Large-N Rat Data Enables Phenotyping of Risky Decision-Making: A Retrospective Analysis of Brain Injury on the Rodent Gambling Task

Abstract

Decision-making is substantially altered after brain injuries. Patients and rats with brain injury are more likely to make suboptimal, and sometimes risky choices. Such changes in decision-making may arise from alterations in how sensitive individuals are to outcomes. To assess this, we compiled and harmonized a large dataset from four studies of TBI, each of which evaluated behavior on the Rodent Gambling Task (RGT). We then determined whether the following were altered: (1) sensitivity to overall contingencies, (2) sensitivity to immediate outcomes, or (3) general choice phenotypes. Overall sensitivity was evaluated using the matching law, immediate sensitivity by looking at the probability of switching choices given a win or loss, and choice phenotypes by k-means clustering. We found significant reductions in sensitivity to the overall outcomes and a bias toward riskier alternatives in TBI rats. However, the substantial individual variability led to poor overall fits in matching analyses. We also found that TBI caused a significant reduction in the tendency to repeatedly choose a given option, but no difference in win- or loss-specific sensitivity. Finally, clustering revealed 5 distinct decision-making phenotypes and TBI reduced membership in the “optimal” type. The current findings support a hypothesis that TBI reduces sensitivity to contingencies. However, in the case of tasks such as the RGT, this is not a simple shift to indiscriminate or less discriminate responding. Rather, TBI rats are more likely to develop suboptimal preferences and frequently switch choices. Treatments will have to consider how this behavior might be corrected.

Exploring decision-making strategies in the Iowa gambling task and rat gambling task

“…One study found that animals with high and low motor impulsivity differed in their preference of the different choices; animals in the low motor impulsivity group chose P2 more, animals in the high motor impulsivity group chose P1 and P4 more, while choice of P3 did not differ ( Barrus et al, 2015 ). This relationship between choice preferences corroborates our finding that options P1 and P2, that in standard scoring approaches are combined, does not seem to represent the same thing for the animals ( Vonder Haar et al, 2022 ). A recent study used cluster analysis to find choice phenotypes and found five distinct clusters: one with strong P2 preference, one with moderate P2 preference, one P3-preferring, one P4-preferring, and lastly a group with P1 and P3 preference ( Vonder Haar et al, 2022 ).…”

“…This relationship between choice preferences corroborates our finding that options P1 and P2, that in standard scoring approaches are combined, does not seem to represent the same thing for the animals (Vonder Haar et al, 2022). A recent study used cluster analysis to find choice phenotypes and found five distinct clusters: one with strong P2 preference, one with moderate P2 preference, one P3-preferring, one P4-preferring, and lastly a group with P1 and P3 preference (Vonder Haar et al, 2022). Once again this indicates that the relationship between the choices in the rGT is not as simple as P1 and P2 being perceived as advantageous and P3 and P4 as disadvantageous.…”

Statistical power and false positive rates for interdependent outcomes are strongly influenced by test type: Implications for behavioral neuroscience

“…A publicly-available dataset of behavior on this task [21] collected by our laboratory, which we refer to as the "control set" was used to inform simulation parameters. For the current study, only stable post-injury data (i.e., collected outside the initial task-learning and acute injury phases) were considered, and data involving manipulations (e.g., treatments) other than brain injury were excluded.…”

“…A second dataset assessing this task, which we refer to as the "validation set" was obtained from Dr. Catharine Winstanley and is fully described elsewhere [22]. In brief, this dataset consisted of 14 studies of rats learning the RGT.…”

“…To evaluate the accuracy of these statistical methods for interdependent data, we applied Monte Carlo techniques to simulate concurrent 4-choice data. Our simulation modeled real behavioral data from the Rodent Gambling Task (RGT) [20] with parameters derived from a publiclyavailable dataset [21]. 1,000 datasets were simulated for each of four sample sizes and four effect sizes (N = 16,000 datasets).…”