<p>Cardiovascular disease (CVD) remains a leading cause of global morbidity and mortality. This has driven growing interest in automated methods for improving cardiovascular risk prediction. While data-driven machine learning techniques show clear advantages over traditional statistical methods, they also introduce ethical challenges in high-stakes settings; accordingly, we ground our work in AI ethics principles, emphasising digital equity and embedding active bias mitigation strategies throughout the entire model lifecycle. Remarkably, such systems risk perpetuating biases against specific patient groups, particularly across ethnicity, gender, and socio-economic status. In this study, as a use case, we examine mortality prediction in CVD using an array of artificial intelligence models, from traditional machine learning to deep learning architectures such as Transformers and Large Language Models. We conduct a thorough and systematic evaluation of these models to identify potential sources of biases that may disadvantage specific groups. To address this, we adopt de-biasing techniques at every stage of the prediction algorithm’s design, pre-processing, in-processing, and post-processing, aiming to improve balanced data distribution and fair model predictions across all patient groups. Our experiments using electronic healthcare records from the MINAP dataset reveal biases affecting specific patient groups, including some unexpected disparities. By exploring methods to mitigate biases and improve fairness metrics compared to baseline models, our study demonstrates the potential to enhance trust among healthcare professionals in automated decision-making tools and equitable triage, ultimately contributing to improvements in the healthcare sector.</p>