FDA CDER statisticians are designing trials with adaptive features to make clinical evaluation of new drug treatments more efficient and informative.
A New Trial Design for a Medical Emergency
CDER statisticians and their collaborators at NIH and in West Africa faced the challenge of designing a trial to assess treatments for the deadly Ebola virus that could be conducted in a medical emergency. In their design, they used a Bayesian approach that include options for adaptations of the trial due to rapidly changing events on the ground and newly acquired information about the disease. Data at shown are from 2014 as compiled by the World Health Organization.
A few years ago, countries in West Africa faced the most serious outbreak of Ebola ever recorded, and at the time, no therapies had been shown to be effective for this disease. In response, CDER statisticians worked in collaboration with NIH, several academic centers, and investigators in Liberia, Sierra Leone, and Guinea to design a trial to evaluate treatments for this deadly viral disease.
A randomized trial is essential for collecting conclusive evidence about the relative benefits of treatments, but it is difficult to conduct trials in the context of an epidemic, in part because it would be unethical to withhold potentially effective treatments from patients. To meet this challenge, the Ebola trial randomly assigned patients to what was then the optimized standard of care (oSOC) or to the oSOC plus investigational treatments for which promising nonclinical evidence existed. To further address ethical concerns of withholding effective treatment from patients, the trial was designed so that if a treatment was proven efficacious, it could become the new standard of care for all trial participants. In addition, the investigators could add or drop new investigational treatments as information became available.
Adaptations to a trial based on accumulating data may increase the probability of committing a type I error, which is falsely concluding that a treatment is superior when it is not. This is because an adaptive design provides multiple scenarios under which the investigated treatments might appear more effective by chance alone. Under typical non-emergency conditions, such as during a large clinical trial at a university health center, it is possible to control type I error by using a study design with no adaptations, but this becomes much more problematic in a medical emergency where adaptations are needed to accommodate rapidly changing circumstances.
To address this challenge, the Ebola trial was designed as a Bayesian trial, which is an approach that readily allows for trial adaptations as new information accrues. In a Bayesian trial, an initial step is to quantitatively specify prior beliefs about the unknown quantity of interest based on previous experience. For example, one could assume that the probability of a patient dying within a specific period of time while on either the control or the experimental treatment varies uniformly from 0 to 1, which means all possible probabilities are equally likely. As data become available—in this case, when information is received about how many more patients on each treatment arm have died—the distribution of probabilities for each treatment can be updated to more precise estimates. This new information allows researchers to determine the probability that one treatment is superior to another.
CDER statisticians and collaborators carefully examined the statistical implications of the Bayesian trial design compared to other possible designs. They considered how readily trials might achieve a statistically significant result under various scenarios and how the type I error probability would be controlled. Careful analyses and comparisons to other designs indicated that the Bayesian design was scientifically sound and would be advantageous for the Ebola trial.
As the public health response gained ground, the epidemic declined and fewer new eligible cases of Ebola became available for the trial. This development made it difficult to reach definitive conclusions about the efficacy of any treatments studied in the trial. However, the adaptive trial design successfully illustrated the feasibility and desirability of innovative trial designs to meet the demands of a public health crisis.
CDER Statisticians are Promoting and Developing Other Innovative Trials that Meet the Challenges of Precision Medicine
Designing an efficient and informative trial during medical emergencies is only one of the many challenges in clinical trial design that CDER statisticians are addressing. As in disease outbreaks, flexibility and adaptive features are critical for evaluating new treatments for cancer and other diseases. In this area, CDER statisticians are helping to facilitate better use, understanding, and acceptance of adaptive designs and other innovative designs, which have had limited use to date.
One such design is a master protocol, which allows multiple treatments or multiple diseases or disease subgroups to be evaluated within one overarching protocol that addresses numerous questions. As cancers are subdivided based on genetic information, and the genetic basis of response to cancer treatment becomes better understood, it is increasingly evident that new treatments may only be effective in subgroups of patients with specific identifiable genetic profiles. CDER researchers are helping to develop methods to identify these subgroups and to make valid conclusions about their response to treatment in the context of a trial with adaptive features.
These examples show that CDER researchers are committed to fulfilling the goals of the 21st Century Cures Act of 2016, which directs FDA to “assist sponsors in incorporating complex adaptive and other novel trial designs into proposed clinical protocols and applications for new drugs and biological products in order to facilitate more efficient product development.” CDER is also developing guidance for sponsors on the use of these complex innovative designs, facilitating exchanges within the scientific community on issues pertaining to trial design and trial simulations, and providing statistical expertise to sponsors at critical stages of their drug development programs.