menu

covid19-scenarios

COVID-19 Scenarios is a project to model how the novel coronavirus is spreading and how much strain this will put on the health care system.

Channels
Team

Questions, discussions

March 21, 2020 at 4:20pm

March 21, 2020 at 11:01pm
Don't hesitate to post your questions and ideas here. We will be glald to help!
  • reply
  • like
Hi, I tried to implement the model you describe on your website https://neherlab.org/covid19/about. However I got stuck on understanding the differential equation for S_a, in particular what the parameter I_b stands for. I was unable to find the equivalent in the code of this repo. In my case the susceptibility decreases rapidly, as the formula I use is dS_a/dt = -beta * S_a * I_a. I am certain that this formula is wrong. Any help would be appreciated. Best regards, Richard
Edited
  • reply
  • like
Hi Richard, Fantastic! We would love to have some peer validation. The science part is handled by Nick and Richard , they will reply to you as soon as possible!
You've also posted on Github, so here is a back-link, just in case: https://github.com/neherlab/covid19_scenarios/issues/18#issuecomment-602118407
Edited
  • reply
  • like
I think I figured it out now, I changed the formula to dS_a/dt = -beta * S_a * I_a / population. Now it seems to work kind of. I am no expert on those matters, it just seems to look right.
Edited
  • reply
  • like
I_b stands for the number of infectious individuals in the "b" age-group. The sum of I_b over b is the number of total infectious individuals. This particular line of math is modeling the infection chance that any given individual in the susceptible population has of becoming infected, each time step of dt. If you assume each person interacts with one person per time step, then the probability they come into contact with any infectious individual is the total number of infectious individuals over the population size. What you have written means individuals can only infect those in their age group. I hope this was helpful -- this is a bit hard to write without latex!
Edited
  • reply
  • like

March 22, 2020 at 8:07am
Hi 👋! I was playing with the scenario modeller after seeing https://twitter.com/richardneher/status/1241458447457157127 which made me then wonder: how hard would it be to get the data needed for mainland china and have a scenario based on that? would be interesting to see how well the model fits with a country where (for now) new cases are much lower than during the peak
like-fill
1
  • reply
  • like
Hello, I'm new and a specialist of some sort. I'm trying to estimate the SIR or extended SIR parameters from data. Afaiu, you don't directly fit your model to the observations. Is this true and if yes, what are your reasons for not doing it?Here's my git of what we're trying to do: https://github.com/sechseck/corona-epidemics-model. Currently, we're having problems with convergence and strong dependencies of the outcome on the first few unobserved cases.
  • reply
  • like
Hi , you are right, we don't fit at present. When we started this 3w ago, there wasn't much data to fit (other than China and we were using parameters estimated from the Chinese data). We are thinking of how to use existing data now and will hopefully come up with a solution soon!
  • reply
  • like
you are right, there is an 1/N missing in the formula (but only on the about page, not in the code...) Will fix. Thanks for pointing it out!
  • reply
  • like
I'd be happy to contribute
  • reply
  • like
How is importsPerDay calculated? Is it the number of current total cases divided by days that have passed since March 1st?
Edited
  • reply
  • like
The imports per day are currently just guessed. They only matter before substantial virus circulation is established. We are thinking about estimating these from travel patterns and cases around the world. But travel patterns are changing by the day...
  • reply
  • like
we first have to be a little more flexible regarding the initial date. Once we have this, we can make scenarios for the outbreaks in China and also the past dynamics in Italy.
  • reply
  • like
I started a (very) rough PR here that just fits a simple exponential to death. Have a look: https://github.com/neherlab/covid19_scenarios_data/pull/27 if you have ideas. let us know.
Edited
  • reply
  • like
I started a (very) rough PR here that just fits a simple exponential to death. Have a look: https://github.com/neherlab/covid19_scenarios_data/pull/27 if you have ideas. let us know.
I'll get back in two hours, need to attend something
  • reply
  • like
Hello, I'm new and a specialist of some sort. I'm trying to estimate the SIR or extended SIR parameters from data. Afaiu, you don't directly fit your model to the observations. Is this true and if yes, what are your reasons for not doing it?Here's my git of what we're trying to do: https://github.com/sechseck/corona-epidemics-model. Currently, we're having problems with convergence and strong dependencies of the outcome on the first few unobserved cases.
If you doing that please mind under reporting or under-diagnosed cases since those depend on testing level
  • reply
  • like
do you have any details on your agent based approach for mitigation mesures that you could share? I'll move my comments on testing here.
  • reply
  • like
we first have to be a little more flexible regarding the initial date. Once we have this, we can make scenarios for the outbreaks in China and also the past dynamics in Italy.
Cool! Would be nice in order for non experts (i'm a ex-particle physicist) to get a feeling for how the model works, how reliable it is in terms of dealing with behaviour/measurement changing, etc. How far out can the model predict and get close to what actually happens? So this is mostly for my education and curiosity :)
like-fill
1
  • reply
  • like
Cool! Would be nice in order for non experts (i'm a ex-particle physicist) to get a feeling for how the model works, how reliable it is in terms of dealing with behaviour/measurement changing, etc. How far out can the model predict and get close to what actually happens? So this is mostly for my education and curiosity :)
ps. very cool work and +100 for putting it all on github!
  • reply
  • like
ps. very cool work and +100 for putting it all on github!
when it becomes sensible again to meet up outside for a favourite beverage i'd be super happy to get a round for y'all (i live in Brugg, just around the corner from Basel)
  • reply
  • like
do you have any details on your agent based approach for mitigation mesures that you could share? I'll move my comments on testing here.
Yes, that is why I am proposing to fit the death count rather than the case count. this should be less affected by underreporting. An agent based model would probably have to run on a backend...
  • reply
  • like
I would say the big question at the moment is how manage health services overload without compromising too much the economy. Even if effective, strict quarantine measures will be hard to keep in place for a long time. Another tool available is testing that can be used in two diferent approaches: detect infected (pcr+) and quarantine those (good for initial phases, then account for those only household transmission, probably at a lower level), and/or detect immunes (Ac+ and Pcr-) and "release" them to work.
Edited
  • reply
  • like
The effect of testing coverage or diferent risk-based testing approaches on mitigation could be a good tool to provide to decision makers at this moment.
  • reply
  • like
What about suspectedCaseMarch1st? In some countries (in south-eastern Europe), the first cases started to appear after March 10th.
Edited
  • reply
  • like
If you doing that please mind under reporting or under-diagnosed cases since those depend on testing level
Hello pinaunes, i would assume that all individuals that end up in a hospital have been reported as a case. Of course there's going to be many more cases, but we should not be too interested in them as the only way for them to go to the critical stage is via the "reported as infected" channel. I agree that the different densities of testing in the countries makes comparison hard. But as long as the testing regime does not change over time the case numbers should be a good predictor of the dynamics of the process. Just my five cents, correct me if I'm wrong. The death cases are more reliable, i guess but definitely lagging by two to three weeks (my estimate, not having looked at the figures). Plus the population is much smaller and figures derived from that might be less stable
Edited
  • reply
  • like
Show more messages