search

Urban-Analytics / RAMP-UA

https://urban-analytics.github.io/RAMP-UA/
MIT License
15 stars 10 forks source link

Notes about current covid data #279

Open nickmalleson opened 3 years ago

nickmalleson commented 3 years ago

Document current status of covid data availability, e.g. what APIs are available, what the different pillar data are, how this relates to original seeding, what it might mean for new seeding, etc.

I have created an initial document on the data_notes branch:

https://github.com/Urban-Analytics/RAMP-UA/blob/data_notes/experiments/covid_data_notes.md

@HSalat @spoonerf please could you contribute to this?

github-actions[bot] commented 3 years ago

Branch spoonerf-issue-279 created!

github-actions[bot] commented 3 years ago

Branch HSalat-issue-279 created!

HSalat commented 3 years ago

The comment I made together with my edits is a bit hidden, so I'll repeat it here.

Hi!

I've added what I propose to aim for and what we would need to achieve that aim. It's quite the same as what's already been done for Devon in fact.

Essentially, we only need 3 things:

*Ideas: can be at national level and assume similar reporting issues across regions; can eliminate some testing bias by counting symptomatic people directly...

spoonerf commented 3 years ago

I wonder if we could estimate a regional infection fatality rate using age and health data?

We are already using the table below in the model to estimate mortality rate based on a persons age and BMI. We could reasonably estimate a regional IFR based on the age and health of the population.

We could then work backwards using death data to estimate a time-series of infections with an approx 22 day lag. The smallest region for death data availability is Lower Tier Local Authorities.

Do you think that could work @HSalat @nickmalleson ?

HSalat commented 3 years ago

It does not compare well to the reported cases data, which is available at MSOA level and has a more controlled lag.

Although, what could be done is estimating the number of true cases at LAD level the way you suggest and then sort them between MSOAs using the reported case data instead of the risk data. But it would still be simpler to find directly a reported to true coef. if such a thing exists.