We explored the utility of tracking emergency department (ED) visits for febrile illness as a proxy for
influenza surveillance, from both a local and a regional perspective.
In Washington State, near real-time analysis of ED
data presents an opportunity to identify potential improvements to public health surveillance by monitoring disease trends and increasing the speed of casefinding. Syndromic surveillance data collected by
Kitsap County Health District, Public Health – Seattle and King County, and Tacoma-Pierce County
Health Department are shared with the Washington
State Department of Health (WA DOH) to create a
regional snapshot of disease activity.

Three years of ED data (9/2004 – 4/2007) were examined from King (n=19 EDs) and Pierce (n=5 EDs)
counties, and two years (9/2004 – 8/2006) of data
were examined from Kitsap County (n= 1 ED). Records were selected if either the chief complaint or
diagnosis contained one of the following key words
or ICD codes in a common fever syndrome definition: FEV, HIGH TEMP, ELEVATED TEMP, HI
FEBR, PYREXIA, 780.6, 780.99. Fever syndrome
records were counted by day, week, and month for
each county and aggregated as a regional count. Total
visits per day, week, and month were used to calculate proportions for King and Pierce counties and a
regional aggregate of the two; census counts were not
available for Kitsap County. To account for missing
and incomplete historic data in the Pierce County
repository, we replaced the census count and fever
count with an average of the respective counts 2 days
before and after when the census count was less than
65% of the 7-day moving average. Fever counts and
proportions were also stratified by age group. Timeseries data were plotted by time unit, county, and age
group in line graphs and histograms. Trends were
compared visually and by Pearson’s correlation
analysis. ED data were also compared to traditional
data sources (number of schools reporting >10% absenteeism, positive influenza isolates, and pneumonia
and influenza deaths) collected by the WA DOH as
part of the annual sentinel influenza surveillance program.
There was a strong correlation between weekly ED
fever visit counts and positive flu isolates, both for
the region as a whole (r=.89) as well as by county
(King, r=.88; Pierce, r=.83; Kitsap, r=.62). The correlation between positive flu isolates and weekly ED
fever visit counts was highest among 18-44 year-olds
(r=.87) and lowest among adults ages 65 and older
(r=.51). In addition, Washington State school absenteeism trends correlated strongly with weekly fever
visit counts among 5-17 year-olds (r = 0.83). The
correlation was highest for King County (r=.85), followed by Pierce (r=.71) and Kitsap (r=.25) counties.
There was a strong correlation between weekly ED
visit counts at Pierce and King counties (r=.87). The
volume of ED fever visit counts at Kitsap County
was lowest compared with King and Pierce counties
and did not correlate as strongly with data from these
two counties (r=.61 and r=.53, respectively). The
correlation between King and Pierce ED weekly fever visit counts was highest for the pediatric age
groups (<2 years, r= .82; 2-4 years, r= .77; 5-17
years, r=.87), and declined with increasing age (18-
44 years, r=.59; 45-64 years, r=.39; 65+ years, r=.13).
When comparing King and Pierce counties, the correlations between the proportion of weekly ED fever
visits was highest for the 5-17 year old age group
(r=.84); the correlation for all other age groups was
less than .65.
ED fever visits captured through syndromic surveillance correlate strongly across counties and with
other specific and non-specific indicators of influenza
activity, including laboratory influenza test results
and all cause school absenteeism. ED fever visit
trends for pediatric age groups, particularly school
age children (5-17 years), may be good indicators of
seasonal flu activity. Because other febrile illnesses,
particularly respiratory syncytial virus, commonly
co-circulate with influenza, the specificity of syndromic monitoring for fever symptoms needs to be
further established. Additional work is also needed to
evaluate the timeliness and alerting properties of the