Confronting AIDS in Florida with artificial intelligence

A University of Florida research team is playing its part to stop the global HIV epidemic, applying the power of artificial intelligence to medical records to uncover patterns of risk and bias that have left some patients in Florida with poor access to today’s effective treatments — so effective that HIV infections are not sexually transmitted.

Why do some HIV-positive patients miss out on such powerful medicine? Do they lack knowledge, insurance, or some combination of factors that push them into limbo?

To seek answers, the UF team will spend four years studying HIV/AIDS in Florida, where transmission remains stubbornly high. The study is funded by a $3.7 million NIH grant. It will use UF’s HiPerGator 3.0, a centerpiece of the university’s artificial intelligence initiative and one of the most powerful supercomputers owned by a university.

According to the Florida Department of Health, an average of more than 4,000 people each year were diagnosed with HIV during the past decade, and more than 120,000 Floridians have HIV.

HIV remains a persistent health issue, and the most affected U.S. group includes men who are Black, bisexual or gay. Globally, more than 38 million people live with HIV, and infections are highest among females. More than 40 million people have died from AIDS-related illnesses.

High HIV transmission rates in Florida happen for many reasons, but a key problem stems from unequal usage of prevention techniques and antiretroviral therapies, with the latter capable of making HIV undetectable. When viral loads are undetectable, HIV infections are not transferable. Such patients, although still infected, can live healthy lives without fear of developing the final, deadly stage of AIDS.

The Florida study will contribute to a national initiative launched in 2019, “Ending the HIV Epidemic in the U.S. by 2030,” led by the Health Resources and Services Administration, part of the U.S. Department of Health and Human Services.

The UF study is co-led by Mattia Prosperi, Ph.D., of the College of Public Health and Health Professions, and Jiang Bian, Ph.D., of the College of Medicine. They will work with a team of researchers, clinicians, public health professionals and a community-based advisory panel. The team’s expertise will cover epidemiology, health equity, electronic health record processing and the science of implementing treatments.

“The treatment of HIV has progressed to the point where nearly every infected person can become healthy again. We don’t have a perfect cure yet, but we can change the course of the disease,” said Prosperi, a professor in the department of epidemiology and PHHP’s associate dean for AI and innovation.

Today, uninfected people can take the medication PrEP to avoid HIV infections even when exposed. If they do become infected, patients can take a daily antiretroviral therapy.

In Florida, nearly 30% of people with HIV lack treatment, and they are more likely than most Floridians to be poor or Black. But more issues than poverty and race are at play.

“We are looking at more than 9,000 factors that might influence patients’ behaviors, their health outcomes, and general health status, or what we call social determinants of health, that are often the root causes of health disparities and inequity,” said Bian, UF Health’s chief data scientist and chief research information officer, who is based in the department of health outcomes and biomedical informatics. “Once we identify the social determinants of health contributing to the problem, we can begin to find ways to address them.”

Both Prosperi and Bian have expertise in artificial intelligence and high-performance computing. Prosperi combines it with causal inference to tackle bias in epidemiology studies, and Bian leverages it to decipher human language, specifically the jargon in medical doctors’ notes. The researchers will leverage GatorTron™ and GatorTronGPT™ — a large language model similar to ChatGPT but trained on all UF Health clinical notes (more than 86 billion words) and tailored for clinical natural language processing tasks.

The researchers will use anonymous patient data from the OneFlorida+ clinical research network and state HIV data from more than 53,000 patients. Surveys of nearly 1,000 patients provide details about patient behaviors and stigmas they encounter. By combining various data sets, the researchers plan to overcome the social stigma and privacy concerns associated with sexually transmitted diseases that result in extremely limited information provided directly by patients.

When the study ends in 2027, the researchers will recommend clinical and public health interventions to reduce HIV infections and to improve the health of HIV-positive patients. These recommendations will build upon a preliminary list of 18 interventions derived from the Florida Department of Health, such as access to telehealth and counseling for violence.

This research is highly multidisciplinary and explores new methods of analyzing data. The data’s massive size poses problems of storage, searching and the creation of software to manage it. Multiple layers of complexity must be mastered because the data sets come from a variety of sources.

One of the research team members, Robert Cook, M.D., is a professor of epidemiology and the director of the Southern HIV and Alcohol Research Consortium, based at UF. His research has identified broad risk factors for HIV infections that include mental illness, substance abuse and homelessness.

Two members of the research team specialize in understanding health disparities among minorities and underserved populations. Lori Bilello, Ph.D., is a research associate professor in the department of surgery, and Shantrel Canidate, Ph.D., is an assistant professor and social scientist in the department of epidemiology.