Module 5
Lesson 3
Hi and welcome back!
Let's start with a concrete example on how to compute disproportionality. Imagine, you have in your
database three reports containing a specific drug and a specific adverse event. Is three reports a
great number of reports? You will remember from the previous lesson that we can only answer this
question if we know the expectation.
The expectation is the number you should theoretically see if the drug and the adverse event are
totally unrelated, if there is no association between the two and they're just co-occurring by chance.
So, how to compute the expectation? Let me show you step by step. First, let's have a look at our
database. In this imaginary example we have 100 reports that I conveniently re-organized so I can
highlight the important reports to compute the expectation. We have the reports with the drug, the
reports with the event, the reports with both and the reports with neither.
Among the 100 reports in our database we observe three reports with both the drug and the
adverse event, so three is our observed count. Then, to compute the expectation we start by
computing the frequency of the adverse event in the entire database. That is to say we count how
many reports we have with the adverse event. Here we count 8 out of the total number of reports
which is 100 so the frequency of the adverse event in our entire database is 8%. Then we count the
number of reports with the drug. We have 10 such reports. So, if the drug and the adverse event are
unrelated we would expect to see 8% out of these 10 drug reports to also contain the adverse event.
8% of 10 reports, that is 0.8 reports. Note that the expectation is not a round number here, in fact it
is rarely so.
So, we have an observed count of three reports with both the drug and the event in our database
and we have an expected count of 0,8 reports. A simple way to measure disproportionality is to take
the ratio between those two quantities. This is called the “relative reporting ratio”. In this case, 3
divided by 0.8 makes 3.75. So, it basically says that we have seen 3.75 times as many reports as we
would expect if the drug and the event were unrelated. That's almost four times as many reports. In
general, an observed to expected ratio >1 tend to suggest disproportional reporting.
There might be several reasons why we observe such disproportional reporting between the drug
and the adverse event. One of these reasons can be a causal relationship, but it could also come
from other factors such as a co-prescribed drug that would trigger the event or comorbidities in the
patients taking the drug. The key point here is that a value >1 does not prove a causal relationship
between the drug and the adverse event. This computation cannot replace the necessary manual
assessment of the combination by expert reviewers.