Module 3: Causality assessment of case series

Lesson 3.1

The learning objective for lesson 1 is to understand how assessing a series of case reports adds value
to single case report assessment in the process of signal detection.
In this module we look at how we assess a series of case safety reports. In the first module we laid
the foundation by considering assessment of single case safety reports. Case series assessment
builds on this foundation in preparation for signal detection. We will look at what information we
can obtain from a case series that we can't find in individual reports.

First here is a reminder of the purpose of causality assessment, which is to answer the questions
"Did the drug cause the clinical event?", "Does the drug increase the risk of the clinical event?" And
therefore, "How do we decrease the occurrence of this adverse drug reaction or perhaps its severity
by early recognition of the symptoms?"

For this presentation a case series is defined as a group of patients with similar exposure (the drug)
and similar outcome (the suspected adverse drug reaction). The term "drug" refers to a medicine or
a vaccine.

There are problems with single case causality criteria, for example, "plausible" or "reasonable"
timing. Was the onset of the clinical event after the drug was taken and did the clinical event occur
after an expected duration of drug use? A single report might not have the necessary dates. Also we
may not know the expected duration of drug use to onset. This is usually referred to as the "time to
onset". If we combine the information in case reports we might start to see a consistent pattern of
time to onset.

We may or may not have de- or rechallenge data in an individual report but some reports in a case
series might have this information.

Excluding other causes is not easy from one report, for example, many people have diabetes or take
statins and it is not easy to determine from individual reports what role they might have played in
the development of the clinical event. Were they alternative causes? Were they likely to have
contributed? Assessment of a case series increases the possibility of finding drugs or clinical
conditions that increase the risk of the suspect drug causing the clinical event.
If a single report is for a condition that is a typical drug reaction such as Stevens Johnson Syndrome
or agranulocytosis that adds strength to the report but there is only a small list of these conditions.

So assessing a case series is likely to supply additional information that is missing or hard to assess in
individual case reports. A logical analysis is applied that is a development from single case
assessment. However, it is important to first carry out single case assessment of the reports in the
case series.

Many case series are identified through observations by the staff at national pharmacovigilance
centres as they process reports, especially in smaller centres. However, if a case series has been
highlighted through a statistical disproportionality method then the clinical assessment of the single
case reports and the case series is still essential to find out if there is a signal. Statistical
disproportionality methods highlight case series for clinical assessment. They do not replace clinical
causality assessment.

Lesson 3.2

The learning objective for lesson 2 is to understand the importance of collecting together or collating
the data from each single case report before applying a guideline for case series assessment. It is
essential to consider each patient, the drug, the suspected adverse drug reaction and the outcomes.

So, what guidelines do we have for clinical assessment of a group of reports? First of all we need to
consider what each report tells us about the patient, just as we did for single case reports. Also the
suspect drug, and any other drugs the patient was taking, including prescribed, over the counter and
herbal and other complementary and alternative medicines; the reported ADRs and the outcome for
the patients, including what happened on dechallenge and rechallenge, if reported. Seriousness
should also be recorded since this influences what action will be taken if a signal is detected.

What does each report tell us about the patient's demographics, their gender, their age, also their
other clinical conditions, relevant genetic predispositions, if known, other non-drug exposures,
reactions to other medicines.
For the drugs, suspect or interacting, we need to note their Anatomic Therapeutic Classification (the
ATC group), the dose, the duration, the start and stop dates, and as much information as possible
about concomitant drugs.

For the adverse drug reaction the System Organ Class it belongs to, the specific symptom or
condition reported, the date of onset and recovery. And also co-reported adverse reaction terms
which may provide important information about the suspected adverse reaction in the case series.
For example a series of reports of jaundice would be difficult to interpret because jaundice occurs
with a number of hepatic or biliary disorders. What is the drug actually causing? However, if many of
the reports in the case series also list gall stones as an adverse reaction term then a reason for the
jaundice becomes apparent.

Once we have found out what is already known about the suspect drug or drugs and the suspected
adverse reaction, we can look for patterns in the collated reports that lend support to a causal
relationship and perhaps evidence for risk factors too. We can then go onto using the causality
assessment criteria developed by Sir Austen Bradford-Hill.

Before we consider these criteria, for those of you who want to read more deeply you will also find it
interesting to read about two approaches to classifying adverse reactions which help with causality
assessment. The EIDOS system focuses on the mechanism by which the adverse reaction may occur
and the DoTs system on clinical pharmacology, dose responsiveness, time course and susceptibility.
Professor Edwards article deals with the principles of causality assessment in depth and all of the
authors recommended the use of the Bradford-Hill criteria.

Additional reading:

Aronson JK, Ferner RE. Joining the DoTs: new approach to classifying adverse drug reactions. BMJ
2003;327(7425):1222-5.

Ferner R, Aronson JA. EIDOS: a mechanistic classification of adverse drug effects. Drug Safety
2010;33:15-23.

Edwards IR. Causality assessment in pharmacovigilance: still a challenge. Drug Safety, on-line, 2017.

Lesson 3.3
The learning objective for lesson 3 is to understand a group of the Bradford-Hill criteria and how
they apply to pharmacovigilance.

So, who was Sir Austen Bradford Hill? He was one of the first medical statisticians and worked on
what is thought to be the first randomised clinical trial evaluating the use of streptomycin in
tuberculosis. He went on to work with the famous epidemiologist Sir Richard Doll on a series of
observational studies on tobacco smoking and lung cancer in the 1950s. The results of these studies,
not surprisingly, caused considerable controversy.

Eventually, in 1965, Bradford Hill proposed a set of criteria which he said would indicate in what
circumstances an observed association between an exposure and an outcome could be considered a
causal association. These criteria have been applied most often to the findings from epidemiological
studies but in recent years we have begun to use them in pharmacovigilance to look at the evidence
for causality between a drug and a clinical condition (or the suspect adverse drug reaction) in a case
series of individual case safety reports.

So what are the criteria? They are:

Strength of association, temporal relationship, consistency of reporting, biologic plausibility,
coherence, dose response, specificity, experimental evidence and analogy.

The purpose of using the criteria is to help us develop judgement when applying the WHO-UMC
system or other causality assessment tools to a series of case reports. All of the criteria will not
usually be fulfilled. They are a guide, not a scoring system.

Strength of association in pharmacovigilance relates to the observed number of reports compared

with the expected number. So, if a statistical disproportionality method shows that the observed
number of reports is significantly greater than expected there is strength in the association and this
increases with greater disproportionality. However, as we noted earlier the reports still need to be
assessed clinically.

A temporal relationship supports causality too. What do we mean by this? As we discussed in the
last module the clinical condition (that is the suspected adverse reaction) should obviously have
commenced after the drug was started. Beyond this, if there is a consistent pattern in the reports of
a time to onset, that fits with the expected time for such a reaction to occur because of the
pharmacology of the suspect drug or the time taken for the clinical condition to develop (the host's
response), then this is supportive.

How do we know what is a reasonable time to onset? A study carried out at the UMC by Gazaleh
Khodabakhshi examined a large number of reports of serious disorders that are often drug related:
agranulocytosis, angioedema, hepatitis, Stevens Johnson Syndrome (SJS), toxic epidermal necrolysis
(TEN). And what she found was that agranulocytosis usually develops a week or two after the
suspect drug is started, whereas the onset of angioedema is usually very abrupt, like anaphylaxis.
Hepatitis usually takes a little longer than angioedema as does serum sickness. SJS and TEN can
occur in the first few days but usually a little later, so you start to see a pattern.

But we don't always know what a reasonable time to onset is. How do we act in this case? We look
at the case series for consistency. Is there a similar time to onset across the reports?

Lesson 3.4

The learning objective for lesson 4 is to understand the meaning of the second group of the
Bradford-Hill criteria and how they apply to pharmacovigilance.

Consistency is also one of the criteria. It means that reports are received from a range of
reporters or, in international databases, a range of countries, and there are similar
observations in the reports. If reports are from only one source, it will be important to
decide if there is a local problem with a medicine, such as a batch difference or poor storage
etc. If not then a lack of similar reports from elsewhere reduces the likelihood of a causal
association.

Then we come to biologic plausibility which means that the suspected reaction fits in with
what we know about the drug's actions. However, we may not know that the suspect drug
can act in a way that causes the adverse reaction, especially if it is newly marketed. For
example tendonitis with ciprofloxacin was unpredictable. Therefore the criterion is
supportive if it is present but it is not a strong requirement for causality.
Here is an example of biologic plausibility. We know that anticholinergic drugs may cause
urinary retention because the bladder outlet sphincter can't relax normally. It is most likely
to occur if the bladder outlet is already compromised, for example with an enlarged
prostate gland. If a new drug is reported to cause urinary retention, if it has some
anticholinergic activity, then the retention is biologically plausible.

Now we come to coherence. Does the suspected reaction fit with existing knowledge? For example,
we know that furosemide can cause loss of potassium. If it isn't being very effective in a patient, it
might not affect the potassium levels at all but it can't retain potassium. So if you have a report of
furosemide and high blood potassium levels, it is highly likely that another medicine being taken is
the cause, such as spironolactone, or that the patient may have developed renal failure leading to
high potassium levels. It would also not be coherent with existing knowledge for a drug that is not
absorbed from the gastrointestinal tract to cause organ damage.

A dose-response relationship is good evidence of causality but may not be observable in some
situations. For example the very small doses needed for an immunological adverse reaction are
considerably below the therapeutic dose of the drug and so any dose response will not be observed
with normal clinical use.

When we think about specificity we consider the terms used for the suspected adverse reaction.
Many adverse reactions have multiple causes, for example headache, abdominal pain, renal failure.
Generally drugs cause ADRs through specific mechanisms so that an adverse reaction is more likely if
a specific cause of the condition is reported such as interstitial nephritis occurring in the reports of
renal failure. We also consider the drugs, are a number of drugs suspect? There is much stronger
evidence for causality if it is just one or two.

Then we come to experimental evidence. This may be from previous animal or human studies. For
example many drugs are now checked for prolonged QT interval.

Analogy is when similar reactions have been observed with other members of the suspect
drug's ATC group, for example:
- Combined oral contraceptives and venous thrombosis
- Angiotensin converting enzyme (ACE) inhibitors and angioedema

Lesson 3.5
The learning objective for this lesson is to be able to collate the data from a series of case reports to
prepare for applying the Bradford-Hill criteria. So we are going to talk about data gathering.

Let’s remind ourselves that causality assessment is not just an intellectual exercise. Our aim is to
detect signals in order to minimise harm to patients. National centres publish their own signals.

Signals from the UMC based on international reports in VigiBase are circulated in this Signal
publication to the national centres so that they can consider the relevance of the signal to them.

In this example we will look at reports of omeprazole and acute renal failure. This is a known
reaction but it is interesting to see how evidence developed starting with individual case safety
reports.

In 1996 a national centre had three reports for the first time of omeprazole and acute renal failure
reported since 1994. This is a simple method of regularly screening a database, to promptly detect a
suspected adverse reaction when it has been reported three times. In this series the reports were
for two males and one female, the ages were from 77-78 years. Omeprazole was the sole suspect
medicine in all three reports. Looking at outcomes, one patient died. They had rheumatoid arthritis
and it was thought they might have had rheumatoid vasculitis affecting the kidneys, but there was
no evidence for that. In the other two reports there was no evidence for why the patients developed
renal failure.

One patient had not recovered at the time of reporting, for the other one outcome was unknown.

The next step for a national centre is to search its database for any reports for omeprazole in the
System Organ Class urinary tract disorders.

There were no other reports of abnormal renal function but there were two reports of interstitial
nephritis (IN) affecting two males aged 59 and 72 years. One had symptoms suggesting renal failure,
and omeprazole again was the sole suspect in both reports. Time to onset was eight months and
four months, and these people had had biopsies. One of them had IN and one had fibrosis which was
thought possibly to be due to repeated attacks of IN. One recovered with sequelae and for the other
the outcome was unknown.
So you might start to ask: well was IN the cause of the renal failure in the other reports? This is a
condition we discussed in the single case report assessment module. The tissue between the
functioning parts of the kidney becomes inflamed and swollen, probably due to an immune reaction,
and compromises renal function.

A literature search is the next step. There was nothing in the product information at that time, but
there was a very small number of published case reports of interstitial nephritis and omeprazole
since 1992. Was there sufficient evidence that omeprazole cause IN? An examination of the reports
in VigiBase to the end of 1995, actually revealed 15 reports of IN with omeprazole, more than were
published.

Using the UMC IC method the number of observed reports were significantly greater than expected
since the IC025 value (the lower border of the credibility value) was greater than zero. The IC value
was 2.17 and had been increasing for the previous three years which is also supportive of causality.
Data mining, including the IC method will be discussed in another module.

It is important to understand how to search and interpret the data in VigiBase. First, find all the
reports for the drug of interest so search for the drug as a substance. View all the SOC group, ensure
you find all the reports of the diagnosis, some people might have just entered nephritis. If it is an
important issue you may want to follow up with the reporting country. On the other hand, be careful
not to include conditions that are not relevant, for example glomerulonephritis is an entirely
different condition and it should not be included. So learning how to search is an important aspect of
assessing a case series.

Lesson 3.6
The learning objective for lesson 6 now that we've collated the data, is to analyse the data in an
actual case series in order to apply the Bradford-Hill criteria.
So let us now look in more detail at all the evidence we have from the Vigibase reports. Who were
the patients, what was the indication for omeprazole, what other illnesses did they have? There
were 15 patients (sic! correct number is 14) from 7 countries.

In this slide you can se that 5 were males and 9 females, the age was late middle age or elderly. The
indication was upper gastrointestinal disorders, so nothing that would cause renal failure or
interstitial nephritis. None of the patients were reported to have other illnesses that might have
caused or contributed to IN.

As you can se here, omeprazole was the sole or only suspect medicine in 12 of the 15 reports. You
will note from this slide that information was not available in all of the reports for all of the variables
such as dose, time to onset etc. This is usual and we have to be aware of this limitation.

There were sufficient reports with information to show that there was not a strong dose-response
relationship, the time to onset ranged from 14 days to 42 months (a wide duration), but the median
duration indicated that most were a much shorter duration than 42 months which was an outlier.
Normally we don't exclude any reports but in one case the time to onset was minus 4 months, so
obviously we can't count that.

What about concomitant or co-suspect drugs? There was no consistent pattern suggesting an
interaction. Five medicines that can cause interstitial nephritis were prescribed for four patients
– bendrofluazide, indomethacin, diclofenac, dicloxacillin, azathioprine. Recovery for two patients was
clearly related to omeprazole cessation and not the other medicines that might cause IN. They were
continued. No concomitant medicines suggested the patients had renal diseases already.

After considering the patients and the drugs we now look at the co-reported adverse reactions.
These include four reports of fever and one of renal failure as well as other symptoms that might
accompany immune reactions or renal failure.

Finally we look at patient outcomes. Half of the patients recovered, one recovered with sequelae,
this was on stopping the medicine, three hadn't recovered and for four the outcome was unknown.

If we consider dechallenge, recovery occurred on dechallenge with no clear alternative explanation

in six patients, although for two I was not entirely certain (so I put this in brackets). Sometimes it is
difficult to interpret the information. There was biopsy evidence for interstitial nephritis for one
patient and you remember that the national centre had another that was not yet entered into
VigiBase in 1996.

So you'll see that now that we have the data together that some is missing and the Bradford Hill
criteria are not completely filled but even so, a picture is emerging.
Let's look at how our data fits with the criteria:

We have strength of association - we have significantly more observed than expected case reports in
Vigibase, and that is statistical disproportionality.

The temporal relationship - there is a consistent time to onset in the interstitial nephritis during
omeprazole use occurred within one to seven months of starting treatment in most patients for
whom this data was available. Also in terms of temporal relationship there was a reasonable time
to recovery. Over half of the patients had recovered or improved at the time of reporting.

The reports were also specific - no concomitant medicines or histories indicated interactions or
conditions that could have been an alternative explanation. Omeprazole was the sole suspect in
most reports. In two of four reports where other potentially causal medicines were prescribed the
patients recovered when only omeprazole was stopped.

We also had consistency - reports with similar details were sent from seven countries.

And then, biologic plausibility - IN is a recognised typical adverse drug reaction.

So, now we consider the simple question: Is there any other way of explaining this set of data? Is
there any other answer equally or more likely than cause and effect? Take a minute to reflect on this
question.

The logic is probabilistic, we weigh up all the factors, assess the probability or likelihood.

In the reports we've looked at there is no clear alternative explanation, apart from omeprazole, for
interstitial nephritis in most of the reports in this case series.

Lesson 3.7
In lesson 7 the learning objective is to begin to learn to recognise when there is an alternative
explanation to direct cause and effect for the drug or clinical condition in an actual case series.
Here we consider an unexpected adverse reaction highlighted by statistical disproportionality:
diltiazem and rhabdomyolysis.

In the second quarter of 2010, diltiazem (a calcium channel blocker) and rhabdomyolysis was
statistically prominent from the background data in VigiBase and there were 55 reports. Does
diltiazem cause rhabdomyolysis? It certainly wasn't listed in product information at the time and it
was a drug that had been used for some years.

If we look at these case reports we can see a common pattern. If we look at the medicines the
patients were taking with diltiazem, they were all taking simvastatin and some were taking some
other medicines as well.

We know that rhabdomyolysis is a severe myopathy and often fatal. It's a known dose related
reaction to simvastatin, but it occurs very rarely at standard doses. The risk is increased with
interacting medicines increasing exposure to simvastatin through inhibition of the CYP3A4 enzyme.
The reports suggest that simvastatin was the cause of the rhabdomyolysis. So was diltiazem just an
innocent bystander that is commonly prescribed with simvastatin? That is the question. Did
diltiazem have a role? Now, we do know that diltiazem is a weak CYP3A4 inhibitor but it is not
thought to interact to a clinically important extent with standard daily doses of simvastatin.

In this slide we have some more information. Take a minute to reflect on it. Think of the ways in
which rhabdomyolysis might have been triggered in these reports.

Firstly consider the simvastatin daily doses. In all except one report the doses exceeded the standard
daily dose of simvastatin which is 20 mg. At the time of this signal cardiologists were encouraging
the use of increasing doses of statins to reach ideal low density lipoprotein cholesterol blood levels.
At these doses diltiazem may have interacted to a clinically important extent. Also in the first report
the combination of gemfibrozil and simvastatin would have increased the risk of rhabdomyolysis. In
the last report rhabdomyolysis occurred at a surprisingly low dose of simvastatin but ciclosporin is a
potent CYP 3A4 inhibitor and colchicine can also cause rhabdomyolysis so diltiazem probably did not
have a role in this patient. In four of the reports an interaction between simvastatin and diltiazem is
possible.
In this case series there is an alternative explanation to diltiazem causing the rhabdomyolysis. So
again to ask our question, is there any other way of explaining this set of data? Yes there is. Evidence
suggests that diltiazem didn't directly cause the rhabdomyolysis but it occurred because there was a
clinically relevant interaction with it when simvastatin was taken at greater than standard daily
doses.

And so, in conclusion, the case series re-emphasises that we must consider every variable, that is;
the characteristics of the drugs implicated, the patients affected, the ADRs reported and the
outcomes. We must gather and evaluate all available data and with the help of guidelines weigh up
the evidence to reach the most probable answer to "Did the drug do it?"