of exposure. It is sensible to consider that the occurrence of cases in the exposed population and their reporting, both correspond to a pseudorandom process which can be described by an ad hoc probability model. Given that, in pharmaco-vigilance, the source population is generally extremely large and the probability of occurrence very low, the Poisson distribution is expected to be quite a satisfactory model (Snedecor and Cochran, 1989). In these conditions, the calculation of the 95% two-sided CI for the reporting rate consists of considering the lower and upper limits for the Poisson parameter read in a table such as Table 21.1.
In the above example, the 95% Poisson CI for the observed number 18 is [10.67; 28.45]. The CI for the reporting rate is thus: 10.67 to 28.45 for 87 719 months, i.e. 1.2 to 3.2 per 10 000 person-months of exposure. When the number k of reports is large enough, i.e. 15 or preferably 30, the CI can be calculated by using the normal approximation for a Poisson count (Daly et al., 1991):
In both cases, this CI defines the set of values which could be observed because of the sampling variation, all parameters remaining identical.
Was this article helpful?