A review of artificial intelligence (AI) in cardiovascular (CV) care predicts its widespread use, with the technology aiding the implementation of multimodal data at the point of care to personalize treatments and eventual outcomes. The review, which appears in the July 2 issue of the Journal of the American College of Cardiology, discusses an AI-powered future that spans novel diagnostic modalities, new digital-native biomarkers of disease, and tools evaluating care quality and prognosticating clinical outcomes. However, the review also highlights the potential pitfalls of the technology, warning of the need for equitable and regulated adoption that prioritizes fairness, equity, safety, and partnerships with innovators, communities, and society. “A key barrier is that the development and deployment of technologies aimed at improving cardiovascular health still rely on many data streams only obtained in a subset of the population,” said the paper’s authors, led by Rohan Khera, MD, from the Yale School of Medicine in New Haven, Connecticut. “While ensuring broad access to high-quality care, there will also be a need to build trust in communities that have historically mistrusted medical innovations both caused by systemic malpractices against their communities as well as challenges with health and technology literacy.” Current uses of AI in CV care One of the main talking points of the review, which was also published Monday online, was the current landscape of AI-powered innovation in cardiovascular care. Here, the technology focuses on developing tools that rely only on structured, tabular data derived from examination findings, imaging and test interpretation. The review offers a glimpse into AI’s future potential, commenting on its ability to identify sources of phenotypic variation and risk in complex signals rather than tabular data. The paper’s authors commented that with advances in deep learning models, AI algorithms can process raw, unstructured biometric signals and images, learning new representations from these data beyond what is manually inferred and encoded. They go on to cite several studies that demonstrate that an AI electrocardiogram (ECG) can screen for the risk of paroxysmal atrial fibrillation using ECGs in sinus rhythm as well as detect structural heart disease through electrocardiographic signatures. Discussing AI’s role in developing digital biomarkers of disease risk, the research team outlined current limitations that include the difficulty in accurately assessing and calculating individual risk factors in a busy clinical setting. “Digital biomarkers can be passive and computable using existing or easily obtainable clinical data and promise to enable risk-informed care through deployments in existing data ecosystems,” the paper’s authors said. “The relevant data streams are already available in many individuals before the onset of clinical disease or can be easily acquired.” AI to aid Integration Envisioning AI’s impact in this research area, the review adds that as clinical practice is inherently multimodal, AI would increasingly integrate elements from multiple distinct modalities to refine cardiovascular risk stratification. This includes composites of tabular electronic health record data with imaging, genetic determinants integrated with dynamic cardiometabolic phenotypes, environmental parameters, including fine particulate matter data, and images and videos extracted from the patient’s local environment. Looking ahead, the review highlights that key aspects of populations and disease could enable models to perform better if they are explicitly adapted to unique care settings and local populations. Deep learning models, for example, could be explicitly fine-tuned, said the research team, as they encounter more data and, therefore, could adapt to local populations. The team also pointed out the presence of key safeguards applied to data sources in these new settings that would not reflect a bias in care that was encoded into the models. However, they did acknowledge that significant resources would be needed to steward these AI models and evaluate and adapt models to local populations. Data privacy and security Other safeguards and considerations in enabling AI in cardiovascular care include the need to address data privacy and security. The team highlighted the “richness” of the data, which means that even anonymized patterns could be reidentified by malicious actors if they gained access to training databases. In equal measure, the methods of data exchange or data sharing used in the AI development pipeline were vulnerable, increasing the risk of data breaches. The review comments on the possibilities of models being exploited once released with generative AI models fed prompts that could force the release of personal health information. As preventative measures, a host of technical solutions have been developed to preserve privacy and enhance security that include model-to-data techniques, the addition of synthetic data noise, and model inversion attacks that can be conducted to identify security weaknesses. The review concludes with a prediction for the next decade that places AI at the forefront of the diagnostic, prognostic and therapeutic cardiovascular toolkit. Here, the prospect of unimodal and task-specific models will likely be superseded by task-agnostic, multimodal, and semi-autonomous systems, effectively augmenting rather than replacing human intelligence, said the review’s authors. “…As the speed of innovation continues to outpace the reflexes of the regulatory environment, the need for ethical, equitable, and trustworthy AI will need to be embedded into the development and validation processes,” they added, “and the scope of AI should expand further to include its optimal implementation within real-world, dynamic systems.” Source: Khera R, Oikonomou EK, Nadkarni GN, et al. Transforming Cardiovascular Care With Artificial Intelligence: From Discovery to Practice: JACC State-of-the-Art Review. J Am Coll Cardiol. 2023;84:97–114. Image Credit: Electro Unicorn – stock.adobe.com