Skip to main content

More on Case Numbers

By May 6, 2020Commentary

The first paper is pretty abstract, attempting to reconcile infection and death rates.  The authors assume that expected mortality in relation to  infection rates should be similar across countries, at least those with similar socioeconomic status.    (Medrxiv Paper)   Without going into all the detail, he ends up finding a linear relationship, after some data manipulation, and a case fatality rate of around .15%, which seems potentially reasonable.  But this paper seems sketchier than some others.

The next study was an antibody survey in Atlanta, which a lab did among 142 contacts of employees and other individuals.  So this is not a random study.  The authors found a range of antibodies, and used a cutoff of values that was likely to exclude some true positive results.   (Atlanta Paper)   They found that in several Atlanta neighborhoods, there were positive test results in almost all of them, although small in number.  Most of the positives were asymptomatic persons.  Some older archived blood samples from before the epidemic, which were used for validation and calibration, showed a response, again indicating the possibility of cross-reactivity with previous strains of coronavirus.  Antibody responses among asymptomatic individuals were lower than those from hospitalized individuals, suggesting that antibody strength may be related to seriousness of illness.  Other interesting results were that of 15 people who felt strongly that they had had the disease, only one had a positive antibody response.  Three positive individuals were in households in which the other members were not positive, suggesting that asymptomatic persons may not spread the disease very aggressively.

Next is modeling paper from Japan which attempts to reconcile an apparent high transmissivity of the virus with relatively low peaks, and often multiple peaks in cases.  In other words, given how infectious the virus appears to be, there should be a lot more reported cases.  The authors focus on asymptomatic cases as a possible explanation.   (Japan Paper)   They used Wuhan and Japan as a whole as test bed and used a SEIR type model.  They assumed asymptomatic patients were as efficient at transmission as symptomatic ones.  They found that in Wuhan the likely ratio of asymptomatic (and unreported) cases to reported ones by testing was 148 to 1 and in Japan it was 8941 to 1.  These seem high, but perhaps directionally consistent with other research.  Japan may have a very high rate of asymptomatic patients because of lower levels of pre-existing conditions.  Having been in Japan last fall, I can testify that there are extremely low rates of obesity for example.

The next study comes from Brazil, where in one state researchers are conducting regular random household antibody testing.   (Brazil Paper)   Every two weeks they go out and attempt to test 4500 persons.  The first surveys began shortly after the epidemic appeared to begin in Brazil.  Again, the test used was much more likely to miss true positives than true negatives.  4188 people were tested in the first round and there were two positive results and 10 inconclusive.  The second round tested 4500 people, and there were 6 positive results and two inconclusive tests.  In a household with a positive test, other members were tested and about half those tested were positive.  Based on the second wave, there would be about 1,333 cases per million inhabitants compared to 128 actually reported cases.  So about ten times as many cases as determined by infection testing alone, and recall that were likely some number of missed positives.

This next paper is also one focused on how to reconcile what models predict with reality.   (Medrxiv Paper)    The authors used a SEIR type model with some revised equations and tested their model on data from China, South Korea, France, Spain and Italy.  But they completely ignore missing infections, which is now widely viewed as multiples of reported ones.  I am not sure how you test models without using more accurate assumptions or data.  A good example of dozens of papers at this site that play with models and formulas but don’t seem to actually be connected to reality, even if they can twist formulas to make it seem like the model works.

And here is another paper trying to figure out how to estimate how many cases there really are, in the absence of comprehensive infection and antibody testing.  (Medrxiv Paper)  The authors used a county in Washington for their work.  They used a model developed for population census work to try to estimate positive but unrecognized coronavirus cases.  They used a case fatality rate from the village in Germany that did some extensive antibody testing and they used CDC data on cases and deaths.  They used the death rate from Germany to estimate uncounted cases assuming a similar case fatality rate in the US.  This gives them an estimate of the total number of cases in the US.  They find about ten times as many unconfirmed cases as confirmed.


Leave a comment