Evaluating Google Flu Trends during the H1N1 Pandemic
Author Information
Author(s): Samantha Cook, Corrie Conrad, Ashley L. Fowlkes, Matthew H. Mohebbi
Primary Institution: Google, Inc.
Hypothesis
Can Google Flu Trends provide accurate estimates of influenza activity during a non-seasonal outbreak like the H1N1 pandemic?
Conclusion
The updated Google Flu Trends model performed better during the H1N1 pandemic compared to the original model, especially during the summer months.
Supporting Evidence
- The updated model included more search query terms directly related to influenza.
- Both models were highly correlated with ILINet data during the pre-H1N1 period.
- The original model underestimated ILI activity during the H1N1 pandemic.
- The updated model performed better during the H1N1 pandemic, especially in Wave 2.
Takeaway
This study looked at how well Google search data predicted flu activity during the H1N1 outbreak, finding that the updated model was more accurate than the old one.
Methodology
The study compared weekly estimates of influenza-like illness from Google Flu Trends with official CDC data, calculating correlation and root mean square error for different time periods.
Potential Biases
Potential bias due to reliance on search query data, which may not fully capture actual health-seeking behavior.
Limitations
The original model underestimated ILI activity during the initial wave of H1N1 due to changes in search behavior and the timing of the outbreak.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website