Clinical evaluation of virus safety and inhibitor incidence: statistical considerations.

Abstract

The basic statistical issue in pharmacovigilance is to claim, with reasonable certainty, that the incidence of an event of interest in a population of subjects is less than a certain value. How many subjects and events must be observed before such a claim can be made? A first situation of practical importance is when a product has been on the market for some period of time, and the safety of this product regarding some outcome of interest is questioned (for instance, the viral safety of blood products). Having observed a few occurrences of the event of interest, how confident can we be that the product is responsible for an elevation of the incidence of this event compared with the baseline incidence in a reference population? This issue will increasingly need to be addressed prospectively: how many subjects need to be treated and how many events observed, to be reasonably certain that a product is safe? Multi-stage designs are appropriate to address this question, yet they do not seem popular in pharmacovigilance. Such approaches could complement the standard recommendations to assess the safety of blood products, which are adequate as a first screen against major safety problems, but wholly inadequate for the long-term surveillance of subjects at risk of rare events. It will be argued that, from a regulatory perspective, the implementation of prospective protocols for pharmacovigilance, using appropriate statistical tools, would permit a tight control of the safety of new products, while making these products available as early as possible to the patients who need them.