Analysis of Joint Models for Mixed Outcomes: Some Capabilities with SAS Software

• Activity Number: 679
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Joseph Gardiner*+
• Companies: Michigan State University
• Keywords: random effects ; endogeneity ; mixed models

Abstract

Outcomes with different attributes, of continuous, count and categorical types are often encountered jointly in many settings. For example, two widely used measures of healthcare utilization, length of stay (LOS) and cost can be analyzed jointly with LOS as a count and cost as continuous variables. Occurrence of an adverse event (binary) would impact both outcomes. The challenge is specifying a joint model for the three outcomes. For fitting marginal distributions and assessing the impact of explanatory variables on outcome SAS offers a number of procedures. Correlation and clustering are additional features of these outcomes that must be addressed in analyses. We survey some SAS procedures that can be applied to modeling multivariate outcomes of mixed types. Examples from the extant literature are used to demonstrate the application of the procedures.