In 2002, Medicare implemented a prospective payment system (PPS) for inpatient rehabilitation facilities (IRFs). The PPS works by assigning patients to groups according to how well patients function. These groups, called function-related groups (FRGs), are then used to predict the cost of treating particular Medicare patients according to their ability to function in four general categories: transfers, sphincter control, self-care (e.g., grooming, eating), and locomotion. Patient functioning is measured according to 18 categories of activity-13 motor tasks, such as climbing stairs, and 5 cognitive tasks, such as recall. As part of a contract to monitor how accurately the IRF PPS is predicting treatment costs, the Center for Medicare and Medicaid Services (CMS) asked RAND to examine possible refinements to the FRGs to identify potential improvements in the alignment between Medicare payments and actual hospital costs. Several developments make it likely that significant refinements can be made: a new recording instrument, known as the IRF Patient Assessment Instrument, containing questions that improved the quality of the patient information available to us; more recent data on a larger patient population that describe the entire universe of rehabilitation patients; improvements in the algorithms that produced the initial FRGs, which should improve prediction of treatment costs; and the two years that have passed since the initial FRGs were created, during which changes in the cost structure of IRFs have occurred. Our analysis had two specific objectives: (1) to explore whether the new data enable better prediction of treatment costs and (2) to assess possible refinements to the FRGs based on the new data. To address the first objective, we reexamined assumptions about whether particular indicators that an activity was not observed, or “missing,” indicated a lack of functioning or simply absent data. We also looked at the usefulness of some new indicators in the IRF PAI data for predicting costs. To address the second objective, we also performed two tasks: First, we considered whether alternative indices that included weighting for patient functioning might predict costs more accurately; second, we ran the algorithm used in 1999 to derive FRGs with the new IRF PAI data to see whether the FRGs would look substantially different. Our analysis identified several potential areas of refinement for the payment system, assuming the analysis effects we observed hold up on 2003 data: missing indicators, importance of “function modifiers,” indices and weighting, and refinements to the FRGs. For example, the earlier data assumed that no report about a particular function meant that patients were unable to perform it, an assumption that held true for most activities. However, we found that a lack of data for “transfer to toilet” and “transfer to tub” should be interpreted less strongly than for the other missing indicators. The more-nuanced information about patient functioning provided by “function modifiers,” such as distance walked, adds information to the basic functional independence measurement, or FIM™, category, such as “walking.” By using a motor index that does not equally weight all components, some improvement in explanatory power could be expected. Moreover, using the 2002 data in an algorithm that produced the 1999 FRGs, we found many fewer payment groups across the various conditions.
Available at: http://works.bepress.com/melinda_buntin1/22/