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

Area of Science:

• Rheumatology
• Health Informatics
• Epidemiology

Background:

• Claims-based algorithms using International Classification of Diseases (ICD) codes are frequently used to identify ankylosing spondylitis (AS) in research.
• Potential misclassification bias exists with these claim-based algorithms, necessitating validation of their accuracy in representing AS diagnoses.

Purpose of the Study:

• To validate the accuracy of existing claims-based algorithms in diagnosing ankylosing spondylitis (AS).
• To assess the positive predictive values (PPV) of different International Classification of Diseases (ICD) code algorithms for AS identification.

Main Methods:

• Retrieved patients with ICD-coded AS diagnoses from a Taiwanese medical center's electronic health records.
• Randomly sampled and stratified patients by age and sex.
• Evaluated medical information against the 2009 ASAS guideline and calculated PPVs for various ICD code algorithms.

Main Results:

• Included 387 patients from an initial cohort of 4160 with claim-based AS diagnoses.
• The PPV for an algorithm requiring at least 4 outpatient or 1 inpatient ICD record was 72.77%.
• Restricting diagnoses to those made by rheumatologists increased the PPV to 85.64%.

Conclusions:

• Researchers must acknowledge the variable positive predictive values (PPV) of different algorithms when defining ankylosing spondylitis (AS) in database studies.
• Algorithms with higher PPVs are recommended to mitigate misclassification biases and improve study accuracy.
• Validation of diagnostic algorithms is crucial for reliable epidemiological research in rheumatology.