Targeted Maximum Likelihood Estimation of Treatment-Specific Survival Curve with Right-Censored Data and Covariates in Observational Studies

• Activity Number: 20
• Type: Topic Contributed
• Date/Time: Sunday, August 2, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ori M. Stitelman and Farid Jamshidian*+ and Alan Hubbard and Mark J. van der Laan
• Companies: University of California at Berkeley and University of California, Berkeley and University of California, Berkeley and University of California, Berkeley
• Keywords: Survival Analysis ; Machine-Learning ; Targeted Maximum Likelihood

Abstract

The treatment specific survival distribution is a common parameter of interest in many observational studies. The targeted maximum likelihood estimate of the survival curve presented here improves on previously used estimating equation based methods for this parameter of interest in the following ways: the estimate is a substitution estimator, the estimate follows the natural bounds of being a probability, and the method creates a criteria targeted to the parameter of interest on which to base choices between models of the censoring and treatment mechanisms. The existence of these criteria allows one to create black box algorithms that use machine-learning methods to choose between models of the censoring and treatment mechanisms.