The Research Domain Criteria (RDoC) initiative prompts us to consider ?dimensional constructs integrating elements of psychology and biology? interrogated with multiple units of analysis toward better understanding mental illness (1). However, such an approach, agnostic to traditional discrete diagnostic classification, does not come with prescribed analytical approaches. Rather, part of the RDoC challenge is to determine appropriate statistical techniques for identifying alternative, meaningful subtypes or clusters defined by bio- psychological profiles. At the same time, there are inherent methodological challenges to clustering with the multiple units of analysis proposed for RDoC approaches. These include 1) highly skewed biological and psychological variables, 2) multi-dimensional, and often high-dimensional datasets, and 3) multiple statistically plausible clustering solutions. Traditional clustering methods most readily accessible and most frequently used in psychiatric research cannot accommodate these issues; their application can produce misleading findings. We propose to utilize cutting-edge finite mixture modeling clustering approaches, robust to these issues, in a large sample of anxiety spectrum disorder patients and matched control participants (n=518) who completed psychophysiological assessment during narrative imagery. Consistent with RDoC constructs within both the Negative and Positive Valence System Domains, participants imagined unpleasant (e.g., threat) and pleasant (e.g., affiliation/reward) as well as neutral narratives while autonomic, facial expressivity, acoustic startle reflex responding and subjective ratings of engagement were recorded. On the basis of the DSM, defensive hyper- reactivity might be expected 1) across anxiety disorders and 2) across response channels. In fact, defensive hypo- and hyper-reactivity were often observed in different channels, within individuals (e.g., exaggerated startle reflex, facial expressivity, and subjective arousal coupled with blunted heart rate and skin conductance responding). The hypothesis of this study is that utilizing cutting-edge clustering approaches will reveal novel transdiagnostic subtypes based on multimodal, dimensional response coordination (or lack thereof) during emotional engagement. These subtypes are expected to cut across traditional diagnostic boundaries, while revealing that the extent of coordination among certain channels is most predictive of emotional health and functional status. The long-term goal is to identify potential novel targets for tailoring interventions, particularly emotion-focused psychotherapy, to coordination among specific response channels. The shorter-term goal is to identify analytical approaches that leverage the rich information in multimodal datasets?applicable to a wide range of RDoC approaches.
In the context of mental health care, there is a growing appreciation that individuals with the same diagnosis may have very different emotional experiences captured both in physiology and self-reported engagement. While this appreciation has burgeoned, our statistical approaches for identifying subtypes based on these patterns have lagged. We propose to utilize cutting-edge clustering approaches that would aid in the interpretation of not only this dataset, but other RDoC style data sets with multiple units of analysis (e.g., genes, circuits, molecules, behavior).