Module 2: Causality assessment of single case safety reports

Lesson 2.1

Hello, I am Dr Ruth Savage from the Uppsala Monitoring Centre. Here I contribute to medical advice
and research, especially signal detection and pharmacovigilance training. I have experience in both
national and international pharmacovigilance and in clinical medicine.

This module addresses causality assessment of single case safety reports. Looking at the pyramid
you can see that this is the foundation. In another module we go to the next level assessing a series
of case safety reports each containing the same drug and suspected adverse reaction. We then have
the basis to move on to detecting and managing signals of previously unknown adverse reactions or
other problems with medicines that need to be addressed.

The learning objective for the first lesson is understanding why a logical process is needed. Logical
thinking in causality assessment means that we apply structured reasoning to the complex data we
find in individual case safety reports. This is not always straight forward.

Let's think about causal relationship reasoning with this example: Did the boy break the window?
Imagine that you are walking along a path at the side of a house, and as you walk along you see a
boy throwing a stone towards the front of the house and you hear a crash. You walk forward and see
glass on the ground and a broken window. There are some angry people saying to the boy, "You
broke the window!", and you think, "Yes he did, I saw him throwing the stone and I heard the glass
break." Is that the only answer?

Would the window have broken if he had not thrown the stone? We know, of course, that there are
other possible explanations. There might have been another boy you couldn't see also throwing
stones.

Perhaps the boy you saw was too small to throw the stone as far as the window. Like this one, my
grandson, although he looks perhaps as if he could have done. Perhaps the stone he threw was too
small to break the window. Somebody in the house may have become very frustrated and thrown an
object at the window.
In summary, there are several explanations. When we are looking at a case report, the thought
processes are very similar. The question is simply "Did the drug do it?" and the answer is only
occasionally, "Yes, definitely." Take a minute to think of some other possible answers.

In this next example, we think about "Did thalidomide cause the phocomelia in this boy?" Many of
us know that this is a very serious condition, a limb reduction defect caused by thalidomide. This
adverse reaction was the trigger for the first national pharmacovigilance centres to be established in
the mid-1960s.

But is this a perfect causal relationship? When you look at the boy do you say "If his mother had not
taken thalidomide he would not have phocomelia"? We now know that some adults with
phocomelia who were born to mothers who took thalidomide during pregnancy now have babies of
their own and they also have the same condition. This is because it is a very rare familial genetic
disorder and so can occasionally occur even when thalidomide has not been taken. It happens more
often if thalidomide is taken. So this boy's mother may not have taken thalidomide.

What were your answers to the question, "Did the drug do it?". These are suggested
responses. Firstly, "Yes". And then, "Yes, but only in certain circumstances" such as risk factors,
demographics, patient’s genetic predispositions, the dose etc. Or, "Yes, because it interacted with
another medicine" or perhaps the answer is "No, it was another drug prescribed with it," or "It was
due to the patient’s disease,"
or "No, that drug could not cause that reaction."

So you can see there is a need for a logical method for causality assessment.

When people first started to study adverse reactions seriously a problem was quickly recognised.
There was often significant disagreement between physicians regarding whether a patient had
developed an adverse reaction or not.

Logical methods were developed to try to decrease disagreement by making sure that each
physician took all the available data into account and then applied a framework to assess the
strength of the association. That is how likely it was that the drug caused the clinical condition
described.
There will never be a perfect agreement because inevitably there has to be some subjective thinking
but disagreement is reduced. The methods also help to compare reports from different sources eg.
different hospitals or countries.

So what are the systems for standardising causality assessment. For single case reports we have the
World Health Organization-Uppsala Monitoring Centre, or WHO-UMC, system. This is widely used
and well-established. This system allows us to consider the data and then place our conclusions in
various categories. The website for finding it is here on the slide.

https://www.who-umc.org/global-pharmacovigilance/who-programme/resources-and-support/

The categories are "certain", "probable", "possible", "unlikely", "unclassifiable or conditional", and
"unclassified". An alternative approach is to use an algorithm. An algorithm is a scoring system based
on the data. We will discuss the Naranjo algorithm later.

Lesson 2.2

The learning objective for the second lesson is becoming familiar with the categories in the WHO-
UMC system for reports with the strongest evidence.

It is helpful to look at the "probable" category first since most key reports are assigned to it as they
fulfill the following criteria: The suspected adverse reaction is described as an event, or laboratory
test abnormality with a reasonable time relationship to drug intake. It is unlikely to be attributed to
disease or other drugs. The response to withdrawal is clinically reasonable. We will go through each
element.

Some examples of reasonable time relationships are, firstly and obviously, when the onset of the
clinical condition was after the drug was started and not before; or secondly, a drug suspected of
causing a congenital cardiac defect, for example, was taken in the first trimester of pregnancy when
the heart is developing and not just in the last trimester when it couldn't affect the developing heart.

To determine if the patient's clinical conditions are likely to be alternative causes, we need first to
identify them from the medical history if it is provided. They can also often be identified from the
indications for the medicines the patient is taking, including the indication for the suspect medicine.
Other drugs the patient was taking, the concomitants, should be considered. Is the clinical condition
a recognised adverse reaction to any of them? If so, is there a reasonable time relationship with
their intake in this case? If the patient recovered when only the suspect drug was withdrawn then
concomitant drugs are unlikely to be alternative explanations.

Lastly, the response to withdrawal, that is, dechallenge, should be clinically reasonable. That is, the
patient recovered after the drug was stopped or the dose reduced, within an expected time period
for the particular adverse reaction.

The WHO-UMC system category "certain" is rarely used. The requirements are very strict. The event
or laboratory test abnormality must have a plausible time relationship to drug intake. The events
cannot be explained by disease or other drugs. The response to withdrawal, that is dechallenge, is
plausible pharmacologically and pathologically. The event must be an objective and specific medical
condition or a pharmacological phenomenon. Rechallenge is satisfactory if necessary.

We will take a closer look at what these requirements mean and how they compare with those in
the "probable" category in the next lesson.

Lesson 2.3
The learning objective for lesson 3 is to understand the meanings of the criteria in the WHO-UMC
system "certain" category. The "certain" classification requires the timing of event to be "plausible."
"Plausible" is a stronger word than reasonable used in the "probable" category. It means believable
and for that we need more information than for reasonable. The time to onset should fit in with the
known pharmacokinetics of the drug, for example its half life, or for a Type B reaction, the time to
mount an observed immune response would be an example.

A "plausible" response to withdrawal would be, for example, a rapid resolution of a Type 1 allergic
reaction such as urticaria and a longer time for hepatitis.

The "certain" category also requires the event to be a recognised pharmacological phenomenon, or,
a specific observable medical disorder.

So what do we mean by a phenomenon? Well, a phenomenon is like the Northern Lights, the Aurora
Borealis, something you can observe that is the result of an underlying physical and chemical
process.
A pharmacological phenomenon might be decreased prothrombin with warfarin leading to
haemorrhage due to its effect on vitamin K activity; respiratory depression with morphine, or
serotonin syndrome manifested as symptoms such as agitation, increase sweating and myoclonus
due to drugs that inhibit serotonin reuptake by its receptors.

With a new drug we may not be aware of all its pharmacological actions. Therefore an unexpected
reaction will rarely be classified as "certain".
Alternatively, the event should be a specific, observable, medical disorder. This is an important
difference compared with the "probable" category. To classify a suspected reaction as "certain" it
needs to be something you can observe or measure. If you have definite evidence of, for example,
hepatitis or tendonitis then that's an observable clinical condition. It’s pathological.

In contrast, for a "probable" reaction symptoms that are not observable, such as headache or
abdominal pain, can be included.

For the final criteria in the "certain" category, we say that it is rarely used since a rechallenge is
almost always required and this is often unethical. There are strict criteria for a drug administration
to be called a rechallenge. If rechallenge is carried out, the patient should have first recovered from
the clinical event on stopping the suspect drug after the first occurrence. The rechallenge should be
with the same drug, at the same dose, and by the same route.

Skin prick testing for allergy is an accepted form of confirmation although there are false positives.
Rechallenge may not be needed for a "certain" classification in a small number of situations such as
when a cytotoxic drug extravasates and causes tissue damage. Rechallenge may well be dangerous
and the majority of reports that have rechallenge data arise from lack of recognition that an illness
following the exposure to the drug previously was an adverse reaction until it happened again or a
lack of a proper record of the previous reaction.

Now that we have discussed the "probable" and "certain" categories consider this image. These
renal stone images show why the pain of renal colic is so severe as they pass down the ureter.
You can see why they are spiky from this microscopic view showing they are composed of crystals.
This patient had a history of renal stone formation. He developed a HIV infection and was treated
with antiretroviral agents. He again developed renal stones but this time when their composition
was examined they were 80% ritonavir.

Take a minute to consider whether you would classify this reaction as "certain" or "probable".

I think the patient with renal stones had a "certain" reaction even though there was no rechallenge.
The renal stones were the observable clinical and pathological condition and they were mainly
composed of the suspect medicine.

Lesson 2.4

For lesson 4 the learning objective is understanding the criteria for the remaining WHO-UMC system
categories and how the Naranjo Algorithm compares with this system.

Looking at the remaining categories we start with "possible". A "possible" ADR report may be
explained by other drugs or diseases. Like a "probable" reaction, we have an event or a laboratory
test abnormality with reasonable time relationship to drug intake. Beyond that there are several
reasons for classifying a reaction as "possible". The suspected reaction could also be explained by
other drugs the patient was taking. Disease might also be an explanation. For example, if we have a
report of pancreatitis with a drug used to treat diabetes there is a problem because diabetes itself
can lead to pancreatitis.

Furthermore, a "possible" ADR report does not require dechallenge. Another reason for a "possible"
classification is lack of information about outcome when the suspect drug was stopped, that is on
dechallenge. In some cases that information will never be available. Stopping the suspect drug does
not always lead to recovery even if it did cause the clinical event. Maybe it was so serious the patient
died or was permanently harmed. Maybe the event was pregnancy due to an interaction with an
oral contraceptive. The pregnancy will of course continue even if the drug is stopped.

An "unlikely" ADR report may have an improbable time to onset or a more likely alternative
explanation. An event or laboratory test abnormality, with a time to drug intake that makes a
relationship improbable for example before for example the drug was started, or too long after it
was discontinued or it was not in keeping with the known time to onset for a recognised adverse
reaction. For example it would take longer than one day for hepatic failure to occur after exposure
to a drug. And then disease or other drugs provide plausible explanations that are more likely than
the suspect drug.

"Unclassified" ADR reports usually contain very little information. Sadly, we receive a number of
reports that really have very little information in them. For example, we might have reports that say
that the patient took the drug on an unknown date, they had the event on an unknown date. We
don’t even know if the event occurred before or after starting the drug. We suspect it occurred
afterwards, because it was reported, but we can’t be sure. So we say these are "unclassified". Some
people also use this as a "conditional" category for unexpected reactions that don't fit with our
knowledge about the drug and the information in the reports is limited. They may add to evidence
later if similar reports are submitted.

Finally, we have "unclassifiable" reports. These reports have insufficient information and no more is
expected.

An algorithm, such as Naranjo's is an alternative standardised way to assess causality. This is rather a
different approach since scores are given for different elements rather than the more reflective
approach of the WHO-UMC classification. The Naranjo algorithm is one that is widely used. It is
made up of questions about the data you have and the answers result in positive, negative or no
scores. The scores are then used to assign the reports to four categories similar to those in the WHO
system but not the "unclassified" and "unclassifiable" categories. The algorithm was developed for
clinical trials and so has some questions that are not usually relevant to adverse reaction reports
such as blood levels of the drug and response to placebo. However, many of the elements in the
WHO-UMC criteria are included and the results of assessments are usually similar. If the final score
does not fit with your overall impression of the report you should use your judgement as to how it
should be classified.

Lesson 2.5

The learning objective for lesson 5 is: applying the logic to specific reports typical of those submitted
to national pharmacovigilance centres.
We will look at the probable and a possible example and how we might assess them.
We will start with two examples where the suspected adverse reaction is interstitial nephritis with
omeprazole.

Interstitial nephritis is an inflammation in the kidneys, in the tissue between the functioning parts. It
is also one of the typical immune responses to a drug although it can occur spontaneously. The
interstitium becomes inflamed and therefore swollen and the functional elements in the kidney are
compromised leading to renal failure.
This image shows normal renal histology with glomeruli and tubules, the functioning parts, packed
closely together with very little tissue between. In the next slide we see an enlarged interstitium
with many inflammatory cells in it. In recent years, this has been identified as an adverse reaction to
omeprazole and then other proton pump inhibitors (PPIs). The incidence is about 1:12000 patients
exposed, but it is frequently reported because PPIs or protone pump inhibitors are so widely used.

So here is a probable and a possible report. I wont tell you which I think is which.

Here we have a 76 year old female. She was taking a standard dose of omeprazole 20mg daily, which
she took for 4 months. She was taking diltiazem and bendrofluazide and the report said she had
interstitial nephritis, poor appetite and fatigue that developed 3 months after she started the
medicine. She continued her other medicines, stopped the omeprazole and she recovered.

And then we have a 68 year old male patient also taking 20 mg omeprazole daily. He took it for 5
months and developed the same disorder. He was also taking azathioprine. He stopped omeprazole
and continued azathioprine and had not recovered at the time of the report sent three days after
omeprazole was stopped.

Take a minute to consider how you would classify each of these reports and why.

I would classify the first report as probable given the reasonable time relationship to onset, the
response to dechallenge and the continuation of other possibly suspect medicines after she
recovered, making them unlikely causes. However, I would follow up how the diagnosis was reached
unless this was a well known adverse reaction.
I would classify the second report as possible. It was sent too early for recovery to have occurred
after stopping omeprazole so there are two suspects, omeprazole and azathioprine as azathioprine
is a known cause of interstitial nephritis

Lesson 2.6

Lesson 6 is about pitfalls, where you might fall down. The learning objective is recognising reports
with insufficient or misleading information.

So here we are looking at the outcome information. Here is a patient who was taking 40 mg of
omeprazole daily. After a month she developed interstitial nephritis. She was not taking other
medicines. The outcome was reported as "not recovered". The outcome of dechallenge was
unknown. On rechallenge the reaction was reported to have recurred. Is that a certain report? There
was recurrence on rechallenge. There were no dates in the report. I would be very reluctant to make
this a certain report. Why is that? There is no evidence that the patient recovered on dechallenge. In
fact the outcome said they weren't recovered. So the statement about rechallenge is suspect. We do
receive reports like this so we need to be cautious.

And then there is duplicates. Be wary of duplicates, look very carefully at the reports you receive. It
is good if a national pharmacovigilance centre has a system for detecting duplicates but checks are
still needed when reports are assessed.

Next, you might have too much information or not enough information. If a report only lists the
suspect medicine and no others and the patient has clinical conditions that would usually lead to
other drugs also being prescribed then, if the reported reaction is serious, especially if unexpected,
ask the reporter for more details.

And there's more. A female patient, age unknown, was prescribed omeprazole. The dose, date of
administration and indication were unknown. On an unknown date the patient developed interstitial
nephritis. No concomitant medicines were reported. No medical history was reported. The patient
recovered on an unknown date. If you are building up evidence, you have a report, but it is not
useful and should be in the "unclassified" category.
There are also reports with many details but without much value. They are often from
pharmaceutical companies, This is one example with 58 medicines, two suspects and no dates. It is
not clear whether these were current or past prescriptions. And then there are reports with many
adverse reaction terms, which are probably the whole of the patient’s medical history. So be wary of
those reports, they usually do not provide reliable data.

In this slide, consider if this report is complete.

Here we have a 55 year old male patient. The suspect medicine was an iodinated contrast medium.
He was having a coronary angiogram because he’d had a myocardial infarction the previous day. He
developed a florid maculopapular rash several hours after he received the contrast medium.

Take a minute to think if the report was likely to be accurate in only listing one medicine. What other
medicines is the patient likely to have been taking and when would they have been started?

I assessed this report and approached the hospital pharmacist. The patient was already taking
aspirin, simvastatin and metformin when he was admitted for the myocardial infarction. He was
then treated with enoxaparin, clopidogrel and metoprolol. So at least all of the medicines started
the day before his angiogram as well as the contrast medium were suspect for the rash.

Lesson 2.7

For lesson 7 the learning objective is understanding what to do when it is hard to place a report in a
category, what should go into the national pharmacovigilance database and what causality
assessment can and can't do.

Here are two frequently asked questions:

What if I find it very hard to classify a report?

Even experienced pharmacovigilance professionals sometimes find it difficult to classify a report. In

All reports, as long as they contain a drug and a suspected reaction, should be entered. They may
provide support if there are better documented reports of the same suspected reaction in the
future. I initially classified reports of omeprazole and hypocalcaemia occurring years after the drug
was started as unlikely. Later we received more reports and found supporting reports in the
literature for omeprazole causing hypomagnesaemia, low blood magnesium levels, after long term
use. The hypocalcaemia was a consequence of the low blood magnesium levels. The "unlikely"
report was then upgraded to "possible".

In conclusion, let's consider what causality assessment can and can't do.

It can classify the relationship, it can mark key reports and it can improve the scientific evaluation by
systematic use of the available data. It can decrease disagreement between assessors.

What it can't do is prove the connection between the drug and the event. In the next module we will
see how we can gather more evidence by assessing a case series.