Age-Structured Influenza Prediction based on Kalman Filter

Influenza epidemics are non-negligible threats to human health, causing three to five million cases every year, with deaths ranging from 250,000 to 500,000. The spreading of influenza imposes a considerable economic burden, with a loss of 71-167 billion US dollars every year. Early surveillance and prediction play a key role in preventing influenza activity. ILINet surveillance data offers detailed and accurate information about influenza, especially in each age group, but the reporting lag postpones it from giving a real-time detection. For compartmental model, influenza trends can be simulated by differential equations. Theoretically speaking, influenza trends in each age group can be achieved. However, its ability of simulating decreases as errors accumulate when model runs. Google Flu Trends offers a relatively accurate estimation about the overall influenza cases, but it fails to provide demographic data, for example, the influenza trends in each age group are not available. These three sources of detecting have different advantages. By combining them we can obtain a more comprehensive detecting of influenza. The objective of this research is to use Kalman Filter to assimilate multi-sources of data and provide an accurate, real-time detection about influenza trends in each age group (0-4 years, 5-24 years, 25-64 years, older than 65 years) with optimized results which tend to be closer to the true values. Result shows that, this method can give a real-time influenza detecting in each age group with correlation ranging from 0.93 to 0.95 from year 2005 to 2010.