COVID19 by the numbers — Take 2
How deadly, and thus how prevalent, and thus how long will it last? And should we stay shut down?
Disclaimer — I am not an epidemiologist. I took liberty with assumptions. I could very well be wrong.
Because most of the world has not implemented widespread testing, we don’t know how prevalent COVID-19 truly is. Therefore, we don’t know how deadly it is. By not knowing the true prevalence, we cannot accurately model when herd immunity will develop, and regular life can resume. Furthermore, the overall mortality estimates will guide us on how strict of measures are appropriate to mitigate the spread. If the mortality numbers are low and the disease is actually far more widespread than we knew, the cost benefit of shutting down the economy may be viewed differently.
This all hinges on widespread testing. There is talk of increased testing with Abbott‘s five minute test and an upcoming serological assay capable of detecting antibodies from blood, however, in most of the impacted world, only those who are seriously ill with the hallmark symptoms get tested.
In the meantime, we can take our clues from the one place where widespread, random testing has occurred, Iceland, courtesy of decode genetics. Decode is randomly screening the population, claiming to literally pick random names from the phonebook, and offering COVID-19 testing. As of April 2, they’d screened over 10,000 individuals, most of them asymtpomatic, representing approximately 3% of Iceland’s population, which is 376,000. 92 individuals, or 0.9%, were found to be positive. Interestingly, the proportion has remained steady over the last 15 days, suggesting effective containment measures. In addition, the national hospital has tested symptomatic individuals and contacts, screening an equally sized population.
According to worldometers, Iceland has reported 1,417 cases and just 4 deaths, yielding a crude mortality rate of 0.3%, which is far lower (10–30x) than other impacted countries, likely owing to screening less affected individuals. Twelve patients are in serious or critical condition. If we assume that half of those will eventually succumb to the disease, we would arrive at an adjusted infection fatality rate of 0.7%. However, that assumes that all infected individuals have been identified, which we know is not true given that only 7% of the population has been tested, and in random screening, 0.9% of patients were positive. If we extrapolate the 0.9% to the population of 376,000, we arrive at an estimated number of total cases of 3,400. Furthermore, if we assume that only half of those who were infected in the past will test positive at present, as many may have already cleared the virus, that would yield a new estimated case number of 6800. If we now assume that all 12 of those in serious or critical condition will die of disease and that none of those with mild disease will die, we would arrive at a new adjusted mortality rate of 16 out of 7800, or 0.2%, which is twice the widely reported mortality rate of the seasonal flu. Again, this is 15x lower than the original estimate put out by the WHO and what is currently seen in the U.S., given our very stringent, and thus limited, testing practice.
If we assume a fixed mortality rate across geographies, we can estimate the number of total infections 2–4 weeks ago by dividing total deaths by 0.2%. For example, in Italy, with 15,362 deaths, we would estimate Italy’s infected individuals at 7.7M, or 13% of the population, and that is likely the number 2–4 weeks ago, given the average time from infection to death. As the northern regions of Italy have been far more impacted, it may be that certain areas have already reached a state of >50% infected (i.e. herd immunity), and thus can begin to return to normal life. Only widespread serological testing can confirm this hypothesis.
Similarly in New York City, the epicenter of the US outbreak, where 2,624 deaths have been reported, we would estimate 1.3M infections 2–4 weeks ago. Given that daily deaths has not yet peaked, we have to assume that the number of deaths will more than double just from those who have been infected to date. This assumption yields upwards of 6,000 total deaths and thus, 3M current or past infections, or 34% of all New Yorkers! Similar to regions of Northern Italy, areas of NYC may have already reached herd immunity.
The moral of the story is that COVID19 may be far less deadly AND far more prevalent than we think. The true test will be widespread serological testing in hard hit areas like Italy and and NYC. If the prevalence is much higher than previously realized, that means that herd immunity is building, and this may all be over much sooner than we previously anticipated. If the numbers play out this way, then the estimates of dramatically over-running the healthcare system may have been exaggerated and the cost benefit calculation of shutting down the economy will need to be reconsidered.
(Note: This is not to say that hard-hit areas like Italy and certain parts of New York City have not had overwhelmed health systems, it’s just to say that they’re unlikely to be 10–30x overrun, like some projections had anticipated.)
Potential flawed assumptions: some or many of Iceland’s mild cases may progress and become severe and then lead to death, which would have led me to an under-estimate. On the contrary, the time it takes an individual to clear a mild infection and become test negative is unknown, making it challenging to estimate the true proportion of the Icelandic population previously infected (vs. currently infected). Lastly, Iceland may have better protected vulnerable populations or may be healthier overall leading to a underestimate in the fatality rate.