During the 1980s, the American College of Rheumatology (ACR) (formerly, the American Rheumatism Association) published several papers on classification criteria for rheumatic diseases (1-5). This issue of Arthritis and Rheumatism includes classification criteria for 7 forms of vasculitis. The ACR has also developed separate classification criteria for osteoarthritis of the hand ( 6) and for dsteoarthritis of the hip (manuscript submitted). Subcommittees of the ACR are developing "prognostic stratification" criteria as well.
Over the years, researchers have asked us about alternative statistical methods of deriving classification rules, about what concepts are considered by statisticians to be important in evaluating such rules, and about how to compare rules that have been derived using different methods. The perceived need and apparent timeliness of presenting and discussing statistical issues and different approaches to deriving classification rules motivated the writing of this article.
We illustrate several approaches to classification by applying them to the same ACR vasculitis data set, with the task of classifying patients as having or not having giant cell (temporal) arteritis (GCA).