J Whittle, VA Pittsburgh Health Care System; J Chang, VA Pittsburgh Health Care System; JK Mickelson, Cincinatti VAMC

Objectives: Because of its prevalence and cost, congestive heart failure (CHF) is a focus of many quality improvement efforts. Although many efforts focus on patients with severe CHF admitted to the hospital, the majority of patients have less severe disease. Although ICD9 codes can be used to identify patients, many of these patients likely have other conditions that were confused with CHF, or CHF may have eventually been ruled out. At least 4 investigative teams have published definitions of CHF based on constellations of history, physical exam, chest x-ray (CXR) and/or electrocardiogram (EKG) findings, but the accuracy of these definitions has not been validated. We therefore studied, agreement among these clinical definitions of CHF, and their sensitivity and specificity for identifying systolic dysfunction, defined by an ejection fraction < 36%.

Methods: We identified all patients admitted to one VAMC with an inpatient or outpatient ICD9 code consistent with CHF. Two trained abstractors independently reviewed the same 75 inpatients and 75 outpatients, drawn randomly from this sample, to identify elements used to define CHF, including finding on history, physical exam, CXR and EKG. Discrepancies between the reviewers were resolved by consensus. Using each of 4 definitions, we classified each of 150 patients as either having or not having CHF. Pair-wise agreement among them was measured with Kappa and simple agreement. We classified the 116 patients (52 outpatients) for whom EF was available as having (EF < 36, n=50) or not having (n=66) systolic dysfunction, then calculated sensitivity, specificity and positive and negative predictive value (PPV and NPV) for detection of systolic dysfunction.

Results: The percent of patients defined as CHF was 57, 59, 63 and 64% for definitions 1 to 4, respectively. Agreement was highest between definitions 1 and 4 (simple agreement 87%, Kappa 0.73) and worst between 1 and 2 (simple agreement 74%, Kappa 0.47). The sensitivity for detecting systolic dysfunction was 0.70, 0.84, 0.72 and 0.70 for definitions 1-4. Thus, false negative rates range from 16 to 30%. Specificity was 0.33, 0.35, 0.38 and 0.27 for definitions 1-4. PPV was 0.44, 0.49, 0.47 and 0.42 and NPV was 0.59, 0.74, 0.64, and 0.55 for definitions 1-4. Restricting the study to inpatients raised the sensitivity (0.79 to 0.93), but lowered specificity (0.11 to 0.26). When applied to outpatients, conversely, the definitions had lower sensitivity (0.48 to 0.71) but higher specificity (0.45 to 0.58).

Conclusions: We have shown that a substantial portion of patients with CHF ICD9 codes do not have CHF by standard definitions. There is reasonable agreement among the definitions. However, none of these definitions reliably identifies patients with CHF associated with systolic dysfunction. Thus, if one relied on any of the standard definitions, a substantial number of persons with impaired EF may be mistakenly omitted from a CHF improvement project.

Impact: There does not seem to be reason to prefer one of the standard definitions of CHF. Because of significant false negative rates, attempts to validate coded data with standardized chart review will miss persons who might benefit from quality improvement efforts directed at systolic dysfunction.