AI tech applied to improve care and research at UCLH

Technology that can interpret ‘unstructured’ text in patient records is being tested across UCLH to improve care and research.

The software called CogStack – developed by a team from the UCLH Biomedical Research Centre, the UCL Institute of Health Informatics, NIHR Maudsley BRC and King’s College London/Hospital (KCL/H) – uses a type of AI called natural language processing (NLP) to interpret this information which would otherwise need to be interpreted manually by NHS staff.

Electronic health records include a large amount of text in the form of typed up information that includes clinical notes, letters and imaging reports, but while rich in information, this text can be difficult to make use of because it is unstructured, narrative form.

To make use of this information, NHS staff assign standardised codes to patient records based on particular words, conditions or treatments mentioned in the text. This process is known as clinical coding, but it is time-consuming, expensive and carries a risk of mistakes.

A paper published in the journal JMIR Medical Informatics outlines ways in which CogStack is being applied at UCLH to avoid the need for manual clinical coding. Examples include:

Treatment of atrial fibrillation

Blood thinning medication is commonly given to patients with the cardiovascular condition atrial fibrillation (AF). But 1 in 5 patients with AF are not on the most effective medication or on no medication at all.

CogStack is being tested at UCLH to identify patients who are on suboptimal medication, so that clinical teams can then review and optimise treatment. NLP via CogStack can analyse unstructured information in patient discharge summaries.

Clinical trial recruitment

Researchers assessed how CogStack could have been used in a trial at UCLH which looked at a time-sensitive treatment for sepsis.

CogStack was used to identify potential participants for the study based on free text information contained in patient records. The system successfully identified patients who in real life were manually recruited to the trial – and identified more potential participants.

The team concluded that the AI could aid recruitment in trials in future removing the need for manual identification of patients.

Immediate use of NLP at the point of care

UCLH is helping to develop a system whereby an NLP system translates clinical notes into standardised information codes as soon as it has been typed into a patient record by a clinician.

So far this system has been tested using data from patients with Covid-19 and patients with heart failure. In each case, the system was trained to extract all diagnoses and symptoms and to code these appropriately in the form of structured data.

Professor Richard Dobson, Group Lead for CogStack and Professor of Health Informatics at the Institute of Health Informatics, UCL, and Dept of Biostatistics and Health Informatics, KCL said:

“Electronic health records hold a wealth of information such as patient symptoms, tests, investigations, diagnosis and treatments, that can be used to improve healthcare. CogStack is a system that can understand this information, reducing the need for manual coding of free text, and provide real insight for clinical teams. Ultimately there is potential to scale this technology across the NHS.”

Dr Wai Keong Wong, UCLH Chief Research Information Officer, said: “We are seeing clearly how powerful this technology will be in terms of improving care, research and how we run hospital operations. Patients will benefit from this technology through better care, higher quality research, and more efficient hospital operations. NHS staff also stand to benefit hugely from the technology which will enable them to offer better care and reduce the burden of manual processing of information processing.”

Read the paper.