Processing of Events

PROCESSING OF EVENTS

CLINICAL REVIEW OF REPORTS

All events are assessed by a physician using the same process as for reviewing ADR reports in the spontaneous reporting programme. A ‘relationship’ is established between the drug and the event following the protocol for causality assessment recommended by the World Health Organisation Collaborating Centre for International Drug Monitoring (Meyboom and Royer, 1992). The events are classified according to system/organ class using the IMMP events dictio-nary, which is a hierarchical terminology based on the WHO adverse reactions terminology (WHOART). The hierarchy has five levels, and events can be sorted at each level into their clinically related groupings or individually. There are approximately 3000 event terms in the dictionary.

DRUG-EVENT RELATIONSHIP ASSESSMENT

Each drug-event relationship is coded as one of the following: definite, probable, possible, unlikely or unclassified. These assessments are based mainly on duration to onset of the event and the response to withdrawal and/or re-challenge. They are not regarded as ‘causality’ assessments and are made without prej-udice. Judgements on causality for many of the events can only come later when epidemiological evidence using aggregated IMMP and other available data can be considered along with biases and confounders and pharmacological plausibility. With this back-ground thinking, and to facilitate further evaluation, the assessed events are divided into two categories: those events with a relationship of certain, probable or possible are classified as ‘reactions’ and those with a relationship of unlikely are called ‘incidents’ (because they are likely to be incidental to the use of the drug and represent the background noise of the condition being treated or community morbidity). These two groups are then evaluated for signals of previously unidentified adverse reactions. This is largely under-taken by observation of the nature and pattern of events being reported, examining comparative rates controlled for age, gender, indication and severity of disease as appropriate and differences in profiles. Signals arising from this process may be investigated further by special studies.

INCIDENTS AS CONTROLS

Unless the incident group contains unrecognized adverse reactions, it should represent the background noise, and this should be generally similar for drugs of similar indication. If the incident rates are simi-lar for comparator drugs, then it can be assumed that reporting bias is not present. If there are statistically significant differences between the incident profiles of comparator drugs, then this may be because of the presence of an unrecognized adverse reaction or confounding, e.g. by indication or reporting bias. Any such differences are therefore investigated.

Incidents are also used as within-drug controls for characterizing adverse reactions. The variables associ-ated with, respectively, the reactions and incidents for a drug and also the patient characteristics are compared, and should there be differences these may indicate risk factors – e.g. a gender or dose difference – for the reaction under study. An example of the use of inci-dents as within-drug controls is as follows. The rates of adverse reactions were higher in women than men for moclobemide [relative risk (RR) 1.7; 95% confidence interval (95% CI) 1.4–2.0] and fluoxetine (RR 1.7; 95% CI 1.3–2.2). There was no significant gender differ-ence seen in the incident rates. It would appear there-fore that the gender difference seen for the reactions is a true risk factor and not because of reporting bias.

For these two drugs, which were monitored concur-rently, the incidents were also used as between-drug controls. The reaction rate for fluoxetine was 50% higher than that for moclobemide (RR 1.5; 95% CI 1.2–1.7). There was no significant difference between the incident rates, suggesting an absence of reporting bias and strengthening the finding of greater risk with fluoxetine (Coulter, 1996).