Covid-19 Modelling

What is the probability ANU classes will be cancelled in the next 4 weeks? Today we examine a simple epidemiology model to answer this question.

Live data from John Hopkins University:

---

S E I R

SEIR uses a graph to model and simulate the spread of viruses in a population. The acronym SEIR comes from the possible states of each node, being Susceptible (S) to Exposed (E) to Infected (I), and Removed (R).

We initiate the graph with non zero infected nodes, each capable of causing a susceptible node (S) to become exposed (E). Exposed nodes differ from infected nodes in that they are asymptomatic and so will not be removed/quarantined but is still capable of infecting others. After the asymptomatic period, exposed nodes (E) become infected (I) which now have a chance of being removed (R). The only transition allowed are S→E→I→R.

Using the python library Ndlib I ran some simulations of how the disease will spread on campus.

The statistics surrounding Covid-19 are still unclear but under the following assumptions:

1. 16000 students on campus.

2. 1 student currently infected.

3. Each student interacts with 30 other students per day.

4. During each interaction, the susceptible student has a 3.1% chance of being infected.

5. An incubation period of 5 days.

6. A 50% chance of being removed each day after being infected.

We have the following simulations:

We can reasonably expect that classes will close if there are 3 or more cases.

In 62.5% of simulations we reach 3 confirmed cases within 4 weeks.

In 25% of simulations the infected is quarantined without spreading to anyone else.

There's another thing to consider which is that it may only take the threshold for closure to occur at one institution to cause the government to close all unis.

To conclude, even this simple model is enough to give us some perspective on the flood of numbers coming out about the virus. Social distancing can lower the interactions, whilst quarantine measures increases the removal probability. Feel free to do some sensitivity analysis to see what policies would be most effective.

Python Code