Relaxed Estimating Equations for Sparse Mark Variable Regression

• Activity Number: 148
• Type: Invited
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Brent Johnson*+
• Companies: Emory University
• Keywords: Censored data ; Dantzig selector ; Lasso ; Penalized estimating functions ; Survival analysis

Abstract

Lung cancer is the leading cause of cancer death in the United States and leads to billions of dollars in medical expenditures each year. Lifetime medical cost for lung cancer patients is an example of a mark variable which is collected over the survival time duration. Because we only observe the mark if the lifetime is uncensored, the joint survival-mark model induces a complicated dependent censoring problem. Here, we propose new regularized estimators of the regression coefficients in a semi-paraemtric model for lifetime medical cost through a theory of relaxed estimating equations. The estimators simultaneously estimate regression coefficients for active variables while setting coefficient estimates for inactive variables exactly to zero. We illustrate the methods through simulation studies and application to clinical data from the Southwest Oncology Group.