Search
Close this search box.
Search
Close this search box.

Research Portfolio

GRANT TITLE:

Using Natural Language AI to identify predictors of refractory epilepsy in NHS Electronic Health Records

GRANT TYPE:

EPILEPSY RESEARCH UK & ANGELINI PHARMA INNOVATIONS IN HEALTHCARE PROJECT

grant amount:

£148,278 over 36 months Awarded in 2022

lead investigator:

Prof Mark Richardson

Co-Investigators:

- Prof James Teo (KCL)
- Dr Joel Winston (KCL)
- Dr Anthony Shek (KCL)
- Prof Deb Pal (KCL)
- Prof Parashkev Nachev (UCL)
- Dr Beate Diehl (UCL)
- Dr James Pickett (Health Data Research UK)

institution:

Kings College London

Background

‘Refractory epilepsy’ refers to epilepsy in which seizures continue despite treatment. This affects one in three people living with epilepsy. Several research studies have identified risk factors for developing refractory epilepsy, but a large proportion of cases are still unexplainedRisk factors remain unidentified because only small numbers of patients have been studied (a few hundred, even in the biggest studies), and the information used to identify the factors is very limited.

"Over the last decade, NHS hospitals have started to adopt electronic systems for collecting medical records. As a result, there is detailed information about health status related to tens of thousands of people with epilepsy. We have developed systems to extract highly detailed information from electronic health records, maintaining confidentiality and privacy of the information. We can analyse this information using artificial intelligence to track the course of epilepsy in thousands of people, and make a much better attempt to identify the factors that predict refractory epilepsy.

The Study

In this study, Professor Richardson’s team will access information available through electronic health records (EHRs) held in NHS hospitals for patients with epilepsy. The amount of information in EHRs is so vast that it would be impossible for a person to find it all, so automated computer-driven methods are needed. However, the data is held in many different formats, meaning it is difficult for a computer to access all the information. It is also difficult for a computer to understand  all the information because much is written in so-called ‘free text’. ‘Free text’ means the data is in whatever form of words a healthcare professional has used.  

The team has developed a system using state-of-the-art Artificial Intelligence (AI) tools to access EHR data held in different formats, and automatically annotate it in rich detail. These annotations provide standardised detailed information about each individual patient. They have also developed AI tools to predict outcomes based on this information. The team will use these tools to identify risk factors for refractory epilepsy from EHRs of up to 30,000 people in three London hospitals.

Significance

The team expect to identify new information about the risk factors for developing refractory epilepsyIf the causes of refractory epilepsy are better understood, this could ultimately enable us to develop new strategies to prevent it.