3 Classification criteria for rheumatoid arthritis

A L E X A N D E R J. MACGREGOR

The need to develop a standardized method of case recognition in the study of rheumatoid arthritis (RA) has been recognized since the early 1950s. The earliest approaches involved developing consistent methods of clinical appraisal based on the opinion of the examining physician as to the presence of disease (de B16court, 1953; de Graaf, 1959). An example was the Manchester clinical grading system used in population studies by Kellgren et al (1953). There were also attempts to develop a more objective approach to classification. An example is Miall's study of RA and pneumoconiosis in a Welsh mining community (Miall, 1955). Cases of RA were classified on the basis of a characteristic history of polyarthritis involving the peripheral joints, whether residual physical signs were present or absent, supported by either radiological evidence of RA in the hands and feet, o r by a positive differential agglutination test (Ball, 1952). These criteria were felt to produce a reproducible diagnosis by only including cases of severe RA. Other investigators examined the ability of individual clinical findings to distinguish RA from other rheumatic disorders. Cobb et al (1955), in a preliminary analysis of the Pittsburgh Arthritis Study, recognized that radiological changes identified only a small fraction of individuals with joint disease and that relatively minor differences in definitions led to large differences in the numbers in the population that could be classified as having RA. In this analysis, morning stiffness was noted to differentiate RA from degenerative joint disease. Kellgren and Lawrence (1956) studied the results of surveys of rheumatic complaints among adult inhabitants of a sample of the population living in Leigh, Lancashire. Their analysis demonstrated that the presence of radiological erosions and a positive direct antiglobulin test (Miall's criteria) were insufficiently sensitive. Additional features such as the occurrence of rheumatoid nodules, sweating palms and cold blue extremities were found to be predictors of clinical disease. In contrast to the findings of Cobb et al (1955), morning stiffness was not found to be a good discriminator for RA, except in younger age groups. Bailli~re's Clinical Rheumatology287
Vol. 9, No. 2, May 1995 ISBN 0-7020-1939-9 Copyright 9 1995, by Baillirre Tindall All rights of reproduction in any form reserved

288 C R I T E R I A IN R E C E N T USE ARA criteria 1956-1961

A. J. M A C G R E G O R

In 1956 a committee of the American Rheumatism Association (ARA) was established to obtain uniformity in cases listed as RA (Ropes et al, 1956). The committee recognized that it would be difficult to determine the limits of the diagnosis of RA and attempted to classify the disease into three categories: 'definite', 'probable' and 'possible'. The first two categories were intended for studies of the course, characteristics and treatment of RA while the latter category of 'possible' disease was intended as a means of detecting early, atypical cases. It was emphasized that the criteria were not designed as diagnostic aids, but were meant simply as a tool for classification. The criteria selected were developed on the basis of the clinical experience of the members of the committee. A list was formed of all the manifestations of RA which might have sufficient diagnostic value to be worthy of consideration. The analysis included data from the Pittsburgh Arthritis Study (Cobb et al, 1955; Cobb et al, 1956) in addition to data collected from a number of physicians interested in rheumatic diseases in various parts of the United States and Canada. Physicians were asked to select from their clinics or practices the five most recent eases of classical RA, five cases of probable RA and five with no evidence of RA. In all, 332 case reports were obtained from 19 different cities. Disease characteristics were chosen which had the highest specificity (lowest false positive rate) for classification as 'definite' and 'probable' disease and with a high sensitivity (low false negative rate) for 'possible' RA. The final criteria set included serum rheumatoid factor (RF) and evidence of erosive disease on hand and foot X-rays (Table 1). In addition, data on the presence of a mucin clot, and characteristic histological features both on synovial biopsy and biopsy of rheumatoid nodules were included. A number of characteristics were eliminated including weight loss, vasomotor symptoms, paraesthesiae, splenomegaly, pericarditis, myocarditis, pleurisy, reversal of the albumin/globulin ratio in the plasma and thickening of the palmar fascia. Five of the listed eleven criteria were required for inclusion as 'definite' RA and three for classification as 'probable' RA. For classification as 'probable' and 'definite' disease it was a requirement that soft tissue swelling should have been observed by a physician and must have been present for at least 6 weeks. For 'possible' RA the criteria were less strict. The required duration of symptoms was 3 weeks and a history rather than the observation of joint swelling was adequate. Interestingly, iritis was also included as a criterion of possible RA. The problem of including patients with other rheumatic diseases was considered. It was felt important not to exclude cases of degenerative joint disease that had progressed to develop RA. The rigid criteria for 'probable' and 'definite' disease were thought to be sufficient to have excluded uncomplicated degenerative joint disease. The criteria were, however, felt to be insufficiently specific to distinguish

CLASSIFICATION CRITERIA FOR RA Table 1. 1958 ARA diagnostic criteria for rheumatoid arthritis.* 1. 2. 3. 4. 5.

289

6. 7. 8.

9. 10. 11.

Morning stiffness Pain on motion or tenderness in at least one joint (observed by a physician) Swelling (soft tissue thickening or fluid, not bony overgrowth alone) in at least one joint (observed by a physician) Swelling (observed by a physician) of at least one other joint (the interval between two joint involvements must be no more than 3 months) Symmetrical joint swelling (observed by a physician) with simultaneous involvement of the same joint on both sides of the body (bilateral involvement of the mid-phalangeal, MCP or MTP joints is acceptable without absolute symmetry). Terminal IP joint involvement will not satisfy the criterion. Subcutaneous nodules (observed by a physician) over bony prominences, on extensor surfaces or in juxta-articular regions X-ray changes typical of rheumatoid arthritis (which must include at least bony decalcification localized to or greatest around the involved joints) Positive agglutination test--demonstration of RF by any method which in two laboratories has been positive in not over 5% of normal controls---or positive streptococcal agglutination test (1956 criteria: positive sheep cell agglutination or positive streptococcal agglutination test) Poor mucin precipitate from synovial fluid Characteristic histologic changes in synovial membrane Characteristic histologic changes in nodules

Classical RA requires 7/11 criteria with 1-5 present continuously for 6 weeks (not included in 1956 criteria) Definite RA requires 5/11 criteria with 1-5 present continuously for 6 weeks Probable RA requires 3/11 criteria with at least one of 1-5 present continuously for 6 weeks (1956 criteria: 4 weeks) Possible RA requires two of the following criteria and a total duration of joint symptoms of at least three weeks: (1) Morning stiffness (2) Tenderness or pain on motion (observed by a physician) with a history of recurrence or persistence for three weeks (3) History or observation of joint swelling (4) Subcutaneous nodules (observed by a physician) (5) Elevated sedimentation rate or C-reactive protein (6) Iritis. Exclusions- (1) Typical rash of SLE (2) High concentration of LE cells (3) Histologic evidence of periarteritis nodosa (4) Weakness of the neck, trunk and pharyngeal muscles or persistent muscle swelling of dermatomyositis (5) Definite scleroderma (6) A clinical picture of rheumatic fever (7) A clinical picture of gouty arthritis (8) Tophi (9) Acute infectious arthritis (10) Evidence of joint tuberculosis (11) A clinical picture characteristic of Reiter's syndrome (12) A clinical picture characteristic of shoulder-hand syndrome (13) A clinical picture of hypertrophic pulmonary osteoarthropathy (14) Neuroarthropathy (15) Homogentisic acid in the urine (16) Histologic evidence of sarcoid or a positive Kveim test (17) Multiple myeloma (18) Erytbema nodosum (19) Leukaemia or lymphoma (20) Agammaglobulinaemia (not included in 1956 criteria). * Where the 1958 criteria differ from the 1956 criteria, the variations in the 1956 criteria are shown in italics Adapted from Ropes et al (1956) and Ropes et al (1958) from the Bulletin on the Rheumatic Diseases, copyright 19. Used by permission of the Arthritis Foundation.

RA in the presence of certain other rheumatic diseases, such as systemic lupus erythematosus (SLE) and gout. It was felt best to classify individuals with these diseases separately for comparative purposes. A list of exclusions was published with the criteria set (Table 1). It is of interest to note that patients labelled as rheumatoid spondylitis, psoriasis, ulcerative colitis and those with an onset of disease under the age of 12 years were recommended to be included in the rheumatoid groups, although it was suggested that these should be separately listed. After 2 years of use, the ARA criteria committee reviewed their original

290

A . J . MACGREGOR

criteria. There was agreement that the criteria were not sufficiently rigorous (Ropes et al, 1958). Consequently, an additional category of 'classical' disease was added for which seven of the 11 criteria had to be satisfied. All of the first five criteria had to be present continuously for 6 weeks for the individual to be classified as having 'classical' or 'definite' disease. Agammaglobulinaemia was added to the exclusion criteria. A difficulty had been encountered in assigning positivity for RF as many assay methods were in use. The committee recommended the use of any method that had been demonstrated by at least two laboratories to be positive in less than 5% of normal controls. The final criteria set (the 1958 ARA revised criteria, Table 1) gained widespread acceptance and was used extensively. The criteria were reviewed at a symposium on Population Studies in Relation to Chronic Rheumatism, sponsored by the Council for International Organizations of Medical Sciences (CIOMS) in Rome in 1961 (Kellgren, 1962). Two problems had emerged in their use. The final three criteria based on invasive tests had proved to be impractical. Additionally, it was recognized that the RA criteria failed to encompass features of
Table 2. The Rome criteria for inactive RA.

1. 2.

3 4.

A past history of polyarthritis Symmetrical deformity of peripheral joints consisting of ankylosis or irreducible subluxation, especially of the lateral MTP or MCP joints; there must be some involvement of one hand or foot; involvement limited to large joints, such as the elbows of knees does not satisfy this criterion Radiological changes of RA of grade 2 or more Positive serological test for RF

Definite RA requires 3/4 criteria Probable RA requires 2/4 criteria Adapted from Kellgren (1962) from the Bulletin on the Rheumatic Diseases, copyright 19. Used by permission of the Arthritis Foundation.

inactive disease. New criteria sets were proposed: the Rome (active) and the Rome (inactive) criteria. The criteria for active disease were equivalent to the 1958 criteria, except that the final three elements of the criteria set were removed. Criteria for 'definite' and 'probable' inactive disease were devised (Table 2). An additional change suggested at the symposium was that ankylosing spondylitis should be added to the list of exclusions.

New York Criteria, 1966

Shortcomings in the criteria for inactive disease were discussed at the Third International Symposium on Population Studies of the Rheumatic Diseases held in New York in 1966 (Bennett and Burch, 1967; Bennett and Wood, 1968). It was apparent that, in population studies of RA, the method of grouping of criteria listed in the Rome criteria was inadequate. The frequency of occurrence of many of the characteristics included had not been established in normal population samples. A study by Lawrence and Wood (1968) had examined data from surveys in Northern England during the years 1954-1961 and had demonstrated that, in comparison with a clinical grading of RA, the 1958 criteria had

CLASSIFICATION CRITERIA FOR RA

291

been both insufficiently sensitive and specific, and had led to important misclassification of cases of osteoarthritis. Their analysis indicated that identifying the pattern of joint involvement was critical for case recognition in the population. Using these data, the 13 most important clinical variables that resulted in the clinical classification of RA were incorporated into the format of the Rome (inactive) criteria. Emphasis was placed on the specificity of these variables such that the new criteria included few false positive cases. Criteria with a low population prevalence, such as rheumatoid nodules, were not included. The result was a set of four criteria designed to identify cases of both active and inactive disease in the population setting (the New York criteria, Table 3). Recommendations on summation of the criteria were not made. Summation was considered unnecessary as long as the
Table 3. The New York criteria. 1. A history, past or present, of an episode of joint pain involving three or more limb joints, without stipulation as to duration. Each group of joints (e.g. PIP) is counted as one joint, each side scores separately Swelling, limitation of motion, subluxation or ankylosis of at least three limb joints (excluding: DIP, 5th PIP, 1st carpometacarpal joint, I st MTP and hips) with symmetry of at least one joint pair. There must be involvement of one hand, wrist or foot, as involvement limited to large joints such as elbows or knees does not satisfy this criterion. Sublnxation of the lateral MTPs must be irreducible. Radiological features of grade 2 or more erosive arthritis in the hands, wrists or feet. Positive RF by a method controlled by periodic testing of reference sera and by exchange of sera with other laboratories.

2.

3. 4.

No recommendations for summation were made. Adapted from Bennett and Burch (1967) from the Bulletin of the Rheumatic Diseases, copyright 19. Used by permission of the Arthritis Foundation.

distribution of each criterion was reported individually. Exclusions were felt to be inappropriate in the population setting where there was an emphasis on including all possible features of disease.

ARA revised criteria, 1987

In 1983 an ARA subcommittee was appointed to revise the 1958 RA criteria. Their recommendations were subsequently published as the 1987 revised ARA criteria (Arnett et al, 1988). The modifications were developed in recognition of the changes in the clinical definition of RA over the preceding 30 years. In particular, the committee cited the HLA B27-associated spondyloarthropathies, the pseudorheumatoid form of calcium pyrophosphate dihydrate deposition disease, polymyalgia rheumatica and Lyme disease as clinical entities hitherto included as seronegative RA, which would no longer be included in the definition. It was also recognized that, in practice, the distinction between 'classical' and 'definite' RA had not proved useful and that patients previously classified as 'probable' RA most commonly did not have the disease. Criteria involving invasive procedures were again noted to be impractical. There was concern that the 'exclusions' proposed with the criteria were difficult

292

a.J. MACGREGOR

to apply. The ARA criteria subcommittee set themselves the task of defining new criteria that would be more sensitive and specific and required a stricter definition of RA. The revised criteria set was developed by comparing clinical features recorded in patients and controls recruited from university hospital clinics and private practices. Patients were recruited consecutively and comprised 366 adults over the age of 16 years with established RA. Controls were taken as the next consecutive patient with a rheumatic disease other than RA. Patients with localized rheumatic conditions were not included. The 'gold standard' used for analysis was the clinical opinion of the panel of physicians. Data were collected in the format of the 1958 ARA criteria and the New York criteria. Groups of criteria were divided into components in an attempt to establish the most discriminating definitions. Two different statistical analyses were performed on the pooled data. In the first, combinations of criteria were established that were the most sensitive and the most specific for distinguishing RA and non-RA in cases and controls. The second method involved a process of recursive partitioning (Breiman et al, 1984) in which a hierarchical classification tree is established to assign a diagnosis based on data from the group as a whole. In this analysis, the clinical criteria with the highest combined sensitivity and specificity were those which reflected the characteristic pattern of joint involvement in RA. Thus, wrist or metacarpophalangeal (MCP) involvement had an accuracy (mean sensitivity and specificity) of 86.1% and involvement of three or more joint areas had an accuracy of 87.4%. Symmetrical arthritis in any region had an accuracy of 84.3%. These individual criteria had a similar accuracy to the New York criterion of joint swelling (87.4%). The analysis also confirmed that the specificity of the morning stiffness criterion was greatly enhanced if the duration of morning stiffness was specified as greater than one hour. The final formulation of the criteria (Table 4) replaced the former categories of 'classical' and 'definite' disease with the single disease category of 'rheumatoid arthritis'. 'Probable' and 'possible' RA were removed. The criteria considered only data from 14 specified joint areas (seven on each side). Information on involvement of the shoulders, which was accepted in previous criteria sets, was no longer included. The 1987 criteria set also replaced the second, third and fourth criteria of the 1958 criteria with 'arthritis of three or more joint areas' and 'arthritis of the hand joints'. Rheumatoid factor, the presence of radiological erosions and rheumatoid nodules were retained. The radiographic requirement was restricted to interpretation of a single posteroanterior radiograph of both hands. The criteria involving joint aspiration, synovial biopsy and rheumatoid nodule biopsy were deleted. Exclusions were not specified in the new criteria set to render them more suitable for population studies. In the classification tree formulation (Figure 1), the population of cases and controls was divided into eight subsets, five representing patients classified as RA. The scheme allowed swelling of the wrists to substitute for RF, and swelling of the MCP joints to substitute for radiological

CLASSIFICATION CRITERIA FOR RA Table 4. The revised 1987 ARA criteria. 1. 2. 3. 4. Morning stiffness Arthritis in three or more joint areas* Arthritis of hand joints Symmetrical arthritis

293

5. 6. 7.

Rheumatoid nodules Rheumatoid factor Radiographs changes

Morning stiffness in and around the joints, lasting at least one hour before maximal improvement Soft tissue swelling or fluid (not bony overgrowth) observed by a physician, present simultaneously for at least 6 weeks Swelling of wrist MCP or PIP for at least 6 weeks Simultaneous involvement of the same joint areas (defined in 2) on both sides of the body (bilateral involvement of PIPs, MCPs or MTPs is acceptable without absolute symmetry) for at least 6 weeks Subcutaneous nodules over bony prominences, or extensor surfaces, or in juxta-articular regions, observed by a physician Detected by a method positive in less than 5% normal controls Typical of RA on posteroanterior hand and wrist radiographs which must include erosions or unequivocal bony decalcification localized to or most marked adjacent to the involved joints (OA changes alone do not qualify)

* Proposed areas: right or left PIP, MCP, wrist, elbow, knee, ankle, MTP. At least four criteria must be fulfilled for classification as RA. Patients with two clinical diagnoses are not excluded. Adapted from Arnett et al (1988). Adapted from the Primer on the Rheumatic Diseases, 10th edn, copyright 1993.

I Arthritisof threeor morejoint areasI

no .............

I I'.wellinO:MoPI I X'mychanges I R(hsEwUit~]i:dfraCtt~ I e~l~ r 9 9

I .heumato,d,actor I (.swelling:wrist)
~'~/ I Symmetrical I i
!

/ ', I swellin~l I ~e~l ~ o / . I ~ yesI "no /ISwelling MCPI / ~# I I andwrist I / vl / / "". / I SwellingMCP

ye

n~"~.

"" "-

i orwrist I / "' o /
;

RA

notRA )

Figure 1. 1987 ARA decision tree algorithm for RA classification. * Variables in parentheses can be used when data on the first listed variable is unavailable. Adapted from Amett et al (1988).

294

A.J. MACGREGOR

changes when these data were not available. Data on morning stiffness and rheumatoid nodules were not required. Interestingly, by this method the majority of individuals are correctly classified by the simple requirement of arthritis affecting three or more joint areas with typical findings on hand radiographs (or with MCP swelling if these data are not available). Swelling of two or less joint areas in the absence of RF excludes RA in the majority of individuals. VALIDATION Requirements of disease classification The historical development of classification criteria in RA has reflected the different requirements for case recognition in clinic and population settings. In the clinic, the requirement is to distinguish RA from other inflammatory arthritides. Given that few cases of mild or inactive disease are likely to be represented, criteria that have been developed in clinic populations emphasize features of active disease. Mitchell and Fries (1982) have demonstrated that for the 1958 ARA criteria there is a close association between the number of criteria satisfied and the severity of clinical disease. By contrast, case recognition in the population needs to include milder and inactive disease and must distinguish these from cases of osteoarthritis, which will comprise the majority of non-RA joint disease in this setting. Only the New York criteria (Table 3) were developed specifically for use in population studies. Although elements of the New York criteria were incorporated into the 1987 revision of the ARA criteria, the latter criteria were not derived by studying a population sample. The following two sections consider contemporary studies where criteria in current use have been validated (a) in the setting of hospitals and outpatient clinics and (b) in population and family studies. Case definition in the clinic In the development of the 1987 ARA classification schemes, the new criteria (both in 'four from seven' (4/7) and decision tree (DT) formats) were found to be of comparable sensitivity but enhanced specificity when compared with the 1958 criteria (Arnett et al, 1988). The investigators also analysed the certainty of diagnosis in those that had been misclassified. A score was allocated to each patient based on the opinion of the examining physician using a scale from 0 (not RA) to 100 (definite RA). The average score among those classified as RA was 91 and among those classified as not RA was 5 by both ARA(4/7) and ARA(DT) methods. Of those patients that were misclassified, the certainty scores were close to the midrange for both cases and controls, suggesting uncertainty about the original diagnosis in these individuals. The authors concluded that the measured sensitivity and specificity of the new criteria sets had understated their true accuracy.

CLASSIFICATION CRITERIA FOR RA

295

~,

I o~

oooo

~o

~o.~

o~

u~ ~a
01

I~ ~ = ~

~'~
~o.o ~ ~ o ~

296

A.J.

MACGREGOR

It is to be expected that the most optimistic assessment of the 1987 criteria sets would be obtained by examining their performance in the population from which they were derived. Data are now available from a number of studies which have compared the performance of the 1958 and 1987 criteria in hospital clinic settings (Table 5). In general, the findings confirm that the 1987 criteria are of enhanced specificity but reduced sensitivity. The lack of sensitivity of the 1987 ARA criteria when compared with the 1958 criteria appears to be more pronounced for newly diagnosed cases of RA. This has been examined further by Moens et al (1992) in two groups of individuals presenting to the outpatient department at the Jan van Breeman Institute in Amsterdam. Records were examined (a) retrospectively from 1338 patients with a clinical diagnosis of 'definite' or 'classical' RA and (b) prospectively from newly referred cases of whom 93 had a certain diagnosis of RA and 1357 had neither certain nor possible diagnoses of RA. The latter acted as controls. Using data from the first visit of the retrospective patients, the sensitivity of the 1987 ARA(4/7) criteria was 66% compared with 71% for the 1958 criteria (Table 5). These values were increased by incorporating data from the follow-up visits. In the prospective group, the sensitivity of the 1987 ARA(4/7) criteria was again lower than the 1958 criteria (82% compared with 86%). The lack of sensitivity was attributed to the failure of the 1987 criteria, unlike the 1958 criteria, to specify explicitly the time interval during which criteria should be fulfilled. Furthermore, the radiological criterion of hand involvement failed to identify erosive disease in 12 out of 17 cases classified as false negative in whom there was clinical or radiological evidence of metatarsophalangeal (MTP) involvement. The value of including MTP involvement for case recognition in early RA has also been noted by others (Paimela, 1992; Symmons et al, 1994). The utility of excluding named diagnoses was also examined by Moens et al (1992). Their analysis of false positive cases in the prospective group suggested that any appropriate list of exclusions would need to have been at least as long as that given with the 1958 criteria. The comparable specificities of the 1958 and 1987 criteria suggested that the use of such exclusion criteria was, however, unnecessary. These findings contrast with those of Levin et al (1989) who reported that the specificity of the 1987 criteria was reduced in comparison with the 1958 criteria. In their study, a high proportion of false positives was found among controls with SLE, psoriatic arthritis and gout. These results suggest a greater need for exclusions in settings where a high proportion of non-RA cases have other inflammatory arthritides.

Case definition in population and family studies

There have been few studies comparing the performance of existing criteria sets in populations with disease and none as yet comparing the performance of the 1987 criteria with the New York criteria to assess whether they are an adequate substitute in this setting. Recently, as part of a larger study of

CLASSIFICATION

CRITERIA FOR RA

297

outcome assessment in RA, Jacobson et al (1994) have compared the performance of the Rome criteria and the 1987 ARA criteria. Their analysis was based on cross-sectional and longitudinal data from repeated health examinations of 3509 Pima Indians. Clinical diagnosis was used as the 'gold standard' for comparison. They demonstrated that, while both criteria sets were of comparable specificity, the 1987 criteria had substantially reduced sensitivity. The 1987 ARA criteria identified only a subset comprising 50% of those identified by the Rome (active) criteria. It was noted that the sensitivity of both sets was enhanced if they were combined with the Rome criteria for inactive disease or equally if data from serial examinations were taken into account. These results indicate that the 1987 ARA criteria may fail to identify mild or inactive disease. Hakala et al (1993) have also reported that the 1987 criteria only recognize severe RA in a population sample. The importance of incorporating data on remitted disease in population studies to assess the occurrence of disease during the lifetime of the individual from a single examination has recently been stressed (MacGregor and Silman, 1992). Two studies by Cathcart and O'Sullivan in Sudbury, Massachusetts (Cathcart and O'Sullivan, 1970; O'Sullivan and Cathcart, 1972) illustrate the importance of this issue. The disease status of a group of 4552 individuals was assessed over a 5-year interval. Definite RA was identified at baseline in 0.87% using the 1958 ARA criteria but only in 0.37% using the New York criteria. At follow up 65% of those who originally fulfilled the New York criteria continued to be classified as having RA whereas only 36% of those originally classified as having definite RA by the 1958 criteria remained in that category. The findings demonstrate that the 1958 criteria, unlike the New York criteria, are affected by fluctuations of disease activity. The results also suggest that the New York criteria may be too specific and may fail to recognize cases of mild disease. We have recently examined the performance of the New York and 1987 ARA criteria in a twin study of RA where there was a clear requirement to include as cases those individuals with inactive disease (MacGregor et al, 1994). The sample of twins was derived from (a) a prospective outpatient survey and (b) cases recruited from the general population by a media campaign (Silman et al, 1993). The 'gold standard' for comparison was the diagnosis of the referring physician. The importance of incorporating data on past evidence of disease activity on the performance of the 1987 4/7 and DT criteria was examined by adapting the format of the criteria (a) to recognize only current features of disease and (b) to incorporate data on disease activity 'ever'. The latter adaptation allowed joint deformity to substitute for joint swelling and incorporated past data on RF seropositivity and the presence of rheumatoid nodules (Table 6). Our analysis showed that criteria which only recognize currently active disease substantially underestimate the lifetime cumulative prevalence of RA. However, when the 1987 criteria were adapted to incorporate evidence of past disease, their sensitivity was greatly enhanced. The modified 1987 criteria had a similar accuracy (87%) to that of the New York criteria using 'three from four' as the cut-off point (86%).

298

A. J. MACGREGOR

Table 6. Format for application of the revised 1987 ARA criteria to population and family studies.

1. 2.

Morning stiffness Arthritis in at least three joint areas* Arthritis of hands Symmetrical arthritis

3. 4.

5.

Rheumatoid nodules

6.

Rheumatoid factor

7.

Radiograph changes

Morning stiffness in and around the joints lasting at least one hour before maximal improvement at any time in the disease course Soft tissue swelling or fluid observed by a physician, with swelling at current examination or deformity and a documented history of swelling Swelling of wrist, MCP or PIP with swelling at current examination or deformity and a documented history of swelling Simultaneous involvement of the same joint areas (defined in 2) on both sides of the body (bilateral involvement of PIPs, MCPs or MTPs is acceptable without absolute symmetry) with swelling at current examination or deformity and a documented history of swelling Over bony prominences or extensor surfaces, or in juxta-articular regions, observed by a physician and present at current examination or documented in the past By a method positive in less than 5% normal controls at current examination or documented to have been positive in the past by any assay method Typical of RA on posteroanterior hand and wrist radiographs which must include erosions or unequivocal bony decalcification localized in or most marked adjacent to the involved joints (OA changes alone do not qualify)

* As defined in Table 4. At least four criteria must be fulfilled for classification as RA. Adapted from Arnett et al (1988) and MacGregor et al (1994).

The New York criteria have not found widespread acceptance over the years because of their cumbersome formulation (Arnett et al, 1988). The results from this analysis of twins have indicated that the 1987 ARA criteria, applied in a way which incorporates past occurrence of disease, would form an adequate substitute for the New York criteria in studies in which there was a requirement to measure the lifetime cumulative prevalence of disease. This method of application of the 1987 criteria was endorsed by the American College of Rheumatology (ACR) criteria committee.

DISCUSSION One criticism of the present approach to classification is that it results in assignment of the label RA to individuals who represent many different parts of a wide spectrum of clinical disease. This criticism is, in particular, levelled at methods of classification in which any combination of criteria from a given set may be satisfied to permit classification as RA as long as a certain threshold number is attained (for example, 4 in the 1987 ARA criteria) (Abramson, 1967). Edwards (1988) has criticized this 'deliberate heterogeneity' and has suggested that there are compelling reasons for using quite different selection criteria when studying, for example, the natural history of early peripheral joint synovitis or the place of auranofin in the treatment of chronic synovitis. Mau et al (1989) recognized this issue

C L A S S I F I C A T I O N CRITERIA FOR RA

299

as a particular problem in the study of early RA and have advocated the use of detailed descriptive diagnoses as an alternative method for disease classification when studying the natural history of the disease. Silman (1988), however, argues that it is unlikely that there would be agreement on such descriptive classification schemes. It is of interest to note that the New York classification scheme, which has a descriptive format (Table 3), has been found to be too cumbersome for practical use and has not been adopted widely (Arnett et al, 1988). A further consequence of achieving greater homogeneity through more descriptive disease definition would be to reduce the size of the available population for study, hence reduce statistical validity. A related concern is that the method of disease classification is based on circular reasoning (Healey, 1990) and therefore will be maximally efficient at classifying as RA cases that conform to a stereotype. This approach may lead to inclusion of subsets into the classification of RA, which may represent distinct disease entities if their natural history were followed for prolonged periods. Seronegative RA, for example, may represent a discrete disease entity, although this continues to be disputed (Calin and Marks, 1981; Masi and Feigenbaum, 1983). Late onset steroid-responsive symmetrical wrist synovitis may also represent a distinct disease that would be classified as RA by the current criteria (Healey and Sheets, 1988). Conversely, juvenile rheumatoid arthritis is conventionally excluded by an arbitrary age cut-off, yet may more correctly belong to the spectrum of RA. The ARA classification schemes, however, do not preclude the identification of subgroups of RA. Mitchell and Fries (1982), for example, have analysed the linkages between the 1958 ARA criteria in a clinic-based series of 840 patients over the age of 16 and identified seven major clinical syndromes. More recently Sellick et al (1990) have also used nonhierarchical cluster analysis methods to examine the distribution of clinical features amongst cases classified as RA by the 1987 criteria. Their method classified patients into five groups on the basis of biochemical and clinical variables. The 1987 classification tree itself gave rise to five discrete end points in which subsets of disease could be classified as RA (Figure 1). Other authors have also recognized different subsets of RA based on the patterns of disease progression (Gordon et al, 1973; Masi et al, 1983; Pincus and Callahan, 1994). Concern has been expressed that different criteria for case recognition are required in different settings (Allander, 1973; Edwards and Snaith, 1988; Silman, 1988). This has been addressed to an extent in the development of the ARA criteria where a distinction has been drawn between case recognition in the clinic and in population samples. The validation studies discussed above indicate that the 1987 ARA criteria set are superior to existing criteria sets as a method of case recognition in the clinic. In population samples, the 1987 criteria also perform well in comparison with the more cumbersome New York criteria, if they are specifically adapted to incorporate evidence of past disease. It is important to stress, however, that the 1987 ARA criteria set were developed in patients and controls recruited from a sample of hospital clinic attenders. Important variation in the

300

A.J.

MACGREGOR

sensitivity and specificity of criteria may therefore occur when these characteristics are measured in different groups. One group, highlighted above, in which the 1987 criteria perform poorly is in cases of early RA where the radiological criterion may be insufficiently sensitive. Lichtenstein and Pincus (1991) have recently shown that the RF seropositivity criterion may have reduced sensitivity in population studies of RA. A review of five cross-sectional population studies of RA showed the prevalence of RF in individuals who satisfied the 1958 criteria for RA was substantially lower than the 70-90% figure reported in hospital clinic-based studies. Potential differences in the utility of individual criteria are an important consideration in studying the ethnic and geographical variation in the occurrence of RA. A high background prevalence of RF has been reported in a number of African and Asian populations (Abramson, 1967). A low prevalence of RF has been reported in certain African patients with RA (Adebajo et al, 1993). The morning stiffness criterion has also been found to be a particularly poor discriminator of RA in African populations (Beighton et al, 1975; Adebajo, 1988). The reliability of individual classification criteria is a further important issue. In the development of criteria sets, attention has been given to standardization of assays for serum rheumatoid factor. The assessment of radiographs has also been subject to standardization (Larsen et al, 1977). Despite this, important inter-observer variation in the interpretation of radiographs is recognized to occur (Abramson, 1967; Wood, 1970). Wide inter-observer and within-patient variation has also been documented in the clinical assessment of morning stiffness (Kellgren and Lawrence, 1956) and of joint swelling (O'Sullivan et al, 1968). In the 1987 ARA criteria, the reliability of the individual criteria based on clinical observation is likely to have been enhanced over previous sets by incorporating the pattern of joint involvement and by restricting the number of eligible joint areas to 14. The definition of symmetrical joint involvement, however, continues to be very much open to interpretation. There also remains substantial within-patient variation in the morning stiffness criterion as defined in the 1987 criteria set (Hazes et al, 1994). The search to find more objective methods for quantifying clinical observations in RA has a long history. Suggestions such as the use of dolorimetry in the quantitative appraisal of joint tenderness and radiological assessment for measurement of swelling (Abramson, 1967) have not found a place in classification. More recently, however, practical modifications have been proposed to the criteria for joint swelling and morning stiffness (Rigby and Wood, 1991; Hazes et al, 1993) which may prove to be of value and merit further exploration. There is also the hope that serological and genetic markers of disease may in the future provide tools for disease classification. As yet, proposed new serological markers such as antikeratin antibody and antiperinuclear factor have not proved to be alternatives to RF (yon Essen et al, 1993). The results of immunogenetic studies are, however, more promising as the association between RA and genes encoding class II antigens is clarified

CLASS|FICATION CRITERIA FOR RA

301

(Gregerson et al, 1987; Weyand et al, 1992). Recent studies have shown a 200-fold increased risk of RA in certain genetically defined subgroups (Ollier et al, 1993). This magnitude of association, which approaches the strength of the association between HLA-B27 and ankylosing spondylitis (Brewerton et al, 1973), may assist disease classification in certain subgroups. Family data on RA may also in the future contribute to classification (Walker et al, 1987). Classification methods for RA will require continual reappraisal as the clinical concept of the disease evolves and as greater knowledge of aetiological mechanisms develop. For the present, however, the 1987 criteria set best reflect the current understanding of RA and provide the most appropriate basis for case recognition of established disease both in the clinic and in the population.
SUMMARY

The development of classification schemes for RA in the last 40 years has followed the increasingly precise understanding of the nature of the clinical disease and the recognition of the different requirements of classification methods in clinic and population settings. In published studies of RA in clinic patients the most widely used criteria sets have been the 1958 ARA (ACR) criteria and its 1961 adaptation (the Rome (active) criteria). These sets classified disease as 'classical', 'definite', 'probable' and 'possible' RA based on criteria comprising clinical, serological, radiological and histological features (the latter were dropped from the Rome criteria set because of their impracticality). More recently, a new criteria set (the 1987 ARA criteria) has been developed using statistical techniques. This set was derived using RA cases and controls attending hospital clinics. It is based on the earlier criteria sets but accommodates the characteristic pattern of joint involvement in RA more precisely. The criteria recognize only the single disease category of 'rheumatoid arthritis'. In validation studies, the 1987 criteria set has been found to have enhanced specificity over earlier schemes in clinic-based studies of RA. The sensitivity may, however, be reduced, in particular in studies of early disease. The application of classification criteria for case recognition in the population and family studies of RA has proved more problematic. In these settings, there is the additional requirement to recognize individuals with remitted and inactive disease as RA cases. The 1966 New York criteria were developed for this specific purpose, however their format proved cumbersome and they have not been widely adopted. The 1987 criteria set is insufficiently sensitive to recognize inactive disease if the criteria are applied exactly as they have been defined. The sensitivity of the 1987 criteria set is, however, substantially enhanced if the criteria are adapted to incorporate features of past disease activity, for example by allowing deformity to substitute for swelling and by incorporating data on the past occurrence of rheumatoid factor and rheumatoid nodules.

302

A . J . MACGREGOR

Developments in the immunology and genetics of RA may in the future provide more accurate tools for classification and may lead to recognition of more precise disease subsets. At present, however, the 1987 ARA criteria provide the most appropriate basis for case recognition in both clinic and population-based studies. REFERENCES
Abramson JH (1967) On the diagnostic criteria of active rheumatoid arthritis. Journal of Chronic Disease 20: 275-290. Adebajo AO (1988) ARA criteria for rheumatoid arthritis (letter). British Journal of Rheumatology 28: 177. Adebajo AO, Charles PJ, Hazelman BL & Maini RN (1993) Serological profile of rheumatoid arthritis in West Africa. Rheumatology International 12: 235-238. Allander E (1973) Conflict between epidemiological and clinical diagnosis of rheumatoid arthritis in a population sample. Scandinavian Journal of Rheumatology 2:109-112. Arnett FC, Edworthy SM, Bloch DA et al (1988) The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis and Rheumatism 31: 315-324. Ball J (1952) Sheep cell agglutination test for rheumatoid arthritis. A clinieopathological study. Annals of the Rheumatic Diseases 11:97-111. Beighton P, Solomon L & Valkenburg HA (1975) Rheumatoid arthritis in a rural South African Negro population. Annals of the Rheumatic Diseases 34: 136-141. Bennett PH & Burch TA (1967) New York symposium on population studies in the rheumatic diseases: new diagnostic criteria. Bulletin on the Rheumatic Diseases 17: 453-458. Bennett PH & Wood PHN (eds) (1968) Population Eudies of the lgeumatic Deases. Amsterdam: Excerpta Medical Foundation. de Bl6court JJ (1953) La lutte contre le rheumatisme aux pays bas. Revue du Rhumatisme et des Maladies Ostdo-Articulaires 20: 105. Breiman L, Friedman JH, Olshen RA & Stone CJ (1984) Classification andggression tees. Belmont, CA: Wadsworth. Brewerton DA, Caffrey M, Hart FD et al (1973) Ankylosing spondylitis and HL-A27. Lancet i: 904-907. Calin A & Marks SH (1981) The case against seronegative rheumatoid arthritis. American Journal of Medicine 70: 992-994. Cathcart ES & O'Sullivan JB (1970) Rheumatoid arthritis in a New England town. New England Journal of Medicine 282: 421-424. Cobb S, Merchant WR & Warren JE (1955) An epidemiological look at the problem of classification in the field of arthritis. Journal of Chronic Disease 2: 50-54. Cobb S, Thomson D J, Rosenbaum J e t al (1956) On the measurement of prevalence of arthritis and rheumatism from interview data. Journal of Chronic Disease 3: 134-139. Dugowson CE, Nelson JL & Koepsell TD (1990) Evaluation of the 1987 revised criteria for rheumatoid arthritis in a cohort of newly diagnosed female patients. Arthritis and Rheumatism 33: 1042-1046. Edwards JCW (1988) ARA criteria for rheumatoid arthritis (letter). British Journal of Rheumatology 28: 176-177. Edwards JCW & Snaith ML (1988) Different criteria for different purposes. Journal of Rheumatology 15: 537-538. Gordon DA, Stein JL & Broder I (1973) The extra-articular features of rheumatoid arthritis: a systematic analysis of 127 cases. American Journal of Medicine 54: 445-452. de Graaf F (1959) International Congress Series. Number 23, p 7 Proceedings of the ISRA symposium on the social aspects of chronic rheumatic joint affections. New York: Exerpta Medical Foundation. Gregerson PR, Silver J & Winchester RJ (1987) The shared epitope hypothesis. An approach to understanding the molecular genetics of susceptibility to rheumatoid arthritis. Arthritis and Rheumatism 30: 1205-1213.

CLASSIFICATION CRITERIA FOR RA

303

Hakala M, Prll~nen R & Nieminen P (1993) The ARA 1987 revised criteria select patients with clinical rheumatoid arthritis from a population based cohort of subjects with chronic rheumatic disease registered for drug reimbursement. Journal of Rheumatology 20: 1674-1678. Hazes JMW, Hayton R & Silman AJ (1993) A reevaluation of the symptom of morning stiffness. Journal of Rheumatology 20:1138-1142. Hazes JMW, Hayton R, Burt J & Silman AJ (1994) Consistency of morning stiffness: an analysis of diary data. British Journal of Rheumatology 33: 562-565. Healey LA & Sheets PK (1988) The relation of polymyalgia rheumatica to rheumatoid arthritis. Journal of Rheumatology 15: 750-752. Healey LA (1990) Classification, clinical assessment, clinical features, and complications. Current Opinion in Rheumatology 2: 441-445. Jacobson LTH, Knowler WC, Pillemer S e t al (1994) A cross-sectional and longitudinal comparison of the Rome criteria for rheumatoid arthritis (equivalent to the American College of Rheumatology 1958 criteria) and the American College of Rheumatology 1987 criteria for rheumatoid arthritis. Arthritis and Rheumatism 37: 1479-1486. Kellgren JH, Lawrence JS, Aitken-Swan J (1953) Rheumatic complaints in an urban population. Annals of the Rheumatic Diseases 12: 5. Kellgren JH & Lawrence JS (1956) Rheumatoid arthritis in a population sample. Annals of the Rheumatic Diseases 15:1-11. Kellgren JH (1962) Diagnostic criteria for population studies. Bulletin on the Rheumatic Diseases 13: 291-292. Kobayshi S, Sugawara M, Takahashi H et al (1989) Evaluation of the 1987 revised ARA criteria for rheumatoid arthritis in Japan. Ryumaehi 29: 97-104. Larsen A, Dale K & Eek M (1977) Radiographic evaluation of rheumatoid arthritis and related conditions by the use of standard reference films. Acta Radiologica Diagnosis 18" 481--491. Lawrence JS & Wood PHN (1968) Criteria for rheumatoid arthritis in population samples. In Bennett PH & Wood PHN (eds) Population ~tdies of the Reumatic Deases, pp 164-174. Amsterdam: Exerpta Medica Foundation. Levin RV, Ostrov B & Bomalaski JS et al (1989) Clinical assessment of the 1987 American Rheumatism Association (ARA) criteria for rheumatoid arthritis. Arthritis and Rheumatism 32: S15. Lichtenstein MJ & Pincus T (199 l) Rheumatoid arthritis identified in population based cross sectional studies: low prevalence of rheumatoid factor. Journal of Rheumatology 18: 989-993. MacGregor AJ & Silman AJ (1992) A reappraisal of the measurement of disease occurrence in rheumatoid aithritis. Journal of Rheumatology 19:1163-1165. MacGregor A J, Bamber S & Silman AJ (1994) A comparison of the performance of different methods of disease classification for rheumatoid arthritis. Results of an analysis from a nationwide twin study. Journal of Rheumatology 21: 1420-1426. Masi AT & Feigenbaum SL (1983) Seronegative rheumatoid arthritis: fact or fiction? Archives of Internal Medicine 143 9 2167-2172. Masi AT, Feigenbaum SL & Kaplan SB (1983) Articular patterns in the early course of rheumatoid arthritis. American Journal of Medicine 75: 16-76. Man W, Raspe H-H & Mersjann H (t989) Early arthritides: nosography, nosology and diagnostic criteria. Scandinavian Journal of Rhematology 79: 3-12. Miall WE (1955) Rheumatoid arthritis in males. An epidemiological study of a Welsh mining community. Annals of the Rheumatic Diseases 14: 150-156. Mitchell DM & Fries JF (1982) An analysis of the American Rheumatism Association criteria for rheumatoid arthritis. Arthritis and Rheumatism 25: 481-487. Moens HJB, Mart AFJ, van der Lanr MAFJ & van der Korst JK (1992) Comparison of the sensitivity and specificity of the 1958 and 1987 criteria for rheumatoid arthritis. Journal of Rheumatology 19: 198-203. Ollier WER, MacGregor AJ, Jawaheer D et al (1993) HLA DRBl*0401/*0404 genotype and rheumatoid arthritis: increased association in males, young age at onset and disease severity. Arthritis and Rheumatism 36: $86. O'Sullivan JB, Cathcart ES & Bolzan JA (1968) Diagnostic criteria and the incidence of rheumatoid arthritis in Sudbury, Massachusetts. In Bennett PH & Wood PHN (eds) Population Fudies of the l?eumatic Deases, pp 109-113. Amsterdam: Exerpta Medica Foundation. O'Sullivan JB & Cathcart ES (I 972) Prevalence of rheumatoid arthritis: follow-up evaluation of the effect of criteria on rates in Sudbury, Massachusetts. Annals of Internal Medicine 76: 573.

304

A. J, MACGREGOR

Paimela L (1992) The radiographic criterion in the 1987 revised criteria for rheumatoid arthritis. Arthritis and Rheumatism 35: 255-258. Pincus T & Callahan LF (1994) How many types of patient meet classification criteria for rheumatoid arthritis? Journal of Rheumatology 21: 1385-1389. Rigby AS & Wood PHN (1990) A review of assignment criteria for rheumatoid arthritis. Scandinavian Journal of Rheumatology 19: 27-41. Rigby AS & Wood PHN (1991) The lateral metacarpophalangeal/metatarsophalangeal squeeze: an alternative assignment criterion for rheumatoid arthritis. Scandinavian Journal of Rheumatology 20: 115-120. Ropes MW, Bennett GA, Cobb S e t al (1956) Proposed diagnostic criteria for rheumatoid arthritis. Bulletin on the Rheumatic Diseases 7: 121-124. Ropes MW, Bennett GA, Cobb S et al (1958) 1958 revision of diagnostic criteria for RA. Bulletin on the Rheumatic Diseases 9: 175-176. Sagawa A & Nakagawa S (1992) Clinical evaluation and application of new ARA criteria for rheumatoid arthritis. Nippon Rinsho 50: 473-478. Sellick K, Littlejohn G, Wallace C & Over R (1990) Identifying subclasses of patients with rheumatoid arthritis through cluster analysis. Journal of Rheumatology 17: 1613-1619. Silman AJ (1988) The 1987 revised American Rheumatism Association criteria for rheumatoid arthritis. British Journal of Rheumatology 27: 341-343. Silman AJ, MacGregor AJ, Thomson Wet al (1993) Twin concordance rates for rheumatoid arthritis: results from a nationwide study. British Journal of Rheumatology 32: 903-907. Symmons DPM, Chakravarty K & Bankhead CR (1994) The 1987 ACR criteria for rheumatoid arthritis: evaluation of the radiology criterion. British Journal of Rheumatology 33:131. yon Essen R, Kurki P, Isom~iki H et al (1993) Prospect for an additional laboratory criterion for rheumatoid arthritis. Scandinavian Journal of Rheumatology 22: 267-272. Walker DJ, Stephenson AM, Griffiths ID & Roberts DF (1987) Classification of rheumatoid arthritis: the importance of including data from siblings. Genetic Epidemiology 4: 249-253. Weyand CM, Hicok KC, Corm DL & Goronzy JJ (1992) The influence of HLA-DRBI genes on disease severity in rheumatoid arthritis. Annals of Internal Medicine 117: 801-806. Wood PHN (1970) Epidemiology of rheumatic disorders: problems in classification. Proceedings of the Royal Society of Medicine 63: 189-193. Yang JL, Zhang NZ, Dong Y et al (1991) The 1958 and 1987 ARA revised criteria for rheumatoid arthritis in Chinese patients. A comparative study. Chinese Medical Journal 104: 649-652.