Background/Purpose:

The prevalence of rheumatoid arthritis (RA) in primary care electronic health records (EHRs) is much lower than patient self-reports from population surveys, which may be due to under-diagnosis. The objectives of this study were to identify possible RA cases using a diagnostic algorithm developed from the 2010 EULAR/ACR criteria, and to compare them with EHR RA diagnosis cases, in order to estimate potential under-diagnosis rates.

Methods:

Patients’ EHRs in the UK Clinical Practice Research Datalink (CPRD) contain diagnostic, clinical and test information. We extracted relevant data to identify patients with diagnosed RA and possible additional RA cases using an algorithm consisting of four parts, including joint involvement and symptom duration, acute phase reactants, and serology tests. Clinical findings, diagnostic labels and results of tests ordered by consultant rheumatologists are not available in CPRD. In the EULAR/ACR classification criteria patients can score up to five points for any joint involvement, up to three points for positive serology, one point for positive acute phase reactants and one point if the duration of symptoms is greater than six weeks. In line with the classification criteria we defined those with a score of six or more as “algorithm-diagnosed RA cases”. The incidence of RA was calculated from the number of CPRD patients receiving a new RA diagnosis in that year divided by the denominator. The prevalence was calculated from the cumulative number of patients receiving an RA diagnosis up to and including a given year (as far back as CPRD records allow, with patients who have died being removed).

Results:

A total of 88,299 patients had a primary care EHR diagnosis of RA, and an additional 12,928 were defined as algorithm-diagnosed RA cases. The prevalence of RA was 0.49% for EHR-diagnosed RA, rising to 0.58% if algorithm diagnosed cases are included (Table 1.). This has risen steadily over time, possibly because of more complete diagnostic coding. The incidence of EHR-diagnosed RA is approximately 30.4/100,000 over the period 2005-14 compared with 6.0/100,000 algorithm-diagnosed cases, an overall incidence rate of 36.4/100,000. There were 3,091 patients diagnosed by both a doctor and the algorithm. The mean age of EHR-diagnosis patients was 60.2 years compared with 57.7 years for the algorithm-diagnosis group, which may suggest diagnostic delay.

Table 1. Incidence and prevalence of RA in the CPRD. Figures are per 100,000 people.

Conclusion:

The algorithm developed from the 2010 ACR/EULAR criteria identified a significant number of possible RA cases, or patients who may be at high risk of developing RA, in addition to those with a coded diagnosis. Insufficiently accurate coding of joint involvement by general practitioners prevented more precise scoring and classification. Delayed or under-diagnosis has implications for prognosis in these patients.

« Back to 2015 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/estimating-under-diagnosis-of-rheumatoid-arthritis-in-primary-care-data-from-the-uk-clinical-practice-research-datalink/