Recent studies, as published in The Lancet, a journal focused on digital health, have found that generative AI research can utilise previous administrative health records to identify individuals at imminent risk of emergency hospital admission.

This study used data from electronic health records (EHRs) to identify predictive patterns in healthcare events such as routine GP visits. AI utilised machine learning techniques to transform these daily healthcare interactions into a tool for predicting future health risks. The University of Swansea hosted a dataset which comprised anonymised nationwide primary care records from 1.37 million patients in Wales. Findings were notably promising, with AI having a high level of accuracy in predicting the risk of emergency hospital admissions based on these daily healthcare interactions.

The approach has several advantages: it relies solely on routine administrative data, making it cost-effective, scalable, and efficient. Additionally, it is potentially deployable across international health systems. The impact of this tool could be profound as it is able to provide early warnings to GPs and patients—therefore improving hospital capacity management and enhancing population-level planning and resource allocation.

Experts have weighed in on the model’s potential. Benjamin Post from Imperial College London noted that while health data can be messy, AI can make it coherent at scale by combining machine learning with human expertise. Aldo Faisal, Professor of AI & Neuroscience, emphasised that, unlike traditional models requiring numerous complex variables, this approach uses simple time and date labels, highlighting the untapped predictive power of temporal patterns in administrative data. He believes this could revolutionise risk assessment, healthcare planning, and capacity management. Stephen Brett from Imperial College Healthcare NHS Trust added that early identification of deterioration risk would enable timely interventions and care planning, benefiting both individual patients and system-wide healthcare management.

Back in 2019, one issue that was flagged by Imperial College London was that the NHS (UK National Health Service) had critical gaps in their use of EHRs. They found a wide variety of NHS hospitals using different and incompatible electronic health record systems. This, therefore, led to missed or incomplete medical records for nearly four million patients. However, following several proposals from policy influencers as well as think tanks, the Department of Health and Social Care are now working on introducing a single patient record in the English NHS.

Many other countries have also adopted similar systems. Finland has implemented nationwide digital health record systems, such as Kanta, which centralise patient data and connect all health service providers. The system integrates AI to predict patient risks, manage chronic conditions, and optimise hospital resources, leading to improved health outcomes and better forecasting of future healthcare needs. Other examples are Canada, where multiple provinces have implemented digital health systems like Ontario Health Records and Pan-Canadian Health Records, integrating AI to analyse patient data for early warning signs of deterioration, identify at-risk patients, and optimise hospital capacity management.

This article explores how generative AI is being used to analyse routine health records to predict imminent hospital admissions. Drawing on large-scale electronic health record datasets, this approach is both economically sustainable and readily extensible, and it holds the capacity to transform healthcare planning, patient risk stratification, and the management of hospital resources.