Online Program

Return to main conference page

Thursday, January 11
Thu, Jan 11, 2:00 PM - 3:45 PM
Crystal Ballroom F
Use of CMS Data and Tools to Help Researchers

Outcome-Dependent Sampling in Cluster-Correlated Data Settings with Application to Hospital Profiling (304000)

Sebastien Haneuse, Harvard University 
*Glen W McGee, Harvard University 
Sharon-Lise Normand, Harvard Medical School 
Jonathan S Schildcrout, Vanderbilt University 

Keywords: outcome-dependent sampling, cluster-stratified case-control, generalized linear mixed models, hospital profiling

Hospital readmission is a key marker of healthcare quality used by the Centers for Medicare and Medicaid Services (CMS) to determine hospital reimbursement rates. Analyses of readmission are based on a logistic-normal generalized linear mixed model (LN-GLMM) that permits estimation of hospital-specific measures while adjusting for case-mix differences. Recently, a bill was introduced to Congress that would require CMS to include currently unobserved measures of socioeconomic status in their case-mix adjustment without further burden to hospitals. We propose that detailed socioeconomic data be collected on a sub-sample of patients via an outcome-dependent sampling scheme: the cluster-stratified case-control design. Towards valid estimation and inference for both fixed and random effects components of an LN-GLMM, we propose an inverse-probability weighted likelihood approach. The methods are motivated by and illustrated with data on N=889,661 Medicare beneficiaries hospitalized between 2011-13 with a diagnosis of congestive heart failure. Small-sample operating characteristics and design considerations are evaluated via comprehensive simulation study.