Multiple Treatment Groups: A Case Study with Health Care Practice and Policy Implications

• Activity Number: 410
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Alexandra Hanlon*+ and Karen Hirschman and Beth Ann Griffin and Mary Naylor
• Companies: University of Pennsylvania and University of Pennsylvania and RAND Corporation and University of Pennsylvania
• Keywords: generalized boosted modeling ; propensity score ; observational studies ; policy ; RCT ; R

Abstract

The traditional RCT has been the gold standard for minimizing bias associated with observed and unobserved baseline characteristics between 2 groups, thus enabling an unbiased estimation of treatment efficacy. Frequently randomization is not possible due to ethical or logistical reasons. In such situations, propensity scores (PS) have been used to control for baseline imbalances on observed characteristics. The PS was originally proposed, and is most commonly used, as a method for producing balance on many covariates between 2 groups. Currently available software tools and guidance to perform 2-group analyses can be found in readily available software packages (SAS, Stata, R). However, tools and guidance for analyses of >2 treatments are only beginning to emerge. Here we present a case study, one with tremendous health care practice and policy implications, designed to compare 3 interventions on the outcome of rehospitalization for hospitalized cognitively impaired older adults. We describe the use of an R package to implement PS weighting for multiple treatments via generalized boosted models for estimation of the necessary PS weights. Outcome models will also be presented.