Validity of claims-based definitions of left ventricular systolic dysfunction in Medicare patients.

PURPOSE: Ejection fraction (EF) is crucial information when studying the use and effectiveness of therapies in patients with heart failure (HF) and myocardial infarction (MI). We aimed to assess the validity of claims data-based definitions of systolic dysfunction (SD).
METHODS: We identified 1072 patients with EF recorded for an HF/MI hospitalization in Medicare linked with pharmacy data and national HF/MI registries in 1999-2006. Thirteen claims-based definitions for SD were developed using a single or combination of ICD-9 diagnosis codes and cardiovascular medications use. We calculated sensitivity, specificity, and positive predictive values (PPVs) using recorded EFs as the gold standard.
RESULTS: Using an EF cutoff of 45%, the definitions based on digoxin use and no atrial fibrillation or flutter had the highest PPVs (76% to 84%) and specificity (>97%) but low sensitivity (6%-14%). As we varied the EF cutoff between 50% and 25%, the specificity decreased by 3%, but the PPVs decreased by 52%. We observed potential differences in the PPVs by patients' characteristics. In a hypothetical study assessing implantable defibrillator effectiveness, using our definition to identify patients with SD would underestimate the effectiveness by 3% to 24%. In another hypothetical study comparing two classes of angiotensin system blockers where SD was considered confounding, our definition introduced ~43% misclassification bias.
CONCLUSIONS: Claims-based definitions for SD had excellent specificity and good PPV but low sensitivity. The definitions with good PPV could be used for cohort identification or confounding adjustment by restriction and would result in relatively small misclassification bias albeit limited generalizability.