Outpatient influenza antivirals in a distributed data network for influenza surveillance

Electronic data collected from routine health care can be used for public health surveillance. To examine the Sentinel System, a distributed data network of health plans, as a source for influenza surveillance, we compared trends in outpatient prescription dispensings of influenza antivirals in Sentinel to trends in CDC's ILINet and NREVSS systems over five seasons. There were 2 102 885 dispensed prescriptions of oseltamivir capsules, 494 188 of oseltamivir powder, and 7955 of zanamivir. Across all seasons, the magnitude and timing of peaks in drug utilization were highly comparable to those in ILINet and NREVSS. Oseltamivir capsules and powder were well correlated with ILINet and NREVSS. This lays the foundation for further exploration of Sentinel's utility for influenza surveillance.


| BACKG ROU N D
In the United States, public health authorities utilize numerous data sources to monitor the annual influenza season (October through May). Because most people with influenza or other respiratory viruses that co-circulate do not seek medical care-and those who do seek are not necessarily tested-sources ranging from syndromic surveillance to laboratory testing data are necessary to gauge the timing and severity of the season. Simonsen and colleagues have previously recommended the use of claims data to enable, among other things, local-level estimates of influenza activity. 1 Viboud and colleagues also demonstrated that an algorithm for influenza-like illness (ILI) applied to medical claims data correlated very well with local and regional ILI data as well as laboratory data. 2 The US Food and Drug Administration (FDA)'s Sentinel System is an active surveillance system that uses electronic healthcare data, primarily administrative claims from commercial insurers, from multiple sources to monitor the safety of regulated medical products. 3 FDA has envisioned Sentinel to be a national resource for evidence generation, and for it to be leveraged for other public health, uses as well as research. 4 We examined whether claims data for outpatient dispensings of prescription influenza antiviral drugs in the Sentinel System might serve as an additional source of influenza surveillance data. We calculated the rate of incident influenza antiviral drug prescriptions dispensed over multiple influenza seasons. We then compared trends in Sentinel to those in the Centers for Disease Control and Prevention  We plotted the prescription dispensing rate data, the proportion of ILI among encounters in ILINet, and the proportion of positive tests among those tested for influenza from NREVSS by month-year and visually compared timing and magnitude of trends in the two systems for each season. We calculated Pearson correlation coefficients for the full study period for the three different dispensing types with respect to ILINet and NRVESS data, separately, using SAS version 9.4. This work was conducted under the Sentinel System as a public health activity, not research, and is therefore not under the purview of IRB. 5

| RE SULTS
Over the study period, among 101 947 808 eligible members, there were 2 102 885 episodes of oseltamivir capsules, 494 188 of oseltamivir powder, and 7955 of zanamivir in the participating Sentinel Data Partners. As shown in Table 1, the influenza antiviral prescription dispensing data in Sentinel yielded expected results. For example, of the little zanamivir use, almost none was in children under 5 years for whom the drug is not approved. Adults were more likely to receive oseltamivir capsules, and children were more likely to receive powder. On average, there was one dispensing per user and 6-7 days supplied per dispensing, depending on formulation.
When the monthly rates of outpatient oseltamivir prescription dispensings were compared with outpatient ILI trends and laboratory test results, we observed excellent overlap with respect to timing ( Figure 1). Further, the general magnitude-or severity-of each season, as depicted in the ILI data, is also evident in the Sentinel dispensing data. For example, the 2011-2012 season was mild and we correspondingly observed a lower rate of oseltamivir use compared   Sentinel System has previously demonstrated the capability to obtain "fresher" data with a lag time of 6 weeks to assess influenza vaccine safety. 6 Outpatient pharmacy data are routinely available within 2 weeks. Finally, Sentinel data have the potential to provide locallevel data on drug dispensings, vaccine utilization, and outcomes like influenza illness as member zip code is available. The ability to examine large-scale surveillance data on a regional or local level could be a major addition to national influenza surveillance efforts, although in this feasibility analysis we looked across all available regions.

| D ISCUSS I ON
One limitation to consider when interpreting our results is that CDC tracks ILINet and NREVSS data by MMWR week. Because the Sentinel data are routinely available by month-year, we transformed the ILINet and NREVSS data into months; we assigned weeks that We therefore could not make exact monthly or weekly comparisons between the sources in this analysis. However, the overlap in trends supports the validity of our methodology. In addition, more granular data are technically feasible in Sentinel and might be used in subsequent work.
In conclusion, we have shown that influenza antiviral dispensing data in the Sentinel System may be a new source of national influenza surveillance data. This analysis lays the groundwork for additional studies to explore and utilize Sentinel in new and important ways.