Integrating frequency and magnitude information in decision-making in schizophrenia: An account of patient performance on the Iowa Gambling Task

Abstract

Background: The Iowa Gambling Task (IGT; Bechara et al., 1994) has frequently been used to assess risky decision making in clinical populations, including patients with schizophrenia (SZ). Poor performance on the IGT is often attributed to reduced sensitivity to punishment, which contrasts with recent findings from reinforcement learning studies in schizophrenia.

Methods: In order to investigate possible sources of IGT performance deficits in SZ patients, we combined data from the IGT from 59 SZ patients and 43 demographically-matched controls with data from the Balloon Analog Risk Task (BART) in the same participants. Our analyses sought to specifically uncover the role of punishment sensitivity and delineate the capacity to integrate frequency and magnitude information in decision-making under risk.

Results: Although SZ patients, on average, made more choices from disadvantageous decks than controls did on the IGT, they avoided decks with frequent punishments at a rate similar to controls. Patients also exhibited excessive loss-avoidance behavior on the BART.

Conclusions: We argue that, rather than stemming from reduced sensitivity to negative consequences, performance deficits on the IGT in SZ patients are more likely the result of a reinforcement learning deficit, specifically involving the integration of frequencies and magnitudes of rewards and punishments in the trial-by-trial estimation of expected value.

Keywords: Decision-making; Iowa Gambling Task (IGT); Reinforcement; Reward; Risk; Schizophrenia.

Figure 1

Weighted averages of individual deck choices on the Iowa Gambling Task (IGT) by patients and controls across 8 previous studies reporting individual deck choices. Error bars reflect one standard error. Relative to controls, patients show an increase in the number of selections from both disadvantageous decks (for Deck A, mean Cohen’s d = 0.421; 95% CI of Cohen’s d: [0.23, 0.61]; for Deck B (mean d = 0.670; 95% CI of d: [0.48, 0.86]). Relative to controls, patients show a decrease in the number of selections from both advantageous decks (for Deck C, d = −0.202; 95% CI of d: [−0.39, −0.01]; for Deck D, d = −0.703; 95% CI of d: [−0.90, −0.51]).

Figure 2

Performance measures from the IGT: current study participants. (a) Percentages of choices from advantageous decks (C and D) across blocks of trials by healthy controls and schizophrenia patients. As participants performed 100 trials of the task, each block represents 20 trials. (b) Percentages of choices from Decks with infrequent punishments (B and D). (c) Percentages of choices from each individual deck of the IGT. Error bars reflect one standard error. (*p<0.05)

Figure 2

Performance measures from the IGT: current study participants. (a) Percentages of choices from advantageous decks (C and D) across blocks of trials by healthy controls and schizophrenia patients. As participants performed 100 trials of the task, each block represents 20 trials. (b) Percentages of choices from Decks with infrequent punishments (B and D). (c) Percentages of choices from each individual deck of the IGT. Error bars reflect one standard error. (*p<0.05)

Figure 3

Performance measures from the Balloon Analogue Risk Task (BART): current study participants. (a) Average numbers of balloon pumps by patients with schizophrenia and healthy controls, across the 3 experimental blocks of trials. (b) Number of explosions experienced and (c) total money earned on the BART by patients with schizophrenia and healthy controls. Error bars reflect one standard error. (**p<0.01)

Figure 3

Performance measures from the Balloon Analogue Risk Task (BART): current study participants. (a) Average numbers of balloon pumps by patients with schizophrenia and healthy controls, across the 3 experimental blocks of trials. (b) Number of explosions experienced and (c) total money earned on the BART by patients with schizophrenia and healthy controls. Error bars reflect one standard error. (**p<0.01)