FOCUS in Sound - Joshua Weitz

Welcome to FOCUS In Sound, the podcast series from the FOCUS newsletter published by the Burroughs Wellcome Fund.  I’m your host, science writer Ernie Hood.

In this edition of FOCUS In Sound, we meet a quantitative biologist who studies the structure and dynamics of complex biological systems.  Most recently, he organized and chaired a two-day, so-called “rapid response” workshop called Modeling the Spread and Control of Ebola in West Africa.  The meeting, which drew 180 participants from all over the world, was held in January, 2015 at the Georgia Institute of Technology in Atlanta, where Joshua Weitz is a tenured associate professor of biology.

Joshua received his Ph.D. in Physics from MIT in 2003, and from 2003-2006 he was an NSF Postdoctoral Fellow and Associate Research Scholar at Princeton University.  He joined the Georgia Tech faculty in 2007. 

In 2007, Joshua received a half-million dollar, five-year Career Award at the Scientific Interface grant from the Burroughs Wellcome Fund, to fund his research on the evolutionary ecology of bacterial viruses.

Joshua Weitz, welcome to FOCUS In Sound…

Thanks for having me.

I’d like to initially focus on the Ebola workshop that you chaired – what exactly is dynamic modeling, and how does it relate in practical terms to the Ebola epidemic in West Africa?

Dynamic modeling is an effort by many to try and take the processes on the ground involved in an epidemic, say, and translate those into a mathematical form that we can then predict both the future behavior, and in some sense understand the mechanisms by which an epidemic may increase or decrease over time or the effect of interventions.  These sort of scenarios requires, to some extent, assumptions, approximations, but also really a conceptualization of the epidemic process.  So the word “dynamic” means that it’s something that describes a process that changes in time, and “model” means that it’s a model.  It’s not perfect; on the other hand it’s one of the few things we have.  It’s a resource that we use along with interventions on the ground to, first of all, evaluate their efficacy, and also to predict how much of a particular kind of intervention we need, over what time scale, and what our interpretation should be once we get changes or we see changes in the number of cases that come in through other on the ground reports.  So it’s really part of the toolbox of what epidemiologists can use to combat infectious diseases. 

How are you and others able to gather the data necessary to conduct that type of modeling?  Is that one of the major challenges?

So there are different levels of data, and certainly the data problem is a challenge.  And the degree to which it’s a challenge, I think, depends on the circumstances.  Frankly, the challenge of Ebola Virus Disease in West Africa makes the problem of data collection rather acute.  First of all, it’s not a disease for which, at least in these particular affected countries in West Africa, which they had much experience with.  Their infrastructure in terms of, let’s say, governmental health organizations, clinics, availability of physicians as well as monitors to record data, was quite limited.  Very, very starkly different to what we might expect to be at norms in the United States or western Europe or elsewhere.  So there is this issue of missing data.  In fact, to some extent, that’s where models can shine in helping to understand and predict circumstances even when the data is poor, and then when data becomes available, to try to reconcile and test against the predictions. 

In the case of Ebola, there are many kinds of data that we might want to have, and are not always made available to the modeling community for various reasons, because they’re not accessible or they’re hard to get, etc.  But certainly this notion of data limitation is one that made the response efforts to Ebola Virus Disease quite difficult. 

How have these recently developed dynamic modeling tools contributed to the response to the Ebola epidemic?

So I think we have to kind of go back a little bit in time to understand the ways in which dynamic models in particular have contributed to the effort, and frankly we should also point out that they remain controversial.  So moving back to August or September, when there were increasing news reports that Ebola was spreading unlike in prior epidemics, meaning it was not limited to a few hundred at most cases in an isolated rural area, but rather affecting hundreds if not thousands of people in increasingly urban areas, and mixing between rural and urban zones.  And so models at the outset were meant, I think, if we look retrospectively, to some extent in a way to raise alarm.  To point out the fact that the current trajectory of the epidemic was not something in which the international response could be to say, well, yes this is a very bad disease if you get it, because the mortality rate is approximately 70%, but rather at a population scale, this epidemic is going to be something that if we don’t stop [it] now, it’s going to get worse, the numbers will increase, and this could lead to systemic problems, not just to those affected because they’re infected, but rather even for those who are infected by other things like malaria, or just for governmental institutions for the overall health of many individuals infected by Ebola and otherwise in the economy.  

So for all of these reasons, I think the early dynamic models were able to say, let’s take the data we have, these trajectories, these case counts that were coming out of Liberia, Guinea, and Sierra Leone, and say, if this continues unabated, how many people will be affected, and over what time scale?  And so the early dynamic models are able to take this limited data and say, what mechanism might produce this trajectory up to this point, in some sense training the model on the data we have, and then projecting out into the future what the future case load might be.  But also, not in such a negative view, in such a hopeless view, but rather also in a hopeful view to say, what would be the benefits of intervention?  And so those early models also pointed out the values of intervention, and the value of intervening early. 

The earliest models, and these are really, I’d say, produced largely by one group from the CDC, but there are other academic workers involved too, and also from the WHO aligned with Imperial College in London.  So there’s a number of groups that around the same time, August, September, were starting to say, look, we need to do something, we need to act quickly.  If not, this is going to get worse.  In fact, it did.  But one could also at least give some credit to these models for alerting the community that large scale, on the ground interventions, not just models, needed to take place. 

There’s also critique of these models, because their forecasts said that in September, if no interventions took place, the CDC reported up to a million infected cumulatively by January.  That obviously did not take place.  The WHO predicted tens of thousands by November, and more beyond that.  So these sort of dire predictions did not take place.  There are many reasons for that.  Models are a part, I think that they’re frankly a small part, there’s a lot of changes in behavior, but they were part of that effort to raise awareness that specific kinds of interventions that governments could do, but maybe they need a justification [or] explanation of why such an intervention is necessary.  Part of the direness of the situation has to do with this concept of exponentials, which we may want to talk about as well.

Sure, could you describe that aspect?

When I say exponential, you often hear that, one can hear that sometimes in the popular press, this notion of exponential growth.  And this is commonly attributed to bacteria growing inside of you or a virus multiplying inside of you, where you have one, then two, then four, then eight, and that doesn’t seem so bad, but soon sixteen, thirty-two, and after a few doublings you get quite a lot of whatever the pathogen is.  This same concept applies to the spread of an infectious disease amongst individuals: that each individual who is infected can give rise to a number of new individuals who are newly infected.  And when each infected individual basically causes or transmits forward the disease that they have to someone else, and an additional person or two, then you can see how that could have a multiplicative effect.  If, for example, one person infects three other people, then you get one to three, if that happens to the next three people, to nine, to twenty-seven, and soon you have hundreds if not thousands of individuals who are infected.  This is very different than a kind of steady growth, a linear growth in the number of infected cases.  And so the concern in August/September is that these early trends looked exponential.  There was a rapidly expanding epidemic.  And that’s something that the modeling community, and obviously the Center for Disease Control, WHO, and many groups, for various reasons realized needed to stop if we were going to avoid these dire scenarios. 

Absolutely.  Well Joshua, is that where the “dynamic” aspect of dynamic modeling comes in, as the models affect outcomes and change on an ongoing basis – and vice versa?

That’s right.  So the models themselves don’t treat the current trend as being a static trend.  They also account for the fact that when a disease spreads, it intrinsically modifies the rate of growth of the dynamic. For other diseases, there’s this concept of susceptible depletion, that is, as a disease spreads, there are less people to spread to, because many of the people one could infect, for a disease like the common cold, are maybe temporarily immune, or for flu, could be immune.  In the case of Ebola, we don’t think we got to a point of – there’s some dispute there – of susceptible depletion, at least on a population scale, but maybe locally.  But there still is an issue of how an intervention might change the spread.  How changes in behavior might change the spread.  How local population structure, networks of contacts, and people begin to change their behavior.  Hospitals are starting to be built, EPUs [Ebola Prevention Units] are built which isolate people, people get word that in their village there’s a burial, but they shouldn’t touch the body.  So there’s many different ways in which the rate of transmission, which is not just a factor of the virus, but rather a confluence of the virus and behavior, which itself has socioeconomic factors as well, and all of those determine how many new cases occur per infected individual, and that itself changes over time.  So the dynamic models try to capture that infection process, the recovery process, but also the intervention process, and make reasoned predictions over what would happen without certain interventions, and the value of the interventions themselves.

What were some of the most significant findings that emerged from the recent workshop?

The workshop, to put in context, was hosted here at Georgia Tech in January 2015, and the objective was to bring together modelers, decision-makers, policymakers, from different fields, different backgrounds, who usually don’t communicate with each other on a regular basis, but certainly because of the epidemic and the crisis in West Africa all had common cause and concern, and wanted to communicate, but we needed a venue and a setting to do so.

Our objective was to try and facilitate having academics, as well as government officials and even modelers from different spheres, from institutions like the CDC and elsewhere, communicate the principles by which they made their predictions, why certain types of outcomes may or may not be expected to occur, and also potentially influence ongoing decisions like vaccine trial design.   So I think there were a number of conclusions that came out of the meeting, though our objective was not to reach consensus but rather to communicate.  In fact, a lot of different views on the right way or the right ways, I should say, to try to intervene. 

One of the things, though, that did come out at the outset was the importance of thinking about Ebola in terms of local population structure.  The nature of the disease transmission process requiring contact with bodily fluids suggests that very close contact is necessary, and therefore, social chains, funeral events, anything that puts people into close contact with bodily fluids would be a mechanism by which there could be a forward transmission event.  And because of that, some of these dynamic models assumed what is called a well-mixed population, one in which many people interact, essentially everyone. You have a common cold, you’re on the bus, you sneeze – that’s not the mechanism by which Ebola spreads.  And therefore one of the major conclusions was that in responding, in building dynamic models, we need to think very carefully, whether it’s Ebola or the next emerging infectious disease, about the nature of transmission and how to build the appropriate model that reflects population structure and the transmission process, because conclusions about how fast the disease spread depend critically on the nature of the transmission. 

The other thing that I think arose was in fact an issue of communication.  That what a model wants to communicate, and how they communicate, is often different than what a policymaker needs, and the time scale over which they need that information.  And so the modeling community, to the extent that we’re often waiting to kind of do things, in a good sense, quite rigorously, over long time scales, we were told by many in fact that there is a value to giving an incomplete answer, as long as also the uncertainty in our predictions is also communicated; that even some information would be important to report back.  And I think the modeling community in fact in some sense is doing that.  There are now journals and venues that…One, for example, is called PLOS Current Outbreaks, that is meant to facilitate reporting academic findings on a rapid time scale in the midst of an epidemic, which is not something that one thinks of academics doing.  I don’t think that’s the conventional view of the academic time scale.  In the course of a few months, we can do that in this world, to build models, make predictions linked to data, and we are finding that this is of value to that community. 

And finally, I think another conclusion is that there is also a greater need for interfacing of models that include epidemic transmission dynamics, but also one of logistics.  Because in West Africa, there really is a problem of logistics, and how to bring, whether it’s supplies or people or information, to affected areas, and models for the most part make certain simplifications or assume certain kinds of processes that may not be taking place, particularly on the ground, particularly in these countries with poor infrastructure.  So this really was less of a conclusion that we know the answer to, but rather an area that we need to pursue if we want as modelers to contribute to response efforts in the developing world, particularly in West Africa. 

Was the 2014 Ebola epidemic an opportunity in a sense for lessons learned going forward?  Will new knowledge emerge from all of this?

So I think the answer is yes, to the extent that there’s obviously a lot still to be learned about the best way to integrate dynamic models into the response process.  Obviously we would have liked to have learned these lessons without having to respond, but to the extent that we now know, first of all, for diseases like Ebola, the importance of considering spatial local structure.  The importance of coordinating early, and there are now, unlike prior epidemics, mechanisms of collaboration and communication, at least in the United States and in other countries as well.  For example, in the United States, there’s a group called BARDA, which it coordinated essentially groups of modelers to communicating, even pre-publication, so that the information and the predictions we were making could be essentially communicated and trickled up to policymakers.  And so to the extent that that is a communication and collaboration tool, as a lessons learned, that’s important. 

In terms of response, I think obviously earlier response is needed and in fact getting information out, particularly in integrating local customs.  With Ebola, the notion of the burial transmission was a key factor in many places in the initiation of spread, and so how the next time Ebola appears, especially in urban or rural areas, how we can communicate quickly to people to be cognizant of the risks, and also one of the big issues is how does one build infrastructure to isolate people quickly, and distinguish them from people who may have the same symptoms but have very different disease with different etiologies and different transmission risks? 

You and one of your colleagues also recently published a paper about modeling post-death transmission of Ebola. Tell us some of the findings that you report in that article.

Because of the nature of Ebola Virus Disease, it is possible to contract it even from a dead individual, someone who died due to Ebola.  And the reason is, again, in part, a combination of the etiology of the virus, the way that the virus persists in an individual, as well as cultural practices.   In these affected countries, it is, to some extent, common practice to have burial ceremonies in which bodies are physically touched.  Now that is in some sense, we might think of that as a very strange and unusual behavior or cultural practice.  But, if one takes a look at common burial practices in the United States, we are not so far away from that.  Common practice involves unveilings of bodies, being close, maybe not touching, but certainly being close.  And bodies are prepared for burial, often by funeral homes, yet it’s not a community effort, by and large.  So to the extent that this is a community effort to prepare the body, as well as touch the body as a form of marking respect, that practice, overlapping with a disease for which transmission can be mediated by contact with bodily fluids, leads to the possibility, and in fact evidence suggests that this happened many times, and may be responsible for about a quarter of the transmission of Ebola forward on to previously susceptible, therefore unaffected, individuals. 

The other risk here is then, how do we treat properly Ebola Virus Disease at the population scale?  And so my colleague and I, Jonathan Dushoff from McMaster University, asked the question, how much benefit would the response you could get in terms of controlling the disease from focusing on funerals, in addition to the focus on hospitalization and isolation of people who are alive?  So if post-death transmission is rare, then yes, it’s important, but it’s not going to have major impacts on the spread of the disease.  And to the contrary, our analysis suggests that based on the time series data at the very start of the outbreak, this is in the summer of 2014, that it was possible to fit the data with models in which a significant portion of the risk could be attributed to post-death transmission, and therefore we speculated that controlling post-death transmission could lead to the reduction of the forward transmission, about one-half of that which is necessary to lead at the population scale to stopping the epidemic.

We released this on what is called the arXiv, a public server that academics use to share information in advance of publication, because our hope was to try to get this information out to the policymakers early, and in fact in October and November, December, this sort of response was ongoing in West Africa.  There were extensive efforts to have burial teams, safe burial teams, to make them culturally sensitive, to explain why these burials needed to be safe, and to try to reconcile norms, whether religious or cultural norms, with the key public health need to stop forward transmission of Ebola during funeral ceremonies.

Do you think that increased attention to the extreme danger of post-death transmissions contributed to getting the outbreak under control?

I think the answer to that is yes.  There are multiple lines of evidence for it.  The CDC has released one of its case study reports based in Sierra Leone, where they reported the number of safe burials that took place in counties, and this was substantial.  And if each one of those safe burials reduced by three or four or five, and there are cases of dozens of people, in Mali there was initiation of, think about it, a dozen new affected cases from a burial ceremony.  So if each one of these burial ceremonies became a safe burial, you can imagine that at a population scale, you have a substantial impact on the transmission of the disease.  Fewer people were infected, therefore infect fewer people moving forward, and you move more quickly to an endgame scenario, we hope, rather than a continued epidemic crisis.  So to summarize, yes, I think that these safe burial teams were a key part.  The other issue there is that you can implement a safe burial team on a much more rapid time scale than one can build and staff an ETU [Ebola Treatment Unit] or a hospital, given circumstances, meaning the requirements from a materials and personnel and training perspective are significantly less, and yet, a person who is not infected is still a person who is not infected, either because you’ve moved them out of the community and isolated them, and they’re alive, or because after death you treated them with dignity and respect, but also gave them a safe burial. 

I know we’ve spent much of our time discussing the workshop and dynamic modeling, but you and your lab group also conduct very valuable research in other aspects of quantitative biology and theoretical ecology.  Would you give us just a quick thumbnail sketch of some of your other work?

Sure.  The focus of our research is on the interactions of viruses and their hosts.  Ebola is an example, where the virus infects humans.  But in fact much of our work focuses on viruses and their microbial hosts.  Microbes, whether they’re bacteria or archaea, or even small eukaryotes, meaning small organisms, get diseases.  These diseases are often viral diseases.  And my group is interested in the effect that these viral diseases have on the fate of an individual microbe, whether it lives or dies or enters a new state in which the virus in some sense takes over the, what’s called the nanobrain of the microbe, meaning alters its physiology and behavior.  And we’re also interested in how viruses affect the structure of entire populations and communities.  How many microbes are there?  Which kind?  What genes do they have?  And what sort of functions they perform at larger, ecosystem scales. 
We are a theory group, meaning we build models.  We interact with experimentalists and empiricists, and we work in very different environments, ranging from funded research on the viruses of archaea living in hot springs in Yellowstone National Park to studies of viruses that infect ubiquitous bacteria that live in the surface ocean in the North Pacific Ocean in Hawaii.  So common to these studies is really our effort to bring a non-linear dynamic perspective, similar to that which we try to bring to bear on Ebola epidemic dynamics, but also on the epidemic dynamics and the interactions that take place between viruses and their microbial hosts. 

Joshua, you’re making some very significant contributions to understanding and controlling epidemics large and small such as Ebola, and we wish you the best of luck for continued success. Thanks for joining us today on FOCUS In Sound…

Thank you, Ernie.

We hope you’ve enjoyed this edition of the FOCUS In Sound podcast.  Until next time, this is Ernie Hood.  Thanks for listening!

Note from Dr. Weitz:  

The workshop report can be found at http://bit.ly/ebm_gt_report

The report is 22 pages long and represents the collective efforts of the student rapporteurs.

The report is an intent to document the findings and discussions that took place in talks, panels and breakout groups. In doing so, there are many ideas and discussions that could not be explored at sufficient depth, even given the current length. Moreover, the ideas and discussions that are reported on may not necessarily be represented precisely in the way that authors would have done so.

Nonetheless, I hope that the report can provide a useful record of the workshop that helps facilitate modeling efforts and dialogue moving forward.

The report has been permanently deposited in an open access format on figshare.com.  Figshare is a repository to cite, share and discover information.  The report has a permanent doi:

http://dx.doi.org/10.6084/m9.figshare.1301267