How do you handle the outbreak of a highly infectious disease?
Whether it's a terrorist use of smallpox or the Avian Flu, public health officials worldwide are fearful that we will see a deadly pandemic in the not-too-distant future. The right way to respond to such an event so as to limit deaths is a big question. Are immediate quarantines the right answer? Mass vaccinations? Vaccination efforts targeting only the "super-spreaders" who have a multitude of social connections? Epidemiologists often have little more to go on than theory and best-guesses. Researchers Chris Barnett, Stephen Eubank and James Smith at the Los Alamos National Laboratory decided to take a page from urban planning and crafted a simulation system called EpiSims, which maps the spread of infectious diseases and models different containment plans. They wrote about their findings in the current issue of Scientific American.
EpiSims is a program based on TRANSIMS, an urban planning/transit system simulation. The software models the behavior of residents of a virtual city, following their movement and contacts, building maps of their social networks, then charting the spread of infection based on different containment scenarios. TRANSIM was based on Portland, Oregon, and the first EpiSims research (on the spread of smallpox) also used that city. EpiSims followed 1.6 million virtual citizens traveling to over 180,000 different destinations, and determined how the disease would spread through the city. The image to the upper right is from a movie available at the EpiSims website tracing a smallpox attack under two different scenarios.
One of the team's findings was that targeted vaccinations of social hubs was less effective than hoped:
If infected "hub" individuals, such as the most gregarious people in a population, could somehow be identified and treated or removed from the network, the reasoning goes, then an epidemic could be halted without having to isolate or treat everyone in the population. But our analyses of the social networks used by EpiSims suggest that society is not so easily disabled as physical infrastructure.
The network of physical locations in our virtual Portland, defined by people traveling between them, does indeed exhibit the typical scale-free structure, with certain locations acting as important hubs. As a result, these locations, such as schools and shopping malls, would be good spots for disease surveillance or for placing sensors to detect the presence of biological agents.
The urban social networks in the city also have human hubs with higher than average contacts, many because they work in the physical hub locations, such as teachers or sales clerks. Yet we have also found an unexpectedly high number of "short paths" in the social networks that do not go through hubs, so a policy of targeting only hub individuals would probably do little to slow the spread of a disease through the city.
In fact, another unexpected property we have found in realistic social networks is that everyone but the most devoted recluse is effectively a small hub.
Their key determination was that the most important determinant of success for disease containment efforts was the speed of response, both of authorities and of infected individuals. Unsurprisingly, the faster people infected with a disease withdrew to isolate themselves from others, the smaller the outbreak. But the difference in deaths from a 4-day delay in intervention vs. a 10-day delay was considerable, and sometimes shocking, a few hundred dead vs. tens of thousands dead. Conversely, the strategy chosen (mass or targeted vaccinations) was much less important than the time of implementation, and was more an issue of cost and vaccine safety.
As we've discussed, simulation results are based on their underlying rules, and if those rules are wrong, the results will be useless. Just as the sim is not the city, the sim is not the sickness. The Scientific American article notes the need for accurate rules by explaining why previous models have proven less useful:
...One reason is that modelers have often lacked detailed knowledge of how specific contagious diseases spread. Another is that they have not had realistic models of the social interactions in which people have contact with one another. And a third is that they have not had the computational and methodological means to build models of diseases interacting with dynamic human populations.
As a result, epidemiology models typically rely on estimates of a particular disease's "reproductive number"--the number of people likely to be infected by one contagious person or contaminated location. Often this reproductive number is a best guess based on historical situations, even though the culture, physical conditions and health status of people in those events may differ greatly from the present situation.
In real epidemics, these details matter. The rate at which susceptible people become infected depends on their individual state of health, the duration and nature of their interactions with contagious people, and specific properties of the disease pathogen itself. Truer models of outbreaks must capture the probability of disease transmission from one person to another, which means simulating not only the properties of the disease and the health of each individual but also detailed interactions between every pair of individuals in the group.
Does EpiSims capture these differences better? Almost certainly. But I wonder what it's missing. Do the sims wandering through the virtual city have varying infection rates arising from different cultural behaviors -- how close they stand to each other, sharing of personal articles, kissing in public, etc.? If use of public transit is included in the pathogen spread model, does it account for which demographic groups are able to afford taking a car in to work instead (as many public transit riders would do once rumors of an outbreak emerged), and which are not? Does it include populations without insurance who use the emergency room for their healthcare, unintentionally spreading the disease to others waiting with them? Such elements, while complex, are not impossible to simulate, and would add greater accuracy to the model.
The use of free/open source methods for the development of simulations would help greatly in that regard, providing opportunities for contributions of more accurate behavioral models. Open source models do exist, and such applications are not uncommon in academia. Unfortunately -- but not surprisingly -- EpiSims is not among them.
Free/open source simulations would have another benefit: transparency. As computers get faster and modeling tools get better, we are nearing a world in which all sorts of decisions affecting society will be made using computer simulations as input. This is not inherently a bad thing -- well-constructed simulations can be of great help in understanding complex systems. But those who rely on models, especially models of human community behavior, need to be explicit about what is included in the simulations and what isn't. The more we rely on simulations to lay out our options, the more we need to know what choices the simulations can't provide.
(EpiSims found via Smart Mobs)
Simulations of this sort are having huge impact in science, and moderate but increasing impact in public policy discussions. How much impact will they have on political discourse and public behavior? And will that impact come just from quoting models as authoritative sources, or with actually playing with the models and understanding why things happen the way they do?
In principle, models that better capture reality are more useful as predictors. But overly complex models can be harder to understand, analyze, and verify. They may also be more likely to contain subtle errors, which the complexity of the model itself makes difficult to find. And of course, inferring causation from a "black box" model is problematic if one can't explain the model behavior.
In practice there is a sort of art to choosing a good model. Sometimes it will be relatively complex, particularly where abundant data and domain science exist to provide truth-tests. Other times, it gets stripped down to just a few axioms in order to stay useful (physics and its interdisciplinary variants excel at this).
Free/open source simulation methods are definitely the way of the future for discussing and collaboratively planning public policy. But complexity in these models could be just as much a deterrent as being closed-source - as with complex econometric models, whose conclusions are often debated but methodology relatively rarely. You can often tell a really good model by how much it leaves out, while still providing useful explanations!
It seems to me there should therefore be "layered" implementations of such models, with various mixes of omitted or "fixed" variables and quantities plugged in, each compared to the full-blown version to see how much difference there is in the results, and under what conditions the differences are greatest. This would both provide a route to keeping the boxes transparent rather than black, and give a basis for cost-benefit analysis of collecting and inputting various data streams.