Use of a statistical algorithm to classify influenza infection by lung and plasma cytokine and chemokine production profiles

James P. Long, Patrick J. Sabourin, Jeremy L. Craft, Mark S. Kotur, Gregory V. Stark, Manjula Kasoji, Katrina M. Waters, Carol L. Sabourin and Herbert S. Bresler

: J Antivir Antiretrovir

Abstract :

I nfluenza epidemics result in approximately 3 - 5 million cases of severe disease and 250,000 to 500,000 deaths annually. Rapid classification of an influenza infection is of utmost importance in determining the proper treatment regimen. In this study, mice were infected with one of three strains of influenza, 2009 swine-origin influenza A (H1N1) A/California/04/09, seasonal H1N1 A/Texas/36/91, and the highly pathogenic avian (H5N1) A/Vietnam/1203/0 or vehicle control. Levels of thirty seven cytokines and chemokines in lung (6, 12, 24, 72 and 96 hr) and plasma (24 and 96 hr) were measured for 4 days following challenge and analyzed statistically using a support vector machine, a statistical concept using supervised learning methods for classification and regression analysis. Using this method, at a given time point post-challenge, only 2 or 3 cytokines in lung or plasma were needed to predict the influenza type with 100% accuracy at a given time post-infection. Since the exact time of influenza infection is seldom known, the ability of the support vector machine procedure to classify the type of influenza infection independent of the time post-infection was also tested. Combinations of 10 lung or 14 plasma cytokines/chemokines were able to provide a 100% classification accuracy of the influenza type over a 96 hour period post-infection