Professor David Finney is arguably the initiator of global pharmacovigilance, and, in his 90th year, he can still provoke us to thought! A statistician by training, he has recently criticized the WHO for their removal of outliers in producing normative calibration curves (for thromboplastin time). The assertion in his paper is an important one for pharmacovigilance – ‘‘Never discard an apparent outlier unless there is strong evidence that it was the product of a measurement or other form of observation that suffered a gross mistake or accident, this misfortune being unrelated to any experimental treatment under investigation.’’ The statement can be seen as implying that data that ‘fits in’ can broadly be accepted as coherent, and that which does not should be scrutinized for reasons why it does not. I am not in agreement with the first putative implication in the sense that it is likely to be a balance of heterogeneous factors that determines belonging to a group; therefore, some important factors may be masked by a much greater predominance of others. The main point, however, is that one should never make an assumption that an outlier is an error, and there may well be significant gains to be made by looking for the reasons behind outliers. In pharmacovigilance we make assumptions about outliers all the time – after all they are the few people who experience adverse events that are reported as adverse reactions. But the first person to consider these patients is the health professional reporter who, in sending the report, has effectively decided that there is a reasonable likelihood that the drug selected could have been causally related to the event the health professional reports. It would be most interesting to understand the decision process behind each adverse reaction diagnosis: why select such a relatively rare cause as a drug for a particular clinical symptom or disease? We often talk about the possibilities of various biases and confounding such as publicity about a drug reaction and the influence of high background disease rates, but we have very little real knowledge about how these influences operate in an individual case. Reports of events after drug treatment are analysed from trials and studies on a statistical basis, comparing with controls, but serious adverse reactions following drugs are rare, usually at rates <1/1000 for marketed products. For such a small total number of outliers, surely one should be more interested in why they are outliers than trying to differentiate whether they are statistically more or less frequently related to the drug? A doctor makes a diagnosis of myocardial infarction, a common disease. If the patient is young and fit with no family history, the doctor will consider underlying causes such as diabetes, hyperlipidaemia, Prinzmetal angina, cocaine abuse and coronary embolism from a septic heart valve, amongst many others. The possibility that a drug that is taken might be the cause is also given more consideration because the case is generally unusual in the health professional’s training and experience. Any health professional should know, and eliminate, the common causes of myocardial infarction, as well as thinking of a drug as the cause, but if the clinical event is rarer in prevalence, such as agranulocytosis, then the range of causes are less likely to be known by health professionals, and perhaps drugs will be more likely to be the cause than other rare clinical events. The point is that EDITORIAL Drug Safety 2009; 32 (8): 623-624 0114-5916/09/0008-0623/$49.95/0
Read full abstract
Mar 11, 2010
M Remzi Sanver + 1
Open Access
