Developing and validating outcome prediction models for people with First Episode Psychosis using Supervised Machine Learning (Precision Medicine)
Psychosis is an illness affecting mainly young people with an incidence rate of 31 per 100,000 people. Evidence from meta-analysis suggests that more than half of young people with First Episode Psychosis have poor outcomes. Despite the heterogeneity in remission and recovery outcomes, there are currently no validated tools available for early identification of these patients. Identifying individual patients with poor outcomes at initial clinical contact would help in stratifying interventions, which may lead to improved outcomes and utilisation of clinical resources.
Previous studies have identified predictors of outcome at group level including socio-demographic variables, clinical and treatment response variables, comorbidity, functional and cognitive deficits with inconsistent reliability. There is a lack of clarity on how or which of these group level factors should be combined for prediction at the individual level, or as to the performance of prediction models in clinical settings that may be very different to the primary research study site. In order to maximise clinical utility, an approach is needed that can be applied to stratify the risk of poor outcome in an individual patient at the initial clinical contact. One solution to this is the use of machine learning, where algorithms can sift through a large array of predictor variables, looking for combinations that can reliably predict individual patient outcomes.
My research will involve empirically testing clinical prediction models in clinical settings to validate their utility in stratifying interventions based on risk prediction of outcomes and whether, by such stratification, outcomes can be improved.
Dr Pavan Mallikarjun
Professor Georgios Gkoutos
Dr Sarah-Jane Fenton
Professor Stephen Wood