You are here

Planning as Inference in Epidemiological Models

COVID-19 Research Area(s): Epidemiology & Public Health

Recent reactions to the COVID-19 pandemic have shown a lack of preparedness in response to public health emergencies. In particular, there has been a lot of uncertainty in modeling the pandemic and predicting the path it would take depending on the interventions enacted by the government. These predictions are vital to devising the right policies in response to the pandemic. In this project, we wish to use probabilistic programming software developed by the PLAI group to automate the exploration of policy options in response to a pandemic like COVID-19, taking into account not only the public health aspect of the policy options, but also their socioeconomic impact.

In this work we demonstrate how existing software tools can be used to automate parts of infectious disease-control policy-making via performing inference in existing epidemiological dynamics models. The kind of inference tasks undertaken include computing, for planning purposes, the posterior distribution over putatively controllable, via direct policy-making choices, simulation model parameters that give rise to acceptable disease progression outcomes. Neither the full capabilities of such inference automation software tools nor their utility for planning is widely disseminated at the current time. Timely gains in understanding about these tools and how they can be used may lead to more fine-grained and less economically damaging policy prescriptions, particularly during the current COVID-19 pandemic.

Populations per compartment during deterministic SEI3R simulations, both without intervention (top) and with intervention (bottom). Plots in the left column show the full state trajectory, and in the right column are cropped to more clearly show the exposed and infected populations. Without intervention, the infected population greatly exceeds the limit (0.0145, black dashed line) for a period, overwhelming hospital capacities. With intervention, the infected population always remains below this limit.

Populations per compartment during deterministic SEI3R simulations, both without intervention (top) and with intervention (bottom).

Project website: https://plai.cs.ubc.ca/research/covid-19/

Collaboration opportunities:
We are seeking collaborators on the epidemiology side to verify our results and explore other approaches to automating the inference task in these models.