AI for Heart Disease Prediction Shows Promise, But Lacks Clinical Readiness: Review
AI tools for predicting cardiovascular disease in India show potential for personalized prevention but require more validation before clinical use. A review found these models can match or exceed traditional risk scores but lack real-world accuracy assessments.
Key takeaways
- AI tools for predicting cardiovascular disease in India show potential for personalized prevention.
- A review found these AI models can match or slightly outperform traditional risk scores.
- Current AI technologies are not ready for routine clinical decisions due to validation gaps.
- Researchers identified a lack of evidence on real-world accuracy, calibration, and clinical utility.
- Independent validation is needed before AI tools can guide treatment decisions in primary care.
Artificial intelligence (AI) holds potential for earlier identification and personalized prevention of cardiovascular disease (CVD) in India. However, current AI technologies are not yet prepared to inform routine clinical decisions, according to a systematic review conducted by researchers from the Indian Institute of Science (IISc), M.S. Ramaiah University of Applied Sciences, and the London School of Hygiene and Tropical Medicine. The findings were published in BMC Medical Informatics and Decision Making.
The review examined 30 studies published since 2017 that focused on AI models designed to predict future cardiovascular disease in adults without pre-existing heart conditions. These studies were identified through a systematic search of over 6,700 records in major scientific databases. The majority of the analyzed AI models were developed using data from the United States, the United Kingdom, and South Korea. Common machine learning algorithms utilized included Random Forests, Support Vector Machines, and neural networks, often employing routinely collected clinical information.
Relevance for India
Cardiovascular disease is a significant health concern in India, accounting for nearly one-third of all fatalities and frequently affecting younger populations compared to many other nations. While AI could enhance the precision of cardiovascular risk assessment, the current evidence base is insufficient for broad clinical adoption, stated Denny John, a faculty member at M.S. Ramaiah University of Applied Sciences and a co-author of the review. He noted that AI offers an opportunity for more precise and personalized risk prediction, but the supporting evidence remains incomplete.
Several Indian institutions have developed AI-based cardiovascular risk prediction models that integrate local factors like smokeless tobacco use, psychosocial stress, and physical inactivity. Nevertheless, Dr. John emphasized that these tools require rigorous independent validation before deployment in primary care or public health initiatives. He pointed out that while many models demonstrate strong discrimination, few studies have verified if the predicted risks align with actual outcomes across diverse populations. Substantial external validation, calibration, and assessment of clinical utility are necessary before AI tools can guide long-term treatment decisions, such as initiating therapies for high blood pressure or cholesterol.
Comparable Performance to Conventional Tools
Twelve of the reviewed studies conducted direct comparisons between AI models and established cardiovascular risk calculators, such as the Framingham Risk Score, which estimates a 10-year risk of cardiovascular events like heart attacks or strokes. The review indicated that AI models generally performed comparably to, and sometimes slightly better than, conventional tools in differentiating individuals at higher risk from those at lower risk over five to 10-year periods. However, the researchers cautioned that superior statistical performance alone does not guarantee improved patient care.
Major Validation Gaps
A primary concern highlighted in the review is the absence of evidence demonstrating whether the predicted risks accurately reflect real-world population outcomes. Almost all studies assessed discrimination, which is a model's capacity to distinguish between high- and low-risk individuals. None of the studies evaluated calibration, a measure of how well predicted disease probabilities match observed cardiovascular events. Independent patient population validation was performed in only seven studies. Sensitivity, the ability to correctly identify individuals who later develop cardiovascular disease, was reported in just four studies. Decision-curve analyses, which assess whether using AI models leads to better clinical decisions than existing methods, were not conducted in any of the reviewed studies.
Sources reviewed
Project Chintan independently synthesized and analyzed information cross-checked across the sources listed above.
Related stories
Chennai Corporation Proposes ₹25,000 Fine for Pet Abandonment
Supreme Court Mandates Cancer Notification for Early Detection in 19 States
Mumbai Doctors Warn of Leptospirosis After Monsoon Rain Exposure
