The increased attention to design and analysis of randomised clinical trials in small populations has triggered thinking regarding the processes leading to the most appropriate design for a particular clinical research question. In common diseases, this might not seem a pressing problem, given the extensive practice- and theory-based experience in trial designs. In the context of drug development for rare diseases, guidance from the European Medicines Agency [1] states that “in conditions with small and very small populations, less conventional and/or less commonly seen methodological approaches may be acceptable if they help to improve the interpretability of the study results”. However, this advice does not provide practical guidance on how such choices can be made at the clinical trial designing stage. Moreover, it also states that “[n]o methods exist that are relevant to small studies that are not also applicable to large studies” [1]. Hence, such practical guidance is actually relevant for all trials. Thus, there is arguably a need for a design framework, with that proposed by Parmar, Sydes and Morris [2] having particularly strong points. Their framework follows a logical order of the steps one would take in designing a trial, allowing for practical implementation. Secondly, it is driven by what could be termed a ‘step-down approach’: at each step following a non-feasible option, the next potential change or comprise to be considered is the one with minimal impact on the objective of obtaining high quality randomised evidence to improve care for the target patient population. It also appropriately addresses the fact that designing a clinical trial is a complex multidisciplinary and multifaceted exercise, not easily captured in a simple decision scheme.
Frameworks (or even algorithms) for applying particular designs for randomised clinical trials in small populations have been previously proposed, notably by Gupta et al. [3] and Cornu et al. [4], both of which are based on a literature search up to 2010, a particular choice of decision ‘nodes’ and considerations of the pros and cons of (less familiar) designs. The decision nodes are driven by the type of intervention [3], type of outcome versus recruitment time [3, 4], feasibility of sample size [3], prior knowledge and treatment alternatives [3], and certain design desirabilities [4]. Further, they address minimising time on placebo, and/or ensuring that all participants are on active treatment at the end of the trial [4]. In these frameworks, it is actually difficult to ascertain whether a particular choice is (in some sense) the best possible given the circumstances. Indeed, the focus of Parmar et al.’s [2] framework on the ‘best’ randomised evidence to improve care for patients makes the search for this ‘best possible’ far more explicit. We concur that this requires the time and adequate attention of the entire clinical research team. If application of a framework helps to concentrate the team in order to design the best possible trial, this is, in itself, a positive effect not to be underestimated, particularly for investigator-initiated trials.
Herein, the proposed framework and its application are further considered, focusing on the level of ambition for its practical application, a discussion on relaxing type I or II errors, and the methods through which a deeper understanding of novel trial designs may be obtained.