Rapid, reliable mobile assessment of affect-related motor processing

Abstract

Mobile technologies can be used for behavioral assessments to associate changes in behavior with environmental context and its influence on mental health and disease. Research on real-time motor control with a joystick, analyzed using a computational proportion-derivative (PD) modeling approach, has shown that model parameters can be estimated with high reliability and are related both to self-reported fear and to brain structures important for affective regulation, such as the anterior cingulate cortex. Here we introduce a mobile version of this paradigm, the rapid assessment of motor processing (RAMP) paradigm, and show that it provides robust, reliable, and accessible behavioral measurements relevant to mental health. A smartphone version of a previous joystick sensorimotor task was developed in which participants control a virtual car to a stop sign and stop. A sample of 89 adults performed the task, with 66 completing a second retest session. A PD modeling approach was applied to compute Kp (drive) and Kd (damping) parameters. Both Kp and Kd exhibited high test-retest reliabilities (ICC .81 and .78, respectively). Replicating a previous finding from a different sample with the joystick version of the task, both Kp and Kd were negatively associated with self-reported fear. The RAMP paradigm, a mobile sensorimotor assessment, can be used to assess drive and damping during motor control, which is robustly associated with subjective affect. This paradigm could be useful for examining dynamic contextual modulation of affect-related processing, which could improve assessment of the effects of interventions for psychiatric disorders in a real-world context.

Individual Differences in Computational Psychiatry: A Review of Current Challenges

“…Note that another way to obtain an upper bound on test-retest reliability is to compute split-half reliability, which, like parameter recovery, is not affected by longitudinal changes in behavior. Less than half of the studies reviewed here estimated split-half reliability (Price et al, 2019;Shahar et al, 2019;Brown et al, 2020Brown et al, , 2021Bruder et al, 2021;Xu and Stocco, 2021;Hitchcock et al, 2022a;Loosen et al, 2022;Howlett et al, 2022). While most of these studies found split-half reliability to be higher than test-retest reliability, those that used tasks involving trial-by-trial learning found split-half reliability to be similar or even lower for some parameters (Brown et al, 2020(Brown et al, , 2021Loosen et al, 2022).…”

“…For example, monitoring disease course, treatment response, or characterizing dynamic computational phenotypes requires longitudinal testing, likely involving batteries of tasks. To make frequent longitudinal testing feasible, it might be necessary to move exclusively to remote testing strategies such as smartphone-based tasks (Gillan and Rutledge, 2021;Zech et al, 2022;Pronk et al, 2022;Howlett et al, 2022). Another underappreciated challenge is engagement.…”

“…dynamic game difficulty balancing in traditional video games), and could reduce ceiling and floor effects in performance, increasing the range of measurable individual differences. Another novel approach for improving efficiency involves dense sampling of behavior (Howlett et al, 2022(Howlett et al, , 2020. Instead of one or two data points per trial (e.g., response choice or reaction time), tasks could be designed to collect continuous real-time data, resulting in dozens of data points per trial.…”

“…Instead of one or two data points per trial (e.g., response choice or reaction time), tasks could be designed to collect continuous real-time data, resulting in dozens of data points per trial. Using this approach, Howlett et al (2022) showed that good test-retest reliability of parameter estimates (ICC = 0.78 − 0.81) can be achieved under 6 minutes of testing.…”