Quality metrics play an increasingly important role in the evaluation and reimbursement of post-acute providers, and inpatient rehabilitation facilities (IRFs) in particular. The Medicare Payment Advisory Commission (MedPAC) uses aggregate trends in quality measures, such as functional gain, to assess the adequacy of payments to providers. In addition, the Patient Protection and Affordable Care Act (ACA) requires IRFs to publicly report quality data or be penalized in annual payment updates. Eventually, quality data could be used in a value-based purchasing scheme for IRFs, as is being developed for other post-acute settings under ACA.

In order to distinguish IRFs from acute...

Our primary data source is the Medicare Inpatient Rehabilitation Facility Patient Assessment Instrument (IRF PAI) from 2004 to 2009. At admission, patients’ personal and demographic characteristics, admission class, preadmission living status, insurance information, impairment group (i.e., reason for rehabilitation), etiologic diagnosis (i.e., diagnosis leading to impairment group), up to ten comorbid conditions and preexisting complications, functional status, cognitive status, and other information are entered on the IRF-PAI. The impairment requiring rehabilitation, cognitive and functional status at admission, age, and the severity of comorbidities are then used to assign each patient to a case mix group, which determines the amount of...

We consulted the literature to define a set of base independent variables that are likely to influence rehabilitative success, disposition of discharge, and subsequent readmissions, and that are available in the IRF PAI, prior acute hospital claims, and the Medicare Denominator File. All health status related measures must be defined at the time of or before admission to the IRF so they are not influenced by the quality of the IRF. Variables from the Denominator Files include age (controlled using indicator variables for five-year intervals: under age 50, ages 50–54, 55–59, 60–64, 65–69, 70–74, 75...

We used a consistent approach to develop, estimate, and test risk-adjustment models for five patient outcomes: (1) FIM gain, (2) discharge to community, (3) 30-day readmission to acute care, (4) 30-day readmission to a SNF, and (5) discharge directly to acute care. For each outcome, we began by specifying a model that contained individual socioeconomic and demographic characteristics, comorbid condition indicators, IGC indicators, and age-by-sex interactions. We estimated nine alternative specifications for each model and chose the model with the best AIC score. In each case, the best model was either the full model or the model that dropped the...