Abstract
BACKGROUND: Cardioversion of acute-onset atrial fibrillation (AF) via electrical or pharmacological means is a common procedure performed in many emergency departments. While these procedures appear to be very safe, the rarity of subsequent adverse outcomes such as stroke would require huge sample sizes to confirm that conclusion. Big data can supply such sample sizes. OBJECTIVE: We aimed to validate several potential codes for successful emergency department cardioversion of AF patients. METHODS: This study combined 3 observational datasets of emergency department AF visits seen at one of 26 hospitals in Ontario, Canada, between 2008 and 2012. We linked patients who were eligible for emergency department cardioversion to several province-wide health administrative datasets to search for the associated cardioversion billing and procedural codes. Using the observational data as the gold standard for successful cardioversion, we calculated the test characteristics of a billing code (Z437) and of procedural codes 1.HZ.09JAFS and 1.HZ.09JAJS. Both include pharmacological and electrical cardioversions, as well as unsuccessful attempts; the latter is <10% using electricity (in Canada, standard practice is to proceed to electrical cardioversion if pharmacological cardioversion is unsuccessful). RESULTS: Of 4557 unique patients in the three datasets, 2055 (45.1%) were eligible for cardioversion. Nine hundred thirty-three (45.4%) of these were successfully cardioverted to normal sinus rhythm. The billing code had slightly better test characteristics overall than the procedural codes. Positive predictive value (PPV) of a billing was 89.8% (95% CI, 87.0-92.2), negative predictive value (NPV) 70.5% (95% CI, 68.1-72.8), sensitivity 52.1% (95% CI, 48.8-55.3), and specificity 95.1% (95% CI, 93.7-96.3). CONCLUSIONS: AF patients who have been successfully cardioverted in an emergency department can be identified with high PPV and specificity using a billing code. Studies that require high sensitivity for cardioversion should consider other methods to identify cardioverted patients.