Net Benefit-Risk of Diagnostic Tests Under Tree Ordering

The evaluation or comparison of diagnostics tests and biomarkers based on benefit-risk involves both the accuracy of the tests and the clinical consequences of the diagnostic errors. In practice, many diseases can be classified into multiple classes. Besides monotone ordering, another important category of scenarios for multi-classes is tree ordering, in which the diseases consist of several unordered subclasses. In diseases with multi-subclasses without an order, the benefits and risks of the clinical consequences could differ from class to class.  Investigations on the benefit-risk of diagnostic tests with more than two classes are lacking in the literature. This paper extends the diagnostic yield table in binary disease cases to clinical conditions with multi-subclasses. Moreover, a decision process based on net benefit for evaluating diagnostic tests is developed. The proposed decision process provides additional interpretation for rule-in or rule-out clinical conditions and their adverse consequences from unnecessary workups in diseases with multi-subclasses. Simulations and real data examples are presented to illustrate the proposed measures.