PhUSE Wiki

Counting Adverse Events (AEs) per treatment group and per patient might seem a straighforward task at a first glance. However, looking at the topic in a little more detail, it reveals room for interpretation and in fact companies do have their own, slightly differing derivation algorithms for counting AEs.

This is an attempt to formalize the factors that can / could contribute to a single AE count and maybe even come to an agreement on a best practice. 

The most common AE table consists of subject counts. Typically in the CSR it is displayed how many subjects out of a populations has a preferred term within a body system and how many subjects have AEs in a body system as summary. Each subject is counted only once within each category. So when one subject has two headache AEs, then this subject is counted once under headache. Subjects are also only counted once within each body system, so when a subject is having different AEs within the same body system. This means, that the numbers of the preferred terms does not sum up to the body system numbers. Typically the AEs are counted per treatment group and sometimes overall groups additionally.

As the denominator the analysis population which is displayed is used in general. So the percentages display typically the frequency of occurrences of AEs within a population of a treatment group.

Pitfalls for Programming

• 
  The preferred terms and system organ classes subject counts needs to be counted separately, to not count subjects several times within a system organ class.
• The denominator should be investigated over the population and not over subjects having AEs, as in the AE domain only subject with AEs are contained.
• According the study design, the "simple" count can easily become tough, when for example for a multiple dose study it needs to be investigated in which treatment group an AE appears.
• Be aware, that SAS works strange, when you have different codes for the same decode. In Meddra there are some identic preferred terms (text decode) having different codes. A useful workaround is removing, calculate and put back the format or working with the text decode only.

Example

subject 1, SOC 1, PT1
subject 1, SOC 1, PT2
subject 2, SOC 1, PT1
subject 3, SOC 1, PT3

Complexity Aspects

When having different study designs, the "simple" count could result in more discussions. For multiple dose studies, the easiest and probably most common way to count AEs is by treatment sequence instead of the group. So it can be investigated which sequences have AEs. 

This could be sufficient, but sometimes it is of interest to see the AEs of a special treatment and not a treatment sequence. Then it needs to be decided how to assign AEs from a sequence to a single treatment. Apart from the association to the single treatment another aspect is the risk group. Identic subjects are in several treatment groups, but not necessarily in all treatment groups, as due to dropouts or study design. Investigating the denominator could also become a very delicate task.