Software for Targeted Maximum Likelihood Estimation

• Activity Number: 135
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Susan Gruber*+ and Mark J. Van der Laan
• Companies: University of California, Berkeley and University of California, Berkeley
• Keywords: software ; targeted maximum likelihood estimation ; application ; R ; biostatistics

Abstract

Traditional parametric modeling approaches to effect estimation cannot be successfully applied to many data-driven health-related questions of interest to researchers today, and great strides have been made in developing alternate approaches. Targeted maximum likelihood estimation (TMLE), a double robust semiparametric approach to causal and non-causal (variable importance) effect estimation, is being applied in diverse application areas including survival analysis, safety analysis, comparative effectiveness research, and genomics. This talk introduces two publicly available R packages that implement TMLE for additive point treatment effect estimation: "tmleLite," a basic TMLE implementation, and "ctmleLite", for collaborative targeted maximum likelihood estimation (C-TMLE). Essential aspects of TMLE methodology will be highlighted as each package is demonstrated.