2020-04-22 - The COVID-19 numbers are not what you think

Given the seriousness of the COVID-19 pandemic, it's amazing how bad our metrics are. We really don't know much about the epidemiology of this disease.

Following is a laundry list of problems with the various metrics that are bandied about in the press...

Chinese numbers are just propaganda. The Chinese claim to have about 85,000 cases and 4,600 deaths. (Google dashboard) This is the recently increased set of figures, where they increased the number of reported Wuhan fatalities by 50% in the hopes that people (at least Chinese citizens) will believe the new lies, because nobody believed the old lies.

The figures don't make any sense:

• Does anyone really believe that Chinese figures are so much better than in the US, given dense living conditions and 4x the population?
• The number of funerary urns recently delivered to cremation facilities in Wuhan alone suggests that the incremental deaths there (above and beyond what is normal for such a large city) were between 20,000 and 80,000. See vice.com. As an order of magnitude guess, it's reasonable to estimate that COVID-19 deaths in China are 4 times that, or about 200,000 -- at least 40 times higher than reported.
• UPDATE: the first recorded death due to COVID-19 in the US was on February 6. This was due to community spread. Given typical 21 days from infection to death, this means that the virus arrived in the Bay Area by mid-January. An infected individual presumably traveled from Wuhan to the region at that time. Given the relatively small amount of such traffic (three direct flights weekly from Wuhan to SFO - so probably under 1000 people/week coming over), the fact that at least one infected individual came over mid-January suggests a fairly widespread infected population in Wuhan by early January.
  I'd guess that at least 0.1% of the Wuhan population, or over 10,000 people, were infected in Wuhan by January 10. The Wuhan lockdown started January 23, so with a doubling rate of about 2.5 days, Wuhan would have had over 300,000 infections by then. That's just Wuhan -- never mind the hundreds of other clusters across China due to 5 million people traveling out of Wuhan for Chinese New Year just before lockdown.

Confirmed cases are not the same as community cases. Every jurisdiction is struggling with testing capacity, and prioritizes tests for symptomatic people, and often those who have had contact with a known infected individual or recent travel. That means that lots of people who are positive are not tested and do not count towards to the total reported number of cases. Moreover, it seems that lots of people are asymptomatic or pre-symptomatic.

Depending on the test capacity of a given region, there is some multiplier to go from confirmed cases to actual community cases. A recent antibody study with a somewhat random sample in Santa Clara placed this multiplier in that jurisdiction at somewhere between 50 and 85 (MedRxiv.org).

• UPDATE: I just ran across This study from USC.edu where another random sample was done to see who has antibodies. It estimates that 2.8% to 5.6% of the population of LA county has already been infected.
• This means 221,000 to 442,000 infections. They have recorded over 650 deaths, so their implied mortality rate is between 0.14% and 0.29%. Worst case: 3x worse than an average seasonal flu (which has 0.10% mortality).

In regions where a lot of testing has been done, per-capita, this multiplier might be lower. Say 10x or 20x. In regions with essentially no test capacity, such as developing countries (India, Egypt, etc.), the multiplier will be much higher -- say 1000x.

World-wide, there have been almost 3 million confirmed tests as I write this. Much of the world is low income, so the average multiplier will be high. Assume 50x as a very conservative multiplier. That means that there are already over 150,000,000 COVID-19 infections world-wide. There is no way that we can ever put this genie back into the bottle.

The tests themselves are not that reliable. There are two categories of tests: (a) are you currently shedding virus? and (b) are there antibodies for the virus in your body? There are multiple versions of each of these. None of them is super reliable. Someone might test negative for the virus now and positive in an hour, for example.

The mortality rates are not what you think.

• First, There is the obvious problem (above) of not knowing how many people are actually infected. We can see who dies in hospital of the disease (at least in countries whose health system has not been overwhelmed and whose political class are not pathological liars) -- that's the numerator. Unfortunately, the denominator is people who were infected 2 to 4 weeks earlier, and we have no idea what that denominator was or is.
  • To give a sense of this - lets use the above estimate of 150 million infections world-wide today.
  • Assume a doubling rate of 7 days (with all the recent control measures, pretty much everywhere) and time-to-mortality of 3 weeks (21 days). That means that world-wide infections totaled about 19 million cases 3 weeks ago. There is huge uncertainty in this number!
  • Reported world-wide fatalities are currently about 183,000. Given the inability of some countries to report and the proclivity of others to lie, lets assume the real number is 300,000. You should get a sense for the uncertainties by this point...
  • Using these estimates, the COVID-19 mortality rate is 1.6%. That's much worse than seasonal flu. But adjust any of the huge assumptions above and this number shoots up or down -- a lot!
• Second, people who die of the disease are not the only variable. There are also these other variables:
  1. People who get into accidents, are sick, etc. and who delay or avoid medical treatment because they are afraid of being infected while visiting a hospital (where all the scary COVID-19 victims are hiding out, waiting to infect you!). This mechanism is killing at least as many people as COVID-19. My friend is an ER doctor and reports seeing conditions that have never been seen in her career - like appendicitis patients presenting with sepsis because they delay coming to the hospital until too late.
  2. A lower rate of trauma (accidents) due to reduced car traffic, closed schools / playgrounds and outdoor recreation opportunities.
  3. Economic hardship and extended social isolation lead to mental illness, which is often accompanied by violence and death.
  4. In places where hospitals are overwhelmed, higher morbidity from COVID-19 due to people not being treated.
  5. Also in places where hospitals are overwhelmed, higher morbidity from other causes due to hospital congestion.

  Even if we knew the actual morbidity rates of COVID-19 (i.e., percent of infections that lead to death), we still wouldn't have a handle on the actual impact in terms of increased deaths in the population as compared to a normal year.

  Note that in a year we'll know. We can measure total mortality rates in a population and compare to the same time the previous year.

Not knowing is dangerous. Policy makers are making monumental decisions that will impact billions of people for decades, and they are flying blind. Is COVID-19 worse than season flu? Yes, certainly. How much worse? Maybe 2x. Maybe 10x. Maybe 100x. Maybe it depends on isolation measures and healthcare capacity. Nobody really knows. Are different control measures working? Some of them are supported by evidence (closing schools, restricting travel, wearing face masks, contact tracing) and some are only supported by what seems like common sense, rather than hard data (stay at home orders).

We need to gather more and better data. Are control measures effective? Are they causing more harm than good?

• We should test random population samples to gauge the extent of infection currently in the community, rather than just focusing on symptomatic people.
• We should repeat this over time to get real trends.
• We should test individuals for antibodies to gauge how far the disease has already passed through the community.
• We should measure increased and decreased morbidity due to deferred hospital admissions and reduced trauma, respectively.
• Data from police and distress centres should be incorporated too, because of the elevated mental stress in the population.

All of this takes work and will require much larger testing capacity. That has to be a top priority. Surely we can afford it, given the billions already allocated to economic supports.

Blog home