The American Heart Association (AHA) issued the following scientific statement: Circulation About the use of artificial intelligence (AI) or machine learning (ML) tools to improve heart disease outcomes.

Over the past decade, academic institutions, industry, and government organizations have invested resources to incorporate AI into healthcare delivery with the goal of improving patient outcomes. However, this is not happening on a large scale.

This scientific statement describes the best practices for using AI in clinical settings, with a particular focus on image processing, electrocardiograms (ECGs), in-hospital monitoring, implantable and wearable technologies, genetics, and electronic health record (EHR) interpretation. Identify practices, gaps, and challenges.

Imaging is an essential tool for diagnosis and clinical decision-making in cardiovascular (CV) medicine. Becoming an expert in image interpretation takes years and talent is in short supply, further exacerbating inequalities in quality health care. Therefore, the use of AI/ML has the potential to address many of the gaps in imaging care delivery.

The use of AI in ECGs is already impacting care delivery by automating interpretation, enabling scaling, and identifying subtle and nonlinear patterns in ECG output that are invisible to experts. However, clinical validation in large and diverse populations is still required to reduce the risk of overfitting and uncertainty bias.

Traditional in-hospital patient monitoring approaches rely on alarms set to respond to predefined thresholds and assigning scores to vital signs. However, subtle changes in physiological signals may be involved. Using AI to monitor patients reduces false alarms and detects or predicts clinical deterioration, sepsis, hypotension, cardiac arrest, and postoperative atrial fibrillation (AF) earlier than using current standard of care. It may be possible.

AI/ML-based methods can be used with implantable and wearable devices that collect continuous data. For example, multiple studies have used data from wearable devices to predict worsening of atrial fibrillation and heart failure in the short term.

In genetics, multiple disease risk alleles identified in genome-wide association studies have been used to formulate polygenic risk scores (PRS). Combining clinical information from PRS and EHRs has the potential to provide precision medicine in CV medicine using AI/ML approaches.

Data-mining EHRs using AI have been used to predict in-hospital mortality, major CV adverse events, hypertension, and ischemic stroke. However, AI models perform only as well as their input data, and there is a risk of bias as EHR data requires curation and is often missing data.

The authors conclude their statement with a framework for successful AI implementation in CV medicine. We also highlight ethical concerns regarding AI, consider appropriate data handling, identify potential sources of bias in AI, and discuss liability concerns and cybersecurity.

The authors of the statement concluded: “The American Heart Association aims to advance cardiovascular health for all people, including by identifying and removing barriers to access and quality of health…creating manageable and cost-effective workflows. There is an urgent need to develop implementation science for AI/ML tools to address core unmet clinical (or translational) needs in AI/ML-based precision medicine, and the evidence is This process should organically incorporate the need to avoid bias and maximize the generalizability of findings to avoid perpetuating existing health care disparities. .”

Disclosure: Some study authors declared affiliations with biotechnology, pharmaceutical, and/or device companies. Please refer to the original reference material for a complete list of disclosures.