(C-151) An algorithm for identifying observable medication time among US nursing home residents

Background: Nursing home (NH) residents use many medications, with nearly two-thirds using 5 or more. Medicare Part D is commonly used to address drug-related research questions among NH residents, but drugs administered to NH residents during hospitalizations or post-acute care episodes are unobservable due to the bundling of claims under Medicare Part A. Existing approaches account for “unobservable medication time” by crudely categorizing residents as short-stay (≤100 days in the NH) vs. long-stay (>100 days in the NH), but this may miss medication exposures or unnecessarily truncate person-time.

Objectives: To develop an algorithm that identifies all observable medication time among NH residents using Medicare claims data.

Methods: We conducted a cohort study of all NH residents in the US between 2013 and 2020 using Medicare claims (MedPAR), the Master Beneficiary Summary File (MBSF), and the Minimum Data Set (MDS). We constructed episodes of NH time by anchoring on residents’ entry and discharge dates from the MDS. We used the MBSF to identify whether NH residents were enrolled in Medicare fee-for-service (FFS) and Medicare Part D as of each calendar day. We used MedPAR to identify days when medications were unobservable (i.e., the resident was in hospital or receiving post-acute care). We identified “observable medication days” once we removed unenrolled and unobservable medication days. We used descriptive statistics to characterize observable medication days according to periods of time following NH entry (e.g., days 1-30; 31-60, 61-100, and 101+).

Results: We identified a total of 12,336,061 residents and a total of 2,685,903,261 nursing home days. The average age was 76 years (SD=13.0) and 59.2% were women. Among all NH days, 54.7% (n=1,468,869,489 days) were enrolled in FFS and Part D; of which, medications were unobservable for 405,122,647 days (27.6% of enrolled days). Within the first 100 days of NH admission, medications could be observed for: Days 1-30=14.1%; Days 31-60=31.0%; Days 61-100=52.4%; Days 1-100=27.3%. In the >100 days after admission, medications could be observed for 89.4% of enrolled days.

Conclusions: We developed an algorithm to identify observable medication days for NH residents, which avoids the cruder 100-day method used in prior studies and by the Centers for Medicare & Medicaid Services. Our algorithm can be a useful tool for capturing medication exposures among U.S. nursing home residents, for mitigating sources of information and selection bias that can be introduced with 100-day cut-offs, and for increasing total person-time.