We should beware of naive extrapolation but the trend is clear and expected. The selection bias
is reduced with test intensity (size of the sample). With the sample covering the entire population,
the real case fatality rate converges to the value much below 0.4%.
The most comprehensive data comes from the Diamond Princess cruise ship, where selection
bias while significant is not disqualifying.
The cruise ship was quarantined on February 2nd 2020. There were 3711 total people on board
out of which 3011 were tested. By 20th February 2020, there were 634 confirmed cases with 328
of these having no symptoms, and 37 required intensive care. [12, 13, 33]
As of March 1st 2020, 7 passengers died (three of them aged 70-79 and four aged 80-89), what
gives mortality rate 1.1% (7/634) relative to the infected and 0.19% (7/3711) relative to the exposed
(it’s safe to assume that everyone on board was exposed because the food service workers have
been the main route of spread). 37 passengers required intensive care what gives 5.8% (37/634) of
severe cases relative to the infected and 1% (37/3711) of severe cases relative to the exposed.
The cruise ship sample is far from being random or complete for several reasons. There are factors
that contribute to underestimating mortality rate like:
• a number of unresolved cases (as of March 1st 2020)
• likely higher socioeconomic status and better general state of health of passengers
There are also factors that contribute to overestimating mortality rate:
• testing was not complete. 81% (3011/3711) people on board were tested, so we miss an
unknown number of cases with mild or no symptoms
• testing started among the elderly passengers, descending by age
• age structure of the subpopulation on board was dramatically different from the world av-
erage. There were 2165 people aged 60 years or over what makes 58% (2165/3711) of total
number people on board. For the world population people aged 60 years or over constitute
14% of total. Table 2 presents the detailed age structure breakdown.
There are various techniques to adjust for delay from confirmation-to-death and age structure.
Detailed estimates produce mortality rate at 0.5% [48] although margin error is high because the
sample size is small.
To summarize, one cannot infer mortality rate from samples that are not random or complete.
When we look at most available data: the number of deaths related to coronavirus and the
number of confirmed cases, we need to understand that the former number is adequately reliable,
and the latter is meaningless because it depends on the arbitrary testing-sampling strategy.
2.2
Ignoring baseline
Numeric data representing change, scope, magnitude or impact is useless without a baseline that
serves as a point of reference or an initial value. Things have to be compared in similar contexts,
otherwise the comparison is meaningless. Evaluating data without knowing the baseline is like
measuring without specifying units.
Specific claim
Even if we don’t know the true mortality rate, the number of deaths is staggering. From December
1st 2019 to April 15th 2020, there are about 140,000 deaths related to coronavirus.
4