AI Predicts Cardiac Arrest Recovery in Low-Resource Care

After a cardiac arrest, families and doctors are often faced with agonising uncertainty about a patient’s chances of recovery. This uncertainty is even greater in hospitals with limited resources, where access to advanced diagnostic tools and large datasets is constrained.

In one example of how artificial intelligence (AI) can help bridge this gap, researchers from Duke-NUS Medical School and their collaborators have adapted an advanced AI model to accurately predict neurological recovery after cardiac arrest in a resource-limited setting.

Published in npj Digital Medicine, the study used transfer learning, an advanced AI approach that adapts pre-trained models built on large datasets, to new settings with limited local data. This method improves performance in new environments without requiring extensive data collection, making it particularly relevant for low-and-middle-income countries.

The team took a brain-recovery prediction model built in Japan using data from 46,918 out-of-hospital cardiac-arrest patients and adapted it for use in Vietnam, where it was tested on a smaller cohort of 243 patients.

The adapted model performed significantly better, correctly distinguishing high‑risk from low‑risk patients about 80 per cent of the time, compared with around 46 per cent when the original model was used in the Vietnam context.

Expanding AI’s role in limited-resource settings

Beyond outcome prediction, AI also holds promise across a wide range of healthcare applications in low-and middle-income countries. In a separate study published in Nature Health, Duke-NUS researchers and their collaborators, including counterparts from University College London (UCL), examined how large language models (LLMs) – trained on vast volumes of text to understand and generate human language – could be applied to advance global health. In resource-constrained environments, they have the potential to improve access to care, diagnostics and clinical decision-making.

Examples highlighted include a chatbot providing pregnancy-related information to expecting mothers in South Africa and smartphone-based applications used by community health workers in Sierra Leone to detect malaria infections from blood smear samples, a more cost-efficient method than conventional microscope-based systems.

Despite these innovative advances, the researchers noted that AI development and deployment remain heavily concentrated in high-income and upper-middle-income settings.

While 63 per cent of surveyed researchers, clinicians and service providers report actively using AI tools, many low-and-middle-income countries continue to face barriers such as limited infrastructure, insufficient expertise and a lack of existing local knowledge on addressing these gaps.

Charting the path forward: governance and guardrails

While AI tools have the potential to improve healthcare delivery, appropriate governance frameworks are essential for safe and ethical implementation. Existing regulations for medical technologies often do not address AI-specific risks, including privacy concerns and model hallucinations, nor do they clearly refine accountability for safe deployment and oversight of new tools.

To address these gaps, researchers led by Duke-NUS have proposed the creation of an international consortium – the Partnership for Oversight, Leadership, and Accountability in Regulating Intelligent Systems-Generative Models in Medicine (POLARIS-GM).

The consortium aims to establish actionable best-practice guidance for regulating new tools, monitoring their impact, establishing safety guardrails and adapting them for resource-limited settings. Bringing together healthcare leaders, regulators, ethicists and patient groups worldwide, POLARIS-GM will adopt a phased approach, beginning with a review of existing research before working towards global consensus on AI governance in healthcare. 

Reference: Li S, Okada Y, Gu W, et al. Leveraging AI and transfer learning to enhance out-of-hospital cardiac arrest outcome prediction in diverse setting. npj Digit Med. 2025;8(1):716. doi: 10.1038/s41746-025-02088-x

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source. Our press release publishing policy can be accessed here.