Precision medicine, specifically precision psychiatry, aims to improve outcomes for people with mental illness by tailoring treatment type and timing to individual characteristics. Clinical prediction models are widely championed as a way to deliver precision medicine. A prediction model is a set of rules that forecasts an individual’s risk of something happening in the future based on a set of risk factors. However, current prediction models are not fully actionable for individuals. At best, they stratify patients based on group averages to recommend interventions selected based on their Average Treatment Effect (ATE) in a population and tell us nothing with certainty about the individual. Further, as current models are based on associations not causes, we do not know whether acting on a predictor variable will actually change the outcome for an individual. For example, yellow tobacco-stained fingers may predict future lung cancer, but cleaning fingers will not prevent lung cancer – both are confounded by the common cause, smoking. To be fully actionable for individuals, prediction models should be based around intervenable causes and show how acting on these causes will materially change the outcome for that individual. To achieve this we should model the Individualised Treatment Effect (ITE) (also known as Conditional Average Treatment Effect (CATE)) of interventions and predict counterfactuals, the different potential outcomes forecast for an individual with their unique characteristics following an intervention. Causal, actionable prediction could deliver true precision medicine enabling personalised care.
In this session, we first outline limitations of current models and why they fall short of true precision medicine. We then review conventional prediction techniques, their gaps introducing causal concepts, and explore advanced topics including ITE/CATE and counterfactual prediction. We conclude with a real-world example showing how ignoring causal structures can lead to problems, focusing on ethnic bias in clinical prediction models.
Learning objectives
By the end of this webinar you should be able to:
• understand the limitations of current clinical prediction models.
• identify why traditional models based on associations and Average Treatment Effects (ATE) fall short in delivering truly personalised psychiatric care.
• differentiate between associative and causal prediction approaches.
• distinguish between models that predict outcomes based on correlations versus those that incorporate causal reasoning and intervenable factors.
• explore advanced concepts in causal modelling for precision psychiatry.
• gain insight into Individualised Treatment Effect (ITE)/Conditional Average Treatment Effect (CATE) estimation, and counterfactual prediction, and understand their relevance to fully actionable clinical decision-making for individuals.
• evaluate real-world implications of ignoring causal structures in prediction models.
• critically assess how overlooking causal relationships can lead to inequitable or ineffective treatment recommendations using the example of minority ethnic bias.
Speakers
Chair: Dr Rajeev Krishnadas, University of Cambridge, Cambridge
Dr Samuel Leighton, University of Glasgow, Glasgow
Dr Fani Deligianni, University of Glasgow, Glasgow
Dr Damian Machlanski, University of Edinburgh, Edinburgh
Dr Rajeev Krishnadas, University of Cambridge, Cambridge
Funding
RCPsych International Congress is financially supported by the purchase of exhibition stand packages. A full list of confirmed exhibitors can be found here: Exhibiting Organisations at International Congress 2026
Availability
This webinar is part of the Congress webinar 2026 package. If you attended all four days of Congress, you will have access to these as part of your Congress package. Otherwise the Congress webinar 2026 package can be purchased below.
