8/19
Pitfalls of the Bonferroni method:

⚠️puts all hypotheses in one bucket (all or nothing)
⚠️⬇️ the ability to reject the null hypothesis when it should be rejected (⬆️ type II error)
⚠️interpretation of the results differs based on # of comparisons

pubmed.ncbi.nlm.nih.gov/9553006/

9/19
In 1979, Sture Holm suggested a stepwise procedure:

🔸divide 𝛂 by number of comparisons
🔸reject the hypothesis that reaches significance
🔸then divide 𝛂 by comparisons - 1
🔸keep going until you fail to reject the hypothesis

jstor.org/stable/4615733

10/19
There are other methods (Simes, Hochberg, Dunnett amongst others)

🔸They all examine endpoints in a data-driven order i.e. endpoints are tested beginning with the least or most statistically significant one

11/19
The Hierarchical approach utilizes clinical information to develop a pre-defined testing sequence based on clinical importance

The two major hierarchical tests are:

🔸Fixed sequence
🔸Fallback

pubmed.ncbi.nlm.nih.gov/29847757/

12/19
Fixed sequence

🔸tests hierarchically ordered endpoints in sequence at the same 𝛂 until the first non-rejection of the null hypothesis, when you stop 🛑

13/19
Advantages of fixed sequence testing:

🔸simple procedure
🔸performs best when earlier endpoints are associated with large effects

14/19
Pitfalls of fixed sequence testing:

🔸once a hypothesis/endpoint is rejected, you can't test any of the other hypotheses/endpoints

15/19
Fallback Procedure:

🔸tests hierarchically ordered endpoints in sequence
🔸splits the 𝛂 amongst the endpoints according to clinical importance
🔸even if the initial endpoints null hypotheses are not rejected, all endpoints are tested

16/19
Now let's look at Table 2 of CLEAR Outcomes:

▶️Once we fail to meet significance (fatal or non-fatal stroke), subsequent endpoints are not tested
▶️Thus, the fixed sequence hierarchical procedure was used

pubmed.ncbi.nlm.nih.gov/36876740/

17/19
For Further Reading:

www2.cscc.unc.edu/impact7/BretzA…
pubmed.ncbi.nlm.nih.gov/29847757/
pubmed.ncbi.nlm.nih.gov/28664566/

18/19
Now that we understand the methods used, join CardioNerds #HouseJones and our experts at 8PM EST on May 31st (TOMORROW!) for #CardsJC to discuss the takeaways from CLEAR Outcomes

cardionerds.com/cardsjc-clear/