevention and Control, the World Health Organization, and individual countries. However, establishing surveillance systems in hospitals has proven difficult in many countries [4-6].

Ideally, surveillance of complications from influenza virus infection in sentinel hospitals would mimic the well-organized surveillance of influenza-like illness (ILI) or acute respiratory infections and influenza infection by sentinel general practitioners (GPs) in primary care [1]. However, in the European region, very few countries have established syndromic SARI surveillance in combination with testing for influenza virus [7]. In the Netherlands, limited real-time data on SARI are available from a pilot project in 2 hospitals [8], but this does not yet provide a nationally representative picture. Therefore, with a lack of national hospital surveillance data to guide health care measures, individual hospitals had to revert to ad hoc crisis management when patient numbers started to increase beyond capacity.

In the absence of nationally representative SARI surveillance, nontraditional Web-based data sources could improve the SARI surveillance. We could not identify previous reports in which media content was analyzed to assess impact on hospital capacity. However, in recent years, several studies have explored the use of alternative surveillance methods for influenza based on internet search terms. An example of internet-based surveillance system is Google Flu Trends, which monitors health-seeking behavior by using data on influenza-related searches to estimate the incidence of ILI in a specific region [9]. It has shown promising results during regular winter seasons [9]; however, it did not predict the 2009 influenza pandemic [10,11]. Multiple other internet-based surveillance systems have been developed over the years; however, it is often complex and cumbersome for epidemiologists to extract the relevant information from large amounts of data on social media or search engine queries [12]. Furthermore, media reports, in comparison with internet queries, have shown to contain more specific and official data in relation to influenza and other public health problems. In a study based on the 2009 pandemic, media reports were analyzed in relation to several influenza surveillance methods [11]. A study by Olayinka et al [13] showed that media reports were useful as a supplemental data source for the real-time mortality monitoring related to Hurricane Sandy. As the capacity problems that hospitals faced were reported in local, regional, and national media, these could potentially be a suitable supplemental data source to specifically assess the impact of hospitalized SARI patients during an influenza epidemic.

The aim of this study was to assess whether media reports during the 2017/2018 influenza epidemic provided relevant information for estimating the impact of influenza on hospital care as an indicator of the severity of the epidemic in the absence of traditional hospital-based epidemiological data.

A search was conducted using Coosto (Coosto), which is a Web-based media monitoring and analytics program. The search term in Dutch griep EN ziekenhuis (in English: flu/influenza AND hospital) was used. There is only 1 word for flu/Influenza in Dutch used in media reporting, that is, griep. Only articles from the Netherlands published on regional or national news websites during the influenza season (week 40 of 2017 until week 20 of 2018) were selected. Trend in weekly number of news articles from the Coosto search was plotted against trends in the different influenza surveillance systems that were available on a weekly basis. A content analysis was performed on a selection of news articles with data on influenza in hospitals during the 2017/2018 influenza season. The relevance of the news articles and possible duplicates was only assessed for the content analysis.

After the search in Coosto, 2 researchers separately scanned all titles for relevance. Articles were excluded if the title was unrelated to influenza in hospitals in the Netherlands or if it was a duplicate news article. Subsequently, full-text articles were assessed and excluded if irrelevant or duplicate. A qualitative content analysis was performed on the remaining articles. In each news article, the following information was identified: whether (1) the article was recently published, (2) it came from a designated spokesperson, (3) the name and place of the hospital was mentioned, (4) it was about a specific hospital department, and (5) the specific problems pertaining to influenza and the implemented preventive/response measures were mentioned. This was guided by the paper from Groeneveld et al who already listed the most common problems and prevention/response measures during the 2017/2018 influenza epidemic [17]. These were applied into the content analysis by categorizing and counting the number of problems and prevention/response measures reported by each hospital in the news articles. The problems reported by hospitals in the news articles were categorized as (1) hospital admission stops, (2) postponing nonurgent surgical procedures likely caused by the high number of influenza patients, (3) staff capacity problems because of influenza, and/or (4) other influenza epidemic-related problems [17]. Prevention or response measures were categorized as (1) ad hoc crisis management, (2) regional cooperation, (3) point-of-care testing (POCT), (4) cohort isolation for influenza patients, or (5) other prevention and control measures [17].

Using SAS version 9.4, Pearson correlation coefficients (significant at .05 level) were computed between the weekly number of media reports and the weekly ILI incidence, weekly number of pneumonia consultations per 10.000 inhabitants, weekly SARI incidence, weekly number of all-cause deaths, and weekly influenza diagnoses reported in the virological laboratory surveillance.

The Dutch Medical Research Involving Human Subjects Act (WMO) does not apply to this study, and therefore, an official approval by a Medical Ethical Research Committee is not required under the WMO. Furthermore, all data are publicly available. Surveillance data used in this study are available on the Nivel website (ILI incidence and pneumonia), the National Institute for Public Health and Environment (RIVM) website (SARI incidence and virological data), and the website of Statistics Netherlands [18].

The large majority of the weekly number of news articles from the Coosto search (n=730) during the 2017/2018 influenza season in the Netherlands coincided with the influenza epidemic as defined by the ILI incidence (Figure 1). The peak in ILI incidence preceded the peak in the number of articles by 5 weeks. The ILI incidence reached a peak in week 4 of 2018 and remained high until week 10 of 2018, while the media coverage increased later from around week 9 and reached a peak in week 11 of 2018. Both trends show a steep decrease after week 10 and 11, respectively. On visual inspection, similar trends were observed for the number of news articles in relation to weekly pneumonia consultations in primary care, SARI surveillance, all-cause mortality, and influenza laboratory diagnoses (Figure 1). The trends in ILI incidence and media reports are significantly correlated (Table 1). Correlations were even stronger for pneumonia in primary care, SARI in the JBH, influenza laboratory diagnoses, and all-cause mortality (Table 1).

For the content analysis, 147 of the 730 news articles were excluded based on the title (Figure 2). The remaining 583 articles were screened for duplicates, leaving 302 news articles to be assessed in full text. Ultimately, 165 (165/717, 23.0%) news articles were included in the qualitative content analysis (Figure 2).

The 165 news articles included 77 articles citing 1 or more specific hospitals or contained a statement from 1 or more hospitals, often by designated spokespersons. The other 88 news articles contained more general information about influenza and/or hospitals.

Even though this study showed that trends in media reports are not a suitable timely surveillance measure, they do provide relevant information on the impact of influenza on hospitals. Media reporting clearly showed the severity of the epidemic with a large number of hospitals having problems with capacity and staff shortages affecting patient care in terms of admission stops and postponement of nonurgent procedures. Media reports are suitable for retrospective analysis of the impact of an influenza epidemic on hospitals; however, real-time data are still necessary for preparedness and response.

On the basis of media reports, the 2017/2018 influenza season clearly had a big impact on Dutch hospitals throughout the country. It confirms that real-time data on influenza-related hospital admissions are needed at local, regional, and national level to inform decision making for preparedness and public health response. Improving SARI surveillance was also one of the recommendations of the Outbreak Management Team that was convened by the Dutch Centre for Infectious Disease Control in response to the severe influenza epidemic. In the absence of a sustainable and robust SARI surveillance program, retrospective analysis of media reports on hospitalized influenza patients offers useful information about the impact of an influenza epidemic. However, using media reports for near real-time surveillance of hospitalized influenza cases with respect to preparedness and emergency control is less suitable because of the longer time lag between detection of the event and published media reports compared with other available surveillance systems. This is in line with a study by de Lange et al [11] comparing traditional routine ILI surveillance with other systems of surveillance and trends in pandemic-related newspaper and television coverage and showed that the increase and peak in media coverage did not precede increases in ILI incidence.

In the Netherlands, there was a lack of real-time nationally representative data that could have guided hospital management in preparing and implementing mitigating action. No national guidelines were issued, in contrast to the United Kingdom, where hospitals were advised by the National Health Service to defer nonurgent operations [19]. Hospitals already participate in training programs on how to deal with emergency situations, such as an influenza pandemic or Ebola outbreak [20]. However, there seems to have been no concerted effort in dealing with the large influx of patients during the 2017/2018 influenza epidemic. When there is pressure on hospital capacity, it seems that hospitals refer to ad hoc crisis management. It shows that there is a need for influenza outbreak response plans for hospital preparedness in managing outbreaks of SARI. An important problem mentioned in the media reports by 13 hospitals was that elderly patients could often not be discharged because of the social situation of the patient or because no beds were available in nursing homes. Furthermore, many informal caregivers of these elderly patients were possibly unavailable because of influenza [17]. This is a growing problem with an ageing population and the prevailing policy that elderly persons should be encouraged to live independently at home as long as possible. Therefore, the pressure on informal caregivers and hospital care will likely continue to increase.

Most SARI patients admitted to a hospital are not routinely tested for influenza virus infection. Without national hospital guidelines on influenza diagnostics, influenza testing occurs mainly at the discretion of the treating physician. Even in intensive care units, only about half of such patients are tested [21]. Influenza POCT can be performed by nurses in the ED and are available with a turnaround time of 20 min. It was demonstrated that implementation of POCT in combination with a designated ward for influenza-positive patients during the 2017/2018 epidemic led to a marked reduction in length of hospital stay [2]. Such a policy may have additional benefits by a decrease in costs and unnecessary use of antibiotics [22].

A limitation of using media reports for surveillance purposes is the potential selectivity of media reporting and thereby, the introduction of selection bias. Especially, reporting on response measures is of concern because it is unclear whether the report was initiated by a specific health event or by the hospitals’ aim for media attention. Moreover, performing qualitative content analysis on media reports is time consuming, and the choice of items to be analyzed is still rather arbitrary. Developing an automated surveillance system would be better to extract information.

This study showed that the 2017/2018 influenza epidemic caused serious problems affecting hospitals throughout the country. Media reports are not suitable for near real-time surveillance because of the longer time lag compared with other surveillance systems. This stresses the importance of a robust SARI surveillance program to inform decision making, which is especially important in seasons with high or long-lasting influenza activity.