|
Activity Number:
|
159
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #304079 |
|
Title:
|
Quantile Regression for Medical Cost Data
|
|
Author(s):
|
Huixia (Judy) Wang*+ and Xiao-Hua (Andrew) Zhou
|
|
Companies:
|
North Carolina State University and University of Washington
|
|
Address:
|
2501 Founders Drive, 209F Patterson Hall, Raleigh, NC, 27695,
|
|
Keywords:
|
Conditional mean ; Health care cost ; Heteroscedastic regression model ; Skewed data ; Two-part model
|
|
Abstract:
|
We propose a new approach to analyze skewed and heteroscedastic medical cost data through regression of conditional quantiles of the response variable on the transformed scale. Using the appealing equivariance property of quantiles to monotone transformations, we propose a distribution-free estimator for estimating the conditional mean cost on the original scale. The proposed method is extended to a two-part heteroscedastic model to account for zero measurements commonly seen in medical cost studies. The proposed approach applies to a wide class of heteroscedastic models without making any parametric assumptions on the error distribution. Through a simulation study, we demonstrate that the proposed estimator has competitive and more robust performance than existing estimators in a varying number of models considered. Finally we illustrate the proposed method in a real health care study.
|
