In order for adaptive designs to reach their potential, it’s critical that knowledge is effectively dissemirnated within the medical research community – in particular detailed information about the operating and statistical characteristics of specific designs and insights as to their benefits and limitations.
Cytel recently announced the publication of an important article in the New England Journal of Medicine which takes a leap forward in promoting better understanding of adaptive designs particularly in a confirmatory setting. We'll discuss some of the highlights of the article in this blog.
The review article, Adaptive Designs for Clinical Trials (1) is co-authored by Cytel’s President and Co-Founder Cyrus Mehta and Deepak L. Bhatt, M.D., M.P.H of Brigham and Women’s Hospital Heart and Vascular Center and Harvard Medical School (D.L.B.) and Harvard School of Public Health (C.M.). It appeared in the July 7th edition of the New England Journal of Medicine and we highly recommend reading the article in full. The four adaptation types covered are:
- Seamless Phase 2-3 Design
- Sample Size Re-estimation
- Changing the Primary Endpoint
- Biomarker- Driven Adaptive Population Enrichment Designs
Using case studies, the authors highlight the statistical and operational considerations of the designs, as well as the benefits and limitations.
Below we summarize some key takeaways from each approach:
Seamless Phase 2-3 Design
The authors review the example of the INHANCE trial (Indacaterol to Help Achieve New COPD Treatment Excellence). This was an adaptive Phase 2-3 randomized trial of inhaled indacaterol, a once-daily long-acting beta2-agonist bronchodilator for the treatment of chronic obstructive pulmonary disease. The trial was designed with multiple treatment groups, with dose selection at the end of stage 1.
As this seamless design combined data from both phases for final analysis it required fewer patients and had a shorter overall duration. Potential weaknesses of this method include lack of sponsor involvement in dose selection at end of phase 2, and complexity of the final analysis. Careful planning upfront is required for success, including detailed dose-selection criteria, a communication plan for disseminating interim results and detailed simulations of the operating characteristics before the initiation of the trial. It is critical that clear and comprehensive decision rules covering anticipated contingencies are included in the DMC charter.
Sample size re-estimation- Champion Phoenix Trial
The CHAMPION PHOENIX trial was a double-blind, placebo-controlled trial in patients undergoing urgent or elective percutaneous coronary intervention (PCI) for coronary insufficiency. The patients were randomly assigned to receive a bolus and infusion of the intravenous antiplatelet agent cangrelor or a loading dose of the oral antiplatelet agent clopidogrel.. The design used a sample-size reestimation at the interim analysis when 70% of the patients had been enrolled using a three zone approach based on observed lower difference in relative risk – favourable( >21.2%) , unfavorable ( <13.6%), and promising ( >13.6%- <21.2%). In the promising zone there may be a benefit from increasing the sample size.
An advantage of this design was that sample size increased only after results were reviewed and categorized as promising by the data monitoring committee, whereupon there may be a gain in conditional power.
Changing Primary Endpoint- the EXAMINE trial
Anti-hyperglycemic agents must show they don’t carry excessive cardiovascular risks before they can gain regulatory approval. Due to low event rates, cardiovascular outcome trials typically require large sample sizes and long study durations, and The EXAMINE trial( Examination of Cardiovascular Outcomes with Alogliptin versus Standard of Care is one example, enrolling over 5000 patients with median follow up of 18 months.
Non inferiority was demonstrated with an upper boundary of the confidence interval of 1.16. With the upper boundary of the confidence interval of less than 1 the trial would have demonstrated superiority rather than just excluding the possibility of excessive CV risk.
The maximum number of targeted major adverse events was 650, with an interim analysis at 550 events at which the trial could be stopped and non-inferiority claimed with a P value below 0.001 for the between group comparison. The trial was also designed to give a second chance to claim superiority by allowing the trial to proceed to 650 events after the interim analysis if the probability of showing superiority by the end of the trial was >20%.
This approach reduced the risk associated with making an up-front commitment to a large superiority trial. In the event the trial was stopped at 550 events as the conditional power of claiming superiority was actually less than 20%. A non-inferiority claim was successfully filed.
Biomarker-Driven Adaptive Population Enrichment designs
The authors discuss the dilemma inherent in planning a phase 3 trial of a targeted therapy. Should enrolment be open to all patients regardless of biomarker status or should enrolment be restricted to a targeted subgroup based on understanding of the mechanism of action from early data? The outcomes from either approach could be undesirable- a large trial in a heterogeneous population could end up diluting treatment effect and causing the study to be underpowered. On the other hand, without certainty about lack of efficacy in the non-targeted subgroup, a potentially beneficial treatment could be denied to a segment of the population. The adaptive population enrichment design may help to tackle this issue by randomizing patients irrespective of their biomarker status but using an interim analysis to identify if there is there is a difference in the benefits between the biomarker positive and negative patients.
Cyrus Mehta commented of the article:
“There has been considerable misunderstanding and confusion among both clinicians and statisticians about the role of adaptive designs in clinical trials. This article attempts to explain through actual case studies, without technical jargon, the different types of adaptive designs for confirmatory clinical trials, and the opportunities that they create for more efficient allocation of patient resources. An interactive graphic demonstrates how statistical validity is maintained despite data-dependent increases in the sample size of an ongoing study.”
The article is available here to subscribers of the New England Journal of Medicine or upon request from the authors.