Prediction of Future Cost for Congestive Heart Failure Patients Using Heavy-Tailed Data

• Activity Number: 377
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jun Han*+ and Krassimir Latinski
• Companies: MEDai, a LexisNexis Company and Elsevier/MEDai
• Keywords: Statistical Learning ; Heavy-tailed Data ; Healthcare Cost Prediction

Abstract

Congestive Heart Failure (CHF) represents an increasing burden to the healthcare system. According to statistics from the American Heart Association, in 2004, the 1.1 million hospitalizations for CHF in the United States amounted to nearly $29 billion in hospital charges. Consequently, efforts have been made to predict future cost of CHF patients, in order to improve care management and reduce cost. The distributions of the dependent variable and the predictors involving healthcare cost and utilization are usually seriously heavy tailed and asymmetric. This is a cause for a regression-based prediction to become much more difficult. In order to find a proper remedy, we compared the effects of a series of solutions for heavy tailed distribution data modeling, such as variable transformation, outlier truncation, and several robust algorithms, which are designed to overcome the challenges of heavy tailed distribution data. The presented results are based on simulation data as well as real healthcare claim data.