Background: The Women's Health Initiative (WHI) randomized trial of estrogen plus progestin (E + P) was terminated early based on an assessment of harms exceeding benefits for disease prevention. The results contravened prevailing wisdom and a large body of literature regarding benefits of menopausal hormone therapy. The results and their interpretation have been the subject of considerable debate.
Purpose/methods: To describe the process of developing a trial monitoring plan, the key interim and final data, and to explain the choice of statistical methods used in trial monitoring and reporting.
Results: A formalized monitoring plan was developed using statistical methods that acknowledged protocol-defined design and analysis plans, input of monitoring board members especially regarding the role of various study outcomes, and multiple comparisons. Major early departures from design assumptions concerning treatment effects indicated a need for additional flexibility in safety monitoring. When the trials were stopped early, questions arose as to how closely the statistical methods in published reports should correspond to those defined by protocol or used in monitoring.
Methods: were selected to provide a simple and transparent summary of the primary results, with a cautious interpretation promoted by acknowledgement of multiple testing.
Conclusions: Developing a formal trial monitoring plan with a view towards influencing clinical practice is useful for creating consensus among DSMB members regarding the evidence that would justify stopping a trial and the framework to be used to address statistical complexities. Departures from design assumptions typically occur. These reinforce the role of the DSMB in exercising their judgment, and the judicious adaptation of these statistical guidelines in monitoring and reporting trials. In communicating the results in such circumstances, priority should be given to presenting as fair, accurate and transparent a view of the data and findings as current methods and technology allow.