Classification criteria

Classification criteria are for case identification in clinical research and not for the diagnosis of individual patients in the clinical encounter. For the clinician, the value of classification criteria lies with the parameterised, abbreviated and refined nature of the diagnostic work-up contained within the criteria. The items within the criteria reflect the critical features of the disease in question and thus form a useful aide memoire to the key points used in making the diagnosis. However, in making the diagnosis, the clinician is neither limited to the informational database contained within the criteria nor constrained to how the individual items of the criteria are aggregated to form the decision rule. All available sources of information are open to the clinician, including prior experience and intuition and the results of investigations not listed within the classification criteria.

The situation is quite different in clinical research. Here, it is necessary to standardise the entry of patients to the study. It is usually important that the patients form a homogeneous group in some sense and advantageous for different studies to be comparable in the types of patients that were studied. Hence, standardised and clearly defined classification criteria are important. The purpose of classification criteria is thus to mimic a gold standard of diagnosis as closely as possible. For most rheumatic diseases, the gold standard involves expert clinical diagnosis. Accuracy of criteria is usually expressed in terms of sensitivity (proportion of patients who fulfil criteria amongst those with the disease) and specificity (proportion of those who do not fulfil criteria amongst those without the disease). Sometimes it is useful to consider the accuracy in terms of error, which is simply the converse of sensitivity and specificity–the error of making a diagnosis when it is not present (false positive = 1 − sensitivity) and the error of not making a diagnosis when it is present (false negative = 1 − specificity). The aim of classification criteria is to minimise such errors. It is often thought that if the classification error (especially the false-positive rate) is sufficiently small, then the criteria could be termed ‘diagnostic criteria’ in order to demonstrate that the criteria could be used for individual patient diagnosis in a clinical encounter. This is a conceptual mistake. Certainly, the clinician will find more accurate criteria more useful in day-to-day practice than less accurate criteria, but the informational database for the diagnostic process is still not limited to these criteria. Unless the criteria perfectly mimics the gold standard (in which case it has become the gold standard), individual diagnosis in a clinical setting should not rely exclusively upon classification criteria.

Pitfalls and limitations to current classification criteria for PsA and SpA

Psoriatic arthritis

Until the pioneering work of Wright and Baker , an inflammatory arthritis occurring in the presence of psoriasis was felt to represent rheumatoid arthritis (RA) occurring coincidentally with psoriasis. The discovery of rheumatoid factor (RF) in the serum provided an important tool that helped categorisation of polyarthritis, but the distinction between RA and psoriatic arthritis (PsA) was achieved primarily on clinical and radiological grounds. Wright described the frequent involvement of distal interphalangeal (DIP) joints with erosion and absorption of the terminal phalanges, co-existing sacroiliitis, involvement of the proximal interphalangeal (PIP) joints of the toes and a characteristic mutilating arthritis with reduction in bone stock particularly in the digits . The American Rheumatism Association adopted PsA as a distinct clinical entity, including it in a classification of rheumatic diseases for the first time in 1964 .

However, the inclusive Moll and Wright criteria may fail to adequately recognise the possibility that psoriasis can exist independently of co-existent arthropathies. Defining PsA as the co-occurrence of an inflammatory arthritis and psoriasis is likely to over-identify such individuals. For instance, the presence of psoriasis alone barely characterises patients with early arthritis in a clearly distinctive clinical way . In this study, there were no differences in the pattern of joint disease between those with and without psoriasis, and the 1-year outcomes in terms of functional disability, erosions and use of disease-modifying drugs were also similar. Moreover, the use of RF to exclude PsA is likely to misclassify patients with coincidental PsA and false-positive RF tests, since RF is not a perfectly specific test for RA.

Despite clinical, radiological and familial evidence supporting PsA as a distinct disease entity, controversy still exists about which patients to include within this disease group. Some authors have even questioned whether PsA is a separate disease suggesting that psoriasis merely modifies the expression of pre-existing RA .

Unfortunately for clinical research in PsA, validated criteria such as those developed for RA did not exist for PsA until recently. Perhaps reflecting the complex nature of PsA, several authors have suggested a variety of classification criteria and the content of these are summarised in Table 1 . None of these have been widely adopted. Apart from the criteria of Fournié, these suggestions represent theoretical notions of the nature of PsA and had not, until recently, been empirically evaluated. Yet the development of new therapies, particularly biological therapies, has highlighted this deficiency and made the need for such criteria and standardised outcome and response criteria, more urgent. The absence of validated criteria may have limited the ability of researchers to conduct clinical studies in PsA; certainly the number of published articles concerning PsA has accelerated since the publication of the CASPAR criteria.

Before discussing the CASPAR criteria, it is worthwhile reviewing some of the background in more detail.

Pitfalls and limitations to current classification criteria for PsA and SpA

Psoriatic arthritis

Until the pioneering work of Wright and Baker , an inflammatory arthritis occurring in the presence of psoriasis was felt to represent rheumatoid arthritis (RA) occurring coincidentally with psoriasis. The discovery of rheumatoid factor (RF) in the serum provided an important tool that helped categorisation of polyarthritis, but the distinction between RA and psoriatic arthritis (PsA) was achieved primarily on clinical and radiological grounds. Wright described the frequent involvement of distal interphalangeal (DIP) joints with erosion and absorption of the terminal phalanges, co-existing sacroiliitis, involvement of the proximal interphalangeal (PIP) joints of the toes and a characteristic mutilating arthritis with reduction in bone stock particularly in the digits . The American Rheumatism Association adopted PsA as a distinct clinical entity, including it in a classification of rheumatic diseases for the first time in 1964 .

However, the inclusive Moll and Wright criteria may fail to adequately recognise the possibility that psoriasis can exist independently of co-existent arthropathies. Defining PsA as the co-occurrence of an inflammatory arthritis and psoriasis is likely to over-identify such individuals. For instance, the presence of psoriasis alone barely characterises patients with early arthritis in a clearly distinctive clinical way . In this study, there were no differences in the pattern of joint disease between those with and without psoriasis, and the 1-year outcomes in terms of functional disability, erosions and use of disease-modifying drugs were also similar. Moreover, the use of RF to exclude PsA is likely to misclassify patients with coincidental PsA and false-positive RF tests, since RF is not a perfectly specific test for RA.

Despite clinical, radiological and familial evidence supporting PsA as a distinct disease entity, controversy still exists about which patients to include within this disease group. Some authors have even questioned whether PsA is a separate disease suggesting that psoriasis merely modifies the expression of pre-existing RA .

Unfortunately for clinical research in PsA, validated criteria such as those developed for RA did not exist for PsA until recently. Perhaps reflecting the complex nature of PsA, several authors have suggested a variety of classification criteria and the content of these are summarised in Table 1 . None of these have been widely adopted. Apart from the criteria of Fournié, these suggestions represent theoretical notions of the nature of PsA and had not, until recently, been empirically evaluated. Yet the development of new therapies, particularly biological therapies, has highlighted this deficiency and made the need for such criteria and standardised outcome and response criteria, more urgent. The absence of validated criteria may have limited the ability of researchers to conduct clinical studies in PsA; certainly the number of published articles concerning PsA has accelerated since the publication of the CASPAR criteria.

Before discussing the CASPAR criteria, it is worthwhile reviewing some of the background in more detail.

The Moll and Wright criteria and specific features of PsA

Other classification criteria

The other proposed classification criteria have not been widely adopted. They are of interest mainly in terms of the selection of features included, which highlight salient features of PsA. Bennett’s criteria combined the clinical features unique to PsA together with characteristic radiological features. In addition, two criteria, one based on synovial fluid analysis and the other on synovial histology, were included. Vasey and Espinoza simplified the Bennett rule recognising that even though there are two principal manifestations of PsA , only two criteria are required – psoriasis and one manifestation of either peripheral joint disease or spinal disease. Although the European Spondyloarthropathy Study Group (ESSG) developed criteria for the classification of the spondyloarthropathy (SpA) group as a whole, particular types of SpA, including PsA, can be identified from the published classification criteria. For the first time, it was possible for PsA to be classified in the absence of psoriasis, if a family history of psoriasis was present. McGonagle et al. have proposed a definition of PsA based on enthesopathy . As the first proper attempt to define and validate criteria from actual patient data, the study by Fournié represented an important step. The items and weighting were selected using discriminant function and logistic regression analysis. The data were derived from a population of patients diagnosed by rheumatologists from a single clinic as having PsA, ankylosing spondylitis (AS) and RA. A score of 11 points is required for the diagnosis of PsA (sensitivity 95%, specificity 98%, LR+ 47.5) and, although the criteria include human leukocyte antigen (HLA) data, it is possible to attain this threshold on clinical data alone. There are a number of limitations in this development study. The problem of case selection is a central issue of classification studies. Prospective, consecutive identification of all patients with physician-diagnosed (or any other gold standard) disease is important to avoid selection bias. The study by Fournié was retrospective in design. Choice of gold standard is also problematic where no pathological benchmark exists. The usual practice is to identify patients with expert-(rheumatologist-) diagnosed disease. This may be problematic in single-centre studies where local diagnostic practice could perhaps differ from other centres. Finally, the statistical method of logistic regression to exclusively define the classification rule has significant problems. In particular, it makes it possible to define patients with disease in ways that were never actually observed. For example, the rule requires a score of 11 to diagnose PsA; the presence of a family history of psoriasis (score 3), RF negativity (score 4) and HLA-B17 (score 6) gets a score beyond the threshold of diagnosis for PsA, yet the patient has neither psoriasis nor arthritis!

There have been two studies that have sought to identify the best performing classification criteria. We found in a retrospective study using data from clinical records and existing radiographs in 343 patients with PsA and 156 with RA that the criteria of Vasey and Espinoza had the best combination of specificity, sensitivity and feasibility, although there was no statistically significant difference between Vasey, Gladman and McGonagle ( Fig. 1 ) . The CASPAR study also found that the criteria of Vasey and Espinoza performed the best of all the currently proposed classification criteria .

Comparison between different criteria for psoriatic arthritis (data from Ref. , with permission).