We (Sebastian Funk and I) recently wrote up a small case study using simulated data looking at an outbreak setting where partially reported onsets are used to estimate the underlying effective reproduction number. Sharing here as may be of interest and as it makes use of epinowcast.
The reasons for making this were:
- It provides guidance for practitioners who are currently handling this kind of data (for example the current Monkeypox outbreak).
- It provides a comparison with our other tools (i.e
EpiNow2). - It provides a baseline for practitioners to compare other tools against those we discuss.
- It highlights limitations in these tools and provides direction for future development (most of the limitations highlighted have WIP PRs to address them in
epinowcast).
Abstract:
In this report we make use of epinowcast, a nowcasting package under development and designed from the ground up around nowcasting with the aim of replacing EpiNow2 for real-time usage. We first explore the data using tools from
epinowcastalongside others. We then nowcast the latest available data, visualise our results, and discuss potential options for improving performance on real-world data. As a check of our approach we construct some retrospective data, nowcast it, and compare our nowcast to the latest available data. We then extract the estimated delay distribution and compare it to the underlying distribution used to generate the data, the empirical distribution, and the distribution estimated using simpler methods. Finally, we show how the output fromepinowcastmay be used in other surveillance packages such asEpiNow2. Throughout this case study we discuss potential issues with the approaches taken, and highlight areas for futher work. For more onepinowcastand it’s planned development roadmap see the package documentation. An alternative approach to the problem usingEpiNow2, suffering from some limitations thatepinowcastis aiming to address, is also available in this repository.
Report: