Adding contact tracing

[DAM-Philippon]

An important intervention that we haven't implemented yet is contact tracing. For a disease such as COVID-19, with a possibility of asymptomatic and presymptomatic transmissions, adding a policy that will look at the contacts in the last X days of an identified case is very important (even more than it is used by Vietnam, which is our "original" case study).

In order to implement that, we have different solutions:

• either having a map defined in the Individual species, representing its last contact (i.e., step, possibly exposure duration) with another individual,

• or having a matrix in global (NxN size, with N being the number of Individuals in the simulation)

The second idea might be better since defining the map of contacts in the Individual species might increase execution time, while the map would not be used for anything but this policy, wasting resources when the policy is not applied.

[codeanticode]

@DAM-Philippon @AlexisDrogoul Hello, I'm using COMOKIT to generate synthetic individual-level data to parametrize a population-level SEIR model using maximum likelihood estimation. Thanks for making your GAMA COVID models available to the community!

One of my aims is to develop individual risk scores of infection that take contact tracing, demographics and other data sources as inputs, which could eventually be used to alert users in real life. At this time, I'm just testing the idea within GAMA wit simulated populations :-)

Anyways, since you are looking to add contact tracing to your models, I'd like to mention this related work I'm doing because I think it is relevant to that end. In fact, I managed to implement a risk score using an older version of the COMOKIT models. My next step is to simulate the effect of risk-based alerts, and thought it could be done by adding a quarantine policy such that if risk > 0.5 (or some other threshold), then the individual self-quarantines at home for 14 days. What do you think? My fork of the COMOKIT project is here:

https://github.com/broadinstitute/COMOKIT-Model/tree/covid19-ilm/ILM-COVID19

Anyways, I think we can collaborate on this specific topic. Pinging @agrignard, with whom I've been discussing these ideas for the last couple of weeks.

[DAM-Philippon]

Hello @codeanticode ! Thank you for your comment. I've looked at your model and I'm looking forward to see it being extended. The idea of using symptoms prevalence to define a risk score is very interesting. If I understand correctly, the idea is that Individuals would do self evaluation/self reporting of their symptoms, am I correct?

Adding potential confirmed (or with a high score) contacts makes sense for your objective. This intervention looks similar to what was applied in Mainland China and Singapore, at least for the contact tracing and risk profile part. I don't think they included self report on symptoms though.

We were actually thinking of using a map to represent the contacts (a pair composed of the ID of an Individual and the last step of their contact), and use a distribution probability to forget some of them (the older the contact, the higher the chance of forgetting) for a given maximum number of days of records, possibly 21 days, since this was used in Singapore.

However, for your project, if the contacts are traced through something like a phone (as in Mainland China), all the contacts should be remembered (since a phone does not "forget"). I did not consider setting the distribution probability of remembering a contact as a parameter, but thanks to your input, I'll definitely will !

[codeanticode]

@DAM-Philippon sorry missed your reply :-)

Thanks a lot for your comments. Responding to your first question, yes the idea is that the symptom data will come from self evaluation/reporting.

I managed to write a manuscript describing the entire computational experiment, including the data generation with COMOKIT. It's here. My plan is to upload it to a pre-print server soon.