Testing the 'preliminary criteria for classification of SLE'.

Systemic lupus erythematosus (SLE) is a disease with varied manifestations involving multiple organ systems, accompanied by multiple laboratory abnormalities, and characterized by exacerbations and remissions. Noting that 'uniform classification of defined groups of patients is necessary in order to assemble and compare data from different sources concerning natural history, evaluation of therapy, and epidemologic description', the Diagnostic and Therapeutic Criteria Committee of the American Rheumatism Association recently approached the formidable task of assembling criteria for the classification of SLE (Cohen, Reynolds, Franklin, Kulka, Ropes, Shulman, and Wallace, 1971). To increase the utility of the criteria in population studies, the use of 'exclusions' and 'duration of manifestations' was avoided. A distinction between 'major' and 'minor' criteria was also not used. Thus, the form of these criteria differs from that of the Jones Criteria for Rheumatic Fever and the ARA Criteria for Rheumatoid Arthritis (Arthritis Foundation, 1964). The committee collected a data base of 245 patients with unequivocal SLE, 234 patients with rheumatoid arthritis, and 217 patients with miscellaneous nonrheumatic diseases. Data was contributed by 52 investigators throughout the United States, who were requested to submit patients with 'unequivocal' SLE, and with 'classic' or 'definite' rheumatoid arthritis. Patients whose clinical course was atypical or in whom diagnostic uncertainty remained were excluded from these two groups. Data was sought on 74 variables of disease considered of potential value for classification. Adequate data was obtained on 57 potential criteria, and the list was reduced to 21 by exclusion of variables which were least useful in distinguishing patients with 'SLE' from 'RA' and 'other' diseases. The final 21 variables were reduced to fourteen 'criteria' by grouping several variables within a single criterion (Table 1, overleaf). The criteria were then tested against the data base, by simple reclassification. A level offour or more positive criteria best distinguished the populations and was recommended by the committee. Tested against the same populations, this level included approximately 90 per cent. ofthe patients with 'SLE' and excluded all but 1 per cent. of the patients with 'rheumatoid arthritis' and all but 2 per cent. of the patients with 'other' diseases. This study evaluates the proposed criteria in the context of five questions: (1) How representative was the ARA test patient population? (2) How adequate are the criteria when tested against an 'outside' patient population? (3) How much does the sensitivity and specificity of the criteria depend upon the point in the course of the patient's illness at which they are applied? (4) How much 'interdependence' is present between various criteria and how much effect does this have on the performance of the criteria? (5) Which diseases might exhibit four or more positive criteria and what percentage of the time may such 'false positives' be expected? Two concepts useful for the testing of criteria are proposed: (1) Testing against a computer-based clinical databank, (2) Testing against 'simulated' patient populations created by computer from known characteristics of the population.