Spotlight: Dr Anoop Shah
7 September 2023
We speak to Dr Anoop Shah, UCL Institute of Health Informatics, to find out more about his work using AI to improve electronic health records.
What is your role and what does it involve?
I am an Associate Professor at the UCL Institute of Health Informatics and I work on a number of research projects which aim to improve the quality of electronic health records and enable them to be used for research to support better patient care. I combine my research with clinical work as a consultant in clinical pharmacology and general medicine at University College London Hospital (UCLH).
Tell us more about your research in AI?
I lead a NIHR-funded research project called MiADE, which aims to make it easier for clinicians to enter information such as diagnosis codes into electronic health records. Currently, much of the information in health records is in the form of narrative text, which is difficult to use for clinical decision support or research. MiADE uses natural language processing to convert the text into structured data, and allows clinicians to validate the information before it is saved. We are testing out our new system which is integrated with the Epic electronic health record at UCLH.
We have recently been awarded a UKRI grant to extend this work. We will investigate more advanced natural language processing models to extract more detailed information from the text. We will also prototype new designs of a user interface for clinicians, and compare the accuracy of natural language systems trained using different data sources. This project will be a collaboration between UCL, UCLH and Great Ormond Street Hospital (GOSH).
The new tools and knowledge gained from these projects will hopefully enable electronic health record systems to be better designed to meet the needs of patients, clinicians and researchers.
I am also the UCL site lead for a European collaboration called DataTools4Heart, which is setting up a federated cardiology research platform across multiple European countries. This will enable AI methods to be applied to patient data from multiple countries without moving data from its original location, to protect patient privacy. I also work on natural language processing of general practice records from the THIN database, which we are using to study the symptoms and diagnosis of Long Covid (UCL-THIN Long Covid Study).
How will AI change the world of health?
AI is rapidly being developed to assist in many specific tasks in healthcare, such as analysing radiology images or monitoring patient safety. In the area of electronic health records, AI voice recognition is already commonplace, and we are addressing the next challenge of converting text to structured data and improving clinical decisions. I imagine that a future electronic health record might be able to record and automatically summarise clinical consultations, and show relevant information without clinicians needing to search.
What are your thoughts of safely using AI?
AI tools can be extremely powerful but can make errors. They might learn too much from biases and peculiarities of the data they are trained on. The use of large amounts of potentially sensitive data for training AI algorithms also introduces issues of confidentiality and information governance.
AI tools need to be deployed within a framework that ensures there are appropriate checkpoints before AI outputs can influence clinical decisions. For example, in our MiADE system, the clinician sees the result of the natural language processing, and has to actively accept the suggested diagnosis codes before they become part of the patient record.
If you could make one change in the world today, what would that be?
I would like to see a reduction in inequality and better cooperation within our society.