Influenza Forecasting with Google Flu Trends

OBJECTIVE: We sought to develop a practical influenza forecast model, based on real-time, geographically focused, and easy to access data, to provide individual medical centers with advanced warning of the number of influenza cases, thus allowing sufficient time to implement an intervention. Secondly, we evaluated how the addition of a real-time influenza surveillance system, Google Flu Trends, would impact the forecasting capabilities of this model. INTRODUCTION: Each year, influenza results in increased Emergency Department crowding which can be mitigated through early detection linked to an appropriate response. Although current surveillance systems, such as Google Flu Trends, yield near real-time influenza surveillance, few demonstrate ability to forecast impending influenza cases. METHODS: Forecasting models designed to predict one week in advance were developed from weekly counts of confirmed influenza cases over seven seasons (2004 – 2011) divided into training and out-of-sample verification sets. Forecasting procedures using classical Box-Jenkins, generalized linear, and autoregressive methods were employed to develop the final model and assess the relative contribution of external variables such as, Google Flu Trends, meteorological data, and temporal information. Models were developed and evaluated through statistical measures of global deviance and log-likelihood ratio tests. An additional measure of forecast confidence, defined as the percentage of forecast values, during an influenza peak, that are within 7 influenza cases of the actual data, was examined to demonstrate practical utility of the model. RESULTS: A generalized autoregressive Poisson (GARMA) forecast model integrating previous influenza cases with Google Flu Trends information provided the most accurate influenza case predictions. Google Flu Trend data was the only source of external information providing significant forecast improvements (p = 0.00002). The final model, a GARMA intercept model with the addition of Google Flu Trends, predicted weekly influenza cases during 4 out-of-sample outbreaks within 7 cases for 80% of estimates (Figure 1). CONCLUSIONS: Integer-valued autoregression of influenza cases provides a strong base forecast model, which is enhanced by the addition of Google Flu Trends confirming the predictive capabilities of search query based syndromic surveillance. This accessible and flexible forecast model can be used by individual medical centers to provide advanced warning of future influenza cases.