COVID-19 SEIR Simulation

ME233: Data-Driven Modeling of COVID-19

Stanford University - Fall 2020 (Remote learning)

This course taught about epidemiology compartment models in order to generate dynamic models for the progression of a highly contagious disease. As was relevant, we reviewed papers making using of compartments models to model and simulate the progression of COVID-19 in large scale communities. One such compartment model is SEIR , which is composed of susceptible, exposed, infectious, and recovered compartments.

My final project for this course was a simulation of the progression of COVID-19 on Stanford University’s campus in the event that only first- and second-year undergraduates returned to main campus housing for the winter quarter. More specifically, my group expanded on the Python code provided to us by the teaching team to perform explicit forward integration on the differential equations characteristic of the SEIR model.

My collaborators who assisted in writing the code for the simulation are Nefeli Ioannou (M.S. Stanford University 2020) and Chris Skalnik (B.S. Stanford University 2021).

What I worked on

I collaborated with Nefeli Ioannou to adapt the original Python code provided to us in class to simulate the development of SEIR compartment populations at Stanford’s main campus following the start of the Winter 2021 quarter. SEIR compartments for this project refer to the susceptible, exposed, infectious, and recovered student populations on campus. Chris Skalnik used a travel network matrix and publicly available COVID-19 nationwide testing data to approximate the initial number of students on Stanford’s campus belonging to the four aforementioned populations.

Nefeli and I then explored a scenario with the initial conditions provided by Chris: if Stanford were to assign students to social pods / households to allow social interaction while mitigating the spread of COVID-19, how would social pod size and contact rates across social pods impact testing positivity rate? A key assumption in our simulations is that all students residing on campus are being tested, and thus positivity rate is a one-to-one count of the number of new COVID cases on campus.

In summary, we adapted the four standard epidemiology differential equations for SEIR compartments to work for our compartmentalized and discrete student populations. This is in contrast to how we used them in our coursework, where exposures, infections, and recoveries were continuous processes acting on very large populations (+10,000).