Application of Claims-Based Frailty Index to a Structured Electronic Health Record Data.

Abstract

BACKGROUND The Claims-based Frailty Index (CFI) has been developed and validated using Medicare claims data. However, whether CFI can be applied to structured EHR data has not been studied. METHODS We applied the CFI to a structured EHR dataset (Explorys dataset) and a Medicare fee-for-service (FFS) 5% sample data and compared the prevalence of frailty from each dataset, using the cohort of older adults. Then, we assessed the odds ratio (OR) and area under the curve (AUC) of the frailty predicting adverse clinical outcomes, any hospital or emergency room (ER) visit, or any adverse drug events (ADEs) related encounter within 1 year in each dataset. RESULTS A total of 526,681 from the Explorys dataset (64.6% with Medicare insurance (Explorys-Medicare), and 35.4% with non-Medicare insurance (Explorys-non-Medicare)) and 346,070 individuals from the Medicare dataset were included. The prevalence of frailty, defined as CFI ≥0.25, among heart failure patients was 7.4% in the Explorys-Medicare dataset, 7.1% in Explorys-non-Medicare, and 14.2% in the Medicare dataset. The ORs of frailty for any hospital or ER visit were 3.57, 4.37, and 3.76 in Explorys-Medicare, Explorys-non-Medicare, and Medicare datasets, and for any ADE-related encounter, they were 2.61, 3.29, and 2.89, respectively. The AUC of the frailty index were 0.656, 0.676, and 0.697 for any hospital or ER visit and 0.654, 0.676, and 0.654 for any ADE-related encounter, respectively. CONCLUSIONS When the CFI was applied to a structured EHR dataset, it captured fewer frailty cases than the Medicare dataset but had similar performance in predicting adverse clinical outcomes.