Accuracy and validation of an automated electronic algorithm to identify patients with atrial fibrillation at risk for stroke.
Am Heart J. 2015 Jan;169(1):39-44.e2
Authors: Navar-Boggan AM, Rymer JA, Piccini JP, Shatila W, Ring L, Stafford JA, Al-Khatib SM, Peterson ED
BACKGROUND: There is no universally accepted algorithm for identifying atrial fibrillation (AF) patients and stroke risk using electronic data for use in performance measures.
METHODS: Patients with AF seen in clinic were identified based on International Classification of Diseases, Ninth Revision(ICD-9) codes. CHADS2 and CHA2DSs-Vasc scores were derived from a broad, 10-year algorithm using IICD-9 codes dating back 10 years and a restrictive, 1-year algorithm that required a diagnosis within the past year. Accuracy of claims-based AF diagnoses and of each stroke risk classification algorithm were evaluated using chart reviews for 300 patients. These algorithms were applied to assess system-wide anticoagulation rates.
RESULTS: Between 6/1/2011, and 5/31/2012, we identified 6,397 patients with AF. Chart reviews confirmed AF or atrial flutter in 95.7%. A 1-year algorithm using CHA2DS2-Vasc score ≥2 to identify patients at risk for stroke maximized positive predictive value (97.5% [negative predictive value 65.1%]). The PPV of the 10-year algorithm using CHADS2 was 88.0%; 12% those identified as high-risk had CHADS2 scores <2. Anticoagulation rates were identical using 1-year and 10-year algorithms for patients with CHADS2 scores ≥2 (58.5% on anticoagulation) and CHA2DS2-Vasc scores ≥2 (56.0% on anticoagulation).
CONCLUSIONS: Automated methods can be used to identify patients with prevalent AF indicated for anticoagulation but may have misclassification up to 12%, which limits the utility of relying on administrative data alone for quality assessment. Misclassification is minimized by requiring comorbidity diagnoses within the prior year and using a CHA2DS2-Vasc based algorithm. Despite differences in accuracy between algorithms, system-wide anticoagulation rates assessed were similar regardless of algorithm used.
PMID: 25497246 [PubMed - in process]