AI research may identify risk of emergency hospitalisation from admin data

Generative AI research using a country's administrative health records may identify people at risk of imminent emergency, enabling early intervention.

A recent study published in The Lancet Digital Health has unveiled a novel approach to identifying people with a high risk of emergency hospital admissions using only administrative data from electronic health records (EHRs). By revealing easy to understand patterns in routine healthcare events, such as GP visits, researchers have identified an opportunity to convert routinely captured information into a powerful predictive tool using machine learning- a type of artificial intelligence. 

Using anonymised nationwide primary care data from 1.37 million patients in Wales, from a dataset hosted by the University of Swansea, the investigators discovered that the pattern of a patients’ health-associated interactions with primary care can identify people at significant risk of emergency hospital admissions with remarkable accuracy.   

“Health data is inherently messy and complex," said Benjamin Post, Research Fellow at Imperial College London. "Yet, human clinicians successfully interact with this information at an individual patient level daily. Artificial intelligence techniques provide the opportunity to make this large volume of data coherent and understandable, but at scale. By combining the strengths of machine learning with human pattern recognition, our approach demonstrates how AI may be deployed to assist real world clinical practice in the future.”

Unlike traditional models requiring large numbers of sometimes complicated or inaccurately recorded clinical variables, this approach relies solely on readily available date and time labels in big datasets.  

"These temporal patterns we discovered in administrative data reveal an untapped predictive power that could revolutionise risk assessment and healthcare resource planning. Our work shows digital public health at scale, as it has the potential to enhance hospital capacity management while providing early warnings to general practitioners and even patients themselves.” (Aldo Faisal, Professor of AI and Neuroscience and Director of the UKRI Centres in AI for Healthcare)   

Importantly, the approach would work well with different types of data, making it potentially deployable across health systems internationally. This study underscores the potential of machine learning to develop scalable, efficient tools for healthcare systems worldwide. By leveraging routine administrative data, this approach could offer a faster, more cost-effective way to predict patient risk and improve healthcare planning at both an individual and societal level.     

“Being able to identify people at risk of unexpected deterioration would be very useful, especially if we can do this with current routine systems, working in different ways. This AI based approach would potentially allow intervention to stabilise peoples’ conditions or allow them to reflect on their future wishes in the event of an unavoidable downturn in health, and at a population level improve service planning.” (Stephen Brett, Professor of Critical Care & Consultant in Imperial College Healthcare NHS Trust)

This research was supported by UK Research and Innovation. [UKRI Centre for Doctoral Training in AI for Healthcare grant number EP/S023283/1]