We thank Dr. Pratt and colleagues for their interest in our data. Confirmation that the prediction rule performed so well in their cohort of 53 UA patients is a helpful observation. (The authors report a PPV of 100% for scores of ≥8, an NPV of 100% for scores ≤6, and an area under the curve of 0.88.) Pratt et al raise 2 very important issues on which we would like to comment.
The first issue concerns the definitions of UA and RA. This is crucial, and in situations in which different disease states of patients are studied, it can be expected that different results will be obtained. For the derivation cohort (the Leiden Early Arthritis Clinic [EAC]) and the replication cohorts (recruited from the UK, Germany, and The Netherlands), development of RA was defined as disease fulfilling the 1987 American College of Rheumatology (formerly, the American Rheumatism Association) criteria (1). Although considerable debate surrounds whether these criteria are the best method of diagnosing RA in individual patients, we chose this outcome measure because of its near-universal use in trials evaluating the efficacy of treatment strategies and because of the lack of other incontrovertible definitions. Patients who fulfilled the 1958 “probable RA criteria” (2) were not categorized as having RA in either the derivation cohort or the replication cohorts.
The definition of UA is perhaps even more controversial, and there are no widely accepted classification criteria for this condition. We included only patients in whom clinically apparent synovial swelling was observed by a rheumatologist. This is in contrast to the apparent practice of Pratt et al, wherein patients with inflammatory joint symptoms in the absence of clinically observable synovial swelling (a state referred to by others as “inflammatory arthralgia”) could be defined as having UA. The issue of inflammatory arthralgia is an important one, and histologic (3) and ultrasonographic (4) data strongly support the contention that at least a portion of these patients will have subclinical synovitis. The use of sensitive methods to detect subclinical synovitis in these patients and the assessment of other biomarkers may help in the prediction of their outcome. Data from the Newcastle cohort and others, which include such patients, will clearly provide very important data to help future management in what currently is a difficult area.
We also want to emphasize that, in our opinion, the terms “early arthritis” and “undifferentiated arthritis” are neither similar nor interchangeable. For the derivation of the prediction rule, we used only those patients with early arthritis who, at their second visit (generally 2 weeks after enrollment in the study), did not fulfill classification criteria for any rheumatologic disorder. Thus, patients with early arthritis who fulfilled criteria for specific diagnoses at baseline (e.g., reactive arthritis or RA), and in whom outcomes could be predicted reasonably well without the use of additional predictive tools, were not included. In contrast, the disease course of UA is highly variable (∼30% of patients develop RA, and the symptoms remit spontaneously in ∼40%) (5); therefore, our prediction rule was derived specifically using patients with UA, and not all patients with early arthritis.
The second issue raised by Pratt et al concerns the additive value of the whole prediction rule compared with the information provided by the presence of circulating ACPAs alone. In their cohort, they observed that all 8 patients with a score of ≥8 were ACPA positive, and they suggest that the prediction rule is preferentially useful in ACPA-negative patients. Our data are not entirely consistent with this. Although van Gaalen et al observed that, in a cohort of 318 UA patients, 93% (95% confidence interval 87–99) of ACPA-positive patients developed RA over 3 years (6), the frequency is lower in our larger cohort of UA patients. In the 570 UA patients from the Leiden EAC, 69% of those positive for ACPA developed RA after 1 year of followup; this increased to 74% during the total available followup (mean ± SD 8 ± 3 years). This indicates that the PPV of ACPA alone (69–74%) is further improved by having information on other predictive variables; the PPV of a score of ≥8.0 was 84% in the original Leiden EAC cohort was 84% and was 97% in the 3 validation cohorts combined. This notion is strengthened by the observation that, of the 93 UA patients with a score ≥8 in the Leiden EAC, 29 were ACPA negative, and of those 29 patients, the disease in 24 (83%) did progress to RA.
In conclusion, although we agree that the prediction rule is useful in ACPA-negative patients, in ACPA-positive UA patients, application of the whole prediction rule improves the predictive ability compared with the use of ACPA alone.