Capturing trial-by-trial variability in behaviour: people with Parkinson's disease exhibit a greater rate of short-term fluctuations in response times

Abstract

Average response time is frequently used to reflect executive function. Less often studied is intra-individual variability in response times (IIVRT) which reflects within-person consistency. Higher IIVRT in Parkinson's disease (PD) has been associated with poor executive function but almost exclusively studied using standard deviation (SD). SD provides a standardised measure of inconsistency in RTs but is necessarily calculated as an average measure, precluding any trial-level investigation. Such linear measures cannot capture rapid and spontaneous changes in biological systems such as dopaminergic bursting activity. Therefore, nonlinear measures provide important complementary insights into dopamine-related neurocognition. The nonlinear method of graph theory is one viable approach to capture the complex biological changes in PD and their effect on behaviour. Our primary aim was to increase the understanding of RT fluctuations in PD beyond the use of SD by investigating nonlinear IIVRT measures using graph theory, constituting the first use of this approach on RT data. As hypothesized, PD was associated with a greater rate of trial-by-trial IIVRT compared to healthy older adults. The difference between groups could not be explained simply by worse overall RT performance, as average RT was comparable between groups. Instead, the IIVRT findings reflected impaired consistency in performance for people with PD and specifically a greater rate of short-term fluctuations in behaviour. These novel results indicate that a similarity graph algorithm may be a sensitive tool to capture the rapidly varying changes in behaviour that result from dysfunctional dopamine bursting activity in PD.

Fig. 1. Experimental setup for behavioural task [51].

The anticipatory response inhibition task display (top) and participant response (bottom) for a Go (left), Stop Bimanual (middle), and Stop Unimanual trial (right). The participant has successfully lifted from both switches (GG: Go, Go), kept both switches depressed (SS: Stop, Stop) and lifted from only the left-hand switch (GS: Go, Stop), respectively. Other type of Stop Unimanual trial (SG: Stop, Go) not shown.

Fig. 2. Illustrations of the three intra-individual variability in response time (IIVRT) measures showing (top left) the assumed gaussian RT distribution for standard deviation, (top right) which ex-gaussian RT distribution component is captured by tau, and (bottom) nodes and edges in the graph theory analysis.

An edge is created (black line) between two nodes (black circles 1–5) if they are sufficiently similar to each other within a set threshold and within the same time window (e.g. five nodes in the figure). Each node corresponds to a Go RT.

Fig. 3. Group effects on intraindividual variability shown via average measures of standard deviation (left) and tau (right).

Parkinson’s disease (PD) participants showed greater variability than healthy controls (HC) on average across the entire experimental response time distribution and across slower response times.

Fig. 4. Group effects on intraindividual variability shown via trial-by-trial measures from graph theory analysis.

Parkinson’s disease (PD) participants showed reduced response time similarity (smaller number of edges) compared to healthy controls (HC) across time windows of ± 2 trials (left) and ± 10 trials (right).