74
Using Predicted Therapy Visits in the Medicare Home Health Prospective Payment System
We used Medicare home health claims linked to the Outcomes and Assessment Information Set (OASIS) for all home health episodes by Medicare FFS beneficiaries in 2013. We used a two-part regression model to predict the number of therapy visits provided by HHAs during an episode using a large set of patient and HHA characteristics. We compared a simulated payment based on predicted therapy visits to payment based on actual therapy visits provided.
For 42% of home health episodes, predicted and actual therapy visits fell within the same therapy thresholds used in payment. However, among episodes with 14+ therapy visits, only 3% of episodes had predicted and actual therapy visits falling within the same threshold; most high therapy episodes had substantially fewer predicted visits. Using predicted therapy visits to simulate payments resulted in a 7.3% overall decrease in average payments for home health episodes, or a $1.2 billion decrease in total Medicare home health payments in 2013. Over 60% of the decrease in payments resulted from a decrease in the number episodes predicted to have 20 or greater therapy visits. Some episodes had a higher number of predicted therapy visits compared to actual, resulting in an 11% increase in average payment for low (or zero) therapy episodes.
Medicare home health payments could change substantially if reimbursement were based solely on patient characteristics rather than actual therapy services provided. Predicting therapy use is challenging since the data are based upon the incentives currently in place. Our model predicted fewer high therapy episodes, resulting in potential changes in Medicare reimbursement. However, further exploration is needed to understand high users of therapy services, including other home health services they use, and alternatives available for ensuring high needs patients would still receive appropriate care. Additionally, the model predicted some episodes to have a positive number of therapy visits when the patient received none, and it may be inappropriate to reimburse for therapy that did not occur. Thus, abstracting away from therapy use and moving towards prediction of total resources used during an episode may be more desirable, but would likely require significant modifications to the current payment system.