40. Exploration not Perseveration: Computational Modeling of Probabilistic Reversal Learning Impairments in Psychosis
Year: 2017
Type: article
Abstract: Background: Computational psychiatry applying algorithms for basic behavioral phenomena to understand how these parameters changed with symptoms across psychotic diagnoses. Rule generation and selection is a critical function for daily life and impairments in this construct have been associated with the negative symptoms of psychosis (Gold, et al., 2008). The current study used trial-by-trial decisions by people with schizophrenia (SZ), schizoaffective disorder (SA) and bipolar disorder (BP) on a probabilistic reversal learning (pRL) task to model five parameters: beliefs about the correct category, exploratory choices, motoric repetitions, attentional lapses and random responding. Methods: 225 participants in the multisite CNTRaCS consortium (Henderson, et al., 2012) completed the pRL: SZ (n = 65), SA (n = 54), BP (n = 50) and demographically similar controls (n = 56). In the pRL, participants select 1 of 2 abstract images using a button press, after which they are told if their choice was correct. The task was probabilistic because one of the items was positively reinforced 80% of the time (and 20% of the time they were falsely told it was wrong), and the alternative item had the opposite contingency. After meeting the initial acquisition criterion of selecting the positively reinforced item on 10 trials, the reinforced item changed in favor of the alternative item. This simple design allowed us to examine differences on these five learning-based parameters to examine reward-based learning across psychotic diagnoses. Results: All patient groups were more impaired for the initial acquisition than controls, however there were no group differences to obtain subsequent criteria, or in perseverations or false-feedback switches. Patients’ difficulty on the task derived from an excess of win-switch trials, or spontaneous switches (Kruskal-Wallis χ2(3) = 12.71, P < .001, SZ, SA, BP > C), and was correlated with total symptoms (ρ = .20, P = .017), but not symptom factor scores. The computational model including both belief about the correct category and exploratory choices fit the data most parsimoniously, excluding the other parameters. Of these two, patient deficits in all groups were associated with more errors due to exploratory choices (χ2(3) = 13.509, P = 0.004*; SZ, SA, BP > C). The exploratory choice parameter correlated with negative and positive symptom factors, and with the total number of symptoms. Conclusion: Consistent with previous research, we found patients with schizophrenia to be impaired on pRL. This impairment was not specific to schizophrenia and predominated in all patients in the initial acquisition stage. A computational model of rule selection and exploration predicted participant performance and showed patients’ rule selection capacity was unimpaired whereas they appeared to explore reward contingencies more than controls. This capacity was relevant to an array of symptoms across psychotic diagnoses.
Source: Schizophrenia Bulletin
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W2600244167