A technician uses AI in cervical cancer screening in China in 2023 as part of a pilot. The method, which has shown promise, is one of many ways health researchers are exploring AI.
Photo courtesy STR/AFP/Getty Images
“There is a big interest in the larger scientific community for these tools.”
— Carlos Saldana
First in a series on applying AI to benefit public health
Across the U.S., public health is quietly undergoing a technological revolution. From Chicago-area food safety inspectors using algorithms to predict restaurant violations to tribal communities in the Southwest deploying machine learning to prevent suicides, artificial intelligence is transforming work to protect and improve health.
The transformation comes at a critical time. With approximately 60% of the public health workforce expected to retire or leave the field and federal funding cuts looming, health departments face unprecedented challenges in maintaining essential services while confronting emerging health threats. The solution, according to a growing chorus of researchers and practitioners, lies in leveraging artificial intelligence to amplify human expertise and detect problems before they escalate into crises.
Public health leaders say this fundamental shift from reactive to predictive public health represents more than technological advancement: It offers the potential to save lives, optimize limited resources and address health disparities with unprecedented precision.
Photo courtesy Tadamichi, iStockphoto
Amid public health staffing and funding shortages, the appetite for learning how to use AI is strong, said Tatiana Lin, MA, director of business strategy and innovation at the Kansas Health Institute.
“AI differentiates from other technologies because it’s really based on not telling a machine or computer what to do; it tells the computer how to learn,” Lin told The Nation’s Health. “It uses those learnings to improve itself and make predictions based on all the information it’s learned.”
Since 2023, Lin and her colleagues have conducted more than 30 trainings around the country to introduce AI to public health workers. Sessions cover everything from defining AI in lay terms to exploring how it can be applied to surveillance and research. Now, with technology maturing rapidly and tools proliferating almost daily, the efforts are intensifying.
At the federal level, real-world AI applications stepped into the public health spotlight during the early years of the COVID-19 pandemic when the U.S. Centers for Disease Control and Prevention used natural language processing to analyze unstructured text data — from social media posts to COVID-19 policy documents — to allow researchers to assess public sentiment, analyze reports of symptoms, identify misinformation and review thousands of policy articles in record time. The task would have taken months using manual methods.
At the local level, AI is being applied to critical operations such as food safety and restaurant inspections. For example, using more than 10 years of data, the Chicago Department of Health used SAS’s Viya 4.0 — a platform for data analytics, data management and visualization — to review and analyze 92,000 free-form statements from 11,000 restaurant inspection reports.
Public health researchers are using artificial intelligence to collect and analyze data, such as health- and disease-related comments posted by the public on social media platforms.
Photo by PeopleImages, courtesy iStockphoto
The project allowed inspectors to prioritize visits to higher-risk restaurants and better focus their attention on likely violations. Manually going through the reports to identify the main issues mentioned would have taken about 7,700 hours, or four full-time employees working for one year, as compared to a week, according to SAS principal solutions architect Tom Sabo, who shared the project’s success story at APHA’s 2023 Annual Meeting and Expo.
In California, the Contra Costa Public Health department near San Francisco partnered with Stanford University in 2022 to develop a tool called COVID Fast Fax. The software flags the most urgent new faxes using machine learning algorithms. As the health department was flooded with faxes during the height of the pandemic, the system processed and categorized thousands of handwritten case reports, allowing workers to triage high-risk cases quickly. Collaborators have since released the code and methodology for researchers or health departments interested in duplicating the tool.
Health departments are also exploring tools such as Prepper AI, focused on emergency preparedness and disaster recovery. Built with a public health lens, the platform supports planning and resource allocation in the face of increasing natural and human-made disasters.
Tribal public health programs are benefiting from AI tools as well, especially when the tools are paired with culturally grounded strategies. One of the most notable examples comes from the White Mountain Apache Tribe in Arizona, where a collaboration with Johns Hopkins Bloomberg School of Public Health’s Center for Indigenous Health led to the development of an AI-driven suicide risk identification model.
Analyzing electronic health records, the system flagged people at high risk for suicide-related events such as suicidal ideation, self-harm or substance abuse crises. Once identified, the patients received followup from community-based mental health teams.
The approach significantly improved early detection and intervention, particularly in rural or under-resourced areas, said Emily Haroz, PhD, MHS, MA, associate professor in international and mental health at the center.
In Georgia, public health researchers created a machine learning tool that could revolutionize HIV prevention strategies.
Photo by Cofotoisme, courtesy iStockphoto
“We found that it not only continued to be valid and add benefit in terms of identifying the highest risk people, but also helped ensure that those at highest risk were reached with care,” Haroz told The Nation’s Health. “And then, also among those at highest risk, we saw a reduced risk for another suicide-related event.”
That success prompted expansion into three Indian Health Service clinics, where the team is now conducting a clinical trial.
“We homed in on making it a model or a tool that doesn’t replace human judgment and interaction, but kind of is that safety light and you change lanes,” she said. “It’s like a reminder you may be missing something in your blind spot.”
Similarly, researchers in Georgia analyzed a decade of STI data from Fulton County, training their algorithm to identify key risk factors associated with HIV diagnosis. The model examines multiple variables, including the number of previous STIs, diagnostic locations, patient age and social vulnerability indices. The comprehensive analysis enables public health officials to predict who is at highest risk of acquiring HIV and prioritize health interventions.
“There is a big interest in the larger scientific community for these tools,” said Carlos Saldana, MD, an assistant professor of medicine at Emory University who specializes in HIV and STI implementation science.
Working in partnership with the Georgia Department of Public Health’s Division of HIV surveillance, Saldana’s team created an innovative machine learning tool that could revolutionize HIV prevention strategies. The goal is to prioritize tailored interventions for people who need it the most.
“We have understaffed public health workers and contact tracers, so how can we use this technology to help us prioritize who to reach and who to offer testing?” Saldana told The Nation’s Health.
The research has generated significant interest in the scientific community, and the team plans to validate the model in multiple jurisdictions using real-time data.
Using AI ethically crucial to field
Implementing AI in public health surveillance raises important considerations about bias and community engagement, said Saldana, who noted that AI systems can perpetuate existing biases present in historical data. For example, inadequate data collection on transgender populations limits the model’s applicability to the communities, requiring intentional efforts to address the gaps.
An AI program called COVID Fast Fax helped a California health department process thousands of handwritten disease reports, allowing workers to quickly triage high-risk cases.
Photo by PixDeluxe, courtesy iStockphoto
Fearful that those gaps can distort research and trust, public health must be present at the AI policy table, said public health strategist and educator Ashley S. Love, DrPH, DHSc, who cautioned about the risks, including AI bias and “hallucinations,” in which systems make up facts. And while the pandemic fast-tracked AI adoption in public health, governance has not caught up, Love said.
“We need to make sure that public health professionals have enough AI literacy so that we can be at the table to make policies,” Love told The Nation’s Health. “If public health isn’t at the table, there will be selection bias, and not all perspectives, views and scenarios could be thought of.”
A seat at the policy table will ensure the ethical, equity-driven perspective public health demands, said Love, a biostatistics and epidemiology professor who formerly served as Delaware’s state epidemiologist.
With public health workforce departures looming large, Love sees value in using generative AI to help personalize education around AI for people who do not have the resources.
When she was first introduced to OpenAI’s ChatGPT years ago through virtual collaborations, Love seized on the tool’s ability to translate technical code and statistical concepts into plain language. She has since used AI tools like ChatGPT, Claude and Canva AI to help students and public health professionals bridge knowledge gaps.
But educators themselves might be surprised by the way their students use the tools, said Abraham Flaxman, PhD, associate professor of global health at the Institute for Health Metrics and Evaluation at the University of Washington. When Flaxman first encountered ChatGPT in 2022, his first impression was not that it could transform his global mortality surveillance research.
“I said, ‘This is a cheating machine,’” Flaxman recalled, laughing. “Every one of my students could easily get an answer to every one of my lazy assignments. But then ChatGPT took over the world, and everyone in public health started wondering: Is this a threat, or is it an opportunity?”
Collecting data on deaths with AI
Flaxman has since moved from skeptical educator to pioneering investigator of AI’s usefulness to address one of global public health’s most persistent blind spots: the lack of reliable information about how people die.
Globally, nearly half of all deaths occur without an official death certificate. That leaves massive data gaps that hinder health planning, disease prevention and resource allocation. While many countries rely on medical examiners and pathologists to review and certify causes of death, such systems are expensive and logistically unfeasible in many parts of the world.
A long-standing workaround has been the “verbal autopsy,” Flaxman said, which is a method that has existed for more than 50 years. In cases where no medical professional was present at the time of death, interviewers ask a family member or caregiver a series of structured questions about the deceased’s symptoms and circumstances leading up to their death. The interviews, which typically take 20 to 40 minutes, are designed to collect enough detail to infer a likely cause of death.
In the past, interpreting the results of verbal autopsies required trained physicians familiar with local disease patterns, an approach that has always faced constraints in time, cost and scalability. Doctors in low-resource settings are often overwhelmed with everyday demands from patients, leaving little capacity to analyze what led to deaths that occurred weeks or months earlier. Now, thanks to generative AI, that process can be automated.
Flaxman is not yet flooded with inquiries from health departments about using AI in mortality surveillance and epidemiological modeling, but interest is growing — especially among researchers.
“I’m trying to get the technology in place and get the word out that it’s possible,” he told The Nation’s Health.
Artificial intelligence tools are evolving so quickly that researchers must constantly reassess which platforms are best suited to their work, Flaxman said. There is no single tool that will stay on top for long.
“They’re always improving these things, and it’s sometimes hard to tell which is the latest and greatest,” Flaxman said.
Lin and Love are among the many public health professionals who will be presenting on their work with AI during APHA’s 2025 Annual Meeting and Expo in November. Presentations on AI will address topics such as wastewater surveillance, police bias, pesticide use, cancer screenings, digital health literacy in adults and more.
For more information, see the Annual Meeting program at www.apha.org/program.
- Copyright The Nation’s Health, American Public Health Association
