A well-timed interim analysis can generally supply added benefits to the operational and administrative aspects of a clinical trial. Particularly when clean data and safety adjudications provide information that is instrumental for stopping a trial early or informing sensitive interim decisions, the timing of an interim look may play a crucial role in leveraging all of the flexibility that a complex, adaptive design has to offer.
Kye Gilder of Orexigen Therapeutics recently spoke on this topic at Statistical Innovations in Clinical Development, an event sponsored by Cytel and the American Statistical Association. During his presentation, Kye discussed 3 strategies used at Orexigen to plan a recent cardiovascular outcome trial (CVOTs):
(1) East® designs
(3) The Cook Method
While each method can be used independently, Kye’s talk also highlighted the benefits of combining these methods, given their various strengths. Kye also added a dainty fourth strategy for the lower budget - using rulers bought at Target to perform linear regressions. (This method, which Kye calls Target 1.99 for rulers purchased at Target, is NOT recommended.)
The Case Study
CVOTs are time to event studies with MACE endpoints. MACE, which stands for Major Adverse Cardiovascular Events, refers to a composite endpoint consisting of death, myocardial infarction and stroke. Such trials have particular need for adaptive designs as they typically enroll thousands of patients, and are sometimes conducted as a part of post-market studies. However, the Orexigen CVOT was designed to be a combined pre- and post-market study in which the results of the first interim analysis were used for NDA submission, followed by three interim looks for post-market studies. The trial randomized over 9,000 patients.
East: Kye’s team at Orexigen used Cytel’s East® for the flexibility of comparing trial design options with multiple interim looks. In addition to using East® ADAPT to design adaptive trials, East® PREDICT can be used to forecast timely interim looks once data has already accumulated.
Simulations: Complementing the design in East with simulations in R, the team at Orexigen developed numerous simulations to model the patient’s on-treatment duration and the MACE hazard ratio between treatment groups. Due to both study design characteristics and on-treatment patterns, differential MACE hazard ratios also modeled.
The Cook Method: The basic idea behind the Cook method is to be able to predict the number of MACE events in the future based on data regarding MACE events up to a given time, plus suspected MACE (i.e. sMACE events), and MACE adjudication metrics.