A few weeks ago, I started working with David Fisman’s IDEA method for short-term outbreak forecasting. In addition to approximating R0, the method allows us to assess how effective control efforts have been (d) at curbing growth in a given outbreak. (I personally think this is the coolest attribute of the method – and one that’s sorely missing in more conventional approaches.) Once we’ve optimized R0 and d, we can also project the expected magnitude and longevity of the outbreak in question (assuming that both parameters remain constant over time).
Since August, I’ve been using the IDEA method to model Ebola in West Africa. (Boston Globe’s BetaBoston team recently published a few of my remarks on the process.) I reran my model again today with data from the most recent WHO Disease Outbreak News report. Here’s a chart that shows how the model fits against the WHO data and projections for the next two weeks:
According to the model, we’ll see 800 new cases this coming week and 1200 the week after. Needless to say, I really hope I’m wrong… But for the time being, we’ll have to just wait and see.
A couple of notes on parametrization:
(1) Generation 0 is set to December 6, 2013 – index case date of death. I used date of death instead of date of onset to serve as a proxy for date of reporting. From March 23, 2014 onward, case data were plotted exclusively by date of reporting
(2) Serial interval length (generation length) is set to 18 days. I considered both duration of incubation and duration of illness for the serial interval calculation. A proportional weighting schematic was utilized to account for variance in duration of illness among fatal versus non-fatal cases.
(3) Potential impact of under-reporting is not factored into the methodology (for the time being).