Accuracy of Administrative Coding to Identify Reduced and Preserved Left Ventricular Ejection Fraction

Background: Coding of systolic function in heart failure is important, but the accuracy is uncertain. Methods and Results: We used data from a chart review of VA heart failure hospitalizations from 2006 to 2013. Trained abstractors determined the documented diagnosis of heart failure and the left ventricular ejection fraction (LVEF). We compared this LVEF with the primary and secondary International Classification of Disease, 9th edition, codes for heart failure for the same hospitalization. Among 43,044 hospitalizations for heart failure, the primary discharge diagnosis was coded as systolic heart failure in 18%, diastolic heart failure in 17%, and other heart failure codes in 65%. For an LVEF <40%, a systolic heart failure code had a sensitivity of 29% and a positive predictive value of 76%. The code for systolic heart failure was used more frequently over time, with sensitivity increasing from 16% to 37% but at the expense of the positive predictive value, which decreased from 80% to 74%. The overall area under the receiver operating characteristic curve for the relationship between LVEF and the systolic heart failure code was 0.71. Using LVEF >50% to define diastolic heart failure led to a sensitivity of 29% for a diastolic heart failure code, with a positive predictive value of 78%. In multivariate analysis, a systolic heart failure code had an odds ratio for 1-year mortality of 1.1 (95% confidence interval 1.03-1.17) compared to not having a systolic heart failure code. Conclusions: Coding for systolic and diastolic heart failure is associated with LVEF, but the accuracy is too poor to substitute for the documented LVEF in performance measurement.