Acute malnutrition is a killer, especially for children.

The medical condition, also known as wasting, occurs when a person rapidly loses weight because of illness or a lack of food. Worldwide, there are an estimated 36 million acutely malnourished children, according to a 2024 report by the Global Network Against Food Crises. The World Health Organization blames undernutrition for nearly half of all deaths of children younger than 5.

Like much of Africa, acute malnutrition plagues Kenya, with an estimated 4% of children under 5 experiencing it, according to Laura Ferguson, director of research at USC’s Institute on Inequalities in Global Health (IIGH) and an associate professor of population and public health sciences at the Keck School of Medicine of USC. In some parts of Kenya, that number jumps to 20% to 25%, especially in the northern region bordering South Sudan, Ethiopia and Somalia. Because children have underdeveloped immune systems, they are more susceptible to diarrhea, malaria, worm infections and other diseases, but are less able to fight them off, putting them at great risk.

“Malnutrition is a public health emergency in Kenya,” said Ferguson, who along with Bistra Dilkina of the USC Viterbi School of Engineering, among others, co-leads a multidisciplinary team that has developed an AI model to predict malnutrition rates around the country. “Children are sick unnecessarily. Children are dying unnecessarily.”

Acute malnutrition, Ferguson added, costs Kenya about 7% of its annual gross domestic product, or GDP, a staggering amount considering the country’s heavy debt load.

Climate change threatens to exacerbate this already dire situation. That’s because extreme weather like droughts damages crops and destroys grazing land for cattle, thereby increasing food insecurity. Similarly, Ferguson said the now unpredictable rainy season makes it nearly impossible for farmers to know when to plant fruits, vegetables and grains.

Against this backdrop, Kenya does its best to predict which areas are most likely to experience an upsurge of acute malnutrition. It does so to give government officials and nonprofits time to rush food, nutritional supplements and medical personnel to potentially affected areas to stave off malnutrition or mitigate its impact. Unfortunately, current efforts often fall short.

“They’ve been using their best guesses based on historical knowledge of malnutrition,” Ferguson said. “And part of the challenge is that the government and its partners don’t always have a good sense of where the hot spots are going to be, of where they are going to see spikes in malnutrition amongst children.”

The national motto of Kenya is “harambee,” a Swahili word that means “let us all pull together.” The USC team and its partners have done precisely that.

A coalition co-led by Ferguson; Dilkina, who also serves as co-director of the USC Center for Artificial Intelligence in Society (CAIS); the Microsoft AI for Good Lab; the nonprofit Amref Health Africa; and the Kenyan Ministry of Health has come up with a different approach. Together, they have created an AI-powered model that predicts when and where child malnutrition is likely to occur in Kenya in the next one to six months. Their tool integrates health data and satellite imagery, making it far more effective than the government’s efforts, which incorporate no statistical modeling or machine learning.

“By using data-driven AI models, you can capture more complex relationships between multiple variables that work together to help us predict malnutrition prevalence more accurately,” Dilkina said.

 

A forecasting tool that saves lives

The AI-powered predictive model, co-developed by USC researchers, considers exponentially more variables than the government’s more basic tool, which relies on historical patterns over recent months to forecast malnutrition among children in Kenya. The machine learning tool uses governmental health data stored on Kenya’s District Health Information System 2, or DHIS2, a centralized health records system. It also incorporates satellite data from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) about crop health and productivity.

The DHIS2 tracks information from more than 17,000 health facilities in Kenya, making it incredibly rich and robust. Data includes the number of children under 5 who are underweight, wasted and in supplemental feeding programs. It also considers other health indicators with potential predictive value such as infant feeding practices, diarrhea treatment and pregnant women with low hemoglobin levels, said Ferguson of the Keck School.

A simple baseline model that creates one-month projections of where child malnutrition will occur based only on historical trends makes predictions with accuracy rates of about 76%, while the coalition’s AI-powered model that incorporates other data features improves this to 89%. Similarly, over a six-month time horizon, the baseline accuracy decreases to 73%, while the AI model reaches 86%.

The coalition’s AI model, Dilkina added, also does a much better job predicting which sub-counties with current child malnutrition rates under 15% will soon experience dramatic increases.

“It’s very hard to predict areas where they are transitioning from low to high malnutrition rates, but we can do it fairly well,” Dilkina said.

And that knowledge can save children’s lives.

When government officials and nonprofits have advanced warning of impending severe malnutrition, they have time to act, Ferguson said. In addition to emergency food, water and medical shipments, they can also send nurses to local health clinics in potentially impacted regions to screen and care for children and others showing early signs of acute malnutrition. They can also better coordinate with local government leaders to ensure that children and others in need receive the care they need.

“It’s extremely rewarding and inspiring to be able to use our technical skills to solve a problem that actually can change lives,” USC Viterbi’s Dilkina said.

 

It takes a village

Hillary Clinton once said, “It takes a village.” That has certainly been the case for the Kenyan predictive AI model for acute child malnutrition. A multidisciplinary team of academics, nonprofit leaders and government officials worked together to develop the tool, which they have successfully tested nationwide.

Keck’s Ferguson had the initial idea to create it and tapped USC Viterbi’s Dilkina for her technological expertise. The pair applied for and received a grant from the Microsoft AI for Good Lab. In addition to making its Azure cloud resources available to securely store data, Microsoft — working with Dilkina and her team — developed and tested the predictive AI model.

Ferguson also reached out to Amref Health Africa, a nonprofit with decades of experience operating in Kenya and on-the-ground knowledge of community needs. Amref, Ferguson said, has also served as the “linchpin” in the coalition’s dealings with the Kenyan Ministry of Health, helping the partners gain access to the government’s DHIS health and other data.

“One of the things that’s very clear to me is that most global health problems cannot be solved within the health field alone, and this is one of them,” Ferguson said. “So, we absolutely need public health officials. We need medical officials. We need nonprofits. We need engineers. If you take out any single partner, it just doesn’t work and won’t have the impact that we hope for.”

Now, the partners have created a prototype dashboard. Such a tool would allow nongovernmental organization (NGO) and government officials to quickly visualize areas experiencing or likely to soon experience acute malnutrition among children, giving them time to move or pre-position lifesaving resources quickly.

The dashboard must still be integrated into AMREF’s and the Kenyan Ministry of Health’s computers, according to USC Viterbi’s Dilkina, adding that she hopes that will happen in the near future.

Girmaw Abebe Tadesse, principal scientist and manager at the Microsoft AI Good Lab in Nairobi, Kenya, said he believes the predictive AI tool will make a difference.

“Personally, I think this project is important, as malnutrition poses a significant challenge to children in Africa, a continent that is facing a major food insecurity exacerbated by climate change,” he said.

Going forward, USC’s Ferguson and Dilkina believe that the Kenyan government could extend the predictive AI model to forecast other health conditions besides child malnutrition, including perhaps infectious disease outbreaks and climate change-related health impacts.

Additionally, an estimated 125 countries collect similar health data using DHIS2, including about 80 low- to middle-income nations. That means the coalition’s machine learning-powered tool could one day predict acute child malnutrition spikes around the world.

“If we can do this for Kenya, we can do it for other countries,” Dilkina said.

“Hopefully, this can inspire others to also look for ways in which they can use AI to benefit society,” she added. There are just so many different ways in which you can do it right. The sky’s the limit when there is a genuine commitment to work in partnerships.”