Ambulance dispatch calls attributable to influenza A and other common respiratory viruses in the Netherlands (2014‐2016)

Abstract Background Ambulance dispatches could be useful for syndromic surveillance of severe respiratory infections. We evaluated whether ambulance dispatch calls of highest urgency reflect the circulation of influenza A virus, influenza B virus, respiratory syncytial virus (RSV), rhinovirus, adenovirus, coronavirus, parainfluenzavirus and human metapneumovirus (hMPV). Methods We analysed calls from four ambulance call centres serving 25% of the population in the Netherlands (2014‐2016). The chief symptom and urgency level is recorded during triage; we restricted our analysis to calls with the highest urgency and identified those compatible with a respiratory syndrome. We modelled the relation between respiratory syndrome calls (RSC) and respiratory virus trends using binomial regression with identity link function. Results We included 211 739 calls, of which 15 385 (7.3%) were RSC. Proportion of RSC showed periodicity with winter peaks and smaller interseasonal increases. Overall, 15% of RSC were attributable to respiratory viruses (20% in out‐of‐office hour calls). There was large variation by age group: in <15 years, only RSV was associated and explained 11% of RSC; in 15‐64 years, only influenza A (explained 3% of RSC); and in ≥65 years adenovirus explained 9% of RSC, distributed throughout the year, and hMPV (4%) and influenza A (1%) mainly during the winter peaks. Additionally, rhinovirus was associated with total RSC. Conclusion High urgency ambulance dispatches reflect the burden of different respiratory viruses and might be useful to monitor the respiratory season overall. Influenza plays a smaller role than other viruses: RSV is important in children while adenovirus and hMPV are the biggest contributors to emergency calls in the elderly.


| INTRODUC TI ON
Surveillance of respiratory viruses is mainly centred on influenza, for which robust systems have been developed in most countries, generally based on sentinel networks of General Practitioners (GPs, primary care). 1 Comparable surveillance systems do not exist for other respiratory viruses, despite the increasing interest and leadership of the World Health Organization (WHO) in expanding surveillance for respiratory syncytial virus (RSV) now that a vaccine may become available. 2 For most viruses, surveillance is limited to laboratory-based counts, often with unknown denominator, low representativeness or lack of standard sampling criteria.
WHO encourages surveillance of severe acute respiratory infections (SARI) in the context of the Pandemic Influenza Severity Assessment program. 1 This is fundamental to determine the severity of circulating viruses, their pressure on healthcare services and the groups most at risk of severe outcomes. Surveillance of severe infections requiring secondary care is much less developed than surveillance in primary care. In Europe, a few countries have established hospital-based surveillance based on syndromic SARI, laboratory confirmed cases or a combination of both. 3,4 In the Netherlands, a pilot involving three hospitals has been running since 2015. 3 Syndromic surveillance using ready-to-use data has also been explored, mainly in emergency rooms. [5][6][7] Few initiatives have used ambulance data 5,8,9,10 or ambulance dispatch centre data. 5,8,10,11 Ambulance dispatch centres could be an alternative source of readily available data to monitor the occurrence of severe respiratory infections. During the triage process, information is collected and recorded in real time, including the chief symptom in very broad categories, as their objective is to rapidly assign an urgency level and prioritize resources. A recent study in the Netherlands has shown how the variability in respiratory syndromes is correlated with ILI from sentinel GP surveillance, 12 making it a potential source for syndromic surveillance. However, not all respiratory viruses will result in ILI, and although the ILI case definition focuses on detecting influenza infections, ILI can be caused by a wide range of viruses.
In this study, we aimed to assess to what extent ambulance dispatches reflect the activity of different respiratory viruses in order to advance our understanding of their use for the surveillance of severe acute infections by different respiratory viruses. Specifically, we evaluated the association of syndromes compatible with respiratory infections in ambulance dispatches with trends in detections of influenza A, influenza B, RSV, rhinovirus, adenovirus, coronavirus, parainfluenza and human metapneumovirus (hMPV).

| ME THODS
The Netherlands is divided into 25 Regional Ambulance Services We focused our analysis to A1 urgency calls, as we previously found these to have a stronger association with ILI. 12 These calls may better capture variations in acute severe infections by respiratory viruses and be a valid source for their surveillance.
Calls with triage codes that were potentially compatible with respiratory infections (Table 1)  year, and hMPV (4%) and influenza A (1%) mainly during the winter peaks. Additionally, rhinovirus was associated with total RSC.

Conclusion:
High urgency ambulance dispatches reflect the burden of different respiratory viruses and might be useful to monitor the respiratory season overall.
Influenza plays a smaller role than other viruses: RSV is important in children while adenovirus and hMPV are the biggest contributors to emergency calls in the elderly.

K E Y W O R D S
adenovirus, ambulance, coronavirus, influenza, respiratory syncytial virus, rhinovirus

| Respiratory virus data
The number of respiratory virus identifications was obtained from the Weekly Sentinel Surveillance System of the Dutch Working Group on Clinical Virology. Twenty-one virological laboratories voluntarily provide aggregated weekly number of diagnoses; individualized information such as age or sex is not provided. Also, no distinctions are made between primary or secondary care, or different diagnostic methods, although currently the majority use molecular methods or rapid tests. 13 We included weekly reports of influenza A, influenza B, rhinovirus, RSV, adenovirus, coronavirus, parainfluenza and hMPV.

| Statistical analysis
We analysed weekly RSC as a proportion of the total number of calls overall, by age group and by time of the day: office hours Because the trends in RSC might coincide, precede or lag behind the trends in viruses reports, we considered virus reports either in the current week, or lagged up to 4 weeks to the right, that is future in time (+lags), or 4 weeks to the left, that is backwards in time (−lags), for a total of 9 time-lagged variables of each virus. When building the models, the time lag with the lowest P-value was selected, and only one time lag per virus was allowed.
Because one of the viruses with the highest interest in monitoring its severity is influenza A (due to shifts, drifts and its pandemic potential), we forced it into the model, unless its coefficient was negative due to biological implausibility. Subsequently, other viruses were added if statistically significant, had a positive coefficient and did not revert to negative the coefficients of viruses previously added to the model. Finally, because the influenza epidemic size and severity varies by season, an interaction between influenza A and an indicator variable for the epidemiologic year (from week 27 to week 26 of the following year) was tested and retained if P < .05. The indicator variable itself was not included, as we wanted to attribute differences between years to influenza A.
Model assumptions and absence of remaining seasonality and autocorrelation were assessed by residuals diagnostics. We used R, version 3.4.0.  Table 2). The most frequent triage code among RSC was "Abnormal breathing, troubles speaking between two breaths" (Table 1). Weekly average number of RSC was 98 (range 58-138), which corresponds to 2.3 calls per 100 000 inhabitants every week.

| RE SULTS
The proportion of RSC showed a periodic pattern peaking in winter, with lower interseasonal peaks (Figure 1). The periodicity was evident in out-of-office hours and people ≥65 years, but the pattern was less clear in other groups and, in children <15 years, the peak occurred earlier.
Among the included respiratory viruses, the most frequently reported were rhinovirus and influenza A, followed by RSV (Table 3).
Most viruses had a periodic pattern similar to RSC, peaking in winter, except for rhinovirus and parainfluenza, which had a less distinct pattern with peaks in the autumn or the spring (Figure 1). Adenovirus reports showed smaller interseasonal peaks in addition to winter peaks.
Associations between respiratory viruses and the proportion of RSC are reported in Table 4 and Figure 2. In children <15 years, only RSV was associated with RSC, explaining part of the RSC winter  e Adjusted by sine and a cosine term with periodicity of 1 y. f +lags mean that the RSC from the current week are best associated with viruses from x weeks in the past (ie trend in viruses precedes RSC); -lags mean that they are best associated with viruses from x weeks in the future (ie trend of RSC precedes the viruses).
*Calculated applying the model coefficient to the average weekly number of virus reports, and multiplied by the annual number of ambulance calls by epidemiologic year, age group, office or out-of-office hours, as appropriate; for the overall effects, this represents the average per epidemiologic year; for epidemiologic year-specific effects, the numbers for incomplete epidemiologic years are extrapolations to represent complete epidemiologic years if the average weekly ILI incidence and ambulance calls were similar in non-observed weeks than in observed weeks.
**Calculated dividing the number of RSC attributable to each virus (from the previous column) by the number of observed RSC by age group, epidemiologic year, office or out-of-office hours, as appropriate.

| D ISCUSS I ON
Our results show that trends in RSC from highest urgency ambulance dispatches are associated with trends in the activity of common respiratory viruses. Depending on the subgroup 0%-20% of RSC was attributable to a combination of respiratory viruses. The specific viruses contributing to RSC varied by age group, with estimates of 1%-11% of RSC being attributable per individual virus.
Their burden on these 4 call centres covering a 4 million population was 948 of highest urgency calls per year (22/100 000 inhabitants), although this varied by virus and age group.
In emergency departments, 25% of all acute respiratory diseases are attributable to respiratory pathogens, 14 up to 80% in children. 15 In our study, the majority of RSC were incorporated into the unexplained baseline. This is not an unexpected finding, since the categories of symptoms included in AMPDS triage codes are very broad, resulting in high background noise. 16 Nevertheless, variability in RSC above this high baseline was associated with trends of common respiratory viruses, pointing at their potential usefulness to monitor the respiratory season overall (ie irrespective of the causative pathogen), as previously shown by its association with ILI. 12 The different viruses potentially involved in RSC, their individual trends, and their seasonal variation in severity would make it challenging to design indicators and models that will allow us to prospectively use RSC data for situational awareness for specific viruses separately. Conversely, large or unexpected increases in a specific respiratory virus might be reflected to a certain extent in RSC.
Influenza A is a leading cause of acute lower respiratory tract infection, particularly in the elderly. 15,16 By contrast, in our study its contribution to RSC was low, especially among the elderly. In children, influenza was not associated to RSC, consistently with its low to marginal role in SARI in this age group. 13,[17][18][19] The effect of influenza A on RSC (1%-3%) is lower than what we found for ILI, which was attributed 4%-34% of RSC. 12 Influenza B did not show association with RSC in any group, in line with our understanding of its lower, less severe impact and lower clinical burden. Lower representativeness of the laboratory data in our current study may have underestimated the association for influenza, or oppositely, its effect estimated through ILI may be overestimated because ILI is caused also by other viruses.
The effect of influenza A on RSC was found to vary by season only for the overall sample. This is fundamental to assess whether causing severe infections, second to RSV in children, [17][18][19] and after influenza in adults. 15,[21][22][23] Its presentation year-round, with peaks in autumn and winter, 24 also contributes to its high overall impact.
Adenovirus explained a significant proportion of RSC, especially among the elderly. Adenovirus is rarely detected in cases of severe respiratory infection, 15,22 although in a study in Finland, it was the second aetiology in mechanically ventilated patients with community-acquired pneumonia. 21 hMPV had a similar relative effect as adenovirus, although its impact on number of ambulance calls was smaller, since it was less frequent.
In children <15 years the peak in RSC developed earlier in the year, and our model associated this to RSV, consistent with its earlier presentation in the season. 13,16,25 RSV is the leading cause of SARI in young children 13,[17][18][19]23 and has been highly associated to SARI syndromes in emergency departments 6,14 and ambulances. 10,18 The differences between office and out-of-office hours likely reflect that ambulance calls in these two time frames are distinct populations, probably with a different share of clinical pictures and severity. However, we cannot totally rule out a lack of statistical power during office hours, since the number of calls was smaller.
Ambulance dispatches are convenient for syndromic surveillance because they reflect events that are perceived as urgent (and thus potentially severe), they are recorded continuously and they have a virtually universal coverage. 8,26 Moreover, triage algorithms are increasingly standardized internationally. 5,11 However, the true usefulness and added value of ambulance dispatches for infectious disease surveillance needs to be studied and piloted prospectively.
Some challenges for routinely using ambulance dispatch data prospectively include establishing data sharing routines and complying with data protection regulations.
There are limitations to our data. Because we did not include A2urgency level calls in our analysis, our results cannot be interpreted as the burden of respiratory viruses in ambulance services as a whole, but only in the highest urgency services. Since all associations are evaluated ecologically, spurious attribution of RSC trends to respiratory viruses cannot be ruled out. Sentinel laboratory surveillance has several limitations: it is passive and reported trends can include surveillance artefacts; it does not provide information on age, so overall number of virus detections was used; and while often biased to secondary care, it captures patients from both primary and secondary care, and the pathogens underlying their symptoms may differ from patients in ambulance dispatches. Our study covered only 6 RAVs, 25% of the population in the Netherlands, but we do not believe these are fundamentally different from non-included RAVs.
However, because the sentinel laboratory surveillance is widespread throughout the country, it could be possible that intensity or timeliness of circulation of the different viruses nationally is different from specific regional patterns in RAVs included in our study. Finally, as the Netherlands has a comprehensive primary care system where GPs that have a strong gate-keeping role (including out-of-office services), our study results cannot be directly compared to health systems with higher use of emergency medical services.

| CON CLUS ION
Because of its ability to capture variations in respiratory virus circulation, ambulance dispatch data might be useful to signal events and to monitor the respiratory season as a whole, specifically reflecting severe infections and thus complementing existing surveillance systems. It will probably have less potential for drawing conclusions about the separate effect of specific individual viruses when not combined with information from other data sources, due to the low magnitude of some associations, the different viruses reflected in RSC and their proportional variation throughout the year. The true utility of ambulance dispatch data needs to be tested prospectively and faces potential challenges regarding timely data sharing and data protection.

ACK N OWLED G EM ENTS
We appreciate the contribution of the four Regional Ambulance

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