ICD 10 Algorithm Accurately Identifies Rheumatoid Arthritis Patients

September 10, 2020

This is the first time such an algorithm was validated in a large US claims database.

Findings of a new study demonstrate using ICD 10-based codified algorithm to identify rheumatoid arthritis patients in a claims database has high accuracy.

The results, presented at the 14th Annual North American Young Rheumatology Investigator Forum, indicated the possibility of establishing a large and accurate rheumatoid arthritis patient registry for future discovery research.

Typically, a large drawback of using claims databases is the high rate of inaccurate coding or diagnostic errors. In an effort to calculate the accuracy of an ICD 10-based algorithm to identify rheumatoid arthritis in a large administrative claims database, Nicholas Wiemer, DO, and colleagues used a cohort of more than 1.7 million Highmark Health members who were seen between 2015 and 2018. The team used a diagnostic algorithm that included ICD 10 codes M05.xx and M06.xx and those who had at least 2 rheumatology office visits at least 6 weeks apart.

The algorithm diagnosed more than 8000 patients as having established rheumatoid arthritis and 519 were seen at the team’s academic tertiary rheumatology practice. The investigators randomly selected 50 patients and manually reviewed the electronic health record (EHR) for provider’s documentation of tender and swollen joint counts and duration of symptoms and lab results to determine if the patients had definite rheumatoid arthritis compared to the gold standard 2010 ACR/EULAR rheumatoid arthritis classification criteria. The team also captured patient demographics and conventional and biologic disease-modifying anti-rheumatic drug (DMARD) use.

Among the 50 patients analyzed, 90% had definite rheumatoid arthritis according to the ACR/EULAR classification criteria. The median age of the cohort was 65 years old (IQR, 57-75 years old), with a majority female (76%), 78% seropositive, 40% had abnormal inflammatory markers and 98% were on a DMARD therapy at some point, either current, started at the initial consultation, or prior to the first note in the EHR.

Of the 5 patients who did not fulfill the classification criteria, 60% were seronegative, 40% had low positive rheumatoid factor/anti-cyclic citrullinated peptide antibody levels, 60% had fewer than 10 swollen or tender joints, and none had an alternative rheumatic diagnosis. The PPV of the algorithm was 90%.

Overall, the ICD 10 codified algorithm for identifying patients with rheumatoid arthritis from a claims database had high accuracy. The investigators noted it was the first time such an algorithm was validated in a large US claims database. The findings make way for the possibility to establish a large and accurate rheumatoid arthritis patient registry for future research.

The study, “Rheumatoid Arthritis: Validation of a Codified Diagnostic Algorithm in a Large Claims Database,” was presented at the North American Young Rheumatology Investigator Forum.