Keywords: Rare disease, Choice of Control arm
I will give an FDA statistical reviewer's perspective on some of the common statistical challenges arising in rare disease clinical trials, with a focus on the Neurologic setting. For example, the choice of primary efficacy measure may be uncertain. It can be challenging to balance sensitivity to change and yet retain clinical meaningfulness. Primary analyses need to account for non-ignorable events such as death which may not be captured by the primary endpoint.
Obviously, the rare disease setting often limits the sample size attainable in practice. Therefore, although there may be some room for regulatory flexibility in some circumstances, alternative designs such as adaptive and Bayesian methods may be desirable to improve efficiency while leveraging what is already known about the disease.
Natural history studies and historical controls are attractive to sponsors to substitute for concurrent controls when they exist. However, their applicability to confirmatory trials is often limited by possible selection biases and/or if the endpoint varies a lot by rater (e.g., patient reported outcomes rather than overall survival). Matching is still sometimes used to utilize natural history data, but, usually, it is not possible to predict clinically relevant duration outcomes with adequate accuracy or to eliminate the likely biases in order to avoid use of a concurrent control. Biomarkers and surrogates can be used to enhance efficiency of confirmatory trials (e.g., enrichment and/or variability reduction) but are often not available or not well established. More upfront disease research and understanding is often needed. Compiling and sharing data from completed trials can stimulate research and disease understanding needed for better drug development. For example, the PRO-ACT database of ALS trials has greatly stimulated research efforts in ALS.