University Times

A 1929 Rene Magritte painting depicts a pipe floating in air beside a pipe on a miniature canvas. Written below the pipe on the canvas in French: “This is not a pipe.”

“The reason Magritte put that there is because, in fact, this is not a pipe, it is a painting of a pipe. I know that; you know that. It’s an interpretation and a useful way to think about pipes, but it is not a pipe,” said Donald Burke, dean of the Graduate School of Public Health at a July 8 event announcing the opening of a new GSPH-based laboratory.

Pointing to a computer-generated image on a giant screen in Crabtree Hall, Burke said, “Here we have one of our models, a simulation of an epidemic of influenza in the United States. I want to write on it, ‘This is not an epidemic.’ I know it’s not an epidemic and you know it’s not an epidemic. What it is is a representation of an epidemic that allows us to think about epidemics and talk about them and talk about the particulars in ways we couldn’t do if we didn’t make an explicit representation of that. It’s the same idea. We know this is not reality; we know these are representations, but that’s what we want other people to share, to have a way to talk about things.”

That, in a nutshell, is the basis for Pitt’s new Public Health Dynamics Laboratory (PHDL), which aims to develop interdisciplinary computational approaches to understand and solve the world’s most challenging public health issues.

PHDL collaborators will include experts from Pitt’s Center for Simulation and Modeling; Pitt’s schools of medicine, engineering and arts and sciences; the Pittsburgh Supercomputing Center, and Carnegie Mellon University.

Computational modeling in public health typically is associated with the evaluation of strategies to contain infectious disease outbreaks, but it also can be applied to behavioral health, emergency response planning and health policy.

GSPH Dean Donald Burke, left, and John Grefenstette, director of the new Public Health Dynamics Laboratory.

According to PHDL director John Grefenstette, professor of biostatistics at GSPH, the lab will be  a “collaboratorium,” bringing together epidemiologists, biostatisticians, behavioral scientists, public health policy experts and computational scientists to produce the next generation of tools for public health analysis.

Areas of focus will include infectious diseases; vaccine distribution in developing countries; public health response to epidemics and other emergencies; social networks and their effects on obesity, smoking and other health behaviors; racism, segregation and health disparities, and open access to historical and current public health data.

“We try to avoid doing academic exercises. We want our models to be useful to practical decision-making,” Grefenstette said.

“The second principle is that the public health system is a complex adaptive system. We have a lot of moving parts and, rather than studying just parts of the public health system, we want to study the interaction of those parts moving together. What that means is if you try to model those things inside a computational system, it’s going to need a complex computer program as well,” he said.

PHDL experts currently are developing six such computer programs that accommodate select areas of public health, such as vaccine delivery supply chains, spatial and temporal dynamics of disease and open access to global health disease data.

Such lofty goals would have been unthinkable a decade ago, but the computational power exists today to meet those goals, Grefenstette said.

The primary technology employed by PHDL is called agent-based modeling. “The idea is provocative: Let’s build a computer simulation of actual populations,” he said. “Let’s have the agents in that simulation represent the people in an area. The agents can have households, families, they can go to work, they can go to school. We can simulate their activities during the day, and follow what happens if a disease occurs and an epidemic starts in that population. That would have been an outrageous thing to attempt even 10 years ago, to say you could simulate a city, or a state or a nation. In fact, progress in computer science has been so great we can now do that. We have models that are simulating the daily activities of hundreds of millions of people.”

Grefenstette stressed that decisions should never be made solely on a computerized model or scenario, but that models work well with hard data and expert opinion in decision-making.

“What you can do with the models you can’t do any other way is you can ask ‘what if’ questions: What if schools close? What if the vaccine [delivery] is delayed a month? What if the disease turns out much worse than expected, or has some kind of shift? What if we have a mild-case scenario, a medium scenario, a worst-case scenario? The final product is always a comparison of different interventions,” he said.

“The way we want to see modeling go is the inclusion of the environmental dynamic: the geographical, spatial, temporal aspects; the effects of weather on epidemics, for example. So we want our models to possess more and more of the details that have a lot to do with the eventual outcome of a decision,” he said.

Burke added, “Computer simulation models are tools that can help compensate for weaknesses in our models. The time is ripe to take full advantage of computational models in our teaching, our research, our problem-solving in public health.”

For more information on the lab, go to www.phdl.pitt.edu/.

—Peter Hart