[New/Revise] Deprecation of SIR-FV model because vaccination effect should be discussed with SIR-F parameters

lisphilar / covid19-sir

CovsirPhy: Python library for COVID-19 analysis with phase-dependent SIR-derived ODE models.

https://lisphilar.github.io/covid19-sir/

Apache License 2.0

110 stars 44 forks source link

[New/Revise] Deprecation of SIR-FV model because vaccination effect should be discussed with SIR-F parameters #573

Closed lisphilar closed 3 years ago

lisphilar commented 3 years ago

Summary of this new feature

The number of vaccinations is a key factor to end outbreak. Because we have vaccination data (VaccineData), we can add the number of vaccinations as a variable to Scenario class.

lisphilar commented 3 years ago

(Cc. @Inglezos and @andrybicio) Addition of the number of vaccinations as a variable to Scenario class enables us to use the number of vaccinations as a variable in SIR-FV model. Currently, vaccination effect is included in reduction of Susceptible variable, but this is not useful for parameter estimation with SIR-FV model.

SIR-FV considers vacctnation of sussceptible people as follows. ω×N persons will be vaccinated in a day. N is the total population.

lisphilar commented 3 years ago

Alternative: Use the number of vaccinations when calculating "Susceptible". This is better because vaccination impacts on "Susceptible" variable, not I, R, F. ODE model is still SIR-F.

lisphilar commented 3 years ago

New parameters when calculating Susceptible:

inverse of "period to get effectivity of vaccination"
inverse of "period to lost effectivity of vaccination"

lisphilar commented 3 years ago

Alternative: Regard "vaccination" as a control factor of rho/sigma/kappa, not a variable of ODE models. With this idea, we can use VaccineData as a explanatory variable in Scenario.fit_predict() as well as OxCGRTData.

Inglezos commented 3 years ago

Yes, I think the last solution is the more correct approach, because the vaccinations: i. have effectivity property which alters the result and ii. affect infectivity but probably cannot stop the transmission, they simply protect the person itself from developing symptoms and from becoming sick.

This shall be represented in the parameters as control factor. In no case the vaccinations mean that a person is removed from the S compartment and in the future never affects again other people. He could either be infected eventually due to decreased vaccine effectiveness or infect others as well as a carrier by transmitting the disease to non-vaccinated people.

lisphilar commented 3 years ago

Thank you for your confirmation. Yes, "Vaccinated" could not be separated from "Susceptible" and so on. We cannot calculate "Period to get (lost) effectivity of vaccination" with datasets at this time. (We may find clinical studies for that with repeated PCR tests and antibody tests at many timepoints, but the number of subjects is limited.)

In our data-driven approach, I think SIR-F model can be the final ODE model for COVID-19. i.e. All records except for Confirmed/Rrecovered/Fatal/Population will be regarded as controll factors.

lisphilar commented 3 years ago

My idea:

Creating instance of covsirphy.SIRFV raises FutureWaring: SIR-FV model will be removed in the future because vaccinated persons may move to the other compartments. Please use SIR-F model and adjust parameter values of SIR-F model, considering the impact of vaccinations on infectivity, effectivity and safety.
We will discuss how to estimate the impact of vaccinations on infectivity, effectivity and safety.

lisphilar commented 3 years ago

Update: raise NotImplementedError, not warnings.