Skip to main content

Table 3 Approach to multiplicity due to multiple treatment comparisons

From: Approaches to multiplicity in publicly funded pragmatic randomised controlled trials: a survey of clinical trials units and a rapid review of published trials

a) Review: multiplicity approach taken

 

Formal adjustment

Hierarchical testing

Other approach

None

Related treatments

7/15a

1/15b

1/15c

6/15

Distinct treatments

2/8d

0/8

0/8

6/8

b) Survey: responses to posed scenarios

 

Yes

No

Unsure

Would you consider adjusting for multiplicity arising from making multiple treatment comparisons?

24/27 (89%)

1/27 (4%)

2/27 (7%)

Consider a parallel group trial with three treatment arms, where all comparisons are of interest. Would you adjust for multiplicity in the following scenarios?

   

Two of the treatment arms are related, e.g. Group 1 = placebo, Group 2 = low drug dose, Group 3 = high drug dose

22/27 (81%)

1/27 (4%)

4/27 (15%)

The three treatment arms are unrelated, including one placebo arm, e.g. Group 1 = placebo, Group 2 = drug, Group 3 = exercise

16/27 (59%)

7/27 (26%)

4/27 (15%)

The three treatment arms are unrelated, but all are active treatments, e.g. Group 1 = drug, Group 2 = exercise, Group 3 = education

19/27 (70%)

6/27 (22%)

2/27 (7%)

Would you be more likely to adjust for multiplicity if the number of treatment arms was increased?

12/27 (44%)

12/27 (44%)

3/27 (11%)

  1. Notes: a Three trials performed a Bonferroni correction, one a Holm correction, two a Hochberg correction and one used a 1% significance level for all treatment comparisons
  2. b Two treatment comparisons were split into primary and secondary hypotheses and analysed in a hierarchical manner
  3. c A post-hoc Bonferroni correction was performed, although this was not the primary analysis for the trial
  4. d One trial performed a Bonferroni correction and one used Dunnett’s procedure