The Use and Misuse of Classification and Diagnostic Criteria for Complex Diseases

The vasculitides are a group of complex multisystem disorders that are linked by the presence of necrotizing inflammatory lesions in blood vessels. Arteries and other vessels of various sizes and locations in the body may be involved, resulting in a wide spectrum of manifestations [1]. The causes and pathogenetic mechanisms of vasculitides are poorly understood but are probably diverse. Their severity, course, and outcomes vary. Within the vasculitides, there seem to be several distinct clinical syndromes. However, even these have multiple presentations, and the different syndromes have overlapping findings. These factors have made it difficult to define these diseases and diagnose individual cases. A tissue biopsy showing arteritis confirms vasculitis but by itself does not necessarily define a specific syndrome [2]. Furthermore, biopsy is a sampling process and sometimes misses an involved vascular segment or provides an incomplete picture of the pathology. In some instances, biopsy may not be feasible. Thus, diagnosis of vasculitis usually depends on the presence of a combination of findings.

Because of the lack of a single pathognomonic test in most rheumatic diseases, including vasculitis, the development of criteria for diagnosis or classification has been of continuing interest to rheumatologists. In the United States, the American College of Rheumatology has taken a lead in developing classification criteria for rheumatic diseases, the main purpose of which is to standardize clinical definitions for use in research studies [3]. Standardized definitions are needed to allow comparison of results of studies from different centers. The method comprises identification of a set of frequent clinical findings (criteria) manifested in the disease in question that both point toward the disease (sensitivity) and away from others (specificity) [4]. Controls for validation have usually been limited to patients with diseases that may be confused with the one in question and who have a workup similar to that of patients with the disease. Because sensitivity and specificity in classification sets rarely reach 100%, some patients are always misclassified. When a target disease is uncommon in a population, the sensitivity of a classification set remains the same but the positive predictive value decreases because of the increase in false-positive results [5]. Selection of criteria with a higher specificity can reduce this effect.

Classification criteria work best in the study of groups of patients and work less well in the evaluation of individual patients. Meeting classification criteria is not equivalent to making a diagnosis in an individual patient [4]. Physicians make diagnoses by considering all clinical features in a patient, some of which may be classification criteria. Because of the limitations of classification criteria, investigators may include patients whom they believe have the diagnosis in question regardless of whether they fulfill the criteria, as long as the investigators describe those who do not meet the criteria and provide reasons for their inclusion.

In 1990, a committee of the American College of Rheumatology published criteria for the classification of seven forms of vasculitis [6]. The criteria sets included those for polyarteritis nodosa, the Churg-Strauss syndrome, Wegener granulomatosis, hypersensitivity vasculitis, Henoch-Schonlein purpura, giant-cell arteritis, and Takayasu arteritis. These criteria were derived from a detailed analysis of 1000 consecutive cases of definite vasculitis submitted by multiple centers. A wide spectrum of vasculitis cases was included; the seven types of vasculitis made up 678 of the 1000 cases studied. Criteria sets were developed by selecting findings that were characteristic of one of the forms of vasculitis but were less frequent or absent in the other cases [7]. Findings that were typically present in one type of vasculitis but common to several forms were excluded because they did not separate the groups. Statistical analysis helped select the criteria sets that best separated each type of vasculitis from the others as a group. The ability to distinguish patients with vasculitis from those without vasculitis was not studied and would have required a different control group, which probably would have resulted in the selection of different criteria.

In this issue, Rao and coworkers [8] report on the applicability of the 1990 vasculitis classification criteria sets in diagnosis. Noting that physicians often misuse classification criteria as diagnostic criteria, they studied 198 patients referred to their clinic for possible vasculitis. Vasculitis was diagnosed in 51 of the 198 patients (26%) under 14 diagnostic categories. Rao and coworkers found that 38 of the 51 patients (75%) fulfilled criteria of one of the four classification groups they had selected. The other 13 were misclassified as not having vasculitis. Thirty-one of the 147 patients (21%) in whom vasculitis was initially suspected but in whom the disease was not confirmed later also fulfilled one of the classification sets. Only half of the patients given a diagnosis of one of the four types of vasculitis criteria fit the appropriate set. Analysis of operating characteristics showed sensitivities of 50% to 100% and specificities of 89% to 92% but positive predictive values of less than 30%. Negative predictive values were higher and ranged from 89% to 100%.

Although the number of cases in the study by Rao and coworkers is small, their results tend to confirm what the Vasculitis Committee thought would happen: If the criteria sets are used for diagnosis of individual patients, they function poorly. Separate criteria are needed for classification and for diagnosis. In Rao and coworkers' patients, diagnosis of vasculitis was based on a physician's overall judgment after chart review rather than on predefined or standard criteria. The reader does not know, for example, which clinical or laboratory features led to the specific diagnosis in individual patients or whether physicians in other centers would agree with the diagnoses. Standardized criteria based on analysis of a series of cases can answer such questions. Multicenter studies help iron out regional or ethnic variations in disease and provide the numbers of cases needed to analyze uncommon diseases, such as vasculitis. Another, more general problem is the large number of diagnostic terms used in vasculitis. Rao and coworkers listed 14 different diagnoses in only 51 cases. Better understanding of these diseases would help, but formulating standard definitions focuses attention on the problem and moves us toward workable answers [9].

Standardized criteria order our approaches to the study of disease cause, course, and therapy. They focus our objectives in clinical research and enhance the identification of important clinical differences and subsets of diseases. The process of disease description needs to be dynamic, with additions or changes to criteria and definitions as more is learned about these illnesses. We must keep in mind that classification criteria and diagnoses are not diseases. They are descriptors that change as new knowledge is acquired.

• Copyright ©2004 by the American College of Physicians