Biostatistics:
A Critical Component of Pharmacovigilance
Alan Phillips • Director Biostatistics, Europe • ICON Clinical Research
Frequently, when a new medicinal product is introduced to the market, relatively limited information is available on the product’s safety (1). This is due to many factors, including the small number of subjects studied in the clinical development program, restricted populations in terms of age, gender and ethnicity, restricted co-morbidity, and co-medication, restricted conditions of use, and the relatively short duration of exposure and follow-up.
Pharmacovigilance practices have tended to focus on individual case reviews of spontaneous reporting of adverse reactions. However, this approach can be strengthened by greater use of statistical methods, especially techniques associated with the analysis of large volumes of data. Such techniques can be readily used when high quality, accessible databases are available, a not insignificant challenge in itself.
One area of particular interest today is the concept of risk management, which is increasingly mentioned in safety guidelines issued by the FDA and other international regulatory bodies.
The management of risk consists of four steps:
1. Risk detection
2. Risk assessment
3. Risk minimization
4. Risk communication
Biostatisticians have the skills and methodology to help systematically review databases to identify potential safety issues related to medicinal products; that is, assist in risk detection.
For example, take the EudraVigilence Data Analysis System. This database is being developed by the EMEA to support EU pharmacovigilance activities. The EMEA has also issued a guideline on the use of statistical signal detection methods in the EudraVigilence Data Analysis System. This guideline discusses the use of the proportional reporting ratio (PRR) to detect signals of disproportionate reporting (SDRs). The term SDRs is used, rather
(PRR) to detect signals of disproportionate reporting (SDRs). The term SDRs is used, rather than “signal detection,” to reinforce the need for results to be further investigated based on the clinical context. That is, it is possible that results from statistical methods could merely reflect reporting tendencies, which could be a function of numerous non-causal factors (confounding, reporting artefacts, statistical noise or some combination of the above).