By Jill Pease
What if a healthcare provider could instantly consult a virtual panel of specialists to guide HIV treatment decisions? University of Florida researchers are working to make that possible with an artificial intelligence tool designed to help providers make more efficient and informed decisions in HIV care.
Supported by funding from the National Institute on Drug Abuse and the National Institute of Mental Health, faculty members in the UF College of Public Health and Health Professions are leading the development of a new AI system to optimize the HIV care continuum.
With treatment advancements over the past decades, HIV is now viewed as a chronic condition, rather than an often fatal one, said Simone Marini, Ph.D., an assistant professor in the UF Department of Epidemiology and one of the project’s principal investigators. Yet over their lifespan, people with HIV may experience health challenges associated with the disease, including increased risk of heart disease, diabetes, cognitive impairment, mental health conditions and substance use disorders. HIV care management is intensive, with treatment protocols requiring strict adherence to drug regimens and frequent medical appointments and lab tests.
To help clinicians address the complexities of HIV care, Marini and fellow principal investigator Mattia Prosperi, Ph.D., a professor of epidemiology and PHHP associate dean for AI and innovation, are creating an agentic AI model to improve healthcare outcomes among people living with HIV.
Unlike traditional large language AI models that only generate text or code, AI agents are designed to take actions, but they do not replace human intelligence, Marini said. In healthcare, agentic AI systems can act like a virtual team member — pulling together insights from multiple specialties, weighing options, helping clinicians decide what to do next or guiding the creation of personalized treatment plans. They may also give clinicians more time to spend with their patients.
“In a real clinical context, any decisions need to be made by a human, with the responsibility and the accountability of a human,” Marini said. “However, AI agents, or in general, large language models, can be your backup. They can remind you of things that you didn’t think about.”
The new project, SYNchronized Agentic Prescriptive System for Improving Sequentially HIV Care Continuum, or SYNAPSIS-HIV-CC, builds on the team’s previous work to create predictive machine learning tools in HIV disease progression, barriers to care and the optimization of public health interventions.
SYNAPSIS-HIV-CC will integrate these models and others with the OneFlorida+ Clinical Research Network which includes health data for 26 million patients, including about 100,000 people living with HIV. Researchers believe the resulting agentic AI system will surpass current limitations of predictive models and calculations of individualized treatment effects.
SYNAPSIS-HIV-CC will be designed to address six crucial health outcomes for people with HIV:
- Loss of care
- Viral suppression of HIV infection
- Levels of white blood cells known as CD4
- A key health monitoring panel that detects medication side effects, monitors organ function and helps identify new complications
- Development of new medical conditions requiring immediate care
- Development of new chronic conditions
Clinicians could put SYNAPSIS-HIV-CC to work in their practice in multiple ways, Marini said.
“For example, the AI agent could double check a proposed change in therapy for a patient and bring up some risks to consider,” he said. “Or, the agent could spot subtle changes in lab results and raise an alert for the physician or nurse to consider additional testing. The system is not better than a human, it’s just complementary.”
The research team includes scientists at Brown University as well as input from key stakeholders, including people living with HIV, clinicians, infectious disease specialists and community health experts, who will help guide the system’s development.
“You can come up with new tools that on paper, are fantastic and solve the problem, but you also need them to work in the highly complex environment that is healthcare,” Marini said. “These tools are useless if they are not designed for the people who use them and the people who need to benefit from them.”