Updating The Accuracy of Administrative Claims for Identifying Left Ventricular Ejection Fraction Among Patients with Heart Failure

Introduction: Left ventricular ejection fraction (EF) is an important factor for treatment decisions for heart failure. The EF is unavailable in administrative claims. We sought to evaluate the predictive accuracy of claims diagnoses for classifying heart failure with reduced ejection fraction (HFrEF) versus heart failure with preserved ejection fraction (HFpEF) with International Classification of Disease-Tenth Revision codes. Methods: We identified HF diagnoses for VA patients between 2017-2019 and extracted the EF from clinical notes and imaging reports using a VA natural language processing algorithm. We classified sets of codes as HFrEF-related, HFpEF-related, or non-specific based on the closest EF within 180 days. We selected a random heart failure diagnosis for each patient and tested the predictive accuracy of various algorithms for identifying HFrEF using the last 1 year of heart failure diagnoses. We performed sensitivity analyses on the EF thresholds, the cohort, and the diagnoses used. Results: Between 2017-2019, we identified 358,172 patients and 1,671,084 diagnoses with an EF recording within 180 days. After dividing diagnoses into HFrEF-related, HFpEF-related, or non-specific, we found using the proportion of specific diagnoses classified as HFrEF-related had an AUC of 0.76 for predicting EF≤40% and 0.80 for predicting EF<50%. However, 23.3% of patients could not be classified due to only having non-specific codes. Predictive accuracy increased among patients with ≥4 HF diagnoses over the preceding year. Discussion: In a VA cohort, administrative claims with ICD-10 codes had moderate accuracy for identifying reduced ejection fraction. This level of specificity is likely inadequate for performance measures. Administrative claims need to better align terminology with relevant clinical definitions.