Regional and National Influenza Surveillance:
How do They Compare?
Background: Spatial and temporal surveillance on a national level provides many advantages. Local municipalities are able to enhance surveillance and nationally, comparable and timely data allow wide-scale epidemiologic investigations. BioSense, a national system developed by the CDC, has shown promise as a surveillance tool for influenza. We previously demonstrated significant correlation with hospital-based systems in Denver and Baltimore however; validation against health department surveillance systems is required.
Objective: To validate BioSense signals at Denver Health (DH) and Johns Hopkins Hospital (JHH) against influenza surveillance systems used by the Colorado Department of Public Health (CDPHE) and the Baltimore City Health Department (BCHD).
Methods: BioSense signals were generated from chief complaint (text), final diagnoses (ICD-9) and laboratory data fed to the CDC from DH and JHH in near real-time. Information was ‘binned' according to an algorithm designed to detect influenza-like-illness (ILI). The rate of ILI visits detected by BioSense at DH was compared to rates generated by two sentinel programs at CDPHE: outpatient visits reported by providers and the positive ILI lab tests provided by laboratories. BioSense counts generated at JHH were compared to the number of over-the-counter (OTC) medication sales and ILI-related chief complaint visits reported to BCHD. We plotted cases/rates for both regions for the 07/08 season. We splined each season into onset(s) and decline(s) and perform linear correlations to generate quantitative measures of similarity.
Results: In Baltimore, influenza A and B overlapped, providing one peak for analysis. Correlations between BioSense and both BCHD data sources were strong (rho > 0.76) and significant (p>=0.002) for onset and decline. Sentinel provider and BioSense cases peaked during the same week while OTC counts peaked the week prior. Denver experienced two distinct peaks starting with influenza A and ending with influenza B. Correlations between BioSense rates of ILI and both sentinel programs were strong (rho > 0.88) and significant (p> 0.0001) for the onset of A and decline of B. Correlations in the middle of the season were not as strong and all were insignificant: a possible result of a small number of time points. While positive labs lagged all systems for Influenza A, they peaked 3 to 4 weeks earlier for Influenza B.
Conclusions: Biosense trends well with a diverse set of health department surveillance systems for influenza, providing additional evidence that a national model could be a good enhancement or proxy for municipalities lacking an effective surveillance system. Further, in the context of influenza, it provides an ideal data source for large epidemiologic investigations and long-term surveillance previously impracticable due to lack of timely, comparable data sources.