COVID-19 Interventions Simulation

Settings
Simulation
Healthy
0
Recovered
0
Sick
0
Quarantined
0
Max Conc. Sick
0
Graph of virus spread
Information

How to use: This is like the Washington Post COVID-19 simulator but, in addition to watching the epidemic spread, you can fight it with interventions based on existing technologies (as opposed to vaccines). As noted in Ferguson et al. (2020), allowing the virus to infect a large fraction of the population would likely lead to tremendous loss of life and overload of health systems. Try to suppress the epidemic without using too many disruptive lockdowns.

The simulation currently supports some of the interventions from phases 1 and 2 of the AEI's proposed plan for fighting the pandemic.

You can ask a certain percentage of the people in the simulation to wear masks, N95 masks, or to simply wash hands 10 times or more per day. In the real world, please do not buy N95 or other medical-grade masks, or any other medical-grade personal protective equipment, until there is no longer a shortage.

You can also require ongoing social distancing. In this simulation, social distancing prevents a given percentage of interactions from happening at all. If things get out of hand, you can impose a lockdown, forcing most people to stay at home.

You can also choose to test a certain percentage of people in the simulation for COVID-19 periodically. Anyone who tests positive will go into self-quarantine and will not infect anyone else until they recover.

Finally, because we still don't know the exact value of R0 for COVID-19, you can choose a value. R0 is the average number of infections caused by a single primary infection. Estimates for COVID-19 have typically range from 1.5 to 3.5, but a recent estimate puts it as high as 5.7; higher values make the virus harder to fight.

What should I take away from this? Qualitative intuition about how mutiple interventions can work together to slow the spread of COVID-19, even if each intervention is not perfect.

In particular, handwashing and mask usage add to the effectiveness of other interventions like testing and lockdowns. So, in addition to the other tools we use to fight the virus, we should all be diligent about handwashing and masking. These interventions are more effective if more people use them.

What should I not take away from this? This simulation does not make quantitative predictions about the real world. None of the parameters that went into this simulation are known with much certainty for COVID-19. If you find that x% handwashing compliance and y% mask usage is sufficient to prevent the epidemic from spreading in the simulation, it doesn't mean that would be the case in the real world.

About the numbers Effectiveness estimates for masks and handwashing more than ten times daily for SARS (not COVID-19) were taken from Jefferson et al. (2008) whose main results were that "Meta-analysis of six case-control studies suggests that physical measures are highly effective in preventing the spread of SARS: handwashing more than 10 times daily (odds ratio 0.45, 95% confidence interval 0.36 to 0.57; number needed to treat=4, 95% confidence interval 3.65 to 5.52); wearing masks (0.32, 0.25 to 0.40; NNT=6, 4.54 to 8.03); wearing N95 masks (0.09, 0.03 to 0.30; NNT=3, 2.37 to 4.06); wearing gloves (0.43, 0.29 to 0.65; NNT=5, 4.15 to 15.41); wearing gowns (0.23, 0.14 to 0.37; NNT=5, 3.37 to 7.12); and handwashing, masks, gloves, and gowns combined (0.09, 0.02 to 0.35; NNT=3, 2.66 to 4.97)."

To emphasize, these numbers should not be taken as predictive for COVID-19. They were measured for SARS, and also substantial variation was observed between constituent studies. However, they are encouraging.

Because SARS infected only 8000 people in total, it is reasonable to assume that the odds ratios approximated the risk ratios. The odds/risk ratios reported were for the study period, not for each potentially infectious contact. Assuming one or more potentially infectious contacts occurred during each study period, the probability of prevention for each would be at least as high as the probability of prevention over the entire study period, since avoiding infection for a study period requires avoiding infection at each contact. Correcting these underestimates would require some (more) modeling assumptions, and for purposes of this simulation I prefer to err on the side of not encouraging inflated hopes about the effectiveness of interventions.

Jefferson et al. recently published a preprint where they find that the efficacy of population-wide mask usage against influenza-like illness is substantially lower than one would expect based on the numbers above. I will need to review the constituent studies of both meta-analyses to determine which numbers are more appropriate for this simulation. I will try to do so quickly.

The number of ticks per 'day' is chosen so that the length of the infectious period is 20 days. So, if you test every 2 days, the infectious period will span ten tests.

The simulation is calibrated to give the desired R0 as follows. The average collision frequency f that each ball experiences was measured empirically. When an infected ball and a susceptible ball collide, the susceptible ball becomes infected with probability p such that p f T = R0, where T is the length of the infectious period in units of ticks.