Abstract
One of the major goals of basic studies in psychiatry is to find etiological mechanisms or biomarkers of mental disorders. A standard research strategy to pursue this goal is to compare observations of potential factors from patients with those from healthy controls. Classifications of individuals into patient and control groups are generally based on a diagnostic system, such as the Diagnostic and Statistical Manual of Mental Disorders (DSM) or the International Classification of Diseases (ICD). Several flaws in these conventional diagnostic-based approaches have been recognized. The flaws are primarily due to the complexity in the relation between the pathogenetic factors (causes) and disorders: The current diagnostic categories may not reflect the underlying etiological mechanisms. To overcome this difficulty, the National Institute of Mental Health initiated a novel research strategy called Research Domain Criteria (RDoC), which encourages studies to focus on the neurobiological mechanisms and core aspects of behavior rather than to rely on traditional diagnostic categories. However, how RDoC can improve research in psychiatry remains a matter of debate. In this article, we propose a theoretical framework for evaluating psychiatric research strategies, including the conventional diagnostic category-based approaches and the RDoC approach. The proposed framework is based on the statistical modeling of the processes of how the disorder arises from pathogenetic factors. This framework provides the statistical power to quantify how likely relevant pathogenetic factors are to be detected under various research strategies. On the basis of the proposed framework, we can discuss which approach performs better in different types of situations. We present several theoretical and numerical results that highlight the advantages and disadvantages of the strategies. We also demonstrate how a computational model is incorporated into the proposed framework as a generative model of behavioral observations. This demonstration highlights how the computational models contribute to designing psychiatric studies.