Toward dynamic phenotypes and the scalable measurement of human behavior

Abstract

Precision psychiatry demands the rapid, efficient, and temporally dense collection of large scale and multi-omic data across diverse samples, for better diagnosis and treatment of dynamic clinical phenomena. To achieve this, we need approaches for measuring behavior that are readily scalable, both across participants and over time. Efforts to quantify behavior at scale are impeded by the fact that our methods for measuring human behavior are typically developed and validated for single time-point assessment, in highly controlled settings, and with relatively homogeneous samples. As a result, when taken to scale, these measures often suffer from poor reliability, generalizability, and participant engagement. In this review, we attempt to bridge the gap between gold standard behavioral measurements in the lab or clinic and the large-scale, high frequency assessments needed for precision psychiatry. To do this, we introduce and integrate two frameworks for the translation and validation of behavioral measurements. First, borrowing principles from computer science, we lay out an approach for iterative task development that can optimize behavioral measures based on psychometric, accessibility, and engagement criteria. Second, we advocate for a participatory research framework (e.g., citizen science) that can accelerate task development as well as make large-scale behavioral research more equitable and feasible. Finally, we suggest opportunities enabled by scalable behavioral research to move beyond single time-point assessment and toward dynamic models of behavior that more closely match clinical phenomena.

The promise of machine learning in predicting treatment outcomes in psychiatry

“…Cognitive testing is a straightforward method to indirectly as sess brain functioning that has been historically linked to treat ment outcomes. Although such testing can be timeconsuming and costly when performed by a trained neuropsychologist, more recent computerized methods can facilitate efficient digi tal assessments that lend themselves especially well to machine learning, including from passively collecting smartphone mea surements as described above 80,114,174 .…”

Digital tools for the assessment of pharmacological treatment for depressive disorder: State of the art

“…In short, the more variables that can be combined the better the profiles and options to understand depression treatment in its complexity. The options for pharmacological treatment and monitoring can be improved ( Buch and Liston, 2021 ;Ressler and Williams, 2021 ), for clinical trials and research, through machine learning techniques and artificial intelligence, paving the road towards precision medicine ( Germine et al, 2021 ).…”

Incorporating ecological momentary assessment into multimethod investigations of cognitive aging: Promise and practical considerations.

“…Thus, researchers need to adopt a pragmatic approach, weighing the costs and benefits of the number of items and constructs assessed in order to strike a balance between sound measurement practices and participant burden. In the future, computerized adaptive testing approaches may provide a more precise way to assess constructs of interest over short time intervals (Germine et al, 2021), helping to improve the measurement phenomena of interest to cognitive aging researchers.…”