Delusions: Sticking With Conclusions

Katharina Schmack; Philipp Sterzer

Disclosures

Brain. 2019;142(5):1497-1500. 

'This morning, there were three black cars parked in front of my house. This must mean that they are spying on me. I am freaking out because I am sure now that I have become a target for this vast, international, right-wing conspiracy.' At the core of this quintessential delusion lies an unusual conclusion ('target of conspiracy') from a usual sight ('three black cars'). For decades, researchers have assumed that delusions are the result of people jumping to irrational conclusions on the basis of little evidence. However, this view fails to explain why delusions are usually very 'sticky' beliefs that are resistant to new evidence. In this issue of Brain, Baker and co-workers use a fresh computational approach to help address this issue (Baker et al., 2019).

More than 30 years ago, Hemsley and Garety proposed that altered inferences are central to delusions (Hemsley and Garety, 1986). This proposition sparked the idea that people with delusions exhibit a tendency to jump to conclusions, and that this contributes to the premature formation of false (delusional) beliefs. Empirical support for this notion came from the so-called 'beads task' (Phillips and Edwards, 1966). In this task, participants are first presented with two jars of beads, one with mostly blue beads and a few green beads and another jar with mostly green and a few blue beads. Participants then draw beads from a hidden jar until they feel confident enough to guess which of the two initially presented jars they are drawing from. People with delusions were found to make their decisions after significantly fewer draws than controls, thus suggesting a jumping-to-conclusion bias (Huq et al., 1988). While replicated many times, the finding of reduced draws-to-decision shows no consistent association with delusion severity, and has often been criticized for being confounded by cognitive and motivational factors. For instance, participants who make fast decisions may not have a problem with inference but may just be impatient, indifferent, or have misunderstood the task.

Baker and colleagues addressed these problems head-on and came to surprising conclusions. The authors devised a new version of the beads task (Figure 1), in which rapid decisions were penalized with monetary losses to counteract confounding factors such as impatience or lack of motivation. After each draw, participants had to decide whether to draw more beads at a small monetary cost; or to make a guess, which was penalized with a large monetary loss if wrong. In addition, before making their decision, participants had to estimate the probability that they were currently drawing from one or the other jar. In agreement with previous work, a diagnosis of schizophrenia was associated with fewer draws-to-decision. However, analysis revealed that this effect could be attributed almost exclusively to the lower socio-economic status of patients with schizophrenia compared to healthy control participants. In contrast, when examining the role of delusion severity, the authors found an association with increased, rather than decreased, draws-to-decision (Figure 1, right). This increase was specific to delusion severity as compared to other psychotic symptoms, working memory capacity, and other clinical and socio-demographic characteristics. Crucially, the authors then used a computational ideal observer model to analyse the reported probability estimates. These analyses showed that individuals with higher delusion severity gave more weight to information presented early in a given trial, relative to new information revealed by subsequent draws. This 'sticky' belief-updating accounted for the observation that delusional patients drew more beads before venturing a guess.

Figure 1.

Modified 'beads task' and key results. In the task used by Baker et al., participants are first presented with two jars of beads, one with mostly blue beads and a few green beads and the other with mostly green and a few blue beads (left). After being endowed with $30 at the start of the session, participants start drawing beads from a hidden jar. Their task is to find out which jar the beads are drawn from. After each draw, they have to first estimate the probability that they are currently drawing from one or the other jar. They then have to decide whether to draw more beads at a small cost of $0.30, or to make a guess. Incorrect guesses are penalized by deduction of $15 from the initial sum (middle). Patients with schizophrenia with high delusion severity make more draws before they reach a probability estimate at which they decide to venture a guess (schematic illustration, right).

This belief-updating alteration could account for the hitherto poorly explained persistence of delusions. If, as proposed by earlier cognitive models of delusions, people with delusions simply jumped to conclusions, one would expect them to quickly revise an initial delusional belief as soon as they encounter contradicting evidence. However, this is not what is commonly observed in people with delusions. On the contrary, their delusional beliefs usually stand out because of their tenacity and resistance to argument, which can be exasperating for those around them. This persistence of delusions is well explained if, as suggested by Baker and colleagues' new results, people with delusions tend to stick with conclusions reached early in the inferential process.

Were delusional patients in this study perhaps more loss-averse, leading them to wait longer before making a guess in order to avoid large monetary losses? Additional analyses rendered this possibility unlikely. Estimates of subjective valuation in the beads task and delusion severity were not related; nor could loss-aversion as measured in a separate task explain the observed effects on inference in the beads task. The authors conclude that delusions are indeed specifically related to the observed alteration in belief-updating. Interestingly, they found no association of altered belief-updating with perceptual disturbances such as hallucinations. This finding raises important questions about the relationship between delusions and hallucinations, two core symptom dimensions in schizophrenia. Clinically, delusions and hallucinations frequently co-occur, but there are also patients in whom one of these two symptom dimensions strongly dominates, suggesting related but partly distinct abnormalities in inference. Such individual differences may possibly be accounted for by hierarchical models of inference (Heinz et al., 2018; Sterzer et al., 2018; Corlett et al., 2019). Here, it is highly plausible that delusions may predominantly relate to higher levels of inference that are crucial for belief-updating. The tendency to stick with a belief once it is formed, and to interpret new ambiguous information in the light of this belief (Schmack et al., 2013), may account for the fixity of delusional beliefs. Hallucinations in turn may additionally involve an abnormality in the way sensory processing is influenced by beliefs. Indeed, recent findings, including work from the same group, point to a mechanism whereby prior beliefs exert an overly strong influence on sensory information processing (Powers et al., 2017; Cassidy et al., 2018).

The new study by Baker and colleagues can serve as a model case for the emerging field of computational psychiatry. The authors' central insight was only possible by means of computational modelling analyses that allowed the authors to dissect their initial unexpected observation of increased draws-to-decisions in delusional patients. Moreover, the authors took great care to ensure that they were indeed measuring what they intended to measure. Thoughtful task instructions including a comprehension quiz and post-task debriefing made sure that all participants indeed understood the task. Control experiments, such as the loss-aversion task, ruled out alternative explanations for their findings. Finally, formal model comparison and parameter recovery minimized the possibility that their results might have been produced by a correlated, confounding process. Although all these steps might seem obvious, they are not always undertaken with the rigour that is necessary for dealing with complex tasks and models.

This study also illustrates one important limitation of purely behavioural computational psychiatry. Without doubt, the applied model described task behaviour at a resolution that allowed the authors to identify a novel computational process underlying delusions. However, it remains to be shown whether this model indeed provides an accurate estimate of the underlying neural mechanism of delusions. At the same time, the computational approach showcased in this work readily lends itself to the study of neural mechanisms. Model-based functional imaging could be used to identify brain regions and networks that are involved in the belief-updating processes related to delusions. Even more excitingly, high-quality computational modelling of delusion-relevant behaviour might open up new avenues for translational research. Here, breaking down task behaviours into fine-grained parameters could allow for a close mapping between human (pathological) experience and animal behaviour. This would create unparalleled opportunities for psychiatry to benefit from the tremendous methodological progress that has been made in the field of neuroscience. The golden age of computational psychiatry has only just begun.

processing....