Build Multilevel Models to Assess the Length to Inpatient Readmission Using SAS ® PROC MIXED

Multilevel models are known as hierarchical linear models or general linear mixed models. The defining feature of these models is their capacity to provide quantification and prediction of random variance due to multiple sampling dimensions (across occasions, persons, or groups). Multilevel models offer many advantages for analyzing both hierarchical models which analyze the data on individuals nested within hierarchies (e.g., patients within hospitals) and individual growth models which are designed for exploring the longitudinal data over time. Multilevel models become very popular in educational and behavioral research, but they are still new to healthcare research, especially for health insurance outcome assessment. Multilevel models could fit nicely with the nature of patients’ healthcare data (both hierarchical and longitudinal structures). SAS PROC MIXED offers great flexibilities to fit many common types of multilevel models. This paper is to present how to utilize SAS PROC MIXED to model two-level effects on the length to inpatient readmission in healthcare setting. Both level-1 and level-2 predictors are examined to see how long it will take for a patient to be readmitted to the same hospital for the same diagnosis. First, an unconditional means model is fitted as the baseline model. Second, level-1 predictors and level-2 predictors are added into the separate models to estimate fixed effects and random effects. Finally, both level-1 and level-2 predictors are included in the model and the model is assessed through various statistics. The paper also shows some strategies on building multilevel models and how to interpret the outputs.