A new AI platform will combine electrocardiogram (ECG) readings with routine NHS data to predict who will develop heart failure before symptoms appear. Heart failure is a leading cause of death and a major drain on NHS budgets, yet it is often caught too late for effective treatment. Current AI models that analyse ECGs work in isolation, ignoring other clinical information such as blood test results or disease codes that could sharpen their predictions. This project builds a single "multimodal" AI system that fuses ECG signals with electronic health records, using the same kind of foundation-model architecture that powers large language models. The team will first train a model on millions of patient records to learn general patterns of health and disease, then fine-tune it specifically for heart failure risk. If successful, the platform could be deployed across GP surgeries and hospitals to flag high-risk patients years in advance, enabling early lifestyle interventions or medications. It would turn the cheap, widely used ECG—already performed millions of times a year in the UK—into a far more powerful screening tool, without requiring new equipment or extra appointments.
View original technical description
Cardiovascular disease (CVD) remains a significant burden globally, with late detection leading to poor outcomes and increased healthcare costs. Early identification of subclinical disease and high-risk individuals is a priority for the NHS, as it enables precision and preventative medicine. Heart failure is particularly important as a major cause of morbidity and mortality as well as a large portion of the NHS budget. Advances in artificial intelligence (AI) offer an opportunity to improve risk prediction by leveraging large, routinely collected structured clinical data. The electrocardiogram (ECG), a widely available and low-cost test, represents a high-dimensional data source with significant untapped potential for AI applications. Current AI-ECG models, however, often fail to integrate additional clinical data, limiting their utility. This proposal aims to build a multimodal AI platform that combines AI-ECG models with routinely collected clinical data (disease codes, lab results and other clinical data) for heart failure screening and risk prediction. The project will utilise state-of-the-art foundation model paradigms to pretrain models on health and disease representations before fine-tuning. The project consists of two key work packages. WP1 will develop a clinical tabular data foundation model using electronic health record (EHR) data, employing transformer architectures and time-to-event models. WP2 will integrate the existing AI-ECG platform with the EHR foundation model to create a multimodal AI risk estimation platform, capable of heart failure-specific predictions. This novel approach builds on the applicant’s prior work, including the development of AI-ECG platforms, and will be developed and validated using diverse, transnational external datasets.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know