Modelling work presented in a recent Nature Comment by Whitty et al. suggests that, without any intervention, the Ebola outbreak in Sierra Leone could peak early December, reaching a maximum of 1600 weekly cases.
Here, we show that this figure is misleading and provide a fully transparent and reproducible analysis of the same data-set.
All data, models, outputs and figures are provided in this repository (see for instance here for the baseline model).
Alternatively you can re-run our analysis using the instructions below.
A detailed step-by-step description of our analysis can be found here.
At the core of our analysis is the SSM (for State Space Models) library, which allows us to
- implement a transmission model
- fit this model to time-series data and estimate key parameters
- forecast this fitted model by integrating the uncertainty of parameter estimates
To re-run our analysis:
git clone https://github.com/ntncmch/ebola_sierra_leone.git
- Compile the model
cd ebola_sierra_leone/SSM
ssm
- Fit the data
cd bin
cat ../theta.json | ./simplex --iter 10000 --prior | ./pmcmc --iter 10000 --eps_switch 20 --switch 500 --cooling 0.99 | ./pmcmc --iter 100000 --eps_switch 20 --trace --acc --traj --n_traj 1000 -v
- Forecast the fitted model
ssm-predict mcmc_0.json X_0.csv trace_0.csv 2014-09-28 | ./simul --traj --start 2014-09-28 --end 2016-08-01 --freq 7 --id 1
All outputs were processed using R.