AI Predicts Recovery After Cardiac Arrest in Poor Regions

When a patient's heart stops beating, families face an agonizing wait to learn if their loved one will recover. In hospitals with limited resources, that uncertainty is even harder because doctors lack the advanced tools that could help them predict outcomes.

Researchers from Duke-NUS Medical School just changed that equation. They adapted an AI model originally built in Japan using data from nearly 47,000 cardiac arrest patients to work in Vietnam, where they tested it on just 243 patients.

The adapted model worked remarkably well. It correctly identified high-risk versus low-risk patients about 80% of the time, compared to only 46% accuracy when they tried using the original Japanese model in Vietnam without adaptation.

The secret lies in a technique called transfer learning, which adapts existing AI models to new environments without requiring massive local datasets. This approach saves time, cuts costs, and most importantly, extends cutting-edge medical technology to places that need it most.

Associate Professor Liu Nan, who led the study published in npj Digital Medicine, explains the significance simply. "AI models do not need to be rebuilt from scratch for every new setting," she said, noting this method can help health care systems with fewer resources.

The same team looked beyond cardiac care in a separate Nature Health study. They examined how large language models could transform health care across low and middle-income countries through applications like pregnancy chatbots in South Africa and smartphone apps detecting malaria in Sierra Leone.

The Ripple Effect

The implications stretch far beyond a single hospital or country. About 63% of health care workers surveyed report using AI tools, but many developing nations still face barriers like limited infrastructure and insufficient expertise.

Duke-NUS researchers are tackling these challenges head-on by proposing POLARIS-GM, an international consortium focused on creating safety guidelines and best practices for AI deployment in resource-limited settings. The group will bring together health care leaders, regulators, ethicists, and patient advocates from around the world.

Dr. Ning Yilin, a co-author of the study, emphasizes that empowering people matters most. "Strengthening digital literacy and building confidence in using these tools will ensure AI supports, rather than disrupts, the workforce," she said.

The researchers acknowledge concerns about privacy and accountability that come with AI in medicine. But they're not waiting for perfect conditions—they're building the frameworks needed to make these tools safe, effective, and accessible where they're needed most.

Millions of people in under-resourced regions could soon benefit from diagnostic tools once available only in wealthy countries, bringing hope to families facing their darkest hours.