Focussing on secondary endpoints has already been covered, but what do trial investigators do if both the primary and secondary endpoints are unconvincing?
Data dredging may produce a statistically more convincing result. Data dredging is the process of analysing all the trial data looking for outcomes that are statistically significant; it produces post-hoc or tertiary endpoints.
These post-hoc endpoints must be treated with caution as the study was not set up to directly collect, examine and answer any questions relating to this composite data group. The statistically acceptable level for a result occurring by chance is 1 in 20. This means that if we conduct 20 post-hoc analyses one may have been statistically significant purely by chance.
For example, a post-hoc analysis of the ISIS-2 study1 found that aspirin therapy was associated with increased harms in patients born to Gemini or Libra star signs. Clearly this is nonsense and confounds our conclusions.
A more recent study that made use of post-hoc analysis was ASCOT-BPLA2. As already discussed the conclusions reached in this trial should be treated with caution.
- ISIS-2 Collaborative Group Randomized trial of IV streptokinase, oral aspirin, both, or neither among 17,187 cases of suspected acute myocardial infarction. Lancet 1988;2:349-360.
- Prevention of cardiovascular events with an antihypertensive regimen of amlodipine adding perindopril as required versus atenolol adding bendroflumethiazide as required, in the Anglo-Scandinavian Cardiac Outcomes Trial-Blood Pressure Lowering Arm (ASCOT-BPLA): a mulitcentre randomised controlled study. ASCOT Investigators Lancet 2005;366:895-906.