search

grattan / covidReff

Simulating Covid reproduction in a partially vaccinated population
Other
4 stars 1 forks source link

Add a vaccine effectiveness time schedule #27

Open andrewlilley opened 3 years ago

andrewlilley commented 3 years ago

The model currently retains info on when each person was vaccinated with their second dose. This allows them to carry a specific daily immune effectiveness to account for waning. The following two papers have reliable estimates if this feature were to be added. https://www.ndm.ox.ac.uk/files/coronavirus/covid-19-infection-survey/finalfinalcombinedve20210816.pdf (estimates for 0-3 months, Pfizer and Moderna) https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3909743 (Pfizer out to 6 months) Cheers

wfmackey commented 3 years ago

Thanks Andrew. Good points and definitely features that should be included in the model.

This feature will require a small restructuring of the internal structure of the model. For each agent, the model currently has a 'days since vaccinated' variable which counts upwards from an agent's first dose toward their second. This number can be used as an input to their protection against infection/hospitalisation/icu/death, based on their vaccine type and age.

I guess a question of implementation here is: what is the appropriate time-dependent function, for each of the vaccine options (AZ or Pf)? And whether this also varies by age.

andrewlilley commented 3 years ago

Thanks Will. Also I should have tried to contribute a bit more in raising a feature request- sorry about that. Hopefully can fix that today.

I took some time today to try to contribute on the appropriate schedule of efficacy vs time. Here's my suggested schedule for Pfizer and AZ for first and second doses. (Can do a pull request to upload it if that's better but figured the file upload isn't the hardest bit.)

I used four papers to make this schedule. Instead instead of trying to do some complex weighting, I just used the best paper that was able to cover that datapoint.

  1. The best paper I think by far is the Oxford paper by Pouwels et al. mostly because it has compulsory testing: https://www.ndm.ox.ac.uk/files/coronavirus/covid-19-infection- survey/finalfinalcombinedve20210816.pdf However, it doesn't have any hospitalization data and only measures efficacy at up to 104 days after the second dose is given.

  2. The next best is the Lancet paper from administrative data for Kaiser permanente because it does show efficacy at various points of "months after vaccination": https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3909743 We can use this paper to fill in the remaining data for Pfizer for 2x doses for infection and hospitalization. However we are still missing: Pfizer hospitalization with one dose, and Astrazeneca hospitalization (1 and 2 dose)

  3. For these two we can use the Ontario study and the PHE study by Bernal et al., both are well known from this infographic: https://www.ft.com/content/5a24d39a-a702-40d2-876d-b12a524dc9a5 I can't actually find the links right now, but here's a screenshot I made a couple months ago of the results: https://twitter.com/andrewlilley_au/status/1418744590492864515?s=20 Neither are ideal and lack a lot of controls, and in the PHE paper they dropped the hospitalization estimates from their publication in NEJM because they didn't trust them, but we have to play the cards we're dealt here. I just picked on "narrowest confidence interval," so I used the Ontario study for 1x dose estimates for Pfizer, and the Bernal-PHE study for both doses of AZ.

There's not great evidence by age to be honest (though I can share some papers suggesting they're broadly similar, except with little to no waning for people under 20).

wfmackey commented 3 years ago

Thanks Andrew, perfect. I'll add this shortly.