Because clinical research is expensive and slow, the ability to carefully test new therapies and hypotheses in human patients and then refine them further is a major bottleneck for medical advancement. One solution under development by scientists at the Computation Institute (CI) — a joint initiative of UChicago and Argonne — looks to a method used regularly by films and video games: agent-based simulation.
When creating hordes of computer-generated combatants in an epic movie battle or the automated characters that ll a game’s open world, it would be cost and time prohibitive to manually animate or direct each individual. Instead, programmers create reproducible “agents,” which behave according to simple rules, such as “when near an enemy soldier, attack.” When multiplied thousands of times, these automatons can produce incredibly complex, realistic and, occasionally, unexpected behavior.
Gary An, MD, professor of surgery and a CI senior fellow, applies these same techniques to study biological systems and disease. Trained as a trauma surgeon, An first became interested in modeling the complex process of sepsis, the dangerous runaway in ammation that can set in after an injury or medical procedure. To study the process, he built agent-based models of how the immune system responds to tissue damage, assigning behaviors to each cell based on ndings from scientific literature. With this model, An could recreate the process of sepsis and test different proposed treatments for their efficacy in stopping the inflammation.
Subsequently, An worked with University of Chicago Medicine clinicians on building similar models for ulcerative colitis, breast cancer and necrotizing enterocolitis. He built a “virtual gut” for studying inflammatory bowel diseases, using the Argonne petascale supercomputer Mira to simulate three feet of colon at a 10-micron scale. Now he’s looking at more abstract exploratory models to help fill out the missing clinical data needed to make significant strides in personalized medicine.
“We need simulation to get the scale of data we need for deep learning,” An said. “By borrowing simulation and modeling methods from climate and astronomy studies and applying them to biology, we can nd the range of possible disease outcomes and responses to treatment, then plot each individual patient within that range and make individualized decisions about therapy.”
Other UChicago and Argonne scientists are using agent-based models to probe higher-order complexities of health care. John Fahrenbach, PhD, a data engineer with the University of Chicago Medicine Center for Healthcare Delivery Science and Innovation, is working with Thomas Spiegel, MD, and administrator Garrett Larance on modeling the hospital’s emergency department to find bottlenecks that slow patient care. Similarly, An and Fahrenbach are building a hospital-wide model to study ripple effects across units, including how the addition of a trauma center or an influenza outbreak could affect operations in the emergency department, surgical suites, the ICU and inpatient units.
“Agent-based modeling allows us to ask: as we grow as a health system, how will that affect our inpatient volume? For example, how will increased patient volumes in Orland Park affect ER wait times in Hyde Park?” Fahrenbach said. “It’s a phenomenal tool to help hospital operations with these really challenging questions they need to answer for us to thrive.”
Agent-based models are also a powerful tool for studying public health, capable of simulating the spread of disease across thousands or even millions of simulated individuals. Working with researchers from the Department of Public Health Sciences, Jonathan Ozik, PhD, Charles “Chick” Macal, PhD, Nicholson Collier, PhD, and the Complex Adaptive Systems group at Argonne developed a detailed model of Chicago, called chiSIM.
Using building and census data, the nearly 3 million citizens of Chicago can be represented in geographic space and trained to behave normally — going to work, school, the park or even jail. Originally created to study the spread of methicillin-resistant Staphylococcus aureus (MRSA), the platform has also been used to simulate epidemics of influenza, Ebola virus disease, and HIV in the city, and test out different strategies for slowing or isolating infections.
But the model can also be used to study more positive transmissions as well, such as the spread of information about local health resources distributed to patients as part of the CommunityRx program, led by Stacy Tessler Lindau, MD, MA, associate professor of obstetrics/gynecology and medicine. By simulating how this information migrates from study participants to family and friends, Lindau’s team can better assess the full impact of their work.
“This approach allows us to move from understanding the mechanisms through which the CommunityRx intervention produces better health and health care to predicting the impact of the intervention in different settings and conditions,” Lindau said. “The simulation model will allow us to present CommunityRx to organizations and leaders in other cities and demonstrate quantitatively the potential impact of the intervention for their clients or residents. There’s just no way to do that with a traditional model.”
This is the third of a five-part series on data-driven medicine and research at the University of Chicago Medicine, originally published in the Spring 2017 issue of Medicine on the Midway.