Large language models (LLM) or large-scale language models based on artificial intelligence (AI), such as ChatGPT, are increasingly used by the population and in Health. Two Hospital da Luz cardiologists, José Ferreira Santos and Hélder Dores , have just published in the renowned scientific journal “Diagnostics” a study on the use and impact of LLM in the prevention of cardiovascular diseases, advancing a proposal of regulation for its safe implementation in medical practice. Published on January 26 this year, the article is entitled “Large Language Models in cardiovascular prevention: a narrative review and governance framework” . José Ferreira Santos, clinical director of Hospital da Luz Setúbal and Hospital da Misericórdia de Évora, and Hélder Dores, coordinator of Clinical Cardiology at Hospital da Luz Lisboa, performed a review of the studies published over the past decade (2015 to 2025) concerning three domains: applications used by patients, clinical applications to support medical decision, and system applications for automated data extraction, record building, and quality monitoring. These are the main conclusions: “Although LLM models empathetically generate information for patients in conformity with guidelines, they still lack the necessary specificity to provide customized advice and with no need of supervision.” “For physicians, LLM efficiently summarize clinical notes and draft documentation, but are still unreliable for deterministic risk calculation and autonomous decision making.” “Applications focused on healthcare systems demonstrate potential for identifying vulnerable populations, supporting the collection of clinical data for registries and studies, and developing tools that help stratify risk by incorporating various types of tests.” “However, the widespread use of LLM, used in an uncontrolled manner, creates the risk of clinical errors, secondary to hallucinations, outdated knowledge, and automation biases, and jeopardizes data privacy.” “LLM may help mitigate structural barriers in cardiovascular prevention but should be presently implemented solely as a sort of ‘reasoning drivers’ that might contribute, under supervision , may assist in clinical judgment, but not replace medical decision-making. “In preventive cardiology, LLM represent a realistic opportunity to bridge the existing gap between scientific recommendation and clinical practice. By processing unstructured clinical data, they may identify high-risk patients, synthesize complex clinical histories, and customize communication on health, extending specialized experience to routine care.” As guidance “on the path to follow for a responsible integration of LLM” in clinical practice, the authors propose the application of the CARDIO formula (clinical validation, auditability, risk stratification, data privacy, integration, and ongoing vigilance). The article “Large Language Models in cardiovascular prevention: a narrative review and governance framework” can be consulted at here .
