search

epiverse-trace / epidemics

A library of published compartmental epidemic models, and classes to represent demographic structure, non-pharmaceutical interventions, and vaccination regimes, to compose epidemic scenarios.
https://epiverse-trace.github.io/epidemics/
Other
8 stars 2 forks source link

Further fixes for the ebola model #132

Closed pratikunterwegs closed 10 months ago

pratikunterwegs commented 10 months ago

This PR contains further fixes to the ebola model:

  1. Fixes #101 by randomising assignment of exposed, infectious, and hospitalised into the Erlang subcompartments,
  2. Fixes #134 by modifying model structure to invert prop_hospitalised, etu_safety and funeral_safety to parameters representing risk, so that <rate_intervention>s can be used to reduce these values without allowing #109,
  3. Fixes #111 by allowing time-dependence, along with a vignette section on using time-dependence to simulate vaccination.
pratikunterwegs commented 10 months ago

Thanks @jamesmbaazam for the review, I'll take a look tomorrow and aim to get it merged by the start of next week.

pratikunterwegs commented 10 months ago

Thanks @jamesmbaazam, that was helpful. I've merged most of your suggestions, and taken most of the others too.

Thanks, @pratikunterwegs. These are nice changes that will introduce some flexibility in the ebola model. I have left line comments. Below, I'm leaving other comments that could not go on lines that were not changed.

  • In general, the state names vector c( "susceptible", "exposed", "infectious", "hospitalised", "funeral", "removed" ) is repeated in several places in the code. It would be better to define it once and then use it everywhere.

For example, if defined as

states <- c( "susceptible", "exposed", "infectious", "hospitalised", "funeral", "removed" )

then

  • L290 - probably safer to use length(states) in case the initial state changes in the future.

  • L291 - probably safer to do colnames(sim_data) <- states

Thanks, I've defined the compartment names in the function body and reused them for names and lengths

  • Would be nice to have one or two quick examples in the function documentation. You could provide examples with and without time
    dependence just to reinforce the idea in the user.

Broadly agree, however the setup steps for these models are pretty long and are better shown as vignettes to clearly explain setup.

  • For the doc of the time_dependence argument, please link to
    the time_dependence.Rmd vignette.

Linked from all instances of this argument.

  • In the Ebola vignette, I'm a bit confused about why the scenarios with intervention tend to have higher outbreak sizes than the baseline as they take off. I would have expected the opposite.

My guess is that this is an artefact, not a real effect. The interventions mostly take a while to have any effect on the epidemic size.

  • More broadly, it would be nice if you could unify the interface by finding a way around applying the <vaccination> class to these types of discrete-time models. It would remove any kind of confusion in using time-dependence to model vaccination in Ebola but <vaccination> to model vaccination in the other models.

This could be done although it would take some time as the vaccination objects are optimised for the ODE models. More relevant though is that this model doesn't have a vaccinated compartment, as this is not included in Li et al.'s consensus model. We can consider adding one, but my impression is that EVD outbreaks are expected to end before immunity due to vaccination becomes really important.

jamesmbaazam commented 10 months ago

More relevant though is that this model doesn't have a vaccinated compartment, as this is not included in Li et al.'s consensus model.

That's true. I forgot about that.

We can consider adding one, but my impression is that EVD outbreaks are expected to end before immunity due to vaccination becomes really important.

Reactive ring vaccination is possible. Conceptually, these individuals will not partake in transmission dynamics because they are removed into the absorbing state (here, the removed class) and could contribute to herd immunity.

pratikunterwegs commented 10 months ago

Reactive ring vaccination is possible. Conceptually, these individuals will not partake in transmission dynamics because they are removed into the absorbing state (here, the removed class) and could contribute to herd immunity.

Sure, thanks - I would say that that adding that functionality might require building off a different kind of model which is already present in {epichains} or {ringbp}? Happy to include once the scope of the different packages is decided.