Berks County Medical Society Medical Record Summer 2020 - 40

An Agent-Based Simulation Model to Inform COVID-19 Intervention Decisions
continued from page 39
We used the CDC's COVID-19 Pandemic Planning
Scenarios [7] to set the parameters of the model unless otherwise
stated. In our model, we defined three infected states: Exposed,
Symptomatic, and Asymptomatic. The average time from exposure
to symptom onset is six days. Individuals may pass the virus to
others during the last 48 to 78 hours before symptom onset [8].
Therefore, we considered the first phase of exposure to symptom
onset as a non-infectious period (the Exposed state in the model)
for both asymptomatic and symptomatic individuals. According
to the CDC, about 35% of coronavirus cases are asymptomatic. In
our model, asymptomatic individuals continued spreading the virus
until they had positive test results, or they got fully recovered. We
assumed that all individuals were law-abiding citizens who would
self-quarantine after testing positive or observing the common
symptoms of COVID-19. In our model, only symptomatic
individuals developed serious complications that required hospital
care with an average of 3.4%. We also considered two types of
hospitalization, Normal and Intensive Care Unit (ICU), because
the average time in hospital and fatality rates are quite different
between individuals who require ICU and those who do not.
In order to represent virus containment measures, R0 was
determined as the product of two parameters: (Contact Rate)x
(Infection Probability). The Contact Rate parameter is the average
number of contacts per day, which is a function of stay-at-home
and social distancing interventions. The Infection Probability
parameter controls what percent of the contacts will result in
new infections, which represent the effectiveness of preventive
measures such as keeping a safe distance, washing hands frequently,
using facemasks, etc. We assumed that both asymptomatic and
symptomatic individuals had the same level of infectiousness.
In our model, we also included the availability of diagnostic
testing to individuals. All individuals could request testing.
However, the number of tests available to the population was
limited by the capacity of the testing facilities (units/per day). There
is no 100% accurate diagnostic tests for COVID-19. To indicate
the accuracy of the test, we used the true-positive rate of 70% as
given in [9] (i.e., the percent of infected individuals that the test
correctly identified as positive), and there was no delay in receiving
test results.

Figure 2. The state chart of the ABS model to study the impact
of social distancing measures and testing on the diffusion of
COVID-19.
40

|

www.berkscms.org

Example Use Cases of the ABS Model
Using our ABS model, we simulated a virtual town with a
population of 10,000 individuals for 100 days. In our virtual
town, an individuals' social network included randomly selected 50
other individuals with whom they may interact during a day. The
initial rate of infection was 0.01% of the population. During the
simulation, no outside interactions were considered, and the model
parameters were constant. We run our model for the following six
scenarios.
Average Number
Available
of Contacts
Daily Testing
Scenario
Capacity
per Day
Based line: No Social Distancing / Low Testing	
Moderate Social Distancing / Low Testing 	
Strict Social Distancing / Low Testing 	
No Social Distancing / High Testing 	
Moderate Social Distancing / High Testing 	
Strict Social Distancing / High Testing 	

52	
25	
10	
52	
25	
10	

100
100
100
500
500
500

Are stay at home orders and social distancing
measures effective?
According to a Gallup report [10], Americans engage in an
average of 52 contacts per day, and individuals following strict
social distancing rules have reduced their contacts to five per day.
The real-life data showed that strict interventions worked well in
controlling outbreaks in China [11]. A recent report from Columbia
University also suggested that the number of deaths would be
36,000 less if the USA had been implementing social distancing
measures just one week earlier in March [12]. We run our model
using R0=2.4 (average contacts of 52 per day) as the baseline and
then reduced the average contacts to 25 and 10 while keeping
the infection probability constant. Our results demonstrated the
significant effect of social distancing on the spread of COVID-19
as shown in Figure 3. For the baseline scenario, the number of
infected individuals peaked rapidly in two weeks. Our results
supported that flattening infection curve would require strict social
distancing measures and other epidemic suppression strategies
might have to remain in effect in the absence of an adequate testing
capacity.

Figure 3. Simulation results demonstrating the infection curve under
the scenarios studied with the true-positive rate of 70%.
Berks County Medical Society Medical Record Summer 2020

Table of Contents for the Digital Edition of Berks County Medical Society Medical Record Summer 2020
