About 10 years ago David Adam scratched his finger on a barbed wire fence. The cut was shallow, but drew blood. As a science journalist and author of The Man Who Couldn't Stop: OCD and the True Story of a Life Lost in Thought, a book about his own struggles with obsessive-compulsive disorder, Adam had a good idea of what was in store. His OCD involved an obsessive fear of contracting HIV and produced a set of compulsive behaviors revolving around blood.

In this instance he hurried home to get some tissue and returned to check there was not already any blood on the barbed-wire. “I looked and saw there was no blood on the tissue, looked underneath the fence, saw there was no blood, turned to walk away, and had to do it all again, and again and again,” he says. “You get stuck in this horrific cycle, where all the evidence you use to form judgments in everyday life tells you there’s no blood. And if anyone asked, you’d say ‘no.’ Yet, when you ask yourself, you say ‘maybe.’”

Such compulsive behaviors, and the obsessions to which they are typically linked, are what define OCD. Far from merely excessive tidiness, the mental disorder can have a devastating impact on a person’s life. Adam's story illustrates a curious feature of the condition. Sufferers are usually well aware their behavior is irrational but cannot stop themselves from doing whatever it is they feel compelled to do.

A new study published September 28 in Neuron uses mathematical modeling of decision-making during a simple game to provide insight into what might be going on. The game looked at a critical aspect of the way we perceive the world. Normally, a person's confidence about their knowledge of the surrounding environment guides their actions. “If I think it’s going to rain, I'm going to take an umbrella,” says lead author Matilde Vaghi. The study shows this link between belief and action is broken to some extent in people with OCD. As a consequence, what they do conflicts with what they know. This insight suggests compulsive behaviors are a core feature rather than merely a consequence of obsessions or a result of inaccurate beliefs.

The research of Vaghi and colleagues demonstrates the type of research being performed by the relatively new field of computational psychiatry. The work could ultimately lead to tools for early detection of people at risk. The field also may help pave the way for improved diagnosis based on understanding the biological or cognitive mechanisms of mental disorders rather than merely observing symptoms, as psychiatrists currently do. A more mechanistic analysis might also reveal that a tendency to repeat inappropriate actions (a compulsivity “trait”) is shared among multiple disorders such as OCD, substance abuse and eating disorders. And this type of analysis might distinguish among different types of OCDs and give psychiatrists a better idea about who might respond best to particular treatments.

In principle, the fact that beliefs and actions in OCD patients are often at odds could have several explanations. It is possible their ability to learn about the environment might be impaired in some way or they might lack confidence in things learned, despite their being accurate. Inspired by these questions, Vaghi and colleagues decided to investigate the relationship between belief and action during learning in people with and without OCD with the goal of discerning that connection—and what may go awry in OCD. The team—led by graduate students Vaghi and Fabrice Luyckx at the University of Cambridge, and neuroeconomist and senior author Benedetto De Martino at University College London—used an established task to study how beliefs and actions evolve over time during learning. They recruited 24 volunteers with OCD and 25 people without the disorder and had them play a video game in which they had to move a target (the “bucket”) around a circle to catch colored dots (“coins”) emitted from the center of the circle. The participants had to move the bucket to a position they thought most likely to catch the next coin, and give a rating as a percentage of how confident they were of the choice they had made. Most of the time the average destination of the coins was more or less the same, varying only slightly, but there was a one-in-eight chance each time that this position would dramatically shift.

The groups did not differ as far as how many coins were caught, but people with OCD tended to move the bucket toward exactly where the last coin landed more than healthy volunteers did. The actions of healthy participants closely mirrored the predictions of a mathematical model of learning whereas the actions of people with OCD deviated substantially from these predictions. Instead, the OCD group overreacted to what neuroscientists call “prediction error,” which in this case is the difference between where they placed the middle of the bucket and where the coin actually made contact with the circle. The healthy volunteers paid less attention to these errors unless a big shift took place in the average direction of the coin. The control group, instead, made a mental calculation concerning the average direction of the coin over the preceding trials. As a consequence, they tended to move the bucket less.

Crucially, though, confidence ratings (which dropped sharply after a shift, then rose as evidence of the new average direction accumulated) were indistinguishable between the two groups, suggesting the patients developed as accurate a sense of what was going on as the healthy volunteers. But their actual bucket placements showed they were not using this knowledge to guide their actions. “This study shows that [in OCD] actions are dissociated from thoughts, in a sense,” Vaghi says. “It's very much related to the clinical manifestation, when [sufferers] say: “I know it’s unlikely I’m going to get contaminated by touching the door handle, but even so, I will wash my hands.’”

The team also found that the extent to which confidence and action were uncoupled tended to be greater in individuals with more severe symptoms. “The new, exciting thing is the finding of a dissociation between action and belief in OCD that seems critical in this disorder,” De Martino says. “We found a clear correlation between the degree of this dissociation and the severity of symptoms.” These results suggest compulsive behaviors are a core feature of OCD rather than just a response to specific obsessions (washing to relieve anxiety about contamination, for instance). “The orthodox story is it’s all grounded in the obsessions; these drive anxiety and people take compulsive actions to alleviate that,” says computational neuroscientist Nathaniel Daw of Princeton University, who was not involved in the study. “This study supports the alternative idea that the compulsions themselves are a core deficit, not secondary to obsessions.”

De Martino is interested in the mechanisms underlying decision-making in general, and specifically the relationship between confidence and action. These are normally so tightly tied together, it is difficult to study their relationship. But the team saw that OCD might provide a natural way of disentangling them. A standard view is that confidence is calculated by monitoring our actions; think about how much longer you take to act in uncertain situations. “This is roughly the idea of monitoring your own behavior to build confidence estimates, but this is not the only architecture the brain could use,” De Martino says. If confidence is estimated by monitoring behavior, it should be impossible to separate confidence from actions. But it is also possible that confidence is calculated independently (or “offline”) and can then be used both for guiding action and reporting confidence levels—a form of processing known as a “parallel” architecture. The second alternative “is exactly what our data suggest,” De Martino says. “This is the general appeal of this work beyond the specific clinical interest; these patients can help us distinguish between alternative cognitive architectures.”

The results suggest the brain calculates confidence independently of action but healthy functioning depends on linking them tightly together. They are also consistent with a “dual-systems” view of behavioral control that distinguishes between explicit, conscious reasoning and more implicit, automatic behaviors, Daw says. “A number of issues in psychiatry related to compulsion have to do with a disconnect, or imbalance, between these two types of process.”

A major caveat is the study was a snapshot in time, of people who were already ill, and so cannot settle questions of cause and effect. “We don't know if this impairment results from illness or caused it,” Daw says. Figuring out how a general impairment like this relates to patients’ specific obsessions and compulsions will require studies of people with OCD over long periods to see how different aspects of the disorder evolve over time. But if the uncoupling of actions from beliefs is at the root of OCD, it represents a common disease mechanism that potentially unites a wide range of patients with quite different observable symptoms. “The hope is that by understanding the general mechanism, rather than focusing on specific symptoms, we can guide new therapies,” De Martino says. One implication of the current findings is that if compulsive behavior is at the core of the disorder, treatments directly aimed at modifying behavior (like cognitive behavioral therapy) may be more effective than treatments more tailored to obsessive, rumination-style thinking, Vaghi says.

The team next plans to investigate where the mechanism behind this impairment is located in the brain. Researchers already know connections between parts of the frontal cortex, which orchestrates higher functions like planning and problem-solving, and deeper areas, including a region called the ventral striatum, are abnormal in OCD. Further, prediction errors, which are what seemed to determine patients' abnormal actions, are primarily processed in the ventral striatum. These findings suggest circuits between the frontal and striatal areas may be the critical ones underlying this dysfunction. Brain imaging people with OCD doing this kind of task should help solidify this hypothesis. “Mental disorders are brain disorders,” Vaghi says “There’s still a lot of stigma because we think psychiatric patients are crazy and making things up, whereas we wouldn’t dare say a person with cancer is inventing it,” she adds. “Linking these kinds of behavior to brain mechanisms should help.”

The study illustrates the potential of computational psychiatry, Vaghi says. “It’s an example of how integrating computational and clinical aspects is a really powerful approach,” she adds. “Without computational modeling we wouldn't have been able to pin down exactly what this behavior relates to—we were able to understand which component of the model explains the behavior.”