BACKGROUND: Syndromic surveillance systems collect near-real-time, pre-diagnostic data and use algorithms to categorize the information into “syndromes” that can be tracked over time. Syndrome validity and usefulness should be evaluated whenever possible as syndrome definitions can be imprecise. The Pennsylvania Department of Health (PADOH) collects syndromic data, via a vendor, from the majority of hospital emergency departments (EDs) in the state. PADOH also collects ILI data from healthcare providers participating in CDC’s ILINet program, and lab-confirmed cases of influenza (including cases confirmed by point-of-use rapid tests) are reportable per state regulation. The performance of a syndromic ILI (ILI-S) definition developed by a BioSense Syndrome Definitions workgroup was evaluated by comparing ILI-S trends to data from ILINet and our reportable disease surveillance system.
METHODS: Data from the past seven influenza seasons (October 2008 through September 2015) were obtained from our three data sources. To adjust for fluctuations in the number of hospitals submitting syndromic data, we calculated the percent of total ED visits that met the ILI-S definition by week. ILINet data is reported weekly as a percent of outpatient visits due to ILI. We aggregated reported influenza case counts by week. Correlations between weekly ILI-S, ILINet and reported case data were evaluated using Spearman correlation coefficients and linear regression. We also visually inspected time trends from the three data sources.
RESULTS: Visually, all three measures of ILI followed similar trajectories over time. Spearman correlation coefficients showed statistically significant positive correlations between ILI-S and ILINet, in all influenza seasons, ranging from rs=0.71 (season 2008-09) to 0.95 (2009-10). Correlations between ILI-S and reported cases ranged from rs=0.73 (season 2008-09) to 0.96 (2010-11), with the exception of season 2011-12 where rs=0.38. That season was notable for having very few reported cases with a limited seasonal peak. The regression indicated that ILI-S was a good predictor of reported cases (R2= 0.78, p<0.0001).
CONCLUSIONS: Correlation and regression analyses suggest that syndromic surveillance using the workgroup-developed ILI algorithm is a useful method for monitoring influenza trends in moderate to severe influenza seasons. ILI-S also correlated well with ILI data from ILINet, during mild and severe seasons. A major advantage of syndromic surveillance is that it places no reporting burden on providers and as such may be more sustainable, especially during severe influenza seasons, than systems that require providers to report directly.