Transforming big data into personalised psychiatry

Clinical Prediction models (CPM) are developed from complex health care datasets using advanced statistical techniques. They inform clinicians and patients about the individual risk of having or developing a disease, the likely treatment response, and so identify the best interventions for a patient.

However, development of these clinical prediction models in mental health is challenging because, unlike in diseases like cancer, genes and symptoms are not specific for a particular disorder but are shared by several different psychiatric conditions.  

Prediction Modelling Group

The National Institute for Health Research (NIHR) Maudsley Biomedical Research Centre Prediction Modelling Group includes senior and early career researchers, and PhD students. It develops, tests and implements clinical prediction models using large-scale clinical and biological test databases. Co-production with patient groups and collaborations between clinicians and informaticians has now led to over 15 published prediction models.

With the Psychosis theme, examples include the development of the transdiagnostic psychosis risk calculator and, in an industry collaboration, creating a way to predict future treatment resistance in people with first-episode psychosis. The model can be used to pre-identify patients at risk of treatment resistance and invite them to participate in new clinical trials.

With the Mood Disorders theme, our clinical prediction model used existing clinical knowledge to preselect likely predictors, including brain imaging, and provided superior results to a traditional approach to predict the recurrence of depression. Other collaborative work used patients' data collected over time to predict remission of depression; and an inexpensive biological test to identify young people at increased risk of self-harm or suicide.

Adoption of clinical prediction models

We recently demonstrated that many published clinical prediction models are of inadequate quality and rarely replicated, a key issue for our patient partners.

Adoption of a clinical prediction models in practice is limited by the trust of clinicians, patients and other stakeholders. To address these issues, the prediction modelling group promotes good practices including open and reproducible science, by offering external short courses and statistical advisory services, and publishing methodological papers for implementing novel methodologies, such as ways of handling missing data and developing reliable scales.

Our monthly research seminars are attended by over 50 people from across the UK and beyond. Future work in our Biomedical Research Centre includes a clinical prediction model implementation group to overcome obstacles for the successful implementation of clinical prediction models including clinical trials and observational studies to assess their clinical usefulness in routine practice.

IMPACT AREAS:

Developing Resources for Research | Involving Patients in Research | Industry Collaboration | Data and Analytics to Drive Healthcare