As part of the Influenza Vaccine Effectiveness network (Flu-VE) (7
), we conducted active, prospective, population-based influenza surveillance among residents of Davidson County. We included those who had visited Vanderbilt University adult or pediatric emergency departments for acute respiratory infection (ARI) or fever/feverishness for <14 days during May 1, 2009–March 31, 2010. Nasal and throat swabs were tested for influenza with reverse transcription PCR (RT-PCR) by using primers and probes provided by the Centers for Disease Control and Prevention (Atlanta, GA, USA) (8
). Specimens were classified as A(H1N1)pdm09 virus if results were positive on both pandemic subtyping assays (pandemic A and pandemic H1) or positive for influenza A, negative for seasonal subtypes H1 and H3, and positive on 1 pandemic subtyping assay.
We obtained the number of emergency department visits associated with ARI or fever (International Classification of Diseases, Ninth Revision, Clinical Modification, codes 381–382, 460–466, 480–487, 490–493, 786, and 780.6) from the Tennessee Hospital Discharge Data System (HDDS) (9
), which is required to include a record of every hospital-based health care encounter. We combined data from Flu-VE RT-PCRs, influenza test results obtained clinically in the surveillance emergency departments, and HDDS discharge diagnoses to calculate age-specific visit rates attributable to influenza A(H1N1)pdm09. We used 2 epidemiologic methods: surveillance sampling and capture–recapture.
For surveillance sampling, we enrolled 826 (52%) of 1,589 eligible patients in the Flu-VE study who had visited surveillance emergency departments; 88 (11%) had positive RT-PCR results for A(H1N1)pdm09 virus (). We divided the pandemic period into 3 intervals according to prevalence of A(H1N1)pdm09 among Flu-VE participants: prepeak (May–July 2009), peak (August–November 2009), and postpeak (December 2009–March 2010). Within each period, we assumed that the proportion of ARI- or fever-associated visits caused by A(H1N1)pdm09 virus among enrolled county residents was the same as that for such emergency department visits among all county residents. Estimated influenza A(H1N1)pdm09–associated emergency department visits were thus calculated by multiplying age- and time- specific counts of total county ARI- or fever-associated emergency department visits by these proportions (). We divided age-specific counts by age-specific county population estimates for July 2009 (10
) and calculated rates per 1,000 residents (). We used the binomial Wilson method to calculate 95% CIs for the proportions of ARI- or fever-associated emergency department visits caused by A(H1N1)pdm09 virus.
Figure Number of patients enrolled in the Influenza Vaccine Effectiveness study at Vanderbilt University (Nashville, Tennessee, USA) who had laboratory-confirmed influenza A(H1N1)pdm09 virus infection (bars) and number of emergency department (ED) visits associated (more ...)
Estimated total number of emergency department visits for influenza A(H1N1)pdm09, calculated by surveillance sampling method, Davidson County, Tennessee, USA, May 1, 2009–March 31, 2010*
Estimated emergency department visits for influenza A(H1N1)pdm09, Davidson County, Tennessee, USA, May 1, 2009–March 31, 2010*
We developed a capture–recapture model (11
) by linking 2 independent data sources for influenza testing from the same population: the Flu-VE RT-PCRs, performed in a research laboratory and not reported to patients or clinicians, and influenza tests performed as routine care in the surveillance emergency departments. Unlike the research laboratory tests, not all clinical tests included influenza A subtyping. However, all positive influenza A results were assumed to be A(H1N1)pdm09 virus because that strain circulated almost exclusively during the study period (12
). To calculate the total number of influenza A(H1N1)pdm09–associated visits in surveillance emergency departments, we summed the following: the number of such emergency department visits detected by Flu-VE and clinical laboratory testing (a), the number detected by Flu-VE alone (b), the number detected by clinical testing alone (c), and the number missed by both systems (d). For each age group, we estimated the number of emergency department visits for influenza A(H1N1)pdm09 missed by both surveillance systems by using the nearly unbiased estimator equation, a modification of the Petersen estimator that performs well with rare outcomes: d = bc / (a + 1) (11,13
RT-PCR identified 88 persons with influenza A(H1N1)pdm09; 541 patients had positive influenza A results by clinical tests: 506 BinaxNOW influenza rapid antigen tests (Alerei Inc., Waltham, MA, USA), 19 clinical RT-PCRs, and 16 viral cultures. Influenza A(H1N1)pdm09 virus was detected by clinical and research laboratory testing (“a” in the formula) for only 13 patients; age groups were <5 years (3 patients), 5–17 years (3), 18–49 years (7), and >
50 (0). Using the nearly unbiased estimator equation, we calculated 572, 1,000, 528, and 90 surveillance emergency department visits for influenza A(H1N1)pdm09 for each age group, respectively. HDDS data indicated that 62.3%, 48.4%, 18.3%, and 14.2% of ARI- or fever-associated emergency department visits among county residents <5, 5–17, 18–49, and >
50 years of age, respectively, occurred in surveillance emergency departments. We calculated the total number of influenza A(H1N1)pdm09–associated emergency department visits by county residents by dividing the number of influenza A(H1N1)pdm09–associated visits to surveillance emergency departments by the age-specific proportions above. To estimate rates, we divided estimated influenza A(H1N1)pdm09 visits by age-specific county populations for July 2009 (10
) and multiplied by 1,000, yielding rates comparable to those obtained by the surveillance sampling method (). We calculated 95% CIs for capture–recapture estimates by using a bias-corrected bootstrap method (14
). Because no persons >
50 years of age were identified by both surveillance systems, 95% CIs for this group and the entire population could not be calculated.