Quantile Prognostic Scores

• Activity Number: 632
• Type: Topic Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Ben Kelcey* and Chris Swoboda and Jiaqi Zhang and Zuchao Shen
• Companies: University of Cincinnati and University of Cincinnati and University of Cincinnati and University of Cincinnati
• Keywords: propensity score ; prognostic score ; causal inference ; observational design

Abstract

In parallel to the propensity score, recent literature has developed the prognostic score to construct models of the potential outcomes. Whereas the propensity score provides a unidimensional description of how pretreatment covariates relate to the probability of selecting into a treatment group, the prognostic score provides a summary of how pretreatment covariates relate to the expected response under the control condition. Prognostic scores are constructed by estimating the covariate-outcome relationships using only units in the control group and then applying those relationships to treated units to predict their expected response under the control condition. Because prognostic score models often draw on outcomes with continuous distributions, traditional mean-based approaches to estimating prognostic scores may frequently be insufficient in capturing heterogeneity in the covariate-outcome relationships across the outcome distribution. In this study, we develop quantile-based prognostic scores and integrate them with propensity scores to present a more comprehensive assessment of potential causal effects.