Cost Evaluation During Decision-Making in Patients at Early Stages of Psychosis

Abstract

Jumping to conclusions during probabilistic reasoning is a cognitive bias reliably observed in psychosis and linked to delusion formation. Although the reasons for this cognitive bias are unknown, one suggestion is that psychosis patients may view sampling information as more costly. However, previous computational modeling has provided evidence that patients with chronic schizophrenia jump to conclusions because of noisy decision-making. We developed a novel version of the classical beads task, systematically manipulating the cost of information gathering in four blocks. For 31 individuals with early symptoms of psychosis and 31 healthy volunteers, we examined the numbers of "draws to decision" when information sampling had no, a fixed, or an escalating cost. Computational modeling involved estimating a cost of information sampling parameter and a cognitive noise parameter. Overall, patients sampled less information than controls. However, group differences in numbers of draws became less prominent at higher cost trials, where less information was sampled. The attenuation of group difference was not due to floor effects, as in the most costly block, participants sampled more information than an ideal Bayesian agent. Computational modeling showed that, in the condition with no objective cost to information sampling, patients attributed higher costs to information sampling than controls did, Mann-Whitney U = 289, p = 0.007, with marginal evidence of differences in noise parameter estimates, t(60) = 1.86, p = 0.07. In patients, individual differences in severity of psychotic symptoms were statistically significantly associated with higher cost of information sampling, ρ = 0.6, p = 0.001, but not with more cognitive noise, ρ = 0.27, p = 0.14; in controls, cognitive noise predicted aspects of schizotypy (preoccupation and distress associated with delusion-like ideation on the Peters Delusion Inventory). Using a psychological manipulation and computational modeling, we provide evidence that early-psychosis patients jump to conclusions because of attributing higher costs to sampling information, not because of being primarily noisy decision makers.

Keywords: At Risk Mental State; beads; cognitive bias; computational psychiatry; fish; jumping to conclusions; schizophrenia.

Figure 1.

Experimental design of a single trial. In 50% of the trials, fish were coming from the mainly black lake, and in 50%, they were coming from the mainly gold lake. The order was pseudo-randomized so that the same sequences were used for all participants. Feedback, depending on the block, was either of the words “Correct” and “Incorrect” in Block 1 or the number of points won (or lost) during the trial in all subsequent blocks.

Figure 2.

State space schematic. A) Markovian transitions in this task. Top: belief (probabilistic) component of states; middle: observable part of the state (data/feedback). Down arrows: actions (sample, declare). Bottom: true state. For example, let the cost of sampling be very high. Then b₀ may be “equiprobable lakes,” Action 1 “sample,” s₁ “B,” b₁ “60% B,” Action 2 “declare B,” and s₁ “Wrong.” B) In this example, sampling cost is very low. A person has drawn 15 fishes, 7 of them g, hence the position of 15 on the x-axis and +1 on the y-axis as there is a +1 excess of black fish so far. The visible states corresponding to all possible future draws are shown. Looking ahead (example: gray arrow), the agent finds the “sampling” action more valuable in that the current preference for the B lake is likely to be strengthened at very low cost.

Figure 3.

Performance: Draws to decision and accuracy according to block and group. A) Mean number of draws to decision in the four blocks of the task. On mixed-model ANOVA, there were significant main effects of block, F(3) = 94.49, p < 0.001, and of group, F(1) = 5.99, p = 0.017, and an interaction between group and block, F(3) = 4.32, p = 0.006, with group differences in Block 1, p = 0.007, and Block 2, p = 0.03. B) Probability of being correct (accuracy) at the time of making the decision in four task blocks. Here there was an effect for block, F(2) = 93.73, p < 0.001, a marginally significant interaction between Block ⋅ Group, F(2) = 2.52, p = 0.086 and a significant groups effect, F(1) = 4.14, p = 0.047.

Figure 4.

Mean number of points won or lost in Blocks 2–3 across all 10 trials. Patients won significantly fewer points in Block 2, F(2) = 4.65, p = 0.035, but did not differ from controls in Block 3 and 4, both p > 0.3.