Application of a Least Trimmed Squares Robust Regression Model for Estimating Influenza-Associated Deaths by State in the United States

• Activity Number: 83
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Po-Yung Cheng*+ and William Thompson and Eric S. Weintraub and Tuyen Do and David Shay
• Companies: CDC and CDC and CDC and CDC and CDC
• Keywords: robust regression ; influenza ; pneumonia ; deaths

Abstract

To monitor local influenza activity in the US, we developed a state-level robust regression model using the least trimmed squares method (LTS). Instead of arbitrarily setting data points to missing during the winter time, as done for traditional Serfling type regression models, the LTS method prunes outliers based on the minimum trimmed sum of squares. We modeled three death categories for each state from the 1989/90 through 2003/04 respiratory seasons. We compared the LTS estimates to estimates using the traditional Serfling model and the Andrews robust regression model currently used by the Centers for Disease Control and Prevention. Annual averages of influenza-associated deaths using the LTS method for underlying pneumonia and influenza deaths, any-listed pneumonia and influenza deaths, and underlying respiratory and circulatory deaths were 8946, 15566, and 32444, respectively.