Figure 1: Coronavirus naive mortality rate depends on the prevalence of testing
The trait may be a rare thing like being blind or a common thing like being right-handed.
Regardless of what trait we choose, because we took terminally ill patients as a sample, we will
always receive 100% mortality rate associated with that trait.
The selection bias makes a person draw an absurd conclusion that being blind or being right-
handed is related to high mortality rate.
If this example might seem far-fetched, let’s see what happens during the coronavirus crisis.
According to the director of the infectious disease unit at Sacco Hospital in Milan [46], “Italy
focused its testing only on people showing severe symptoms in areas with high epidemic intensity.
This causes an increase in the fatality rate because it is based on the most severe cases and not on
the totality of those infected.”
Also the researchers from Istituto Superiore di Sanità, Rome note the following [45], “After
an initial, extensive testing strategy of both symptomatic and asymptomatic contacts of infected
patients in a very early phase of the epidemic, on February 25, the Italian Ministry of Health issued
more stringent testing policies. This recommendation prioritized testing for patients with more
severe clinical symptoms who were suspected of having COVID-19 and required hospitalization.”
These reports show that selecting candidates for testing against coronavirus in Italy is not far
from our hypothetical absurd setup. Unfortunately data coming from other countries is laden with
similar selection bias.
In case of data coming from various countries we don’t have random samples however, we know
that some countries do more intense testing than others. If any country had tested the entire
population then we would have quite reliable data to calculate mortality rate. This is unrealistic
because, by the middle of April 2020, only four countries tested more than 3% of the population.
It’s not much, but we may expect that data from these four countries is less affected by selection
bias.
Selection bias is not eliminated, but it’s clear that the more comprehensive testing, the more
reliable data. Based on the Worldometers [32] data, we plot Naive Case Fatality Rate as a function
of Test Intensity on Figure 1. We know that the more intense testing, the more reliable data. The
chart also reveals a clear trend, with more reliable data we get the lower case fatality rate.
Four data points stand out in terms of test intensity: Iceland, United Arab Emirates, Luxem-
bourg and Bahrain. Out of all data points they are the most reliable ones and at the same time
show the lowest case fatality rate.
2