Year: 2025
Type: article
Abstract: Karim Lekadir is a research professor at the Catalan Institution for Research and Advanced Studies, (ICREA) of the University of Barcelona, Spain. He investigates new data science techniques for trustworthy and ethical artificial intelligence (AI) in healthcare and recently took part in the European Society of Cardiology (ESC) Spring Summit and Cardiovascular Round Table—which discussed the importance of building trust in AI for healthcare. Prof. Lekadir shares some important insights from these major strategic meetings (Figure 1). Professor Karim Lekadir, Research Professor at the Catalan Institution for Research and Advanced Studies, (ICREA) of the University of Barcelona, Spain Trust is the foundational element in the successful integration of AI into healthcare. When it comes to medical decisions, especially those impacting life and death, clinicians and patients must feel confident in technology. For AI to be embraced in healthcare, it must not only be effective but also safe, fair, and transparent. Without trust, AI tools are unlikely to be used properly, potentially leading to suboptimal or even harmful outcomes. Think about the COVID-19 vaccine rollout—while millions trusted the science, others, like tennis star Novak Djokovic, declined the vaccination due to a lack of trust. In Djokovic’s case, he was excluded from a Grand Slam tournament. For others, that lack of trust may have cost them their lives. This illustrates that AI tools in healthcare need to build good levels of trust to be widely accepted and adopted. As with any major change, there are challenges, but we can look at these through the paradigm of six characteristics that define what makes AI trustworthy. Fairness: It is important that AI works equally well for all patient groups. Unfortunately, some AI tools have shown bias, for example heart failure prediction models that do not perform as well for certain racial or ethnic groups. It is up to developers to actively address these biases and make sure their tools are fair across the board. Universality: AI tools in healthcare must be developed and validated to function effectively across different clinical settings regardless of the local equipment or geographic location—in big cities or rural regions—to enhance access to AI-powered healthcare. Traceability: Even if an AI tool is performing well initially, there is always the risk it might lose accuracy over time. This could happen because of changes in the data, the patient population, or even the way diseases are defined. A good example of this is how AI models struggled with new COVID-19 variants—they were trained on old data and did not adapt quickly enough. So, ongoing monitoring is key to keeping the tools on point and accurate. Usability: AI tools must be user-friendly within clinical workflows to avoid mistakes and ensure clinicians can use them effectively. Robustness: Healthcare data can be messy and inconsistent. For example, when you are dealing with something like an ECG, you might get disruptions from things like patient movement. Artificial intelligence needs to be able to handle all these variations to consistently deliver accurate results. Explainability: For AI to really gain trust, clinicians need to understand how it is making decisions. If the system is too opaque or does not provide clear reasoning behind its results, it risks being dismissed as unreliable. The more transparent it is, the more likely clinicians are to trust and adopt it. We have developed an international consensus guideline for trustworthy and deployable AI in healthcare called FUTURE-AI.1 This includes 30 best practices addressing technical, clinical, socio-ethical, and legal dimensions. The recommendations cover the entire lifecycle of healthcare AI, from design, development, and validation to regulation, deployment, and monitoring. However, the key to making any framework for trustworthy AI effective is collaboration at each stage. For example, at the design stage, we recommend involving diverse stakeholders—clinicians, patients, ethicists, and health economists—to co-develop the AI tools. Clinicians in particular play a vital role in shaping AI, contributing insights on clinical relevance, application, and data biases. In the evaluation stage (where the AI tool is tested for accuracy, fairness, usability, and robustness), feedback from end-users (e.g. clinicians and patients) is gathered to ensure the tool is user-friendly and effective. In the EU funded AI for Heart Failure (AI4HF) project, we aimed to develop AI tools for diagnosing heart failure. In this project, clinicians, patients, and regulators were involved in the design phase, ensuring that the tool met both clinical needs and patient preferences. In Peru, the team discovered that altitude, rather than ethnicity, was a critical factor for bias which would impact AI-driven heart failure predictions. This was an important insight that might have been overlooked by the core project team, without the local stakeholders’ input. Engaging these diverse perspectives helped the project avoid biases and ensure the AI tool would work universally. Healthcare practices, disease definitions, and even risk factors can vary significantly depending on where a person lives (like the altitude factor in Peru) and among different healthcare systems. When developing an AI tool for cardiac MRI analysis, we used data from hospitals in Spain, Germany, and Canada. We ensured generalizability by using data from multiple medical imaging manufacturers and healthcare environments. Additionally, we implemented a continuous monitoring system to track performance over time and detect any potential performance drift. By including diverse regions in AI development, we can ensure that these tools are universally applicable and do not favour one population over another. It is about building trust in the tools’ fairness and applicability. With a solid framework in place, AI can transform healthcare by delivering robust, fair, and transparent solutions that enhance accessibility, enable personalized care, and ultimately improve patient outcomes. However, for AI to reach its full potential, it must earn the trust of all stakeholders—clinicians, patients, and the wider healthcare community. Interested in all things AI? Join us at the ESC Digital Health and AI Summit in November 2025 –and stay connected through the ESC’s AI Hub (www.escardio.org/AI-Hub). Nothing to declare.
Source: European Heart Journal
Author: Judith Ozkan
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W4413370856