New Coronavirus Data

I thought it was worth penning a quick update based on additional infomation which I have stumbled across, or has come to light in the last few days.

The most interesting is this study which examines all the testing, hospitalization and fatality data from Wuhan itself.  The authors recognize that testing in Wuhan was inadequate to determine the true number of symptomatic cases and that many milder cases were never tested and therefore not counted in the statistics.  Therefore they have used testing data from the various expat evacuation flights from Wuhan, where all passengers were tested upon return to their native lands, to estimate the actual number of infections in Wuhan itself in order to more accurately estimate the overall CFR as 1.4%. I must confess that my mathematical skills have decayed somewhat through three decades worth of lack of use and that therefore I have not completely understood exactly how the authors have calculated their results, but as the authors themselves note, there are a number of reasons to believe that they might have overstated the CFR and/or it may be lower elsewhere.

  • The healthcare system in Wuhan was ovewhelmed which
    • would have resulted in inferior care to what could be expected where healthcare systems are not overwhelmed and therefore more fatalities
    • could also have affected the CFR by biasing the data sample (those who received tests) towards more seriously ill patients
  • Some number of cases will be asymptomatic, or exhibit symtpoms sufficiently mild as to escape diagnosis (strictly speaking this will lower the infection fatality rate or IFR rather than the case fatality rate)
  • The expatriates who departed Wuhan in the early stages of the epidemic may have been infected at a lower rate than the general population of Wuhan, for example as a result of moving in different social groups or as a result of already infected expats being too ill to travel.
  • As physicians discover treatments, outcomes should improve.

It is also possible that the number of deaths in Wuhan were undercounted, especially in the earlier stage of the epidemic, which would result in the CFR being understated although the authors consider this less likely to be a significant source of error.

The study also found that

  • compared to those aged 30–59 years, those aged below 30 and above 59 years were 0.6 and 5.1 times more likely to die after developing symptoms
  • the risk of symptomatic infection increased with age at ~4% per year among adults aged 30–60 years; in other words younger people were less likely to catch the disease in the first place as well as less likely to die from it when they did catch it.
  • the ascertainment rate was between 2% and 3% (i.e. the overwhelming majority of cases were not detected, although this is an area where my lack of understanding of how the results have been calculated makes me unsure as to exactly what is being claimed).

And finally the study also calculated R0, and estimated that, using a heterogenous model of societal interactions, we could expect between 25% and 50% of people to become infected.  In epidemiology the basic homogenous transmisson model assumes that everyone in society is identical, where more sophisticated heterogenous models make assumptions about factors such as the stratification of susceptibility to infection by age, and the differing potential for contact within and without specific social groups.  heterogenous models usually predict lower penetration rates for a disease than the simple homogenous model.  One input to such a model is population density and its worth noting that Wuhan is a very densely populated city, so the penetration may be lower in less populated areas.

It’s also worth noting this article from the FT which notes that Germany’s low crude CFR is likely a result of them running 160,000 tests per week, more even than South Korea.  As a result Germany believes it is detecting many infections with few or no symptoms and therefore much higher rates of survival which explains low crude fatality rate reported there.  Germany has also seen a higher rate of infection amongst younger people than older people as has the US which contradicts the both the data from the Diamond Princess and Wuhan, but may be a consequence of younger people not taking the threat seriously and continuing to meet in social groups in which case we should expect to see this trend reverse fairly soon.

Another datapoint is the director of the Ohio Department of Health estimating that 1% of Ohioans (i.e. 117,000 people) were already infected by March 12th despite only five confirmed cases at that time.   She appears to have based this assertion on a rule of thumb contained in a CDC publication from 2017, which strikes me as likely to be an overestimate in this case.  Nevertheless it gives some idea of the extent to which the number of reported cases in the US could possibly be out of whack, and reinforces the lessons from the swine ‘flu epidemic where initial crude CFR measures overstated the risk by as much as 500 times in at least one case based on similar problems with case counts.

Overall I still feel that the Diamond Princess continues to give us the best data, but the Wuhan study should not be ignored and perhaps it would be safer to estimate the CFR range for people with access to good healthcare as 0.2% – 1.0% than my previous 0.1% to 0.5%.

UPDATE  I just found the data from Italy on March 17thScreenshot_2020-03-23 Report-COVID-2019_17_marzo-v2 pdf

99.2% of fatalities were already seriously ill.  48.5% of fatalities were suffering from three serious illnesses.

One thought on “New Coronavirus Data

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s