Discordance Between Two Claims-Based Frailty Indices in a Hospitalized Medicare Population
Author(s)
Michael Vincent Murphy, BS1, Jessica Duchen, MPH2, Pamela Blumberg, MPH, DrPH1.
1Magnolia Market Access, Bridgewater Township, NJ, USA, 2Magnolia Market Access, Hamden, CT, USA.
1Magnolia Market Access, Bridgewater Township, NJ, USA, 2Magnolia Market Access, Hamden, CT, USA.
Presentation Documents
OBJECTIVES: Frailty is an indicator of health outcomes in older adults but is difficult to identify in claims data. Previous work identified discordance between two claims-based indices (rho=0.227). This study further investigates this finding in the same Medicare population.
METHODS: The Risk Analysis Index (RAI) adapted for ICD-10-CM codes (range 0-81) and the Kim Index (range 0-1), were calculated at admission for all hospitalizations in the Medicare 5% SAFs (2017-2022). Higher scores reflect greater frailty. Patients were classified into four and then two categories, “normal” or “frail”, as defined by each index for every hospitalization in the timeframe. Agreement between indices, Charlson Comorbidity index (CCI), length of stay (LOS), discharge status, mortality, and 30-day readmissions are presented.
RESULTS: Among 2,540,925 hospitalizations, agreement between score categories was poor (kappa-0.196). 41.1% were normal and 19.8% were frail for both indices (60.9% agreement). Among those with discordant Kim/RAI classifications, 26.2% were frail/normal, 12.9% were normal/frail. Disagreement by ≥2 of the 4 frailty levels was 50.9%, with 19.2% Robust in RAI but Mildly Frail per Kim. The greatest differences observed between concordant frail and concordant normal groups were CCI (5.3 vs. 2.1), LOS (7.9 vs. 6.2 days), in-hospital mortality (5.9% vs. 2.0%), discharges to hospice (8.2% vs. 0.8%) and home (40% vs. 79%), and 30-day mortality (15.6% vs. 2.2%) and readmissions (28% vs. 21%).
Differences among those with discordant Kim/RAI frailty, frail/normal versus normal/frail, were CCI (4.5 vs.3.7), LOS (7.5 vs. 7.2 days), in-hospital mortality (3.6% vs. 4.7%), discharged to hospice (2.1% vs. 6.3%) and home (60.1% vs 56.7%), 30-day mortality (12.3% vs. 5.5%) and readmissions (32.7% vs. 25.6%).
CONCLUSIONS: Discordance between Kim and RAI indices was significant and outcomes between discordant pairs were inconsistent in direction. Differences may be driven by RAI’s inclusion of age or Kim’s inclusion of procedure codes (CPT; HCPCS) and 1-year look-back.
METHODS: The Risk Analysis Index (RAI) adapted for ICD-10-CM codes (range 0-81) and the Kim Index (range 0-1), were calculated at admission for all hospitalizations in the Medicare 5% SAFs (2017-2022). Higher scores reflect greater frailty. Patients were classified into four and then two categories, “normal” or “frail”, as defined by each index for every hospitalization in the timeframe. Agreement between indices, Charlson Comorbidity index (CCI), length of stay (LOS), discharge status, mortality, and 30-day readmissions are presented.
RESULTS: Among 2,540,925 hospitalizations, agreement between score categories was poor (kappa-0.196). 41.1% were normal and 19.8% were frail for both indices (60.9% agreement). Among those with discordant Kim/RAI classifications, 26.2% were frail/normal, 12.9% were normal/frail. Disagreement by ≥2 of the 4 frailty levels was 50.9%, with 19.2% Robust in RAI but Mildly Frail per Kim. The greatest differences observed between concordant frail and concordant normal groups were CCI (5.3 vs. 2.1), LOS (7.9 vs. 6.2 days), in-hospital mortality (5.9% vs. 2.0%), discharges to hospice (8.2% vs. 0.8%) and home (40% vs. 79%), and 30-day mortality (15.6% vs. 2.2%) and readmissions (28% vs. 21%).
Differences among those with discordant Kim/RAI frailty, frail/normal versus normal/frail, were CCI (4.5 vs.3.7), LOS (7.5 vs. 7.2 days), in-hospital mortality (3.6% vs. 4.7%), discharged to hospice (2.1% vs. 6.3%) and home (60.1% vs 56.7%), 30-day mortality (12.3% vs. 5.5%) and readmissions (32.7% vs. 25.6%).
CONCLUSIONS: Discordance between Kim and RAI indices was significant and outcomes between discordant pairs were inconsistent in direction. Differences may be driven by RAI’s inclusion of age or Kim’s inclusion of procedure codes (CPT; HCPCS) and 1-year look-back.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR20
Topic
Methodological & Statistical Research
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Geriatrics