Validation of obesity coding among newly treated nonvalvular atrial fibrillation patients using an integrated electronic medical record and claims database

Objective: To evaluate the validity of diagnosis codes for identifying obesity and morbid obesity among newly treated nonvalvular atrial fibrillation (NVAF) patients. Methods: An integrated electronic medical record (EMR) and claims database (1 January 2013-31 March 2018) was used. Adult patients with >= 1 claim for an oral anticoagulant (OAC) from 1 January 2014-30 September 2017 were identified (index date). Patients were required to have >= 1 atrial fibrillation diagnosis, no OAC use or valvular disease during the 12 months before index date, >= 12 months of continuous enrollment before and >= 6 months after index date, and >= 1 BMI measurement 6 months before or after index date. Patients with BMI >= 30 kg/m(2) and BMI >= 40 kg/m(2) were classified as obese and morbidly obese, respectively. Sensitivity, specificity and positive predictive value (PPV) were calculated to assess the validity of diagnosis codes for obesity and morbid obesity. Results: A total of 7501 patients met all selection criteria. Forty-six percent of patients had BMI >= 30 kg/m(2), of whom about one-quarter had a BMI >= 40 kg/m(2). Twenty-five percent and 10% of patients had a diagnosis code for obesity or morbid obesity, respectively. Sensitivity, specificity and PPV for obesity diagnosis codes were 48.67% (95% CI: 47.00%-50.35%), 95.24% (94.54%-95.88%) and 89.78% (88.32%-91.12%), respectively, and 62.75% (59.30%-66.11%), 96.46% (95.99%-96.89%) and 67.93% (64.43%-71.29%) for morbid obesity diagnosis codes, respectively. Conclusion: Among newly treated NVAF patients, obesity diagnosis codes had high PPV, high specificity and modest sensitivity. Morbid obesity diagnosis codes also had high specificity, but modest PPV and sensitivity. These findings have implications for case selection and control for obesity as a confounder in studies using a claims database.