COVID-19 Clinical Guideline Browser: A tool for sharing and understanding hospital guidelines

In this project we are applying a combination of natural language processing and data visualisation to UK hospital guidelines for treating COVID-19 patients. This is a collaboration between the Edinburgh Language Technology Group (led by Beatrice Alex), the Visual+Interactive Data group (led by Benjamin Bach) and clinicians at the Royal Infirmary of Edinburgh and across the UK. The main goal is to provide a user interface to clinicians to assist them in accessing, sharing, comparing and writing hospital guidelines.

National NHS guidelines often exist for a particular disease or condition, but hospitals tend to develop their own guidelines containing information specific to the hospital and lessons learned locally. During the COVID-19 pandemic, it is very important that clinicians can have access to as much up-to-date knowledge gained by scientists and their colleagues over the course of the pandemic as possible.

Our work requires the processing of hospital guidelines uploaded via an online web interface by clinicians in different formats, such as Word documents or PDFs. The tools we have developed process and convert each report into a common representation and extract information from their text. Each guideline is processed as a whole document and further broken down into snippets. This multiscale representation of the text allows us to enable keyword search and highlighting in the original documents as well as to apply clustering and additional similarity measures to group both the guidelines and individual snippets.

The information extracted from the texts is presented to clinicians in the web interface where guidelines can be searched, and similar guidelines or snippets can be highlighted in the original documents and inspected. The prototype is currently available to clinicians in the UK and they are able to upload their hospital guidelines. In the next phase, our plan is to gather further guidelines, tune the text processing and clustering, consider additional ways to analyse the text and add further data visualisations. In the longer term this work can be extended to other diseases, and deployed internationally to share hospital guidelines around the world.

Collaborators

Beatrice Alex (CoI), Edinburgh Language Technology Group, Edinburgh Futures Institute, School of Literatures, Languages and Cultures, School of Informatics

Benjamin Bach (CoI), Visual+Interactive Data group, School of Informatics

Kenneth Baillie (PI), RIE and The Roslin Institute

Andreas Grivas, Edinburgh Language Technology Group, Informatics

Arlene Casey, Edinburgh Language Technology Group, School of Literatures, Languages and Cultures

Claire Grover, Edinburgh Language Technology Group, Informatics

Andy Law, Bioinformatics, The Roslin Institute

Evan Morgan, Design Informatics

Clark Russell, Centre for Inflammation Research and The Roslin Institute

James Scott-Brown, Visual+Interactive Data group, School of Informatics

Richard Tobin, Edinburgh Language Technology Group, Informatics

Funding

This work is part of the ISARIC 4C consortium funded by UKRI (MRC) and led by Kenneth Baillie (PI), University of Edinburgh, Malcolm G Semple, University of Liverpool and Peter Openshaw, Imperial College London (co-leads). See full list of investigators and deliverables.