Comparison of Medicare Claims and Physician Adjudication for Identifying Hospitalized Heart Failure: The Women’s Health Initiative

Background : Medicare claims data are increasingly being used to identify incident cases of heart failure (HF) in epidemiologic cohorts. Comparison of protocolized study ascertainment of HF with Medicare data is limited, especially in older women and for HF subtypes. We compared physician adjudicated HF, including HF with preserved (HFpEF) or reduced (HFrEF) ejection fraction, to Medicare data in older women in the Women's Health Initiative (WHI). Methods : In this study HF follow-up was among 10,991 women age-eligible for Medicare from WHI enrollment (1993-1998) through February 2022. Trained physician adjudicators reviewed hospital medical records after participants reported a HF hospitalization on annual questionnaires. Medicare coding data were then obtained for the same hospitalization. We compared sensitivity, specificity, positive (PPV) and negative (NPV) predictive values, kappa scores and c-statistics for predicted HF absolute risk using WHI adjudicated versus Medicare identified HF cases. Results : We identified 872 cases of hospitalized acute HF in WHI, of which when using HF primary diagnosis codes, Medicare included 446 HF cases, but did not include 426 cases. An additional 324 cases were identified only in Medicare. This classification resulted in sensitivity of 51%, specificity of 97%, PPV of 58%, NPV of 96%, and a moderate kappa of 0.51. WHI adjudicated and Medicare identified HF gave similar results for model-based HF absolute risk (c-statistic 0.68 and 0.69, respectively). Conclusions : Using WHI adjudication as the criterion, HF ascertainment from Medicare data in older women has poor PPV from undercounting HF but has high specificity and gives HF absolute risk comparable to study-based criteria. HF subtype identification using Medicare data is prone to misclassification, perhaps reflecting temporal variation in case definitions and diagnostic practices.