Only 0.29% of AI studies in cardiovascular medicine were prospective trials, with 95% meeting primary endpoints, mostly single-country studies in Asia, Europe, and North America.
Do AI-tools improve diagnosis and outcomes in cardiovascular medicine compared to standard of care?
Despite rapid development of AI models in cardiovascular medicine, very few have undergone rigorous prospective clinical evaluation, highlighting a major gap in evidence for clinical implementation.
Abstract Background Artificial Intelligence (AI) holds major promise to improve cardiovascular care. The use of AI-tools in medicine has seen remarkable growth in the last decade. However, it is unclear how many AI-tools for cardiovascular care have been implemented into practice or prospectively evaluated in randomised clinical trials. We performed a systematic review to understand the AI landscape in cardiovascular medicine. Purpose To i) identify prospective clinical implementation studies and randomised controlled trials evaluating the performance of AI-tools in routine clinical care, ii) map the characteristics of these trials, and iii) identify gaps in the use of AI-tools in cardiovascular medicine and identify areas for future research. Methods We systematically searched PubMed, Scopus, CENTRAL, and the International Clinical Trials Registry Platform for studies in English published between January 1st 2018 and January 24th 2025. Abstract screening, selection of full-text articles and data extraction was carried out independently by three investigators. Data extraction was performed for study-level information, including location, participant characteristics, primary endpoint, comparator, as well as the type and origin of the AI used. We followed the recommendations of the PRISMA-ScR guidelines for scoping reviews. Results Our search identified 8,177 studies of interest, resulting in 6,842 records after deduplication. Following title and abstract screening, 128 articles were retained for full-text review. Of these, 108 were excluded, resulting in 20 (0.29%) articles (Figure 1). Thirteen (65%) were randomised controlled trials and 7 (35%) were prospective implementation studies, with 11 (55%) using a superiority and 9 (45%) a non-inferiority design. All trials were conducted in a single country, with Asia conducting most trials (8 40%), followed by Europe (6 30%), North America (5 25%), and Africa (1 5%) (Figure 2). Overall, study populations comprised a median of 1,509 patients (IQR 702–9,730). Subspecialties included cardiac imaging (8 40%), electrophysiology (5 20%), interventional cardiology (4 20%), acute cardiovascular care (2 10%), and to provide decision-support for medical therapy (1 5%). The evaluated AI-tools were compared to standard of care, for example how AI helps to diagnose myocardial infarction earlier. Endpoints focussed on the diagnosis or prognosis of cardiovascular conditions, such as heart failure or atrial fibrillation. The primary endpoint was met in 19 studies (95%). Conclusion Despite the rapid development of AI-models in cardiovascular medicine, the minority undergo clinical evaluation in prospective clinical trials using rigorous methodology. While AI-tools are often hailed as silver bullets, our review highlights the need of prospective international trials to evaluate whether care guided by AI-tools improves diagnosis and outcomes for patients across different health care systems.PRISMA diagram Study types, location and subspecialties
Péquignot et al. (2025) studied this question. Only 0.29% of AI studies in cardiovascular medicine were prospective trials, with 95% meeting primary endpoints, mostly single-country studies in Asia, Europe, and North America.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: