Patients and methods

Participants

We selected subjects enrolled in the Toronto Early Arthritis Cohort (TEACH) for our study. TEACH has recruited people since 2003 as part of an inception cohort of patients with UA. Inclusion criteria are age >16, symptom duration between 6 weeks and 12 months and two or more swollen joints or one swollen metacarpophalangeal or proximal interphalangeal joint with one or more of the following: positive rheumatoid factor (RF), positive anti-cyclic citrullinated peptide (anti-CCP), morning stiffness for >45 min, response to non-steroidal anti-inflammatory drugs or painful metatarsophalangeal squeeze test.

Patients are evaluated at baseline and every 3 months according to a standard protocol that includes baseline radiographs, 44-joint count for tender and swollen joints, laboratory indices, disability measure by the Health Assessment Questionnaire and assessment of disease activity by the Disease Activity Score 28 (DAS28) and Rheumatoid Arthritis Disease Activity Index (RADAI).

If an inflammatory arthritis is suspected, treatment of patients is started with combination DMARD treatment with methotrexate, sulfasalazine and hydroxychloroquine, with the option of intramuscular or intra-articular steroid bridging. Treatment is adjusted at every visit with the aim of remission, defined as no swollen joints.

Data collection

The prediction rule consists of clinical variables, which are scored (range 0–13) and correspond to the percentage chance of developing RA (appendix 1, online only).14 We applied the rule to baseline characteristics of all patients with UA who had completed a minimum of 6 months’ follow-up to allow sufficient time for diagnosis. After 6 months, we examined all patients to determine if RA or another American College of Rheumatology (ACR)-defined condition had developed.

We imputed values for patients with missing baseline variables needed to score the prediction rule. Median values for C-reactive protein were substituted (n = 6), and indeterminate anti-CCP status (n = 10) was scored as zero.

Statistical analysis

Patients with UA who developed RA were compared with those who did not using the χ² test for nominal variables and the Student t test for continuous variables. A p value of <0.01 was considered significant for predictor variables.

Results

Table 1 describes the characteristics of the 105 eligible participants with UA. The mean age was 45 (15) and most were female (77%). All patients had been followed up for a minimum of 6 months or more (18.5 (9.3) months, mean (SD) follow-up). At follow-up, 80 (76%) were diagnosed with RA while the remainder developed another rheumatic diagnosis. Of the non-progressors, one (4%) patient was diagnosed with psoriatic arthritis due to subsequent development of skin psoriasis, two (8%) developed symptoms consistent with fibromyalgia, three (12%) were diagnosed with inflammatory osteoarthritis, three (12%) with spondyloarthropathy and three (12%) with systemic lupus erythematosus. The largest group (n = 13, 52%) of non-progressors remained undifferentiated at follow-up.

At baseline, 78 (98%) patients with RA had features prompting combination DMARD treatment compared with 10 (40%) non-RA patients. By 6 months, all patients with confirmed RA were receiving DMARDs. Of those patients who remained undifferentiated, nearly half (n = 6, 47%) started treatment during the 6-month period.

Patients progressing to RA were more likely to be current smokers, have a positive family history of RA and present with symmetric joint involvement. Patients with RA also tended to have greater morning stiffness and high inflammatory markers, although not statistically significant. Factors significantly associated with developing RA were the number of tender and swollen joints, RF positivity, anti-CCP positivity, poor function and high disease activity (table 1).

Table 2 shows the number of patients who developed RA in relation to the calculated prediction score. No patients with UA who scored <4 progressed to RA, while all who scored >9 did. For those who scored between 4 and 9, higher scores frequently predicted progression. Among progressors, the median prediction score was 8.0 (interquartile range (IQR) 6.9–9.1), while non-progressors’ median score was 5.0 (IQR 3.8–6.7). Sixty-nine (66%) patients with UA scored between 5 and 8, and the majority (n = 56, 81%) had confirmed RA by 6 months. Using these cut-off values, 18 (72%) patients with scores ⩽5 did not develop RA, while 35 (97%) with scores ⩾8 did (fig 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Cut-off values for prediction scores and risk of progression to rheumatoid arthritis (RA).

Discussion

The Leiden prediction rule is a fast and easy tool to help identify patients with UA who may go on to meet criteria for RA. Baseline scores ⩾8 predicted those who developed RA by 6 months while scores ⩽5 corresponded with lower rates. If therapeutic decisions were based solely on these cut-off points, treatment would have been withheld in seven patients with RA and potentially inappropriately prescribed to one non-RA patient during the first 6 months of follow-up.

These findings resemble the original model, where scores of ⩽6 and ⩾8 most accurately predicted outcome. Similar to the Leiden derivation cohort, the number of tender and swollen joints, RF positivity, anti-CCP positivity and poor functional status predicted the development of RA.13 Erosive disease was also high among progressors but not independently predictive and was thus left out of the original model. These findings contrast with other predictive models demonstrating the importance of radiographic erosions in discriminating between forms of self-limiting and persistent disease.15,16

Our study results differed from the derivation cohort in four major ways. First, a large proportion of our patients with UA (76%) developed RA within 6 months, compared with 31% of the Leiden cohort. This large difference may reflect our clinic’s inclusion criteria favouring types of inflammatory arthritis, such as RA, based on presenting symptoms and signs. In contrast, patients with any physical examination evidence of arthritis are enrolled in the Leiden clinic and therefore may encompass more benign forms of arthritis or even self-limiting disease. Despite this difference, we feel that the distribution of high scores among those later diagnosed with RA in our sample only increases the discriminative ability and positive predictive value of the rule. Second, disease outcome in our study was assessed at 6 months and not at 12 months. While this short duration of follow-up may be interpreted as a limitation, it did not appear to affect ascertainment of disease in our cohort as the majority (88%) achieved an ACR diagnosis by this time. A third important difference was the high proportion of patients treated with DMARDs during our study period, whereas data on the proportion of DMARD-treated patients were not included for the Leiden cohort. Treatment was started in nearly 84% of patients at baseline and therefore it is plausible we may have hampered progression of RA in the subset that remained undifferentiated at follow-up. However, such “misclassified” patients who would otherwise be prone to develop RA would still be expected to have high baseline scores, which would only strengthen the operating characteristics of the rule. Lastly, disease activity measured by the DAS28 and RADAI were highly significant in our sample but were not evaluated in the original model.

Based on our findings, the Leiden prediction rule can be applied to other cohorts of patients with early UA. The question is, how advantageous is a model to predict who will meet the 1987 ACR classification criteria for RA? These criteria, the most widely accepted “gold standard”, perform poorly in identifying patients with early RA and those at risk of severe disease.17,18 Furthermore, they do not incorporate new biomarkers such anti-CCP antibodies, which were shown to be predictive of RA in the Leiden sample (thus receiving the strongest weight in the prediction rule) and were equally prognostic in our study.

Therefore, models that better predict which patients with UA will develop persistent, destructive disease or who will be at risk of loss of function or work productivity may be more relevant, both clinically and economically. More importantly, no predictive model has yet dealt with the fundamental problem of identifying which patients will benefit from specific treatment strategies. The PROMPT study nicely demonstrated that methotrexate may postpone the diagnosis of RA in those with UA.19 However, no algorithm exists for the optimal timing, dose or duration of intervention needed to achieve a sustained favourable outcome for patients with UA. Given our results showing that patients with scores between 5 and 8 have a high probability of progressing to RA, research focusing on how to best manage this group to minimise risks of undertreatment and maximise benefits of early intervention is warranted.