Prediction of Future Cost for Congestive Heart Failure Patients: A Second Look

• Activity Number: 466
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Donghui Wu*+ and Xin Deng
• Companies: LexisNexis and LexisNexis
• Keywords: Health Cost Prediction ; Robust Regression ; Local Regression Ensembles ; Outlier Reduction ; Performance Evaluation ; Population Health Management

Abstract

Congestive heart failure (CHF) affects people of all ages, and more than 5 million Americans are currently living with CHF. It is the leading cause for hospitalization among elders and accounts for more than 1 million hospital admission. It is estimated that CHF costs the nation more than $34 billion. Consequently, we would like to predict the future cost for CHF patients for the purposes of chronic disease management and interventions. However, the distributions of the dependent variable and the predictors involving healthcare cost and utilization are usually seriously heavy tailed and asymmetric. In our last year's paper, we demonstrated that there is no significant differences in R-squared among Ordinary Least Square regression, Robust Regression for Asymmetric Tails, Quantile Regression Neural Network, Least-Trimmed Squares Regression, and other robust regression methods. This year, instead of global regressions, we further studied the distributions and effects of outliers, and designed/experimented with a number of local robust regression ensembles with various outlier reduction techniques.