Ranking of Pandemic Influenza Mitigation Strategies

• Activity Number: 244
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Greg Lambert*+ and Andrew Huff and Patrick Finely
• Companies: Sandia National Laboratories and Sandia National Laboratories and Sandia National Laboratories
• Keywords: Random Forests ; predictive modeling ; pandemic influenza ; variable importance ranking ; multinomial logistic regression

Abstract

Pandemic influenza mitigation strategies are ranked based on their effectiveness at preventing infections in modeled populations. Alternative ranking methods are evaluated for predictive power and sensitivity. A new approach to assessing variable importance for Random Forests called k-fold variable importance (KFVI) is applied to quantify the loss in prediction for each classification in the output space. Finally, the KFVI results are compared to both permutations based variable importance ranking and conditional variable importance ranking methods. Analysis indicates that ranking using the KFVI method is better at assessing demographic factors (sick at home, babysitting, children per household, teens per household, and adults per household) that contribute to pandemic influenza mitigation strategies.