"Biological markers" and psychiatric diagnosis: risk-benefit balancing using ROC analysis

Although the clinical interview retains a central role in psychiatric diagnosis, recent research has suggested that "biological markers" may ultimately increase the precision of clinicians' nosologic and therapeutic decisions. Evaluating and operationalizing diagnostic tests require mathematical techniques that reflect the tests' essential features and limitations, and that guide clinicians in particular clinical situations. In this article we describe a technique that combines signal detection theory and utility-based decision theory, and apply the technique to published data in which sleep architecture was used as a biological marker for depression. We show how outcome utilities influence the optimum REM latency cut-off and show how this relationship is influenced by the prevalence of depression in the population being tested. We also make specific calculations of the practical limits that must be imposed on uncertainties in utilities to operationalize a diagnostic test for a specific clinical situation.