About the Group
The Precision Medicine & Statistical Learning Group is a research group of interested statisticians and data scientists within the Department of Biostatistics and Health Informatics at the Institute of Psychiatry, Psychology and Neuroscience headed by Prof Daniel Stahl. The group is part of the NIHR Maudsley BRC theme “Trial, Prediction and Genomics” and established the BRC Prediction modelling group.
In recent years, there has been a shift towards precision medicine. Precision medicine is a medical paradigm that customizes preventative care and treatment strategies based on the genetic, environmental, lifestyle, and clinical characteristics of an individual. It aims to optimize therapeutic efficacy by considering that each patient possesses unique genetic, environmental, and lifestyle factors that can influence their health and response to treatment. Prediction modelling plays a crucial role in precision medicine by using statistical and machine learning methods to analyse large datasets containing detailed patient information. These models help to predict a patient's likelihood of having a disease (diagnostic model), develop a disease (risk model), respond to a particular treatment (treatment response model) or estimate the disease progression (prognostic model). Consequently, prediction modelling enables healthcare providers to make more informed decisions, optimizing treatment strategies and improving patient outcomes through personalized care approaches.
Prediction modelling in psychiatry presents numerous methodological challenges, including dealing with unbalanced groups, population substructure, multi-centre trials, missing data, multicollinearity, scales with measurement error and the validation and implementation of predictive models. Collaborating closely with our clinical partners, we have successfully developed and validated prediction models in the field of mental health, including a transdiagnostic psychoses risk calculator and a model predicting the recurrence of depression.
In the era of 'Big Data', prediction modelling can no longer solely rely on traditional statistical methods, and computer-intensive machine learning methods are increasingly needed. In statistics, the prime focus is usually on understanding the data and relationships in terms of models by estimating parameters and quantifying the uncertainty of these estimates. Machine learning uses computer-intensive learning -algorithms and focuses on prediction and classification and less on mechanisms. Statistical learning theory tries to unify the two approaches and thus studies within a statistical framework the properties of learning algorithms commonly used in machine learning. For example, prevailing interests are the development of (i) novel models based on Machine Learning and Topological Data Analysis to allow for precision medicine, with main works on treatment personalisation for Depression and Hypertension (Raquel Iniesta), (ii) methods to boost predictive performance while preserving interpretability in time to event health outcome (Diana Shamasutinova), dynamic risk models (Daniel Stahl) or (iv) sample size calculators for machine learning applications (Ewan Carr, Diana Shamsutdinova, Gordon Forbes).
Professor Daniel Stahl
Find us on facebook: @predpsychUK