Validation of diagnosis algorithms for ankylosing spondylitis in claim-based database

Background: Claims-based algorithms using International Classification of Diseases (ICD) codes have become a common approach for researchers to define ankylosing spondylitis (AS) in studies. To address potential misclassification bias caused by the claim-based algorithms, we conducted the current study to validate whether these algorithms of medical claims could accurately represent AS diagnoses. Methods: Patients diagnosed with AS based on ICD codes were retrieved from the electronic medical records database at a Taiwanese medical center (Chung Shan University Hospital, Taiwan). After random sampling and stratification based on age and sex, the medical information of participants was appraised based on the 2009 ASAS guideline to evaluate the actual status of ICD codes claim-based AS patients. Positive predict values (PPV) of different algorithms of ICD codes were also calculated. Results: Within the 4160 patients with claim-based AS diagnosis, 387 eligible patients were finally included in the study design after random sampling. The PPV of the diagnostic algorithm of having at least 4 outpatient or 1 inpatient ICD record was 72.77 (95% CI, 66.79–78.75), whereas the PPV increased to 85.64 when the diagnoses were restricted to be made by rheumatologists (95% CI, 80.53–90.74). Conclusions: While performing database studies, researchers should be aware of the low PPV of specific algorithms when defining AS. Algorithms with higher PPV were recommended to be adopted to avoid misclassification biases.