Bear’s post on multiple endpoints got me thinking about things…

 Below, I provide a little bakcground on statistical testing that will help in understanding the information on evaluationg multiple endpoints in the next post. 

There are two types of error that can occur in statistical comparisons of two sample groups for differences:

Type 1 error: rejecting the null hypothesis when it is true.  The null hypothesis is always that there is no difference between your two groups. 

Type 2 error:  accepting the null hypothesis when it is untrue.

Generally, the most important error to avoid is a Type 1 error.  The likelihood of making a type 1 error is estimated for an experimental endpoint and must set to a low value to be acceptable (generally P=0.05, but it can be set lower for a more rigorous test).  For drug development making a type 1 error means concluding that the drug has a beneficial effect when it doesn’t. It could lead to needless loss of life, money spend on development and application of a useless or even harmful drug, etc.

Making multiple tests on a data set leads to an increase in the likelihood of making a Type 1 error.  Therefore, a correction must be made (generally by setting a lower P value that must be met) to avoid making a spurious conclusion that there is a difference between the two groups (in one of multiple factors compared) that doesn’t really exist.  There are many things that must be considered in making multiple comparisons (having multiple endpoints in a clinical trial), but a simplified view of it is that the more endpoints you are testing, the lower the P value that must be achieved.

For the primary endpoint of BETonMace, at a P value of 0.05, we must accept the null hypothesis.  Treatment of patients with apabetalone did not lead to increased time to first MACE event. So, the trial definitely did not lead to a type 1 error for this endpoint. We all hope that a type 2 error occurred and we currently must accept the null hypothesis although it is untrue. 

Jupe