AI model identifies heart failure patients at risk
The period after discharge is critical for many heart failure patients, but it is often difficult to predict who is at risk of needing hospital care again. In the project Heart Failure Readmission Prediction (HaRP), researchers have developed an AI model intended to help healthcare staff identify high-risk patients in time.
“A tool that provides a clear risk assessment and shows which factors carry the most weight can reduce uncertainty and make decisions more consistent among patients.”
Lina Lundgren, Docent in Health Technology
HaRP is a continuation of an earlier collaboration between Halmstad University and Region Halland. The first study, published in 2019, explored whether it was possible to predict readmissions among heart failure patients using the Region’s patient data.
In the subsequent project, carried out between 2021 and 2024 within the CAISR Health research profile, the researchers wanted to take things a step further. The aim was to refine the machine-learning model into a tool that is more useful in practice – one that not only calculates risk, but also explains which factors underpin the outcome.
“We knew that the model had the potential to predict readmissions, but not how it could best support clinical practice. It was important to create something that is both accurate and easy to interpret, so that staff can actually use it in their daily work”, says Lina Lundgren, Docent in Health Technology at Halmstad University and Project Leader for HaRP.
The model is based on data from around 6,000 heart failure patients. By combining the patient’s medical history, test results and previous healthcare contacts, it calculates the probability that the patient will need to return within 30 days and categorises the outcome as either high or low risk of readmission.
Interviews offered insight into healthcare needs
Lina Lundgren.
To understand how a clinical decision support system like this one might be used, the research team interviewed healthcare professionals from different parts of the heart failure care pathway: nurses, physicians, home-healthcare staff and physiotherapists. The conversations clearly showed that a tool of this kind could have significant clinical value.
“A recurring theme in the interviews was the need for more support in identifying which patients require closer follow-up. A tool that provides a clear risk assessment and shows which factors carry the most weight can reduce uncertainty and make decisions more consistent among patients”, says Lina Lundgren.
Those who participated in the interviews also stressed that patient benefit must always be the priority. With better support for risk assessment, fewer patients may need repeated hospital visits – freeing up resources and reducing the uncertainty that often surrounds discharge decisions.
Staff envisioned that a well-designed tool would strengthen discharge prioritisation and help direct resources to where they are most needed. At the same time, they emphasised that AI support should be viewed as a complement to clinical judgement.
User tests indicate strong potential
Based on the interviews and their own observations of the wards, the research group developed both the machine-learning model itself and the interface that healthcare staff interact with. In small-scale user tests, the prototype was well received. It was perceived as clear, easy to understand and generally trustworthy.
“What was most appreciated were the risk factors and how they were presented. This offered tangible guidance on what to look out for – especially if something deviated from what is expected”, says Lina Lundgren.
At the same time, the testers asked for more dynamic functionality, such as the ability to click through to see additional details on the underlying components of the risk.
Although HaRP has not been implemented clinically, the project has provided valuable insights for the future development of clinical decision support systems in heart failure care. The results indicate that an AI-based tool, if developed in close dialogue with healthcare staff and based on existing systems and routines, can become a meaningful support in both discharge planning and follow-up.
“This is a complex field, but we see a clear need. With the right conditions, such a tool could make a real difference for both patients and staff”, says Lina Lundgren.
Text: Lovisa Essunger
Photo: iStock, Magnus Karlsson
