Earlier this week, I did a quick Bayesian analysis on the likelihood of Ebola hopping on a plane and making its way to the United States. There are a number of critical caveats to the method that I used, and while they’re mentioned in the text of the original post, I wanted to take some time tonight and make them as clear as possible:
1. The inputs to the model are static, so the output only applies for the time at which the model was originally calibrated. (However, this is easily addressed by just plugging in new statistics as we get them.) Generally speaking, probability of the event of interest occurring will increase as long as the active (infective) number of people are on the rise.
2. The catchment population considered by the model should be smaller than it is currently. Right now, the model considers the vulnerable population to be the entirety of Guinea, Liberia, and Sierra Leone. For increased accuracy, the catchment population should be re-calibrated to include only the populations of cities that have experienced secondary transmission*.
3. To properly assess the likelihood of an infected individual unintentionally boarding a plane from West Africa to the United States, we need detailed flight-by-flight data that provide deeper insight into passenger itineraries (AKA initial departure, stop over, and final arrival locations)*. I’m hoping to acquire this data soon, courtesy of HealthMap and the CDC! We also need to consider that not every individual is as likely to fly as the next*. (For instance, city-dwellers are more likely to fly than farmers; this is particularly concerning due to the fact that Ebola has hit metropolitan areas pretty hard in affected countries.)
4. [IMPORTANT] This particular model does not evaluate the risk of secondary transmission in the rare event that an Ebolavirus-infected individual unintentionally makes it to the United States…
5. [IMPORTANT] …And on that note, this model does not evaluate the risk of secondary transmission in the likely event that an EBOV-infected individual intentionally makes it to the United States.
*The purpose of the original blog post was to take an initial pass at developing the bare bones of a Bayesian probability model for the event of interest. (Just a fun exercise, really.) If I pursue it any further, I’ll be sure to tighten up the parameters and consult with those much wiser (and more experienced) than me! For the time being, the model only explores the likelihood of an EBOV-infected individual unintentionally flying to the United States. It does not assess the likelihood of secondary transmission in the United States… And I’m not yet sure whether I’ll ever try to create a coupled counterpart model that does do this. I guess we’ll just have to wait and see!