Cholera cell. Illustration by Marina Tsaplina
Modern epidemiology, or how we track the health of populations, is founded on the work of John Snow, who in 1854 connected contaminated drinking water in London’s Soho district to an outbreak of cholera. Without looking at a single bacterium, Snow managed statistically and cartographically to trace an outbreak back to its source, and in so doing set the groundwork for a shift within the scientific community from the concept of “miasma” to modern-day germ theory. Now a new generation of scientists is using Snow’s founding principles, with the help of satellites and computers, not only to understand why diseases strike but also to predict when they will.
In a study published in a recent issue of PNAS, Rita Colwell, of the University of Maryland’s Institute for Advanced Computer Studies, and her team unveiled a predictive model of cholera. It signals a change in the way disease prevention is understood, shifting from a transmission-based model, in which it is necessary to observe the disease in order to stop future cases, to a model that assumes disease emergence is reliant on predictable variables in the environment.
Unlike Snow, whose cholera investigation was based largely on proximity data, Colwell bases her work on the known biology of cholera and its hosts. The disease’s most common host is the copepod species of zooplankton, and so conditions linked to copepod multiplication, most notably the relative presence of chlorophyll in the water, are also favorable to cholera outbreaks. Today cholera is common to the developing world, where citizens utilize untreated water for a variety of purposes, including for drinking. Using remote sensing, or networks of satellite data to determine environmental and geographic factors, the team identified a number of environmental factors to predict likely cholera outbreaks in the developing world.
The predictive capacity of Colwell’s model over time was striking, and its power is further reinforced by a painstakingly controlled three-year study in which her team showed that filtering local water with the cloth of a sari drastically reduced cholera incidence. Even more impressive, those who continued to filter their water have begun to demonstrate a protective effect — other people who live near them get cholera less often. The work has combined the predictive and the prescriptive: It demonstrates not only how to predict an outbreak but also how to use that information to develop interventions. “It presents a way of guiding the effective and efficient use of limited resources in these areas,” says Colwell. “You can distribute filters, or make public postings letting people know that incidence could go up.”
Though it may seem obvious to link host prevalence to disease outbreak, Colwell’s model is representative of a recent radical change in epidemiology, in which a seemingly endless backlog of data is being marshaled in an increasingly effective way by both computer scientists and epidemiologists. Colwell is convinced that this approach is not a temporary fad, but rather a paradigm shift toward a “more math-formulated approach to health resource management.” As another example, she points to Google.org’s recent introduction of Google Flu, in which a number of search terms related to the flu are used as predictive markers of flu outbreak in the US. Early analysis indicates that Google’s program has already bested the CDC’s wait-and-listen approach to data collection in its predictive capacity.
Google’s much-publicized breakthrough is the result of two simple observations: The biological phenotype of the flu — as opposed to, say, Ebola — is amenable to information seeking online, and normal people tend to be right when they think they’ve got it. Similarly, Colwell and her team realized that a massive amount of climate and sea-temperature data, when properly analyzed, could predict cholera. Important for both these groups is the adaptability of their hypotheses to future improvements — Google would have no problem adding a new search word linked to the flu, just as Colwell’s team could adjust its algorithm when new data comes to light. “The more relevant parameters we have,” Colwell points out, “the more exact our predictions will be.” Data collection is already under way in Senegal and Peru, and Colwell expects that the more information she accrues, the more accurate her model will become.
As a result of work like Colwell’s cholera project, a future in which infectious-disease outbreaks can be predicted weeks in advance is now foreseeable, even in the developing world. What makes Colwell’s work so impressive, and what differentiates it from Google’s, is the focus on biologically linked precursors of cholera, a focus that allows her team not only to predict cholera outbreaks but also to intervene effectively, as she did in the Bangladeshi village where her water filters were deployed. This, of course, is the ultimate goal of all medical science: prevention.
Originally published December 17, 2008