Data science and pharmacoepidemiology for outcome improvement in severe mental illness (DS-SMI)

Lead Research Organisation: University College London
Department Name: Division of Psychiatry


People with severe mental illness (SMI), including schizophrenia, bipolar disorder and other psychotic illness often only partially respond to drug treatment. They also experience medication adverse effects, and increased morbidity and mortality compared to the general population. There is a desperate need to improve pharmacological treatment of SMI. Two cost effective approaches to addressing this problem are to i) improve response to existing medication via personalisation, and ii) identify drugs already in existence (with different indications) that can be repurposed to treat psychiatric symptoms.

These complimentary translational research streams will harness the power of large routine health registers, electronic health records and mobile phone applications, along with modern statistical and machine learning techniques for prediction modelling and causal inference. Data will come from the United Kingdom, United States, Sweden, Denmark, Hong Kong and Taiwan.

This research stream will advance my current work on prediction of maintenance treatment response in individuals with bipolar disorder using machine learning.

Despite recent progress in the field of treatment personalisation, psychiatry lags behind other medical specialties. Currently, no validated system of tailoring treatment choices is available and matching treatment to specific patients is often a matter of trial and error. Via prediction modelling clinicians could more precisely select treatment for patients' needs and thus improve their outcomes. This research stream will focus on:

i) Identifying predictors of treatment response in patients during their first illness episode
ii) Predicting which individuals will not have their symptoms adequately treated after trials of two medications (treatment resistance)
iii) Predicting adverse effects, including weight gain, restlessness (akathisia) and excess sedation

Clinical features contained in medical records have been shown to be associated with response, treatment resistance and adverse effects, but these have not been combined in a systematic way. The scale and widespread use of electronic health records globally now allows for use of multiple data sets for external validation of generated models.

There may also be important changes early in the course of treatment that can predict long term outcomes. These changes are unlikely to be captured in electronic health records, but may be available via patients mobile phones. Capture of passive data via phone apps is now straightforward and potentially contains markers of changes in mental state, such as sleep, movement and phone usage. Apps also facilitate remote symptom monitoring and performance of cognitive tasks. This information will be used to further enhance prediction models. The models built during the early stage of this fellowship will be tested at scale in clinical populations via implementation science methods.

This translational research stream builds on my previous work which examined whether a number of drugs identified as having potential for repurposing had effects on psychiatric hospitalisation and self-harm rates in patients with SMI.

There are a number of other drugs which should be examined via similar approaches to validate these signals for potential effectiveness, whilst robustly accounting for potential confounding, these include a range of anti-inflammatory agents. This work will be cross-validated in other international data sets.

The process of pharmacoepidemiological validation optimises the chance of success and provides guidance on which drugs to take forward to randomised controlled trial (RCT). Towards the end of this fellowship I will develop the protocol necessary for a large adaptive RCT and run a pilot to assess feasibility and acceptability.


10 25 50