Estimated rates of influenza‐associated outpatient visits during 2001‐2010 in 6 US integrated healthcare delivery organizations

Background Population‐based estimates of influenza‐associated outpatient visits including both pandemic and interpandemic seasons are uncommon. Comparisons of such estimates with laboratory‐confirmed rates of outpatient influenza are rare. Objective To estimate influenza‐associated outpatient visits in 6 US integrated healthcare delivery organizations enrolling ~7.7 million persons. Methods Using negative binomial regression methods, we modeled rates of influenza‐associated visits with ICD‐9‐CM‐coded pneumonia or acute respiratory outpatient visits during 2001‐10. These estimated counts were added to visits coded specifically for influenza to derive estimated rates. We compared these rates with those observed in 2 contemporaneous studies recording RT‐PCR‐confirmed influenza outpatient visits. Results Outpatient rates estimated with pneumonia visits were 39 (95% confidence interval [CI], 30‐70) and 203 (95% CI, 180‐240) per 10 000 person‐years, respectively, for interpandemic and pandemic seasons. Corresponding rates estimated with respiratory visits were 185 (95% CI, 161‐255) and 542 (95% CI, 441‐823) per 10 000 person‐years. During the pandemic, children aged 2‐17 years had the largest increase in rates (when estimated with pneumonia visits, from 64 [95% CI, 50‐121] to 381 [95% CI, 366‐481]). Rates estimated with pneumonia visits were consistent with rates of RT‐PCR‐confirmed influenza visits during 4 of 5 seasons in 1 comparison study. In another, rates estimated with pneumonia visits during the pandemic for children and adults were consistent in timing, peak, and magnitude. Conclusions Estimated rates of influenza‐associated outpatient visits were higher in children than adults during pre‐pandemic and pandemic seasons. Rates estimated with pneumonia visits plus influenza‐coded visits were similar to rates from studies using RT‐PCR‐confirmed influenza.


| INTRODUCTION
Influenza infections are responsible for substantial morbidity during most seasons. [1][2][3][4][5][6][7] Influenza-associated illnesses are difficult to count because symptoms are non-specific, diagnostic codes associated with influenza-related symptoms are broad, and sensitive and specific laboratory testing for influenza is not routine. Many studies have estimated rates of serious complications of influenza infections-including hospitalizations and deaths-with statistical models. [1][2][3][4][5][6][7][8][9][10][11] Modeling these outcomes is routine because severe influenza sequelae are uncommon, and confirmation of infection in such patients may be difficult, even with modern diagnostics. By contrast, influenza-associated outpatient visits are plentiful, so rates of laboratory-confirmed visits should be easier to document. However, relatively few studies have made populationbased estimates of influenza-confirmed outpatient visit rates. [12][13][14][15][16][17] Most have focused on children, typically in a single site. 12,[14][15][16][17] Often, few influenza-confirmed cases are reported (range, 90-372), 12,13,15,17 resulting in wide confidence intervals (CIs) around rate estimates. While a key characteristic of influenza is season-to-season variability in intensity and severity, most studies have focused on a few seasons, 12,13,17,18 or reported summary estimates from multiple seasons. [14][15][16] Finally, studies testing for influenza per protocol are not common, 12,13,15,17 as prospective studies are resource-intensive. Thus, such studies are rarely conducted in large populations including persons of all ages, in multiple sites, or during multiple influenza seasons.
The 2009 influenza A(H1N1) pandemic highlighted a lack of US population-based rates of medically attended influenza-associated ill-

nesses. The Centers for Disease Control and Prevention (CDC) and
National Institute of Health estimated pandemic-associated illnesses, hospitalizations, and deaths, [19][20][21][22] but few estimates of the incidence of influenza-like outpatient illnesses associated with H1N1pdm09 infection are available. 23 Without consistently made estimates of influenzarelated outpatient visits, the complete health burden of influenza cannot be established.
We used electronic health data from 6 integrated healthcare delivery organizations (hereafter, sites) participating in the CDCfunded Vaccine Safety Datalink (VSD) project to estimate rates of influenza-associated outpatient visits. Our analysis included the 2009 pandemic and the 8 preceding influenza seasons. We compared several estimates of influenza-associated outpatient rates with those derived from 2 concurrently conducted studies testing for influenza with reverse-transcription polymerase chain reaction (RT-PCR) assays.

| Study population
The VSD was established in 1990 to monitor vaccine safety in the US childhood immunization program. 24,25 It has since expanded in size and scope. Currently, there are 8 participating integrated healthcare delivery systems that enroll about 10 million persons of all ages, or ~3% of the US population. Standardized data files with demographic information, enrollment history, healthcare utilization, and mortality data are maintained at each participating site, and accessed via a distributed data model to ensure confidentially; data quality checks are performed weekly to evaluate the quality of vaccination and medical encounter data. 26 An assessment of possible differences between the insured VSD population and the overall US population found no substantial differences by sex, race, ethnicity, or educational attainment; adults aged 55 through 64 years were slightly over-represented in VSD data. 27

| Human subjects
Institutional review boards at each of the 6 sites reviewed and approved the study protocol.

| Models for estimating influenza-associated outpatient visits
We fit age-and site-specific negative binomial regression models to weekly outpatient visits coded for pneumonia or respiratory diseases. [5][6][7] We considered visits coded specifically for influenza to represent acute influenza infections; thus, visits listing 487-488 were not included in regression models. Data for the proportions of specimens testing positive by week for A(H1N1), including H1N1pdm09 and seasonal H1N1, A(H3N2), and B viruses, were included in all models. A model can be summarized as: where Y age,site (i) was the predicted number of outpatient visits by age group and site during week i; α was the offset term, equal to the natural log of the population size for each age group and site; and β 5 through β 8 represented coefficients associated with a standardized estimate of the proportions of specimens testing positive for influenza or RSV during a given week in the region corresponding to a site. 7 To estimate influenza-associated pneumonia or respiratory, we started with visits predicted by a full model incorporating all viral terms and those predicted by a model in which an influenza covariate was set to 0, as previously described. 7 Weekly site-and age-specific numbers of estimated influenza-associated outpatient visits for each site were the sum of predicted pneumonia or respiratory visits and visits coded specifically for influenza. Annual age-specific incidence estimates were calculated as the sum of estimated outpatient visits in that age group divided by the sum of enrollments from the 6 participating sites. We estimated 95% confidence intervals (CIs) for each rate using 2.5th and 97.5th percentiles from a distribution derived from 10 000 bootstrap simulations. 31

| Comparisons of estimated influenzaassociated outpatient visits with RT-PCR-confirmed influenza rates
To check the validity and precision of rates estimated as described above, we compared them with rates derived in 2 studies that tested outpatients for influenza infection using RT-PCR. Marshfield Clinic rate estimates were compared with rates obtained using data from annual influenza vaccine effectiveness (VE) studies conducted there. 32,33 In VE studies, patients presenting with acute respiratory symptoms were approached for enrollment; if consented, a respiratory specimen was collected and tested with CDC-approved real-time RT-PCR assays. Data on RT-PCR-confirmed influenza visits were available among persons aged ≥50 years from 2005-06 through 2009-10; similar data were not available from other sites.
Our pandemic rates were compared with those derived from a US influenza surveillance system, the Influenza Incidence Surveillance Project (IISP). IISP began during the 2009 pandemic. It conducted influenza surveillance in 38 outpatient practices in Florida, Iowa, Minnesota, North Dakota, Utah, and Wisconsin and New York City. 22 Estimates of the incidence of influenza-confirmed influenza-like illness (ILI) outpatient visits among 272 642 outpatients were made from October 2009 through July 2012. The number of influenzaassociated ILI cases each week was estimated by multiplying the proportion of ILI visits testing positive for influenza by RT-PCR and the number of ILI patient visits reported during each week. Incidence rates were calculated by dividing numbers of influenza-associated ILIs by age-specific denominators representing the outpatient practice population. 23 (Table 1). Of these, 16.5% tested positive for non-pandemic viruses. By type and subtype, these pro-   Table 2). The highest rates for outpatient visits coded for influenza occurred in children aged 2-17 years, followed by children aged <2 years; the lowest rates occurred among persons aged ≥65 years. Rates for pneumonia/influenza and respiratory outpatient visits were highest among children aged <2 years, followed by persons aged 2-17 years or 65 years and older, while the lowest rate was among adults aged 18-49 years.

| Estimates of influenza-associated outpatient visits
Among persons of all ages, the annual mean rate of influenzaassociated visits estimated with models using pneumonia-coded vis-  (Table 3). During non-pandemic and pandemic seasons, rates were higher among younger persons.
In pre-pandemic seasons, the highest rate was 67 (95% CI, 49-164) per 10 000 person-years in children aged <2 years. The lowest rate was 25 (95% CI, 20-48) for persons aged ≥65 years. The highest rate during the pandemic was 381 (95% CI, 366-481) in children aged 2-17 years; the lowest rate was 63 (95% CI, 56-86) per 10 000 person-years for individuals aged ≥65 years. We found variability by site as well as by age group in estimated rates, but as no consistent patterns of geographic variability were noted (Table S1), we focused on the clear variability in rates by age group.

| Rate comparisons
Our estimates of influenza-associated outpatient visits among persons aged ≥50 years at Marshfield Clinic were similar to rates of RT-PCR-confirmed influenza visits calculated with VE study data ( Table 4). Rates of RT-PCR-confirmed influenza visits each season fell within the 95% CIs of our model-based estimates in 4 of 5 seasons.
During the mild 2005-06 season, our estimated rate made with pneumonia visits was significantly greater than the rate from the VE study.
During the pandemic, rates estimated with pneumonia were similar to rates of RT-PCR-confirmed Influenza among subjects enrolled in the IISP. Figure 2 plots our estimated weekly rates and weekly rates of influenza-associated ILI from IISP. The timing, peak, and magnitude of these rates were consistent among both children aged 0-17 years and adults aged 18 and older.

| DISCUSSION
We estimated influenza-associated outpatient visit rates in 6 US healthcare organizations enrolling ~7.7 million persons and found that rates were greater among children and young persons during pre-pandemic and pandemic influenza seasons. The 2009-10 pandemic was associated with significant increases in rates of influenzaassociated outpatient visits in each of 5 age groups when compared with rates from preceding seasons. These increases were most pronounced in children, as expected.   Cross-protective immunity in older persons from prior infections with H1N1 viruses more closely related antigenically to the pandemic strain than H1N1 viruses circulating since 1977 contributed to these findings. 35 We estimated influenza-associated outpatient visits by add- Rate is calculated as the sum of estimated outpatient visits divided by the sum of enrollments from 6 participating sites. CI, denotes confidence interval generated by bootstrap simulation.
a Specific ICD-9-CM codes for influenza, pneumonia, and respiratory illness categories provided in the Methods.

T A B L E 2 (Continued)
T A B L E 3 Annual rates of influenza-associated outpatient visits per 10 000 person-years estimated with negative binomial regression models, by age group, among 6 US healthcare systems The second source of PCR-confirmed data was limited to the pandemic season. Estimated rates based on pneumonia-coded visits in the 6 participating sites were broadly consistent with rates of RT-PCR-confirmed influenza in the IISP study, which tested outpatients presenting with ILI. 23  10 000 in these 2 seasons, respectively. Besides the usual caveats regarding differences in study design affecting incidence estimates, it is possible that geographic differences in influenza activity also affected these comparisons. Poehling's study had 3 sites and ours 6; neither had the population size or geographic variation for its estimates to be interpreted as national in scope. We did find variation among our sites in estimated rates (Table S1) The lower confidence limit is defined by a rate established by visits listing the ICD-9-CM code for influenza.
In addition to the issues noted above, we acknowledge other limitations. While we did adjust for RSV activity by including a weekly term for this virus in all models, we could not consider the effects of other viral or bacterial respiratory pathogens on influenza estimates.
Because RSV circulation often overlaps with influenza circulation and it is the leading cause of infectious respiratory disease among in young children, 6,37 we emphasized adjusting for regional RSV activity.
Because the onset, duration, and intensity of influenza virus circulation vary geographically, 38