Performance of Claims-Based Frailty Measures in Medicare FFS and MA: Findings from NHATS

TL;DR

This study compares how well two claims-based methods detect frailty in older adults enrolled in Medicare Advantage and fee-for-service plans.

Contribution

The study evaluates the performance of claims-based frailty measures in Medicare Advantage and FFS populations using real-world data.

Findings

CFI had high specificity but low sensitivity for detecting frailty in both Medicare FFS and MA populations.

HFRS showed high sensitivity but low specificity for frailty detection.

Both models performed better when benchmarked against CGA-FI compared to FP.

Abstract

Frailty is a state of vulnerability in older adults which predicts adverse health outcomes. Although clinical measures, such as a Comprehensive Geriatric Assessment–Frailty Index (CGA-FI) and the Fried phenotype (FP), are reference standards, claims-based measures are often used for research. Given the rapidly growing Medicare Advantage (MA) population and concerns regarding MA data completeness, we assessed performance of claims-based frailty models in detecting frailty by CGA-FI and FP among community-dwelling MA and fee-for-service (FFS) beneficiaries. We measured the Kim’s Claims-based Frailty Index (CFI) and Gilbert’s Hospital Frailty Risk Score (HFRS) in the National Health and Aging Trends Study linked to Medicare claims. In 2019 (FFS n = 2,153, MA n = 1,672), CFI had high specificity (FFS 0.97, MA 0.99) but low sensitivity (FFS 0.31, MA 0.24) in detecting frailty by CGA-FI. The CFI’s PPV was higher for CGA-FI than FP (FFS: 0.84 vs 0.44, MA: 0.90 vs 0.47). In comparison, HFRS had high sensitivity (FFS 0.91, MA 0.91) but low specificity (FFS 0.38, MA 0.43) for CGA-FI frailty, with similar results for FP. Model discrimination was good (C-statistics: CFI 0.81, HFRS 0.79), with both models performing better when benchmarked against CGA-FI. 2021 analyses were similar. In sensitivity analysis using only inpatient claims, HFRS sensitivity decreased while specificity increased. These results suggest that claims-based frailty measures may perform comparably in MA and FFS beneficiaries. However, performance characteristics between models vary – highlighting the importance of model selection for a given population or purpose.