Clinical Pharmacology and Therapeutics

ADVERSE DRUG REACTION MONITORING/SURVEILLANCE (PHARMACOVIGILANCE) 65
analgesic. Following its release, there were spontaneous reports
to the CSM of photosensitivity and onycholysis. Further reports
appeared in the elderly, in whom its half-life is prolonged, of
cholestatic jaundice and hepatorenal failure, which was fatal
in eight cases. Benoxaprofen was subsequently taken off the
market when 3500 adverse drug reaction reports were received
with 61 fatalities. The yellow card/black triangle scheme was
also instrumental in the early identification of urticaria and
cough as adverse effects of angiotensin-converting enzyme
inhibitors. Although potentially the population under study
by this system consists of all the patients using a drug, in fact
under-reporting yields a population that is not uniformly
sampled. Such data can be unrepresentative and difficult to
work with statistically, contributing to the paucity of accurate
incidence data for adverse drug reactions.
Systems such as the yellow card scheme (e.g. FDA MedWatch
in the USA) are relatively inexpensive and easy to manage,
and facilitate ongoing monitoring of all drugs, all consumers
and all types of adverse reaction. Reports from the drug regulatory
bodies of 22 countries are collated by the World Health
Organization (WHO) Unit of Drug Evaluation and Monitoring
in Geneva. Rapid access to reports from other countries should
be of great value in detecting rare adverse reactions, although
the same reservations apply to this register as apply to
national systems. In addition, this database could reveal geographical
differences in the pattern of untoward drug effects.
CASE–CONTROL STUDIES
A very large number of patients have to be monitored to detect
a rare type B adverse effect. An alternative approach is to identify
patients with a disorder which it is postulated could be
caused by an adverse reaction to a drug, and to compare the frequency
of exposure to possible aetiological agents with a control
group. A prior suspicion (hypothesis) must exist to prompt
the setting up of such a study – examples are the possible connection
between irradiation or environmental pollution and
certain malignancies, especially where they are observed in
clusters. Artefacts can occur as a result of unrecognized bias
from faulty selection of patients and controls, and the approach
remains controversial among epidemiologists, public health
physicians and statisticians. Despite this, there is really no practicable
alternative for investigating a biologically plausible
hypothesis relating to a disease which is so uncommon that it
is unlikely to be represented even in large trial or cohort populations.
This methodology has had notable successes: the association
of stilboestrol with vaginal adenocarcinoma, gatifloxacin
with hypo- and hyperglycaemia, and salmeterol or fenoterol
use with increased fatality in asthmatics.
INTENSIVE MONITORING
Several hospital-based intensive monitoring programmes are
currently in progress. The Aberdeen–Dundee system abstracts
data from some 70 000 hospital admissions each year, storing
these on a computer file before analysis. The Boston
Collaborative Drug Surveillance Program (BCDSP), involving
selected hospitals in several countries, is even more comprehensive.
In the BCDSP, all patients admitted to specially designated
general wards are included in the analysis. Specially
trained personnel obtain the following information from hospital
patients and records:
1. background information (i.e. age, weight, height, etc.);
2. medical history;
3. drug exposure;
4. side effects;
5. outcome of treatment and changes in laboratory tests
during hospital admission.
A unique feature of comprehensive drug-monitoring systems
lies in their potential to follow up and investigate adverse
reactions suggested by less sophisticated detection systems, or
by isolated case reports in medical journals. Furthermore, the
frequency of side effects can be determined more cheaply than
by a specially mounted trial to investigate a simple adverse
effect. Thus, for example, the risk of developing a rash with
ampicillin was found to be around 7% both by clinical trial
and by the BCDSP, which can quantify such associations
almost automatically from data on its files. New adverse reactions
or drug interactions are sought by multiple correlation
analysis. Thus, when an unexpected relationship arises, such
as the 20% incidence of gastro-intestinal bleeding in severely
ill patients treated with ethacrynic acid compared to 4.3%
among similar patients treated with other diuretics, this cannot
be attributed to bias arising from awareness of the hypothesis
during data collection, since the data were collected
before the hypothesis was proposed. Conversely, there is a
possibility of chance associations arising from multiple comparisons
(‘type I’ statistical error), and such associations must
be reviewed critically before accepting a causal relationship. It
is possible to identify predisposing risk factors. In the association
between ethacrynic acid and gastro-intestinal bleeding,
these were female sex, a high blood urea concentration, previous
heparin administration and intravenous administration of
the drug. An important aspect of this type of approach is that
lack of clinically important associations can also be investigated.
Thus, no significant association between aspirin and
renal disease was found, whereas long-term aspirin consumption
is associated with a decreased incidence of myocardial
infarction, an association which has been shown to be of therapeutic
importance in randomized clinical trials (Chapter 29).
There are plans to extend intensive drug monitoring to cover
other areas of medical practice.
However, in terms of new but uncommon adverse reactions,
the numbers of patients undergoing intensive monitoring
while taking a particular drug will inevitably be too small
for the effect to be detectable. Such monitoring can therefore
only provide information about relatively common, early reactions
to drugs used under hospital conditions. Patients are not
in hospital long enough for detection of delayed effects, which
are among the reactions least likely to be recognized as such
even by an astute clinician.