I don't understand how people can model epidemics like this. Isn't the infection and spread for something like ebola highly variable, and mostly a function of education and resources?
Unlike the Flu, which is easy to pass on, Ebola is hard to catch, and unlike HIV, has a fairly brief infectious period. The WHO could, with the assistance of local government, basically shut down this epidemic in a matter of months with field hospitals, education, and proper burial/isolation practices.
I'm not sure how you model that kind of response when trying to make future predictions.
You can sure model pretty much anything without knowing internal variables and complexities of the system. In fact that's the entire premise of the field learning from data. Of course you need good quality data and in sufficient quantity.
Your model will almost never be perfect but you can evaluate confidence intervals by using part of the past data (training set) and making prediction for events that is not included in that data (test set) and find out how good is your model. This process is scattered with land mines. People frequently end up overfitting or do not take in to account variability trends or power laws or possible "black swans". So all such models you want to take with grain of salt in any case.
For this particular case, I don't think we have enough data yet to make predictions with significant confidence. From past epidemics in same region, it will very likely that this will end without getting out of control. One of the big deal about current strain is that it doesn't manifest symptoms until 1-3 weeks which gives plenty of time for virus to spread.
One should never discount the power of power law. We are safe only until tipping point is not researched. I think that tipping point is about 10,000 infected people in a city. If this happens, it would take extreme measures to prevent spread (something like completely isolating entire countries). But by that time, virus might have already escaped elsewhere.
I understand what you are saying, and if this were the Flu, then I would totally agree with you. But this is a terrible disease that has a lot of people's attention. The WHO is about to invest $100 Million (http://arstechnica.com/science/2014/07/calling-ebola-outbrea...) in curtailing it.
I have to believe that this disease would spread completely differently, under the scenarios (A) No external intervention, (B) WHO invests $100 mm in stopping it, and (C) World community jumps on this immediately and invests $1B to immediately curtail it.
In each of those three different scenarios, the epidemic would progress differently. Perhaps it's being modeled under (A), and things will work out better than expected. We can only hope.
First, its an entirely reasonable thing to model a disease under the assumptions of no interventions, as there's a great deal of value in knowing "Where is this headed if we don't do anything", and to find out whether or not the time and resources you're spending are actually having an impact.
Second, while the international response to this is good, putting in money doesn't have a linear return on reduction of disease. Consider HIV or Malaria.
HIV and Malaria are both broadly endemic in Africa, including some conflict areas - it's going to be a long tough slog to rid the world of them.
This breakout of Ebola is geographically isolated, and its nature (short infection period, burns out quickly, hard to catch) - makes it something that a strong effort right now would absolutely shut down.
Regarding modeling the disease - yes, of course - from that perspective I can see how useful it would be to model, but the assumption should be pretty explicit. I.E. "Assuming traditional regional responses to this disease, this is what we would expect" - to make it clear that this is a model, and not a prediction.
One of the problems with Ebola is the relatively long incubation period which when coupled with easy worldwide mobility means that cases can spread unnoticed and start popping up in other countries unexpectedly. You can see a similar scenario using Smallpox here:
Smallpox is a very different bug than Ebola (harder to catch,etc) so this kind of worldwide scenario is much less likely. Education and surveillance will certainly help a lot and we did get lucky because we "saw it coming", in the sense that people are already paying attention before it starts spreading on a larger scale.
Thank you for making this. I've been following the outbreak in the news and searching for something better than repeating quotes from the WHO or politicians saying everything is going to be fine.
Even if this models isn't 100%, it's better analysis than other easily accessible analysis I've found. I'm certain it will be of assistance to those trying to help and needing to understand. It may spur others to do similar work.
It looks like something like a sixth of his writeup is devoted to chains of MCMC convergence diagnostics. Is non-convergence really that serious an issue for epidemiological modeling?
Non-convergence can be fairly serious in epidemiological modeling, and crop up somewhat unexpectedly. Beyond that, it's really easy to produce the trace plots to check it.
18 comments
[ 2.7 ms ] story [ 51.4 ms ] threadUnlike the Flu, which is easy to pass on, Ebola is hard to catch, and unlike HIV, has a fairly brief infectious period. The WHO could, with the assistance of local government, basically shut down this epidemic in a matter of months with field hospitals, education, and proper burial/isolation practices.
I'm not sure how you model that kind of response when trying to make future predictions.
Your model will almost never be perfect but you can evaluate confidence intervals by using part of the past data (training set) and making prediction for events that is not included in that data (test set) and find out how good is your model. This process is scattered with land mines. People frequently end up overfitting or do not take in to account variability trends or power laws or possible "black swans". So all such models you want to take with grain of salt in any case.
For this particular case, I don't think we have enough data yet to make predictions with significant confidence. From past epidemics in same region, it will very likely that this will end without getting out of control. One of the big deal about current strain is that it doesn't manifest symptoms until 1-3 weeks which gives plenty of time for virus to spread.
One should never discount the power of power law. We are safe only until tipping point is not researched. I think that tipping point is about 10,000 infected people in a city. If this happens, it would take extreme measures to prevent spread (something like completely isolating entire countries). But by that time, virus might have already escaped elsewhere.
""The patients become contagious once they begin to show symptoms. They are not contagious during the incubation period.""
Source: http://www.who.int/csr/disease/ebola/faq-ebola/en/
It still gives them time to diffuse from where they were infected.
I have to believe that this disease would spread completely differently, under the scenarios (A) No external intervention, (B) WHO invests $100 mm in stopping it, and (C) World community jumps on this immediately and invests $1B to immediately curtail it.
In each of those three different scenarios, the epidemic would progress differently. Perhaps it's being modeled under (A), and things will work out better than expected. We can only hope.
Second, while the international response to this is good, putting in money doesn't have a linear return on reduction of disease. Consider HIV or Malaria.
This breakout of Ebola is geographically isolated, and its nature (short infection period, burns out quickly, hard to catch) - makes it something that a strong effort right now would absolutely shut down.
Regarding modeling the disease - yes, of course - from that perspective I can see how useful it would be to model, but the assumption should be pretty explicit. I.E. "Assuming traditional regional responses to this disease, this is what we would expect" - to make it clear that this is a model, and not a prediction.
http://www.nature.com/srep/2013/130717/srep00810/full/srep00...
Smallpox is a very different bug than Ebola (harder to catch,etc) so this kind of worldwide scenario is much less likely. Education and surveillance will certainly help a lot and we did get lucky because we "saw it coming", in the sense that people are already paying attention before it starts spreading on a larger scale.
Isn't that "differentiation"?
http://en.wikipedia.org/wiki/Compartmental_models_in_epidemi...
Even if this models isn't 100%, it's better analysis than other easily accessible analysis I've found. I'm certain it will be of assistance to those trying to help and needing to understand. It may spur others to do similar work.