Outcome of patients with early arthritis without rheumatoid factor and ACPA and predictors of rheumatoid arthritis in the ESPOIR cohort

Diagnosing rheumatoid arthritis (RA) at an early stage remains a challenge for clinicians. Indeed, no test is sufficiently specific to identify RA with certainty. The diagnosis is based on a spectrum of clinical, biological, and radiographic features, although the latter, despite a high specificity when typical erosions are observed, are rarely present at an early stage of the disease. Thus, testing for auto-antibodies such as rheumatoid factor (RF) and anti-citrullinated protein antibodies (ACPA) is useful to diagnose RA among patients with early arthritis [1].

ACPA such as anti-cyclic-citrullinated peptide (anti-CCP) antibodies are highly specific to RA [2, 3], have good predictive validity for RA in patients with early arthritis [4], and are associated with radiographic progression in early RA [5, 6]. Although RF is considered less specific [3], it is also associated with worst radiographic outcome [7, 8]. ACPA and RF have the same weight in the 2010 American College of Rheumatology/European League Against Rheumatism (ACR/EULAR) classification criteria for RA [9] and have a major impact in the diagnosis and prognosis of RA. In clinical practice, the decision to start a disease-modifying anti-rheumatic drug (DMARD) is more difficult in the absence of RF and ACPA [10] since these auto-antibodies are important features both for diagnosis and for risk of persistency and erosiveness. Nevertheless, 20 to 30% of RA patients do not have ACPA [2] or RF, and erosive RA may exist without these two auto-antibodies.

Distinct genetic risk factors are associated with ACPA-positive or ACPA-negative disease. Anti-CCP–positive RA was found associated with HLA-DRB1, HLA-DP, PTPN22, C5-TRAF1, and TNFAIP3-OLIG3 polymorphisms [11‐13], whereas anti-CCP–negative RA was found associated with genes such as HLA-DR3 and IRF-5 [13, 14], two genes that are also associated with systemic lupus erythematosus and Sjögren syndrome. These data might indicate distinct pathogenic mechanisms underlying ACPA-positive and ACPA-negative RA, with the last entity not well defined. Whether ACPA-negative RA features auto-antibodies binding to other citrullinated proteins (vimentin etc.) not detected in routine care or whether it is true RA without auto-antibodies is unknown. Because the clinical presentation of early arthritis is not specific, we do not know whether RA without ACPA or RF is “true” RA or another undifferentiated inflammatory arthritis. The literature contains few data specifically regarding the follow-up of “seronegative” early arthritis (i.e., negative for RF and ACPA).

The objectives of this study were first to describe the disease course of patients without RF and ACPA in an inception cohort of early inflammatory arthritis patients and second to determine baseline predictors of fulfilling 2010 ACR/EULAR criteria for RA within 3 years in these patients.

When patients present inflammatory arthritis, physicians must identify disease that will progress to RA. Because auto-antibodies such as RF or ACPA are key in the diagnosis, their weight is important (up to 3 of 10 points) in the 2010 ACR/EULAR classification criteria for RA. Thus, obtaining 6 of 10 points required for such a classification of RA in the absence of these 2 auto-antibodies may be difficult in the early phases of the disease, since it requires the involvement of more than 10 involved joints. Although this situation is frequently encountered, especially in early RA, the initial clinical presentation and disease course of seronegative RA is not well known. We first compared the initial features of patients with early arthritis according to positivity for RF and/or ACPA. We used a large, prospective, early-arthritis cohort from the community. This situation reflects clinical practice and allowed us to study the clinical value of RF and ACPA in patients selected by symptoms, not diagnosis. The patients in our 2 seronegative and seropositive groups were well balanced, thus confirming that almost half of the patients were negative for RF and ACPA at inclusion in this primary care–based cohort. The disease was less active based on DAS28-ESR and also less severe in terms of functional index and radiographic score at baseline in seronegative versus seropositive group. These results agree with those of the Norfolk Arthritis Register (NOAR) [18, 19] but not with the Canadian early-arthritis cohort (CATCH), showing seronegative patients with higher mean swollen joint count, DAS28, and erosive disease [20], which suggests that these patients are more frequently referred to rheumatology if they have more active and severe disease. The disease progression was less severe and DMARD or steroid use less frequent in seronegative versus seropositive group during follow-up in the ESPOIR cohort, which agrees with other early-arthritis cohorts [18, 20]. In our cohort, about 10% of patients developed de novo auto-antibodies (RF or anti-CCP2) during the 3 years of follow-up. This proportion was less in another Dutch early-arthritis cohort, in which the switch from negative to positive occurred in 2% for ACPA and 3% for IgM-RF during the first year of follow-up [21].

The secondary objective of our study was to determine baseline predictors of fulfilling the 2010 ACR/EULAR criteria for RA within 3 years in these patients. We were looking for factors other than auto-antibodies that suggest the development of RA. The best independent predictive factors were additive, bilateral, and distal involvement; presence of a trigger event; pain at rest; morning stiffness; number of tender joints; and mTSS. In the prediction model for persistent erosive arthritis published by Visser et al., similar baseline variables were selected, including morning stiffness for at least 1 h, arthritis in at least three joint groups, IgM-RF, and anti-CCP2 positivity, in addition to erosions on hand and foot radiographs that were not selected in our model [22].

ACR/EULAR classification criteria were used to guide the diagnosis of RA, which may lead to incorporation bias, because explanatory variables were part of the reference standard. This observation may suggest overestimation of the discriminative ability of the model [23]. To partly solve this problem, we could have added expert opinion to the ACR/EULAR criteria, although incorporation bias still exists in this case. This bias can explain the important weight of the number of tender joints in the model. Of note, the area under the receiver operating characteristic curve for the stepwise logistic regression analyses was high (0.905), which suggests that these predictors are efficient to differentiate early in the disease between RA and other forms of arthritis in clinical practice. ACR/EULAR classification criteria were built based on cohorts of real-world patients with early arthritis, to identify factors and their relative weights, which were associated with the subsequent decision by a physician to start methotrexate. Consequently, they were better adapted to the classification of patients with early arthritis. Nevertheless, performance of these criteria have also been studied in patients with established disease and had the same sensitivity, but higher specificity than the 1987 ACR criteria for predicting a diagnosis of RA after 10 years, mainly due to the use of exclusion criteria [24].

Nevertheless, some limitations in our study should be noted. First, patients were defined as seronegative if they did not have IgM-RF and anti-CCP2 antibodies. Some seronegative patients may have had other RA-associated antibodies, such as IgA-RF or anticarbamylated antibodies, which were not tested in our study. However, based on currently available studies, these patients likely represent a small proportion of the seronegative population [25]. Among 318 patients with anti-CCP-negative RA (1987 ACR criteria) included in the Leiden early-arthritis cohort, a cluster analysis performed to evaluate whether patients resemble each other revealed no grouping of patients, which suggests that anti-CCP–negative RA patients may be homogeneous [26]. Second, more patients were lost to follow-up in the seronegative group, even after accounting for the 38 patients with auto-antibodies during follow-up. This finding can be explained by patients with severe RA more likely to be followed up. Previous data noted that anti-CCP antibodies were the best predictor of remaining in the ESPOIR cohort for 5 years [27].

Patients with early arthritis and without RF and ACPA have less active disease at baseline and less severe disease during follow-up. They are more likely to fulfill RA classification at 3 years if they have RA typical inflammatory characteristics of pain, symmetric involvement of numerous small joints, and initial erosive disease.