| EPSRC Reference: |
EP/Y018192/1 |
| Title: |
Clinical prediction foundation models for individuals with multiple long-term conditions |
| Principal Investigator: |
Yau, Professor C |
| Other Investigators: |
|
| Researcher Co-Investigators: |
|
| Project Partners: |
|
| Department: |
Women s and Reproductive Health |
| Organisation: |
University of Oxford |
| Scheme: |
Standard Research - NR1 |
| Starts: |
02 October 2023 |
Ends: |
01 April 2025 |
Value (£): |
615,516
|
| EPSRC Research Topic Classifications: |
|
| EPSRC Industrial Sector Classifications: |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary on Grant Application Form |
In recent years, a new approach for building artificial intelligence (AI) systems has appeared. The idea is to build a single AI system using a huge amount of information creating what is known as a "foundation model". This foundation model learns from the data some of the basic rules but is not necessarily useful in itself. Instead, it is a building block that can be taken by someone to then further train and construct an AI to perform a task. A single foundation model can therefore give rise to many AI applications.
In this project, we will explore whether foundation models could be produced for the development of clinical risk prediction models. Clinical risk prediction models take in information about a person's health and then produce output which tells us how likely they might experience some health outcome of interest in the future. For instance, a risk prediction model might use a patient's current health conditions, blood pressure measurements and body mass index to predict their future risk of having a stroke. Risk models are able to do this by learning from historical data about how often people with similar health conditions experienced the same type of health outcome in the past.
When building risk prediction models it is typical that we first find data on a relevant group of patients, analyse their data and then create the prediction model. However, if we change the data (e.g. to include cholesterol measurements) or we want to predict something different (e.g. the risk of heart failure instead of stroke), we will often have to build an entirely new prediction model from scratch. If this is done manually, it would take a large amount of time and resource, and even if it were possible, medical regulators do not have time to evaluate all these possible creations. This is where foundation models could be useful since they could be used to rapidly produce risk prediction models based on any combination of input and output data that is desired.
This is important for patients with two or more chronic health conditions (or multiple long-term conditions). They are often under-served because most medical guidance and practice is often based upon how to treat one condition at a time. It is not always clear to doctors how to treat multiple conditions but using AI to analyse historical medical records, they could identify what has worked best in the past and use this as guidance as to how to best treat patients in the future. The challenge is that one moment doctors could be faced with a patient with type 2 diabetes, high blood pressure and osteoarthritis who is looking for long-term chronic pain management and in the next they might be concerned about the risk of stroke of someone with dementia, obesity and high cholesterol. Do we need to build separate prediction models for both? What if another patients arrives with a different set of conditions, do we need a model for them as well?
Our research proposes to build an automated system, based upon a general clinical risk prediction foundation model, which would allow doctors to build their own customised prediction model on-the-fly. Given an individual's conditions, the system would automatically construct an appropriate prediction model for that person, giving doctors the individualised guidance for that patient. We will investigate how to build such systems safely and whether the performance is at least as good as models which are built manually. The ultimate aim is to use AI to empower doctors with the ability to make better decisions based on the use of health data.
|
|
Key Findings |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Potential use in non-academic contexts |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
|
Impacts |
|
| Description |
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk |
| Summary |
|
| Date Materialised |
|
|
|
|
Sectors submitted by the Researcher |
|
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
|
| Project URL: |
|
| Further Information: |
|
| Organisation Website: |
http://www.ox.ac.uk |