Estimating Causal Effects with Prognostic Scores: An Alternative Use of Outcome Regression Models for Dimension Reduction

• Activity Number: 22
• Type: Topic Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Ingeborg Waernbaum*+
• Companies: Umeå University
• Keywords: prognostic scores ; propensity score ; matching

Abstract

In semi-parametric estimation of an average causal effect model assumptions that partially specifies the joint distribution of the outcome, treatment and the covariates are used. Here, the dimension reducing models are functions of the covariates e.g., the propensity score and outcome regression models. A doubly robust estimator uses both outcome regression and propensity score models in combination. Here, we discuss the use of regression models as prognostic scores as an alternative semi-parametric approach when estimating causal effects. We extend the theory concerning identification of causal effects when using prognostic scores and we show that efficiency can be gained compared to approaches using the propensity score.