Distinct influenza surveillance networks and their agreement in recording regional influenza circulation: Experience from Southeast Michigan

Abstract Introduction In Southeast Michigan, active surveillance studies monitor influenza activity in hospitals, ambulatory clinics, and community households. Across five respiratory seasons, we assessed the contribution of data from each of the three networks towards improving our overall understanding of regional influenza circulation. Methods All three networks used case definitions for acute respiratory illness (ARI) and molecularly tested for influenza from research‐collected respiratory specimens. Age‐ and network‐stratified epidemic curves were created for influenza A and B. We compared stratified epidemic curves visually and by centering at seasonal midpoints. Results Across all seasons (from 2014/2015 through 2018/2019), epidemic curves from each of the three networks were comparable in terms of both timing and magnitude. Small discrepancies in epidemics recorded by each network support previous conclusions about broader characteristics of particular influenza seasons. Conclusion Influenza surveillance systems based in hospital, ambulatory clinic, and community household settings appear to provide largely similar information regarding regional epidemic activity. Together, multiple levels of influenza surveillance provide a detailed view of regional influenza epidemics, but a single surveillance system—regardless of population subgroup monitored—appears to be sufficient in providing vital information regarding community influenza epidemics.

A wide range of surveillance systems is used to track seasonal influenza epidemics, based on everything from over-the-counter medication purchases, to school and workplace absenteeism, and even internet search results. [6][7][8][9][10] The most traditional and representative data are generated by active surveillance for confirmed influenza cases, though there are many methods for this as well. 11,12 In Southeast Michigan, three prospective surveillance systems exist to actively capture influenza cases from regional households, ambulatory clinics, and tertiary care hospitals. While the primary purpose of each network is to assess seasonal vaccine effectiveness in these distinct populations, their data are also used to monitor epidemic trends in the region. A better understanding of how these systems agree or disagree epidemiologically has important implications regarding how the choice of surveillance population influences an overall interpretation of the regional influenza epidemic.
In this investigation, we consolidated surveillance data from each of these distinct networks across five different influenza seasons. For one, we were interested in whether epidemic curves of any of the three networks displayed unique temporal patterns or features (e.g., characteristic timing of an epidemic peak or initial detection of epidemic spread). Second, we assessed whether there were any consistent patterns or relationships between the three networks (e.g., activity in one network foreshadowing activity in the other two).
Finally, we considered how seasonal patterns in network activity aligned with knowledge of past influenza season characteristics.

| METHODS
The Michigan Influenza Center at the University of Michigan operates three large studies which conduct prospective, active surveillance for influenza cases from community households, ambulatory clinics, and hospitals in Southeast Michigan (Table 1). Each network is designed to estimate vaccine effectiveness against a specific influenza outcome (i.e., community-acquired, medically attended, and hospitalized disease). As a result, data captured by each of the three networks reflect the seasonal patterns of regional influenza circulation within three distinct source populations-each of which with unique characteristics like age, underlying health status, and experienced illness severity.

| Household
The Household Influenza Vaccine Evaluation study (HIVE) is an ongoing, prospective cohort study of community households, the majority of which are in Washtenaw County in Southeast Michigan ( Figure 1). [13][14][15][16] While specific eligibility criteria changed over the course of the study, currently eligible households receive care from the Michigan Medicine healthcare system and have at least three members with at least one child <10 years old. 16 Participants are queried weekly about the occurrence of acute respiratory illness (ARI) in any household member. An ARI episode is defined as the presence of two or more age-specific symptoms of ARI, as described previously. 16 Following an illness, participants visit the study clinic where combined throat and nasal swabs (nasal swabs only for children <3 years) are collected by study staff. analysis, a majority of clinics were located in Washtenaw County, while three clinics were located in neighboring Livingston and Wayne Counties (Figure 1).

| Hospital
The Hospitalized Adult Influenza Vaccine Effectiveness Network (HAIVEN) monitors vaccine effectiveness against influenzaassociated hospitalization in adults. [19][20][21] In this study, staff use hospital intake logs and admission notes to find eligible participants, defined as: adult inpatients (≥18 years old) who recently

| Laboratory methods
Respiratory illness swabs collected from participants in all three surveillance networks were tested for influenza using reverse transcriptase polymerase chain reaction (RT-PCR). The Influenza Division of the US Centers for Disease Control and Prevention (CDC) provided all primers, probes, and lab protocol for each study; these were designed for detection of universal influenza A and B, as well as their respective subtypes and lineages. All tests were performed at the Michigan Influenza Center laboratory.

| Seasonality
We included all influenza positive cases that had been

| Epidemic comparisons
We created influenza epidemic curves for each network in each sea-

| Circulating virus
In all but 2018/2019, respiratory seasons were characterized by a single dominant influenza A subtype and B lineage.

| Seasonality
Most ARI in the region generally occurred between January and April ( Figure S1). Seasonality of influenza was pronounced and captured similarly from year-to-year by each of the three networks ( Figure 3).
The influenza epidemic in all five seasons began with influenza A circulation, which tended to reach its midpoint in late February. Influenza B activity followed and generally reached its midpoint 1 to 4 weeks after the influenza A midpoint. The exception to this was in 2014/2015, when influenza A activity was early and prolonged in all three networks, reaching its midpoint in mid-December-much earlier than in other seasons. We saw no evidence of any consistent sequence or pattern in the order of network activity; for example, no one network was persistently peaking prior to the others.
In each season, the middle 50% of each network's total influenza cases was detected over roughly the same interval, which generally occurred 6 weeks after initial epidemic circulation and lasted for around 6 weeks (represented by the colored horizontal lines in after the calculated community midpoint. Regional epidemic patterns were also similar when stratified by age group across seasons ( Figure 4B).

| DISCUSSION
Our study compared epidemic influenza data from three distinct sur- Each of the three networks uses influenza surveillance as a means to an end; the primary purpose of each study is to gauge influenza vaccine effectiveness against different outcomes (communityacquired, medically-attended, and hospitalized disease). 16,20,22 Surveillance data from the household network has also been used to provide estimates and predictors of vaccine uptake, as well as influenza transmission parameters. 15,23 Therefore each individual network is necessary for a better understanding of influenza prevention in these three distinct settings.
A closer look at our results, specifically at the few instances when the three curves did not agree, also attests to the unique information provided by each of the three surveillance systems. When data from the three systems are considered all together, we are provided with a F I G U R E 3 Overall epidemic curves for influenza A and B recorded by three surveillance networks of Southeast Michigan across five surveillance seasons. The left column (A) represents influenza A cases (aggregate of H1N1, H3N2, undetermined subtypes) and the right column (B) represents influenza B cases (aggregate of Victoria, Yamagata, undetermined lineages). The grey bars reflect the epidemic curve of all influenza cases from all networks reported that season. The colored lines reflect influenza A and B epidemics of each of the three networks. Y-axis units represent the network-standardized weekly number of cases reported, as a proportion of all cases reported in the network that season. Horizontal, colored lines are equivalent within rows and represent the period during which the middle 50% of all influenza cases (influenza A and B combined) were reported to a given network that season more holistic view of regional influenza epidemics. Below, we highlight three seasons in which discrepancies in network epidemics helped to retrospectively explain characteristics of seasonal influenza, including vaccine effectiveness, disease severity, and circulating strains.
First, the 2014/2015 season is notable in our results for the early activity and rapid escalation to peak of all three networks.
This matches national summaries of that season. 24  A reliable early warning sign of impending influenza epidemics would be indispensable for public health preparedness and resource allocation-particularly in the event of a novel pandemic strain. 30 Because research has established school as a driver of communicable disease spread and children as important introducers of virus to their households, we initially thought that a season's influenza activity would be first picked up by our household study. [31][32][33][34][35][36][37] This was not the case. While households reported ARI throughout the year ( Figure S1), there was no persistent early reporting of confirmed influenza in our household network (or the other two networks, for that matter). We were also interested in whether a particular age group would experience influenza activity before all others; in one past study, children experienced higher influenza risk before the community epidemic midpoint compared to older age groups. 38 After we stratified epidemic curves into four age groups, we found no evidence that influenza activity was occurring persistently earlier in children or any other age group ( Figure 4B). This is not to say that children are not important purveyors of influenza transmission in a community.
Instead, our regionally-focused data suggest that children did not serve as the proverbial "canaries in the coal mine" for community outbreaks during those five seasons. 39 While considering data from all three systems together provided us with a more comprehensive view of the overall, regional epidemic

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1111/irv.12944.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.