Artificial intelligence (AI) could help India identify people at risk of cardiovascular disease (CVD) earlier and enable more personalised prevention, but the technology is not yet ready to guide routine clinical decisions, according to a recent systematic review 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.

Published in BMC Medical Informatics and Decision Making, an open-access journal, the review assessed 30 studies published since 2017 on AI-based models developed to predict future cardiovascular disease among adults without established heart disease.

The researchers identified the studies through a systematic search of more than 6,700 records across major scientific databases. Most of the models were developed using datasets from the U.S., the U.K., and South Korea, and relied on routinely collected clinical information and machine learning algorithms such as Random Forests, Support Vector Machines, and neural networks.

Relevance for India

The findings are particularly relevant for India, where cardiovascular disease accounts for nearly one-third of all deaths and often affects people at younger ages than in many other countries.

Denny John, faculty of life and allied health sciences, M.S. Ramaiah University of Applied Sciences, and one of the authors of the review, said AI could potentially make cardiovascular risk assessment more precise, but the evidence was not yet adequate for widespread clinical use.

“AI offers an opportunity to make cardiovascular risk prediction more precise and more personalised. But our review shows that the evidence is still incomplete,” he said.

Several Indian institutions have already developed AI-based cardiovascular risk prediction models incorporating locally relevant factors such as smokeless tobacco use, psychosocial stress, and physical inactivity. However, Dr. John said such tools needed rigorous independent validation before they were used in primary care or public health programmes.

“Many models demonstrate good discrimination, but very few studies examine whether the predicted risks correspond to what actually happens in different populations,” he said.

He said that robust external validation, calibration, and assessment of clinical usefulness were necessary before AI tools could guide long-term treatment decisions, such as starting blood pressure- or cholesterol-lowering therapies.

Comparable with conventional risk scores

Twelve of the studies reviewed directly compared AI models with established cardiovascular risk calculators, including the Framingham Risk Score, which estimates a person’s 10-year risk of cardiovascular events such as a heart attack or stroke.

The review found that AI models generally performed as well as, and in some cases slightly better than, conventional tools in distinguishing people at higher risk of cardiovascular events from those at lower risk over five to 10 years.

However, the researchers cautioned that better statistical performance by itself did not establish that an AI model would improve patient care.

Major validation gaps

A key concern identified in the review was the lack of evidence showing whether the predicted risk accurately reflected what happened in real-world populations.

Almost all the studies assessed discrimination — the ability of a model to distinguish high-risk individuals from low-risk individuals. None assessed calibration — which determines whether the predicted probability of disease corresponds to the actual number of cardiovascular events observed.

Only seven studies validated their models using independent patient populations. Sensitivity — the ability to correctly identify people who subsequently develop cardiovascular disease — was reported in only four studies. None conducted decision-curve analyses to establish whether using the AI models would lead to better clinical decisions than existing approaches.

For India, this gap is particularly important because a model developed using populations in the U.S., U.K., or South Korea may not perform similarly in Indian populations with different risk factors and patterns of disease, Dr. John said.

Need for independent testing

The researchers have called for prospective validation of AI-based cardiovascular risk models across diverse populations, including Indian populations, before their integration into routine healthcare.

They have also recommended that future studies follow international reporting and assessment frameworks such as TRIPOD+AI and PROBAST+AI to improve transparency, reduce bias, and enable independent evaluation.

Published - August 12, 2026 06:00 am IST