Evaluating the performance of a genetic test using information from previous studies in place of a gold standard.

ABSTRACT

OBJECTIVES: To evaluate the performance of a diagnostic test, a researcher usually must classify study subjects with respect to (1) whether the test result was positive or negative and (2) whether the test result should have been positive or negative. To classify the subjects in the second manner, the researcher needs to have access to a gold standard (ie, a test that classifies the subjects with 100% accuracy). The authors show here how to evaluate the performance of a diagnostic test that allows researchers to determine whether a disease that is occurring within a family is attributable to one of two newly discovered genes without the use of a gold standard. METHODS: By taking advantage of well-known genetic phenomena and their statistical implications, the behavior of the diagnostic test is mathematically modeled, and its performance with respect to various criteria is shown to be functions of genetic parameters. RESULTS: The performance of the test over a wide range of values of the genetic parameters was evaluated, and cutoff points that would allow the test to perform very well or well with respect to all criteria were found for almost all of the situations examined. CONCLUSIONS: This test can be used effectively under a wide range of conditions. In addition, because the genetic parameters have been estimated in previous studies, the effectiveness of the test for the specific conditions the researcher may need to run the study under can be evaluated before the study is performed.