J Am Med Dir Assoc. 2026 Sep 17;27(11):106466. doi: 10.1016/j.jamda.2026.106466. Online ahead of print.
ABSTRACT
OBJECTIVES: To examine factors associated with variation between actual length of stay (LOS) and expected LOS in inpatient rehabilitation facilities (IRFs), to compare LOS variation between traditional Medicare (TM) and Medicare Advantage (MA) beneficiaries, and to rank the relative importance of these factors in explaining LOS variation.
DESIGN: Cross-sectional study.
SETTING AND PARTICIPANTS: Data were obtained from the IRF-Patient Assessment Instrument among Medicare beneficiaries discharged from IRFs between 2019 and 2020. The study included 242,955 rehabilitation episodes across 4 diagnostic groups: stroke (N = 127,685), hip fracture (N = 64,241), cardiac conditions (N = 33,116), and joint replacement (N = 17,913).
METHODS: The primary outcome was the difference between actual LOS and expected LOS (Diff LOS). Multivariable generalized linear regression models were used to examine associations between Diff LOS and temporal, demographic, facility, insurance, and regional factors. Dominance analysis was subsequently applied to evaluate the relative contribution of different predictor domains to explained variation in Diff LOS.
RESULTS: MA status and pandemic-period indicators represented the largest contributors to explained variation across all diagnostic groups. When analyses were stratified by insurance type, demographic characteristics consistently ranked among the most important contributors within both TM and MA beneficiaries. Discharge month and regional fixed effects also consistently ranked among the leading contributors, whereas facility characteristics contributed comparatively less explained variation.
CONCLUSIONS AND IMPLICATIONS: Variation in LOS within IRFs was primarily associated with demographic, temporal, and regional fixed effects. TM and MA beneficiaries demonstrated broadly similar patterns of contributing factors to LOS variation.
PMID:42753358 | DOI:10.1016/j.jamda.2026.106466

