Influenza‐like illness‐related emergency department visits: Christmas and New Year holiday peaks and relationships with laboratory‐confirmed respiratory virus detections, Edmonton, Alberta, 2004–2014

Background Emergency department (ED) visit volumes can be especially high during the Christmas–New Year holidays, a period occurring during the influenza season in Canada. Methods Using daily data, we examined the relationship between ED visits for the chief complaint “cough” (for Edmonton, Alberta residents) and laboratory detections for influenza A and respiratory syncytial virus (RSV) (for Edmonton and surrounding areas), lagged 0–5 days ahead, for non‐pandemic years (2004–2008 and 2010–2014) using multivariable linear regression adjusting for temporal variables. We defined these cough‐related visits as influenza‐like illness (ILI)‐related ED visits and, for 2004–2014, compared Christmas–New Year holiday (December 24–January 3) and non‐holiday volumes during the influenza season (October–April). Results Adjusting for temporal variables, ILI‐related ED visits were significantly associated with laboratory detections for influenza A and RSV. During non‐pandemic years, the highest peak in ILI‐related visit volumes always occurred during the holidays. The median number of holiday ILI‐related visits/day (42.5) was almost twice the non‐holiday median (24) and was even higher in 2012–2013 (80) and 2013–2014 (86). Holiday ILI‐related ED visit volumes/100 000 population ranged from 56.0 (2010–2011) to 117.4 (2012–2013). In contrast, lower visit volumes occurred during the holidays of pandemic‐affected years (2008–2010). Conclusions During non‐pandemic years, ILI‐related ED visit volumes were associated with variations in detections for influenza A and RSV and always peaked during the Christmas–New Year holidays. This predictability should be used to prepare for, and possibly prevent, this increase in healthcare use; however, interventions beyond disease prevention strategies are likely needed.


| INTRODUCTION
Emergency departments (EDs) can experience especially high visit volumes during the Christmas and New Year holidays. 1 In Canada, this surge has been highlighted by the media 2,3 and by hospitals themselves 4 and, because this holiday period occurs during the influenza season, it is important to consider the contribution that influenza-like illness (ILI) makes to the increase occurring at this time. For example, at one hospital in Winnipeg, Manitoba, Canada, between 1995 and 1999, ED visits related to ILI increased by as much as 61% (128 visits) during the Christmas holidays compared with the week before Christmas. 5 Aside from EDs, in Canada, individuals experiencing ILIrelated symptoms may also visit family physicians, whose offices may be closed during the Christmas holidays, or walk-in clinics.
The influenza season of 2012-2013 was especially intense in Canada 6 and, in January of the following season (2013-2014), an influenza care clinic was opened in Edmonton, Alberta to reduce ED visit volumes following the Christmas-New Year holidays. 7 Quantifying ILI-related ED visit volumes during these comparatively severe influenza seasons and during the Christmas-New Year holidays in particular may improve planning and preparedness for future surges in ILI-related ED use.
We investigated ILI-related ED visits in Edmonton, Alberta, quantifying the magnitude and duration of peak volumes, noting the dates when peak volumes were highest, and compared volumes during the Christmas-New Year holiday period to those during the remainder of the influenza season. To evaluate our definition of ILI, we first examined the relationship between these ILI-related ED visits and laboratory detections for respiratory viruses in Edmonton and surrounding geographic areas.

| Overview
Our study is divided into two parts. In Part 1, we evaluate our definition of ILI-related ED visits by examining the relationship between ED visits for the chief complaint "cough" and laboratory-confirmed respiratory virus detections. In Part 2, we explore these ILI-related ED visit volumes in more depth, investigating the timing and magnitude of surges in volume during the Christmas-New Year holiday and non-holiday periods, similar to others who have looked at total ED visits. 1 Overall, we include a total of 3423 days of data (3 October  as "pandemic-affected" and the remaining seasons (2004-2008 and 2010-2014) as "non-pandemic" seasons. We limit Part 1 to nonpandemic seasons (n=2688 days) while, in Part 2, we include the entire study period but focus our analysis on the influenza season.
We defined a consistent Christmas-New Year holiday period from season-to-season: December 24 of one calendar year to January 3 of the following calendar year (11 days); this is similar to a study by Zheng et al., 1 but they used a longer, 28-day period.

| Emergencydepartmentvisitdata
To quantify ILI-related healthcare use, we used daily data from the Alberta Real Time Syndromic Surveillance Net (ARTSSN), which provides syndromic surveillance data for Edmonton, Alberta, and surrounding areas 9 and included data from nine EDs, three of which began contributing data after the study began (in 2004, 2005 and 2013).
ARTSSN data included the visit date and chief complaint, as well as the patients' demographic information (age and sex) and area of residence. We included ED visits that listed chief complaints for "cough".
Originally, we investigated ILI-related visits with chief complaints of "fever" and "sore throat"; however, these were excluded from further analysis because preliminary analyses demonstrated weaker relationships between these chief complaints and respiratory virus detections in our preliminary analyses (Pearson correlation coefficients for influenza A and RSV, respectively, were 0.35 and 0.27 for fever and 0.10 and 0.058 for sore throat). This may be due, in part, to the way in which chief complaint data are entered into the system: only one chief complaint is allowed and it is meant to be the most prominent symptom. We limited our analysis to ED visits made by residents of the city of Edmonton, based on the forward sortation area (FSA) of their postal code (n=38 FSAs with a total population of 815 812 in 2011 10 ) in order to calculate rates in Part 2. We included all ILI-related ED visits regardless of discharge disposition (i.e. we included visits categorized as "left without being seen"). For context, Edmonton residents made a total of approximately 2.5 million ED visits for any chief complaint during the study period according to the ARTSSN data. as described previously. [12][13][14] For each positive laboratory sample, we used the date of receipt in the analyses and defined the geographic region using a hierarchy of information as provided for the patient (city of residence), physician (location of practice in laboratory information system) and submitting location (e.g. medical clinic or hospital) for the specimen (in descending order). For example, if the patient's region was unavailable, we used the physician's region and if this was also unavailable, we used the region of the submitting location. Based on this definition, we limited our analysis to samples from Region 6, which is the broad geographic area that includes Edmonton (see map: http:// www.ahvna.org/pdfs/RHA-Map-December-1-2003.pdf).

| Part1:comparinglaboratorydetectionswith ILI-relatedEDvisits
For Part 1, we used ARTSSN data (aggregated by patient age and sex) and laboratory data to compare the relationships between the following time series: daily ILI-related ED visit counts and daily laboratory detections for each of the respiratory viruses. We omitted pandemicaffected seasons from Part 1 so that changes in either the laboratory testing procedures or the healthcare seeking behaviour of the population would not distort the overall relationship observed between these two datasets.
We examined the number of laboratory detections for each virus and the variation in detections over time. To compare laboratory data and ED visit data, we began with a descriptive analysis examining joint time-series plots of the number of detections for each of the respiratory viruses and ILI-related ED visits and calculated associated Pearson's correlation coefficients. We used these descriptive statistics as a rationale to decide which respiratory viruses to include in the remainder of the analysis. Then, we investigated lags for each of the respiratory viruses; that is, we allowed the laboratory detections to either precede or follow ED visits. We investigated lags from −20 days to +20 days, similar to van den Wijngaard et al. 15  To characterize variations in daily ILI-related ED visits with respect to respiratory virus detections, similar to van den Wijngaard et al., 15 we examined multivariable ordinary least squares regression models and adjusted for seasonality using two sinusoidal terms, sin( 2π 365.25 day) and cos( 2π 365.25 day) (where day=1, 2, 3, …, 2688) for the n=2688 days included in Part 1. After examining residuals, we adjusted for long-term temporal characteristics of the ILI-related ED visit data using a linear time trend. Additionally, we adjusted for indicator variables for each day of the week and the 11-day Christmas-New Year holiday period. In our final model, we included all lags (0-5 days) for each respiratory virus that we considered for further investigations and all temporal adjustment variables. We examined regression parameter estimates and assessed R 2 . To examine autocorrelation of residuals after model fitting, we used the Durbin-Watson test and added two autoregressive terms to address first-order autocorrelation in our final model.
For comparison, we reran our final model for each lagged respiratory virus separately. Bhaskaran et al. 16 provide an overview of time-series analysis that was helpful in our study.

| Part2:examiningILI-relatedEDvisits
We compared ILI-related ED visit volumes during the Christmas-New Year holiday period to those during the rest of the influenza season

| Descriptiveanalyses
During non-pandemic seasons, 14 955 laboratory samples tested positive for at least one of the respiratory viruses considered, representing 15 262 detections, with the highest number of detections for RSV (40%) and influenza A (26%) ( Table 1). Of the 306 mixed virus samples, 11 viral combinations were seen, most commonly parainfluenza-adenovirus (n=87), RSV-adenovirus (n=73), and RSV-parainfluenza (n=63). A submitting location was provided for 91% of samples, of which most were acute care hospitals and non-urban health centres (n=11 381, 83%). These samples would have been submitted from inpatient wards, EDs, as well as hospital-based clinics. Compared with other viruses, influenza detections (especially influenza A) were associated with an older median patient age (Table 1). Overall, approximately half the laboratory detections were from female patients, but this proportion was somewhat lower for adenovirus (39%) ( Table 1).
Comparing the viruses to each other, influenza A and RSV had the highest number of detections, highest variation in the number of detections, and the strongest, unlagged, unadjusted correlations with ILI-related ED visits (Table 1). For these reasons, and because our goal was to investigate whether ILI-related ED visits were associated with respiratory virus detections rather than how each virus might help to explain variation in visits, we decided to focus the rest of our analysis on influenza A and RSV.

| RelationshipsbetweenILI-relatedED visitsandviraldetections
We visually compared ILI-related visit volumes and respiratory virus detections over time (Fig. 1). In 2004-2005, two distinct peaks in ILIrelated ED visits can be seen, which correspond well with separate peaks for influenza A followed by RSV (Fig. 1). Although the maximum peak in ILI-related ED visit volumes always occurred during the  corresponds to the non-pandemic season with the highest volume of ILI-related ED visits (Fig. 1). In the next season, 2013-2014, influenza A peaked during the extended holiday period to the highest level throughout the study period in Part 1 (i.e. non-pandemic seasons) and RSV also had a small peak during this time; this corresponds to the non-pandemic season with the second-highest volume of ILI-related ED visits (Fig. 1).

| Multivariableanalysis
Our final temporally adjusted multivariable linear regression model with two autoregressive terms explained an estimated 77% of the variability in ILI-related ED visits (  (Table 2). Therefore, an increase in one influenza A detection on each of these six days was associated with an increase of 1.

| Part2
During non-pandemic and pandemic seasons combined, Edmonton residents made 79 431 ED visits for the chief complaint "cough": 57 660 during influenza seasons and 5257 specifically during the Christmas-New Year holiday periods. The majority of these visits, both during the holidays (58%) and the remainder of the influenza season (59%), were made to two large healthcare centres.

| Non-pandemicseasons
During each non-pandemic season, the highest peak in daily ILI-related ED visit volumes occurred during the Christmas-New Year holidays, specifically between December 26 and January 2 (Fig. 1, Table 3).
Additionally, several seasons had comparable, but lower, peaks in  (Table 3).   Fig. 2). Of the 1620 days during non-pandemic influenza seasons,

| DISCUSSION
We examined the relationships between ILI-related ED visits and lab- Others have also found influenza and RSV detections to be associated with syndromic surveillance data for respiratory illnesses, including ED visits among children, 20 calls to a telehealth service, 21 visits to general practitioners and hospitalizations. 15 Of note, in our analysis, we found it important to allow for delays between the syndromic sur- EDs experience an intense and highly predictable surge in ILIrelated visits during the Christmas-New Year holidays. Although the magnitude of this peak differs by season, it always represents the highest daily ILI-related visit volume during non-pandemic seasons and its timing occurs during the same 11 days. This predictability can enable emergency departments to prepare (e.g. by adjusting staffing) and for health authorities to potentially prevent or mitigate this increase in healthcare use (e.g. through dedicated influenza assessment centres).
The ILI-related ED visits examined were associated with detections for respiratory infections and most strongly associated with influenza A; however, the timing of the highest peak in respiratory virus detections is not always consistent with this peak in ILI-related ED visits.
Therefore, interventions beyond disease prevention strategies are likely needed to mitigate holiday pressure at the ED.