The curves on this website are real-time and forecast estimates of flu activity in the USA. The displayed date corresponds to the last day of the week. The plot contains 5 different estimates. These are described below.
The US Centers for Disease Control (CDC) continuously records the percentage of patients seen in clinics who exhibit influenza-like illnesses (ILI) according to physicians' reports. A patient is said to have an Influenza-like illness if she/he has a fever plus a cough and/or sore throat. The traditional clinical verification and distribution of this information takes approximately two to three weeks. This means that we do not have a sense of what is happening now using CDC information, we only have verified information of what happened two or three weeks ago.
This estimate of flu activity uses search data provided by Google, in combination with a dynamic multi-variate approach and historical information about previous flu outbreaks (different from the now discontinued Google Flu Trends), to provide an accurate estimate of flu activity in the US. The methodology behind this algorithm can be found in: Yang S., Santillana M., and Kou S.C. (2015) Accurate estimation of influenza epidemics using Google search data via ARGO. Proceedings of the National Academy of Sciences 112.47, pp:14473-14478.
Flu Near You (FNY) is a crowd-sourced participatory disease surveillance system. It gathers information from users about their health in real-time. The plot shows the percentage of ILI-positive reports with respect to the total number of reports. The data has been smoothed with a spike detection algorithm to remove media spikes. The details of our methodology are presented in: M. S. Smolinski, et al. (2015) Flu Near You: Crowdsourced Symptom Reporting Spanning 2 Influenza Seasons. American Journal of Public Health e1-e7.
athenahealth (ATH) is a company specialized in medical practices management and electronic health records management. Using anonymous aggregated data from athenahealth real-time database, we calculated an estimate of the weekly percentage of patients seeking medical attention with ILI symptoms. The curve shows these values scaled to fit historical CDC data. The details of the methodology behind these predictions are presented in Santillana et al. (2016) Cloud-based Electronic Health Records for Real-time, Region-specific Influenza Surveillance.
By combining data from CDC, Google, athenahealth, and FNY, we produce weekly estimates of flu activity (trained to predict CDC-reported ILI) for last week, current week, next week, and the week after next. Inspired by machine learning approaches, our ensemble methodology is capable of producing more accurate (and more robust) predictions than any of the independent curves shown in the website. The details of the methodology behind these predictions are presented in: Santillana et al. (2015) Combining Search, Social Media, and Traditional Data Sources to Improve Influenza Surveillance. PLoS Computational Biology. 11(10): e1004513.