Reliability, Decision Rules, and the Value of Repeated Tests

Publisher Website
Google Scholar
Abstract
While unreliable medical tests often must be repeated, the value of such repetitions may be elusive. This paper attempts to clarify some influences upon the diagnostic usefulness of repeated tests. Reliability is a complex factor whose effect varies with patient heterogeneity and with the decision rule used to resolve conflicts in test results. Two informal rules and one formal statistical method are examined in this paper. In all of these cases, differences in reliability which depend on the disease status of the patient appear to be critical. These differences may alter the value of tests and provide important information that can help us to distinguish between persons with and without a disease. Illustrations of this phenomenon are provided from published data on the hemoccult test in silent colon cancer and the platelet monoamine oxidase test in chronic schizophrenia. These findings may provide heuristics to refine clinical intuition about the value of repeated tests. They also clarify the need to estimate test reliabilities for both disease and nondisease groups.