Internet-Based Epidemiology

Socioeconomic bias in influenza surveillance

“…Finally, to further evaluate alternative explanations for the observed bias, we conducted a simulation experiment to address the role of other factors, such as reduced rates of ILI primary care in lower socioeconomic groups [ 37 , 38 ], lower correlation between ILI-related Internet searches and actual ILI in lower socioeconomic groups [ 39 , 40 ], socioeconomic differences in vaccination levels [ 41 , 42 ], and/or socioeconomic differences in underlying health conditions [ 43 ]. The results of this simulation experiment demonstrate that, when all else is equal, a higher hospitalization rate should increase statistical power and provide greater prediction precision ( S6 Text , S2 and S3 Figs).…”

“…Sixth, we did not have individual-level patient socioeconomic and/or ZIP Code information from ILINet, BioSense 2.0, and GFT, and thus we were unable to assess directly whether lower socioeconomic groups are underrepresented. However, prior studies suggest that lower socioeconomic groups use the Internet less frequently than higher socioeconomic groups, and that disease-related signals derived from Internet-search data poorly reflect incidence in lower socioeconomic communities [ 39 , 40 , 45 ]. Interestingly, our results suggest that predictions based solely on GFT performed no worse in the lowest income quartile than did other candidate predictors.…”