epiforecasts
Infections in isolation are isolated after symptom onset
Since the model updates in PR #169 which hopefully resolve the issue around how cases in isolation transmit, a model assumption question has arisen.
Currently, if a transmission event occurs when the infector is in isolation, the infectee waits until they are symptomatic to also isolate. This seems sensible if we assume the infectee assumes not to be a case and is in the community. However, we might also want to allow a scenario in which infectees from isolated infectors isolate immediately upon exposure (I believe this would be equivalent in this model to assuming the infectee was already isolating).
Imagining this as a real-world scenario. If we set $R_t^{isolated} > 0$ because we assume cases isolate at home and then can transmit to others in the household. The current implementation assumes susceptible individuals in the household go out into the community (e.g. go to work) and are only treated as a case when they become symptomatic (assuming quarantine is turned off). The alternative scenario, described above, would isolate the infectee as soon as the infector is isolated. This would represent all individuals in a household isolating as soon as one individual in the household is isolating.
This is different from the current implementation of quarantine which affects all cases, whereas this proposal only early isolates cases caused from transmission in isolation.
I believe the change to the code would be relatively minor. A new argument to intervention_opts() and then a new conditional added to the assignment of isolation time in outbreak_step():
https://github.com/epiforecasts/ringbp/blob/088e52678ce90d828305c0537104bf809ba14cc5/R/outbreak_step.R#L145-L159
I've lost track of the model diagram here, but I thought transmission from isolation was definitional precluded (currently - I agree it might be interesting to model e.g. household transmission)?
This makes sense to me. That said, worry a bit about feature creep in an ever expanding model implemented with conditionals and decisions that implicitly preclude future model directions (or complicate them by requiring more forking paths). I could e.g. imagine situation were $R_t^{isolated} > 0$ and this does not represent household transmission (e.g. people have needs in the community, visit a doctor, go to a shop etc.). Is the suggestion informed by a particular need in an application that requires including household transmission? If that is the case, would it perhaps make sense instead to model household transmission explicitly (where you might want to have household size distributions, household secondary attack rates and household-specific interventions)?
This specific feature suggestion aside, I think it might make sense to come up with a design document for future model directions that would help put it into context, ideally solving https://github.com/epiforecasts/ringbp/issues/68 first (i.e., finding a way to accommodate different branching process models without forking or adding ever more conditionals). That would provide context for assessing suggestions like this one.
I thought transmission from isolation was definitional precluded
This was an area of confusion for me and relates to issue #148. I didn't see any documentation in the package that $R_0^\mathrm{iso}$ needed to be 0, nor was there an error or warning when a user set the offspring distribution for isolated individuals to be non-zero. However, I do think that prior to recent updates the package was not handling $R_0^\mathrm{iso}$ correctly. Thinking about how transmission in isolation should work led me to thinking about this issue.
Interestingly, I think this has been realised multiple times independently: https://github.com/timcdlucas/ringbp/issues/3.
Is the suggestion informed by a particular need in an application that requires including household transmission?
No, I'm not thinking about this as a household transmission feature, for that I think it'd need to be more of an agent-based/network model. I was only mentioning household transmission in isolation as an example. I think the feature would apply more broadly like the scenarios you mentioned.
that said, worry a bit about feature creep in an ever expanding model implemented with conditionals and decisions that implicitly preclude future model directions. ... I think it might make sense to come up with a design document for future model directions that would help put it into context
Let's discuss the overall development plan offline and then I'll feedback on relevant issues, including this one, with a development roadmap and links to any design document.
I've lost track of the model diagram here, but I thought transmission from isolation was definitional precluded (currently - I agree it might be interesting to model e.g. household transmission)?
Please see the vignette added in PR #188 for model diagrams. Hopefully this vignette will act as a useful resource for checking our assumptions about the model and can be updated as we further develop the epidemiological model in {ringbp}.