search

igoldsteinh / improving_rt

MIT License
5 stars 0 forks source link

Appreciation and some thoughts. #1

Open seabbs opened 5 months ago

seabbs commented 5 months ago

Hi!

I liked the paper and very glad someone else is talking about this: "Consequently, EpiEstim can be mimicked via Poisson regression with an identity link and no intercept." though it felt a little strange to be talking about EpiEstim in a section titled state of the art methods!

a preliminary study by Abbott et al.

I also enjoyed this as a way of describing my hilarious preprint with no results text!

I just wanted to note that I think (but could be wrong!) your adaption of scaling ascertainment by number of tests could be incorporated into epidemia via the observation model and testing this would make the case stronger for your approach (similarly it would be useful to see the performance of your model without this test scaling). I don't see a discussion in the paper of other common approaches people take to this problem (Binomial and Beta-Binomial obs models come to mind) and the relative trade-offs of those would be a great addition imo.

See this example for extension options in epidemia: https://imperialcollegelondon.github.io/epidemia/articles/multiple-obs.html

Discretizing distributions

Also noting some recent work here that might be interesting for future work (Figure 9 in particular): https://www.medrxiv.org/content/10.1101/2024.01.12.24301247v1.full-text

As an aside, I currently run a real-time analysis community https://epinowcast.org with a seminar series - potentially this or similar work would be a good fit for that if interested. You may find your way to some packages hosted at epinowcast.org and if interested I would read https://arxiv.org/abs/2308.13262 if not already as it does a much better job of describing the overall model! Note that these tools are designed to be community developed and so making sure some of your work could be run within the framework would be really interesting - a post on community.epinowcast.org is usually how we get the ball rolling with that.

igoldsteinh commented 4 months ago

Thanks for reaching out, and for reading the paper! I was on vacation, hence the delayed reply.

It would definitely be interesting to make a test model in epidemia and to compare against other observation models. Thanks for the paper links, I appreciate all the detail in the delay distribution paper, and will keep epinowcast.org in mind.

Giving a talk at the seminar series sounds great, let me know when you have openings.