Administrative Data and the Philosopher's Stone: Turning Heart Failure Claims Data into Quantitative Assessment of Left Ventricular Ejection Fraction

Background Administrative data are widely used in observational assessment of patient-centered clinical outcomes. In studies of cardiovascular outcomes, claims data are limited by lack of quantitative information, such as left ventricular ejection fraction. We aimed to determine whether left ventricular ejection fraction can be assessed from heart failure claims. Methods This observational, retrospective study used administrative and echocardiographic databases to identify heart failure patients (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] code 428.XX) who received echocardiograms. The study cohort included patients with at least one inpatient or outpatient claim for systolic (428.2X) or diastolic (428.3X) heart failure from January 1, 2007, through July 10, 2013, who received an echocardiogram within 30 days of the claim. Receiver operating characteristic (ROC) curves were used to determine the optimal left ventricular ejection fraction cut-off threshold between ICD-9-CM heart failure codes 428.2 (systolic) and 428.3 (diastolic). Bootstrapping was used to determine a 95% confidence interval for the best cut-off. Results A total of 2714 echocardiograms with ascertainable left ventricular ejection fraction were performed within 30 days of a heart failure diagnosis. ICD-9-CM codes 428.2 and 428.3 accounted for 28.9% and 18.2%, respectively, of all heart failure codes. The resulting ROC curve had a best threshold cut-off for ejection fractions of 43.5% (confidence interval 39.5%-44.5%). The area under the curve was 0.812, with positive predictive value 0.72 and negative predictive value 0.81. Conclusions Subject to study limitations, we conclude that assessing left ventricular ejection fraction using heart failure claims is possible.