Biostatistics - continued
Suppose the following data are observed:
|
Event (R) |
All other events |
Total |
Medicinal product |
A |
B |
A+B |
All other medicinal products |
C |
D |
C+D |
A+C |
B+D |
A+B+C+D |
The PRR is computed as follows:
PRR = A / (A + B)
PRR = C / (C + D)
The PRR looks to see whether an adverse event is reported relatively more frequently in association with a particular medicinal product than with other medicinal products. That is, it measures a reporting relationship between a medicinal product and event on the basis of a relative increase of the proportion of individual cases related to an adverse event. The chi-square test can be used to investigate any association. In addition to reporting the statistical association, 95% confidence intervals for the observed PRR can be derived. Note that a significant PRR does not necessarily imply a causal relationship.
One of the potential weaknesses of this approach is at what level should the PRR be set to identify potential statistical signals. At present there is no “gold standard” on the threshold value that should be used. In practice there needs to be a trade-off between generating too many false positive signals if the threshold is too low and missing potential signs if the threshold is too high.
Another weakness with the measure is that it is one-dimensional. That is, the PRR involves the comparison of a reporting relationship for a specific medicinal product for a specific event. The statistical methods can be refined to permit multi-dimensional modelling of potential risk factors; for example taking into account age, gender and ethnicity and co-medication and their interaction. Statistical methods used in other areas of clinical research are readily transferable to pharmacovigilance data.
The guideline also discusses the concept of the dynamic PRR. That is calculating the PRR and 95% confidence interval, and investigating how it changes over time. Again, techniques developed by statisticians for trend analyses can be readily applied to pharmacovigilance data to formal identify statistical trends over time.