“Adaptive designs” in clinical research refer to study designs where certain features are left flexible on purpose, so they can be modified within pre-defined limits during the conduct of the trial. The decision on which modifications are selected is taken on the basis that data only becomes available during the study conduct. A most basic form of adaptive design is a dose escalation trial, where the dose level of a new and potentially risky treatment is selected in response to safety and tolerability observed in previously treated cohorts. However, a wide range of other adaptations is known (Pallmann 2018; Park Ford 2021). For instance,
In basket trials (Park Hsu2020), a common targeted intervention is used in different disease entities with similar molecular alterations (e.g. different tumour types), and recruitment in a certain disease type may be expanded or not, depending on the efficacy and safety observed in that subpopulation.
In an umbrella trial, one single disease is tested in several targeted interventions, and the population in the most successful intervention can be further expanded.
Dose selection: After an initial phase in several parallel treatment arms, each with a different dose level, the arm with the best benefit risk/ratio and a comparator arm are retained whereas recruitment in all the other arms is discontinued.
- Sample-size reassessments examine the effect of a therapeutic intervention in a controlled trial in an interim analysis. If the therapeutic effect is overwhelming, the trial can be ended prematurely for early success. If the therapeutic effect is disappointingly low, then the trial can be ended prematurely for futility. In all other cases, the sample size is optimized, so an adequate power is retained, cost is mitigated, and exposure to an excessive number of patients is avoided.
The statistical methodology to control a type I error under an adaptive design is more than 20 years old (Kieser 2000, Collignon 2018) and well established, so there is no restriction in employing an adaptive methodology in the confirmatory setting. In fact, Colignon et al. 2018 report in their survey on adaptive trials submitted to the EMA is predominantly in phase III trials, with phase II/IIb trials being clearly a minority.
Given the competitive landscape and the reduced life expectancy in many tumor entities, developers in oncology have an understandable interest in reducing development timelines. Some try to do so from the very first I/IIa design. Requests for proposal received in the CRO industry may comprise of a dose escalation for a novel compound as a monotherapy, followed by a dose escalation of the novel compound in combination with a known baseline therapy, followed by an umbrella phase with the new dose levels in multiple tumour types.
Even though timelines may appear tempting on paper, the following hurdles provide good reasons to go for simpler designs in many cases:
Changing too many variables at the same time will lead to too many possible interactions. When therapeutic outcomes are modelled, an interaction is a mathematical term that describes an influence on the outcome variable as a function of treatment exposure and the presence or absence of baseline features that influence one another e.g. in the event your outcome is the probability of treatment success in a bacterial infection, your baseline features is that the bacteria are beta-lactamase positive and that the beta-lactam antibiotic treatment was given in combination with the beta-lactamase inhibitor clavulanic acid. It is obvious that in such a setting, the interaction – clavulanic acid added to an infection with a beta-lactamase positive strain – will strongly increase the probability of success. This simplified example is easy to interpret because the underlying molecular mechanisms are well known. They will not however be fully known in a new compound just entering the clinical stage. Another example could be in an escalation phase, a patient in a low dose group with HPV- head and neck cancer shows a poor response, but a subsequent patient in a higher dose level with HPV+ shows a better response. Would this then be a result of the higher dose level being more effective or HPV+ head and neck squamous carcinoma being more responsive to treatment (Economopoulou2020)? The more baseline features are changed across the entire population the more difficult it will be to tell meaningful interactions from coincidental observations.
Ending a trial, freezing the database and compiling the study report will always allow a more in-depth evaluation of the study data than a pre-defined interim analysis. It will always be possible to take more results and observations into account in designing a completely new trial than in the late phase of an adaptive trial which in part still remains defined by the initial design plan.
In an early phase trial, not only the dose level and a disease entity (e.g tumor type) need to be optimised. Other variables like the dosing mode can play a big role. For instance, teriparatide is more effective in fighting osteoporosis when injected in the morning, as compared to the evening (Michalska 2012). The timing and type of endpoint need to be confirmed and sometimes adapted. As another example, Xgeva /denosumab, was not superior to zolendronate for overall survival in many tumor types, it is however superior in preventing bone metastases and other skeleton-related events (SRE). A developer commencing a clinical program with the standard endpoints of overall survival and progression-free survival will need the time to confirm the true therapeutic benefit that the new product confers and define a more suitable set of endpoints early enough in the program.
Logistic challenges should not be underestimated: in prolonged adaptive trials a sponsor’s study management team may need to be split into blinded and unblinded team members for an unblinded interim analysis; similarly the study drug and protocol-specific baseline medication may need to be provided for a longer time and at a higher dose than expected.
The bottom line is that adaptive designs can shine best in late phases, where relevant features such as the dosing mode and the timing and type of benefit are well known. A certain set of adaptive features, starting with the dose level, can also be chosen in early phase oncology, but other sources of variability should be introduced only sparingly. Despite the understandable desire to cut down development time, there are good scientific and logistic reasons in favour of a series of separate subsequent studies with only a few adaptive features. Phase 0 trials (Coloma 2013) may help increase the amount of molecular data available in a new treatment in which case a slightly higher level of adaptiveness can be handled.
Oncology and hematologic cancer are uniquely demanding areas of clinical research. In particular, early phase research in solid and blood cancers requires a CRO partner that is nimble and responsive to your needs. From consulting on study optimization to navigating the complexities of rare cancers to managing complex cohorts, our team is ready to help you. Our staff’s deep therapeutic expertise combined with the extensive capabilities of industry-leading clinical technology delivers an agile CRO capable of taking your study the distance.