Computational Psychiatry

Question 1

What is computational psychiatry?

Answer 1

Using approaches from computational neuroscience to understand and treat mental health problems

Question 2

What are the 2 categories of computational psychiatry?

Answer 2

Theory-driven
Focused on understanding underlying mechanisms
Builds mathematical models of psychological/neural processes and how they go awry

Data-driven
More focused on practical utility (e.g., diagnosis/prediction)
Applies computational approaches to find patterns in data in service of this aim

Question 3

What is a computational model?

Answer 3

They take hypotheses about the mechanisms underlying mental health problems and put them into a mathematical form

This is useful because it makes it easy to quantify precisely how well our data matches our hypothesis

Question 4

What is the Rescorla Wagner model?

Answer 4

Proposes that we update our expectations based on weighted prediction errors

(not only most recent outcome; but gradually update our expectations based on proportion of prediction error)

Question 4

Prediction error equation

Answer 4

= outcome - expectation

Question 5

What can we use to assess learning irl?

Answer 5

Ask people to complete cognitive tasks where they learn to associate stimuli with an outcome

(find people have wide range of learning rates/weighting)

Question 6

What was found about learning and schizophrenia?

Answer 6

A number of studies have used this exact approach to show that people with schizophrenia learn more slowly than healthy individuals (e.g., Martinelli et al., 2018; Collins et al., 2014)
This might suggest that people with schizophrenia have difficulty learning about the world around them

Question 7

What was found about learning and anxiety?

Answer 7

In contrast, it seems people with high levels of anxiety tend to learn faster (higher learning rate) than people low in anxiety (Huang et al., 2017)

suggest anxiety tend to be faster to update their expactations; linked to tendency to assume negative outcome

Question 8

What was the anxiety experiment?

Answer 8

Subjects choose one of three options, trying to get the biggest reward
Each option has a different (hidden) probability of reward
Learning which option is most likely to give a reward results in greater rewards

Question 9

What was tested and found by Aylward 2019?

Answer 9

Participants choose one of four options, trying to collect reward and avoid punishment (negative reward)
Each option has a different probability of reward and punishment which varies over time

They found that people with anxiety have higher punishment (but not reward) learning rates
May explain how people with anxiety develop a negative view of the world around them

Question 10

What are the 2 systems for making decisions?

Answer 10

Goal directed vs habitual decisions

Question 11

What are goal directed and habitual decisions?

Answer 11

One involves prospective thought about the outcomes of our actions – this “goal-directed” system is more demanding, but more accurate

One involves learning based on the immediate consequences of our actions – this “habitual” system is simpler, but less accurate

Question 12

What is thought about decision making in OCD?

Answer 12

One hypothesis is that there is an imbalance in these systems in conditions like OCD
People with OCD rely on habitual decision-making, which encourages repeated compulsive behaviours

Question 13

What was found about habits in OCD by GIllian et al 2011/14?

Answer 13

This hypothesis found a lot of support from studies using cognitive tasks that assess how habitual or goal-directed people are in their decision-making
In general, people with OCD tended to rely on habits to a greater extent than people without OCD (Gillan et al., 2011; 2014)

Question 14

But what did Alvarez 2016 find out about this?

Answer 14

However, a problem emerged
Similar results were found in studies looking a range of other conditions, such as social anxiety, psychosis, eating disorders, autism…
Clearly a reliance on habit wasn’t specific to OCD

Question 15

What did Gillian propose instead?

Answer 15

Gillan et al. (2016) sought to solve this conundrum
They hypothesised that perhaps an over-use of habitual decision-making might be related to compulsive symptoms, which are in people with all types of mental health problem

They gave people a battery of questionnaires measuring a variety of symptoms, and extracted 3 symptom dimensions that cut across traditional diagnostic categories:

They found that symptoms of most conditions were linked to an increase in habitual decision-making
This replicated what had been in observed in prior studies (except social anxiety!)
Again, this suggested that excessive habitual decision-making wasn’t specific to OCD at all

Question 16

What are the 3 symptom dimensions that cut across traditional diagnostic categories?

Answer 16

Anxiety & depression
Compulsivity & intrusive thought
Social withdrawal

Question 17

What was only found to be linked to increased habitual decision making? What did this suggest?

Answer 17

“compulsivity and intrusive thought” was linked to increased habitual decision-making
This suggested that the reason so many conditions are associated with habitual decision-making is because this symptom dimension is present in all of them
This ultimately supports the hypothesis that compulsive behaviours are linked to increased reliance on habits

Question 18

What would be clinically useful to detect in AD?

Answer 18

deterioration?

Question 19

What did Kambeitz find about depression, what are the limitations of this?

Answer 19

accuracies of 75% depending on data modality

but small samplies and misestimation of acccuracy and publication bias

Question 20

What would be useful to predict in depression?

Answer 20

Treatment response

Question 21

What did Cohen et al 2021 find, what are the limitations of this?

Answer 21

78% accuracy (Cohen et al., 2021)
However, these studies also use small samples
Additionally, even if effective neuroimaging is potentially too expensive for real-world clinical use

Question 22

What did Wager 2013 sought to develop

Answer 22

Objective measure of pain using machine learning

to decode subjective pain rating successfully

Question 23

Mental health problems aren’t unitary concepts instead

Answer 23

comprise multiple subtypes

(e.g. depression highly heterogenous condition)

Question 24

How many subtypes were identified of depression?

Answer 24

3

with distinguishable biological correlates and characterised by different symptom profiles and different treatment response patterns (Drysdale 2017)

but haven’t been able to replicate

Question 25

What is linked to symptoms of mental health problems?

Answer 25

Learning and Decision-making processes

Question 26

What did data-driven approaches demonstrate?

Answer 26

success in distinguishing people with diagnoses from those without – however these studies are limited by small samples and often limited clinical potential

Question 27

Data driven methods have the potential to discover

Answer 27

subtypes within heterogeneous conditions, but this potential is yet to be realised