by Giuliana Miglierini
Single-arm clinical trials (SATs) are often the preferred study design in cases where target populations are very small, and it turns thus impossible to run a standard randomised clinical trial comprehensive of two patients’ arms. In single-arm trials, all recruited patients receive the experimental treatment, without the presence of a control arm. This means the observed effects of the investigational treatment are the sole source of evidence, without possibility to compare them with another available treatment or a placebo.
Rare diseases, including some rare cancers, are typically addressed by single-arm clinical trials. The so gained evidence is often submitted to regulatory authorities as pivotal evidence to demonstrate efficacy in a marketing authorisation application (MAAs). In order to improve single-arm study design and robustness of the collected data, the European Medicines Agency has published a Reflection paper, which is open to consultation until 30 September 2023. Comments will support the drafting of the final guideline, to be published in 2024.
The reflection paper represents the first guidance document produced at the international level to address issues and challenges typical of single-arm clinical trials. The document has been adopted by EMA’s CHMP committee with contributions from the CAT committee, the Methodology Working Party and the Oncology Working Party.
A choice that needs to be always justified
The choice to use SAT to produce the pivotal evidence to be submitted to regulatory authorities has to be always justified by sponsors, as it greatly deviates from the standard approach of double arms randomised clinical trials. The reflection paper suggests always seeking prior scientific advice to verify the acceptability of this type of study design with respect to the specific development programme.
EMA’s document focuses on the assessment of the efficacy of the treatment under study, but similar considerations are valid also to support the assessment of its safety; it can be applied also to SATs not intended to provide pivotal evidence. Trials in which only experimental arms are randomised, but without formal comparisons between the arms (i.e., platform trials) are also considered as SATs.
The direct consequence of the lack of randomisation is that SATs do not support a causal interpretation of the observed effect of the treatment. External data thus need to be used to estimate the average outcome for patients not being treated with the experimental drug.
The choice of the endpoints
The reflection paper addresses different items which should be considered in the planning and execution of a single-arm clinical trial, starting from the choice if the endpoints for the assessment of causal interpretation. According to ICH Q9, this should reflect the variable capable of providing the most clinically relevant and convincing evidence directly related to the primary objective of the trial.
The selection of the primary endpoint should take into consideration its validity, reliability and feasibility, as well was its ability to isolate the treatment effects. This is particularly critical, as SATs cannot exclude biases in the interpretation of results, and thus may lead to uncertainties in their effective acceptability. The reflection paper indicates the need to always discuss on clinical bases the possibility to accept a specific endpoint in a therapeutic area or in order to establish a clinically relevant treatment effect. Examples of some more common endpoints are also provided.
These include time-to-event endpoints (i.e., time to death, progression-free survival, etc.), which are considered not suited to be used in SATs as individual outcomes cannot often be attributed to the treatment. The determination of “time 0”, i.e., the starting point of being at risk for a specific endpoint, is also challenging, as well as the impact of the course of the disease on the selected endpoint.
Continuous endpoints measure changes the patient experience during the trial in comparison to baseline values. This type of endpoint may be affected by random fluctuation over time, systematic change due to the natural course of the disease or measurement errors, thus making it difficult to assign a causal attribution of treatment effect.
Binary endpoints refer to states of disease that cannot change without intervention, for example if life expectancy substantially exceeds that achievable without treatment. In such instances, binary endpoints are deemed acceptable to isolate treatment effect with sufficient certainty. Dichotomised endpoints similarly refer to pre-established thresholds, which are not possible to cross without treatment.
The patients population
A main challenge in single-arm clinical trials is represented by the limited number of patients available for recruitment. This reflects the external validity of study results, when transposed to the general population for routine treatment.
To this instance, assumptions made on the natural course of the disease are very important and must apply to the trial population, so to reflect the hypothetical control group. The reflection paper also mentions the possible bias deriving from selection mechanisms associated with prognosis and highlights the need to always specify and document the subject selection process to support the correct assessment of the benefit-risk balance.
Furthermore, the potential impact of unknown prognostic or predictive variables cannot be controlled in SATs, for example with reference to biomarker-selected populations or to the possible differentiation between subgroup heterogeneity.
External (extra-study) information is critical for the analysis and interpretation of SAT results; it should always be pre-specified in the study protocol, including the precise and a priori definition of control conditions.
The approach to statistical analysis
A specific chapter addresses the principles to be used for the statistical analysis of results. More particularly, if an SAT is intended to provide pivotal evidence, the level of the analysis should be the same as for confirmatory trials, according to ICH Q9.
It is thus important to define a priori a clear success criterion, to be justified on the basis of available external information. To this instance, the reflection paper underlines the high criticality of unplanned changes to the SAT study design once the experimentation had started, as amendments might be considered potentially data driven.
Methods to address multiplicity should also be pre-planned and adhered to in order to control the probability of false positive conclusions. Not less important is the predefinition of the primary analysis set, which should include all subjects that entered the SAT upon providing in-formed consent in order to avoid any bias. The reflection paper addresses the possible cases in which exclusion of some patients from the set is considered acceptable. The approach to avoid the occurrence of missing data is also described.
The pre-defined detailed statistical analysis plan should also be available before the start of the SAT, and it may be based both on the use of non-parametric or parametric statistical methods. The statistical analysis model used for estimation of the treatment effect should be fully pre-specified and justified. Critical points to be considered include the lack of calibration against a control and the distribution of the trial population in comparison to the target population. Sensitivity analyses should always be considered in order to support the robustness of the estimates resulting from the analysis of study results. The reflection paper also discusses how to approach the definition of the threshold values used to assess outcomes at the endpoint level.