AI tool targets death data gaps in poor regions

[NAIROBI, SciDev.Net] Researchers have unveiled an innovative tool that uses Artificial Intelligence (AI) to provide accurate cause of death data where resources are limited.

The Cause of Death Determination Ascertainment (CODA) project, a three-year Gates Foundation-funded initiative, aims to improve mortality data in low-income countries where only eight per cent of deaths have documented causes.

The goal is to support health professionals in accurately assigning cause of death using WHO standards and ultimately inform public health policies to address these causes.

Philip Setel, vice president for civil registration and vital statistics at Vital Strategies, a global health organisation leading the consortium of partners behind the initiative, says a high percentage of deaths in Sub-Saharan Africa and large parts of Asia occur outside of health facilities, with little opportunity to derive an accurate cause of death.

“It is an enormous blind spot,” Setel told SciDev.Net.

“In Africa only about one in ten deaths are even registered to begin with, according to the World Health Organization. The proportion of those one in ten that have a reliable cause of death is even smaller.

“The gap is also prevalent in parts of Asia, where cultural preferences for home deaths complicate data collection.”

Setel says the innovation is an opportunity for countries to combine AI with proven public health approaches such as civil registration and vital statistics systems, to enhance how cause of death is determined, even in cases with limited or no clinical information. This could lead to a better understanding of why people are dying and how such deaths can be prevented.

Data inputs

The tool, which draws on historical data to train the AI algorithms, can be deployed both in clinical and community settings. If necessary, it can work offline, uploading data later when connectivity is available.

In community settings, details from post-mortem interviews with family members, conducted by community health workers, can be combined with information about the deceased such as age and gender, and any available data on common causes of death in the area, such as malaria.

The CODA tool manages the interview in real time, dynamically adjusting suggested questions as information emerges from the conversation. As the interview unfolds, the AI converts spoken testimony into structured data, so that described symptoms are translated into terms the cause of death algorithm can process, Setel explains.

In a health facility setting, physicians familiar with the case can input clinical observations, such as patient history, observations and test results prior to death.

Rather than producing a final cause-of-death verdict, CODA will indicate a level of confidence in its recommendation, allowing humans to make an informed decision.

Together with Northeastern University, the University of Washington, IS Global, RTI International, and the CHAMPS Project Office, Vital Strategies is set to conduct limited trials in South Africa and Bangladesh starting in September.

Two university consortium partners are currently using historical data to train and test the AI models.

“The dataset that is being used to train the AI models consists of rigorously validated deaths in low- and middle-income countries where causes of death have been confirmed through post-mortem investigation,” Setel explains.

“This ensures that models are trained on cases with a high-quality diagnostic standard against which the AI’s performance can be evaluated and measured.”

He says while autopsies and post-mortem examinations in health facilities offer the highest-quality data, they are costly and limited in scale.

Informing policy

Beyond its clinical applications, the initiative aims to equip governments to allocate health resources.

Mary-Ann Etiebet, president and chief executive officer of Vital Strategies, says cause of death data is essential for effective policymaking.

“Less than three per cent of global health financing is spent on non-communicable diseases, despite their significant burden,” Etiebet told SciDev.Net.

“The lack of accurate data hinders policymakers’ ability to allocate resources effectively.”

She highlighted the ethical considerations around sensitively interviewing families about deceased loved ones, as well as legal considerations such as the ambiguity of AI’s status within existing legal frameworks.

Etiebet says a scientific advisory committee will be formed to address such issues, adding: “The goal is to ensure the tool is ethical, legal, and culturally sensitive.”

Laura Ferguson, director of research at the University of Southern California’s Institute on Inequalities in Global Health, observes that a lot of work is underway on the digitalisation of civil registration and vital statistics.

“As always, we have to start with the problem—in this case insufficient data about who is dying, where and why—and assess what tools might be able to help,” says Ferguson who is involved in the CODA initiative.

“In this case, AI might usefully be deployed to improve understanding of causes of mortality, which allows governments to allocate resources to evidence-based interventions that might help reduce mortality.”

Ferguson stressed the importance of safeguards, including transparency, data protection, government engagement, and collaboration with end users. She added: “AI can be a wonderful tool but how it is deployed remains critical to its real-world effectiveness.”

This piece was produced by SciDev.Net’s Global desk.