[07/01/20 Update: Data Tracking on plot is from the NY State COVID Wikipedia page and consistent with NY State COVID Dashboard Data and is differs from data reported by John Hopkins. Our model was calibrated to NY State data and we keep this approach to be consistent]
A case study of New York State is presented with the assumption that the majority of the deaths are in the greater New York City metropolitan area, as opposed to “Up State”. The greater NY metro area is assumed to be approximately 20 million people of whom 8 million reside in New Jersey or Connecticut. For this study we are using a population of 12 million (a 6 multiplier simulation population). Graphic below is from the NY Times USA Covid19 infection map.
Key simulation space parameters are presented below in table 1. These virus infection parameters are largely derived from consideration of a variety of sources listed on worldometer.
Cumulative probability distribution bins and definitions are defined below in table 2 for the general population of the NY metropolitan area and NY state are based on generally published data on Wikipedia. Some of the parameters relative to health and mobility are based on reasonable assumptions.
The infection outcome distribution table is shown in table 3. This table is derived from the Imperial College Covid-19 response team Report 9. Additional scaled outcomes for asymptomatic cases have been added as an input variable. A nominal value is used to represent the present consensus that a high number of cases are asymptomatic:
Figure 2 shows the dynamic input parameters for this baseline run. Test access increases over time, test processing time decreases and transmission factor is scaled due to increased social distancing and stay at home orders. In May the transmission factor is scaled up to 0.25% to reflect a return to normal with reduced social contacts. Test access increased throughout late March and April. Test processing time is reduced as well over the same period. The number of seed infected people in the simulation population is 2000 (.1%) per day for 7 days to accelerate the infection and cut the simulation time. Calibration runs (at this post) verified that the same effect can be achieved with 4 individuals starting on 2/24/20. This model also includes a mortality adjustment with seasonal temperature that rolls off the death rate as temperatures peak through summer (test feature to be calibrated later, used for example purposes).
Figure 3 shows simulation results compared to actual data through April 11th, 2020. Data has been tracked comprehensively on Wikipedia.
Figure 4 shows the view of simulated infections and deaths per day. Note the significant delay between the infections and death curves. March 15th restrictions cut the back end off the infection curve. The loosening of restrictions to return 25% less contacts than baseline reduces ongoing death rate to something akin to the the flu. This reduction is likely easy to achieve since people are likely to remain overly cautious at this point in time as restrictions are lifted. Note that the IHME models for New York do not allow for any return of the infection, they are only looking the near term.
Figure 5 shows the sequencing of the curves for infections, standard hospital care, critical hospital care, recovery and death. Hospital demand peaks before deaths will peak.
Figure 6 shows active infections and cumulative recoveries, health people (no infection) and deaths.
Figure 7 shows the cumulative percentage of susceptible people who are not infected. Susceptible people are those over age 70 and / or people with degraded health immunity (nominal or high risk).
Figure 8 shows the time phasing between actual infections and positive test results in the model. As testing is broadened to general population and is near real time, the curves are more aligned.
Figure 9 shows the relative risk of infection and the death rate in time. Risk at present is significantly less that what it was prior to the lock-down.
Figure 10 shows the derived R0 over time. This is a dynamic indicator of whether the virus spread is growing or dying out. Numbers greater than 1 show growth, numbers less than one show decline. In this scenario, the lock-down effectively kills the growth of the model and a then a modest return to normalcy allows for a controlled uptick in the infection rate.