In a clinical trial, thousands of people defined by disease and demographics test a single intervention for safety, efficacy, and side-effects. While nothing can replicate the accuracy of a real world test, simulation models based on empirical data can help with scalability, ethics, and combining multiple interventions.
Within a session on new approaches at the Mt Hood Diabetes Challenge, where simulation models addressing problems in diabetes go head to head in competition, Jonathan Ozik, Principal Computational Scientist at Argonne National Laboratory and Senior Scientist in the Department of Public Health Sciences at the University of Chicago, will deliver a plenary lecture on agent-based modeling, a method that simulates a complex system by separating it into individual entities, known as agents, that interact with each other within an environment.
“Each agent has attributes, behaviors, and the capacity to learn or set goals,” he explains. “Agents interact within environments—geographic spaces, social networks, etc.—and those interactions can affect the system as a whole. For example, people could exchange information or contend for resources. An agent might even affect the environment, which in turn affects other agents.”
Simulating interactions between individual agents—which can be people, organizations, or even cells—can help researchers understand how systems evolve over time. Using data from the census, health surveys, epidemiological studies, and other empirical sources, it can be possible to synthesize these into a coherent population.
Ozik and collaborators used agent-based modeling to project effects of policy during the COVID-19 pandemic. “We built a model of Chicago with 2.7 million agents—not the real people, but synthetic versions whose aggregated behavior matches the real population,” he says. “The idea was to create a digital twin of the city to test interventions like school closures or mask mandates and see what the simulated outcomes look like compared to the real-world data.”
This work, done as part of a task force convened by the Illinois governor’s office, helped guide decisions around reopening schools and masking policies. “Alderpersons in Chicago sent out our slides to constituents to encourage mask-wearing,” Ozik recalls. Their work also helped the state assess hospital capacity as the pandemic progressed. “It was a good alignment between analysis and policy capability. But we’re very clear—we don’t make recommendations. We provide analysis. It’s up to policymakers to act on it.”
Initially trained in the physics of pattern formation and network behavior, Ozik has been drawn to real-world problems that couldn’t be solved with simple equations. “That led me into complex systems—situations with many people or decision-making entities—where we’re trying to understand behaviors or how to deliver effective interventions. Using an in silico platform for experimentation was a natural extension,” he says. “I’ve been doing agent-based modeling here for almost 20 years now.”
Recent work in collaboration with the CDRP’s Stacy Tessler Lindau and Elbert Huang focuses on applying ABM to health equity and systems change, particularly through the lens of CommunityRx, a project designed to connect individuals with local health and social resources.
“One thing ABM allows us to do is test how information gets exchanged,” says Ozik. “This can help understand the spillover effects of interventions—like combining a drug treatment with an information campaign and seeing if it activates broader community behavior change.” The ability to model information flow, social dynamics, and access to care is especially relevant for chronic conditions like diabetes, which are deeply intertwined with behavior and community support systems.
But how do we know models are accurate? “Agent-based models are complex because they often operate on multiple scales,” says Ozik. “For example, we use the American Time Use Survey to model daily schedules of individuals. That data tells us not just what people do, but also their demographics. In the case of Community Rx, we identified which activities were more or less likely to involve information exchange—going to work, attending church, etc. We also modeled how people retain information—because obviously no one remembers every resource they encounter. We experiment with different assumptions and configurations to see what matters most. If a parameter significantly affects outcomes, we know we need more precise data. If it doesn’t, we can treat it more flexibly. This iterative process helps us prioritize which knowledge gaps are most important to fill.”
The complexity and scale of agent-based models requires an interdisciplinary and collaborative approach. Ozik works with clinicians and epidemiologists who provide crucial real world information on how diseases progress and how patients seek care. “If I’m in a room and understand everything, something’s probably wrong—it means we’re not really pushing boundaries,” he says. “It’s about joint sense-making, improving our understanding, and hopefully delivering better healthcare as a result.”
Jonathan Ozik delivers the plenary lecture “Health Behaviors and Information Diffusion: An Agent-based Modeling Perspective” at the Mt Hood Diabetes Challenge on Wednesday, June 25 at 1:30pm as part of a session on New Approaches to Modeling. REGISTER HERE.