In 1961, Arthritis & Rheumatism published an essay on the history of rheumatoid factor (RF) (1). In the essay, Dr. Charles Ragan wrote that “the significance of rheumatoid factor in the pathogenesis of rheumatoid arthritis remains conjectural” (1). Despite that fact, the utility of RF for rheumatoid arthritis (RA) diagnosis was unequivocal, and RF was already incorporated into the 1958 American Rheumatism Association diagnostic criteria for RA (2). RF, an autoantibody directed against the Fc portion of IgG, remained a criterion in the revised 1987 American College of Rheumatology (ACR) classification criteria for RA (3) and remains a part of the 2010 RA classification criteria developed by the ACR and the European League Against Rheumatism (EULAR) (4).
In the past decade, the development of anti–citrullinated protein antibody (ACPA) assays has improved our ability to diagnose patients with RA. In numerous studies, ACPA positivity has been found to be a sensitive and specific marker for RA, particularly in early disease (5). Compared to RF, ACPA appeared to have a higher specificity for early RA, but the addition of RF may capture additional cases, thereby increasing sensitivity (6). The rising importance of ACPA in clinical practice, and its ability to aid in earlier diagnosis of RA, was one factor motivating the creation of the new RA classification criteria, which include ACPA.
The 2010 RA criteria are the result of a rich collaboration between rheumatologists from many parts of the globe and the two major rheumatology organizations, the ACR and EULAR. These new criteria were created using a data-driven and consensus-based approach, utilizing data from 9 early arthritis cohorts from Europe and Canada, and developed systematically, following rigorous methodology (7) (Table 1). The criteria were developed in 3 phases. Phase 1 used a data-driven approach to identify the important domains and clinical variables that define RA from a list of RA clinical characteristics generated by consensus (8). Phase 2 engaged an international panel of clinical rheumatologists to submit real-life cases representing patients with early undifferentiated arthritis (UA). Included in these cases was clinical information the rheumatologists believed was relevant to ruling RA in or out as a diagnosis. The clinical factors of highest importance were derived using a consensus-based decision-analysis approach (9). Phase 3 incorporated the conclusions from Phases 1 and 2 to create the final 2010 ACR/EULAR classification criteria for RA (4).
|1. Create a list of possible inclusion and exclusion criteria using consensus methodology||Content validity|
|2. Potential criteria should be reliable (reproducible), precise in measurement, easy and feasible to measure, and clinically sensible|
|3. Potential criteria redundancy should be assessed and minimized|
|1. Selection of cases (patients considered to have disease):|
|a. Cases should include the spectrum of disease severity|
|b. If the criteria are to be used for epidemiologic studies, both clinical and community cases should be included|
|2. Selection of controls (patients considered not to have disease): Controls should be chosen with the intended purpose of the criteria in mind, i.e., to distinguish individuals with disease from those without disease versus to distinguish individuals with a particular rheumatic disease from individuals with other rheumatic diseases. Ideally, multiple controls should be used|
|3. An adequate number of cases and controls should be chosen, defined in the context of the sensitivity and specificity of the criteria set|
|4. For each individual criterion, and for combinations of criteria, the sensitivity and specificity for detecting and ruling out disease should be calculated. These results, together with clinical opinion, should be used to reduce the number of criteria for inclusion||Construct validity: convergent and divergent validity|
|5. Included criteria should be those with greatest demonstrated validity, sensitivity, and specificity|
|6. Acceptable statistical approaches should be used to create the classification criteria from the reduced number of criteria|
|7. The final classification criteria should be validated in samples of cases and controls distinct from the patients used to develop the criteria||Face validity|
While the new criteria set a standard for methodologic rigor, they are already being tested against other data sets. In this issue of Arthritis & Rheumatism, van der Linden and colleagues (10) report the results of a study comparing the ability of ACPA and RF to predict the development of RA or persistent arthritis in patients presenting with UA. They focused on the value of different RF titers in predicting the classification of RA if the ACPA status is known. The authors analyzed RF levels in two ways. The first was using the cutoff defined in the 2010 RA classification criteria, where a low-positive RF titer was any level higher than the upper limit of normal (ULN) but ≤3 times the ULN, while high-positive RF referred to levels >3 times the ULN for the laboratory assay. The second method defined a high RF level as any level >50 units/ml regardless of laboratory-specific cutoffs. The authors reported that a high-positive RF added only limited information for predicting the development of RA from UA. Thus, they concluded that using only RF presence rather than RF levels may improve the 2010 RA classification criteria. Based on their findings, they suggest that low versus high ACPA levels had greater value than RF levels and should be retained as part of the new classification criteria.
The study by van der Linden et al has several strengths, including the use of multiple data sets and clear statistical methods. However, the study findings are unlikely to settle the debate regarding whether the presence of RF or RF levels should remain part of the classification criteria. The authors analyzed RF levels and ACPA levels in isolation from other aspects of the 2010 RA classification criteria. Since the classification criteria were developed using methods including multivariate logistic regression and classification tree analysis, the relative importance of each criterion was considered as part of a group of RA clinical predictors. Therefore, it would have been interesting to see these authors' analysis in the context of the full classification criteria. It was unclear how much the classification criteria would be improved with the removal of RF level. During development of the new classification criteria, RF positivity and levels and ACPA positivity and levels were both significant variables in predicting treatment with methotrexate for early UA in the data-driven approach (Phase 1). A similar conclusion with regard to the importance of RF and ACPA in predicting the development of RA was reached using a consensus-based approach (Phase 2), resulting in the inclusion of both tests with defined cutoff levels in the final classification criteria. In addition, the 2010 classification criteria were designed so that only one test result is required for RA classification.
Whether or not one agrees with the findings of this study, it raises another interesting issue: how often should we update classification criteria for any disease? On the title page of the 2010 RA classification criteria article, ACR and EULAR issued a joint statement that “approved criteria sets are expected to undergo intermittent updates” (4). The stated goal of the 2010 RA classification criteria was to facilitate the study of RA subjects at all stages and allow comparison across studies and geographic areas. The design and implementation process for clinical trials and studies can take years to complete. Constantly changing criteria sets would undermine the overall goal of this exercise, which is to harmonize studies' inclusion criteria, facilitating comparison. Clearly, classification criteria need constant evaluation. However, it would be unwise to modify the new RA classification criteria before investigators and clinicians have had a chance to see how they perform in studies and in practice. Recently presented abstracts (11–14) suggest that the new classification criteria will help identify more patients with early RA, facilitating study and hopefully earlier initiation of disease-modifying treatment.
The development of ACPA was a major advancement in the field of RA, and there was no question that the criteria developed in 1987, based on patients with established RA, required updating. It is possible that in the next 23 years ACPA will no longer be needed, supplanted by markers that are directly related to the pathogenesis of the disease. Today, the role of RF in RA pathogenesis remains a conjecture. We believe that it is an epiphenomenon of the processes involved in the development of RA. However, its importance in the classification of RA has weathered 2 subsequent updates and perhaps more to come.