Illustration by Whitney Mgbara Science as narrative in infectious disease biology It’s been a few weeks since we’ve posted a new blog, and that delay is largely due to my tardiness—I was caught up in a grant deadline without a second to spare until now. Apologies! But a bit of time for reflection is rarely a bad thing, and it’s refreshing to revisit this talk now. On Tuesday, September 17, Professor Wayne Getz from UC Berkeley’s Department of Environmental Science, Policy, and Management presented about “Quantitative narratives in disease ecology and the recent outbreak of Ebola in Sierra Leone.” In Wayne’s own words, the talk was heavy on the “quantitative narratives” and less heavy on the Ebola, but it sure gave us plenty to think about. What is a quantitative narrative, you might ask? We don’t often think about narration as quantitative, but it certainly is. Wayne’s talk focused mostly on the fitting of mechanistic models to time series data in ecology and epidemiology, and mechanistic modeling is, by its very nature, narrative. In disease ecology, we distinguish between two broad classes of model: statistical and mechanistic. In statistical models, like the well-known linear regression, data drives the story; we plot two variables against one another and use statistical methods to demonstrate associations between those variables. In mechanistic models, we attempt to describe process, or causation, rather than correlation. This makes the approach much more active, much more narrative. All mechanistic models incorporate a process-based thesis—essentially a narrative—and we test how well each model, each narrative, recaptures the data to which we compare it. As Wayne said himself, “You can never actually prove something is true by fitting a model to data.” But if the model can convincingly recapitulate the data, we tend to accept the narrative on which the model was based. Wayne opened his talk with a broad overview of mechanistic modeling in ecology, touching on many systems that I have not reflected on much since my qualifying exam, including predator-prey models and critical community sizes for population extinction. Then, he dove into the main thread of his talk on zoonoses. Zoonotic pathogens are pathogens that are transmitted from wildlife reservoirs to human hosts, and Wayne’s research group works on several high profile zoonoses: anthrax and Ebola among them. Wayne described another two classes of model—both of the mechanistic type—which his group has used to recapitulate data from zoonotic systems. These model classes—compartmental and individual-based—represent differing approaches to narratives in disease ecology. In compartmental modeling, such as the Susceptible-Infectious-Recovered (SIR) framework, hosts are classed into broad epidemic categories, and modelers use equations to track their movements and define transitions between them. In individual-based modeling, distinct individuals are modeled explicitly, allowing for much greater flexibility in incorporating heterogeneity (i.e. in behavior or immunity) in the system. The decision over what type of model to apply to a problem is a narrative one, as different models tell different stories. My committee member in grad school, Princeton Professor Bryan Grenfell, used to say, “If you apply a complex model to a complex problem, then you have two things that you don’t understand.” Bryan was a big fan of simple compartmental models; simplifying the modeled system down to its fundamental elements tends to give the modeler much greater power to interpret the most important drivers of disease and assess how an effective intervention might be applied. Bryan also worked on a simple disease: measles, a perfectly-immunizing childhood infection that makes it easy to ignore heterogeneities between hosts because most people respond to the virus in very similar ways. In Wayne’s case, he showed us how compartmental models fit to the 2014 Ebola epidemic in West Africa failed to recapitulate the data under the simplest assumptions but did better once he incorporated a time-varying transmission rate. Only by allowing for the transmission rate to slow in the latter half of the epidemic was Wayne able to recapture these data. In his narrative, this is because human behavior changed—contact rates diminished as people stayed home out of fear. It’s a good story, but as Wayne acknowledges himself, you can never really prove it. Wayne closed by introducing us to a new platform in individual-based modeling that his lab group as developed with a software called Nova. Wayne is clearly excited about all of the gadgets embedded in the platform—it allows you to easily incorporate a within-host model (representing immune dynamics, for instance) within a compartmental model and simulate across many layers all via an accessible graphical interface. While certainly offering greater flexibility than a compartmental system, these models are a bit harder to interpret, and fitting these models to data can prove additionally challenging. Ultimately the decision over what type of model to use comes back to the research question and the story that you want to tell. If you are trying to recapitulate data with a credible narrative, then your model will need to produce its own data to compare with reality. Individual-based models hold enormous promise in application to longitudinal cohort studies where the modeler possesses data on individual hosts resampled over time—possibly even including genetic information about pathogen evolution within those hosts. In other cases, where data are less well-resolved, a compartmental approach might be more appropriate. In sum, a mechanistic model is a narrative, the disease ecologist’s way of telling a story. It’s the modeler’s challenge to find the model that is most appropriate for the narrative he or she wishes to tell. Writing by Cara Brook
0 Comments
Leave a Reply. |
AuthorBerkeley EEID Group Archives
February 2020
Categories
All
|