This is the first known empirical study of clinical trial experts’ views on ethical issues in adaptive clinical trials. Previous normative work has debated the ethical construct of clinical trials, and how adaptive clinical trials represent areas where the ethical issues may change based on design features [13,16]. The major concerns raised in these describe a tension between collective ethics (trial validity versus efficiency) and individual ethics (exposure to a better treatment versus fairness of enrolling early versus late in a clinical trial.) A great deal of recent discussion in the clinical trials literature has focused on response adaptive randomization in two-arm trials; however this represents a fairly specific and relatively infrequently used type of ACT [17-22]. Our current investigation builds upon this understanding, and directly examined the opinions of vested stakeholders in the development of ACTs under a special grant from the NIH and FDA to accelerate new discoveries and translate findings into practice [27]. While there were some similar patterns of agreement and disagreement, there were substantive intra-group and inter-group variation. Given that all stakeholders— clinicians, biostatisticians, and others—must work together, understanding the anchor points and values of these groups relative to the potential ethical advantages and disadvantages of ACTS is important for the collaborative efforts needed to make these trials a reality.
Areas of agreement across stakeholders
Although textual data illustrated many similarities in the understanding of ethical issues of ACTs, the VAS scores demonstrate different anchor points among different stakeholder groups on the relative importance of the ethical advantages and disadvantages of ACTs. The constituent groups agreed that ACTs, including response-adaptive randomization and dropping futile arms, would have ethical advantages for patients. Use of ACTs can help avoid exposing some participants to ineffective treatments, thus offering a clear ethical advantage. For example, the constituent groups agreed that “killing bad drugs” sooner—that is, leveraging the strengths of ACTs to evaluate drugs that do not have potential and ending such trials early—provides ethical advantages. In addition, the ethical advantages of dropping a treatment as soon as possible, and stopping the trial as soon as possible when a treatment is found to be superior, merit highlighting. All stakeholders agreed that adaptations need to be prespecified, and that having a clear understanding of what is being changed or “adapted” is prerequisite for conducting a valid, and hence ethical, ACT.
Response-adaptive randomization
The point was raised that under response-adaptive randomization, individual participants enrolled early, before any adaptation, would not achieve any additional benefit (or any harm), compared with a trial that lacks response-adaptive randomization or has fixed-interval analysis. However, it is clear that a greater proportion of individuals over the course of a response-adaptive randomized trial would be expected to receive the effective treatment if there ultimately was a more effective treatment being tested. In short, at the population level, participation in an ACT confers an ethical advantage over the course of the trial.
Unintentional unblinding and biased enrollment
The participant groups identified potential risks and hence potential ethical disadvantages of ACTs. For example, if information about the results of interim analyses were leaked or inferred by clinicians involved in recruiting potential participants, bias could be introduced if the clinician then chose to enroll or not enroll patients on the basis of the information being leaked. For example, in a trial with RAR that is searching for the optimal duration of hypothermia exposure in spinal cord injury, there would be concern for investigator bias if clinician investigators noted that patients at their clinical center/site were being allocated to longer durations of hypothermia. Such risks are greatest if clinicians know what treatment is being assigned, e.g., in an unblinded trial like the spinal cord trial, but also enrollment at their site is relatively high since they won’t necessarily know about enrollment in other sites.
The clinician stakeholders in this study primarily work on multicenter trials in emergency settings where discerning patterns of treatment assignment to different arms of a trial seems unlikely because the actual number of subjects in a given site will be relatively small, even if the trial is unblinded. The concern may be real in open-label trials in which the number of patients in one or more participating centers is large.
Informed consent
Participants disagreed on whether informed consent would be more complex in ACTs. Existing data suggest that when current informed consent procedures are used, participants consistently fail to grasp the difference between research and treatment, even when there is fixed 1:1 randomization [10]. Thus, the addition of further complexity in an ACT featuring response-adaptive randomization of participants to what appears to be the more effective arm or explaining fixed-interim analysis could appear to increase the complexity of informed consent decisions [31]. Discrepancies in views about this may reflect different value judgments about a “forest versus trees” understanding of the study. If a researcher believes that potential participants must understand the details of the allocation procedure (e.g., response-adaptive randomization) or interim analyses, then the informed consent process could be more difficult. If, alternatively, a “forest” (“big picture”) approach to explanation is followed, then the informed consent process may not provide significantly greater challenges. Among the authors with the most experience in trials using adaptive designs (DB, RL), informed consent has not felt to be substantively more challenging with ACT studies, compared with non-adaptive trials. Future research could explore empirically the difficulty of achieving informed consent in trials that do and do not use adaptive designs. The concept of individual fairness (enrolling early versus late in a ACT), was not substantially explored by this group of stakeholders. Since the ADAPT-IT project focused on neurological emergencies (stroke, spinal cord injury, brain injury following cardiac arrest etc.) and others had substantial oncology experience (cancer patients are unlikely to delay treatment to get a better chance in an ACT) this general concept was not frequently discussed as it is not possible to delay the occurrence of one’s cardiac arrest so as to have a better chance at getting a new treatment in a clinical trial.
Respondent group variations
Some considerations may help explain the differences between the two biostatistician groups. First, the consultant biostatisticians may have been focusing explicitly on response-adaptive randomization and frequent interim analyses in the context of clear stopping rules, as these elements of high-quality trials are the primary focus of their work. In contrast, the academic biostatisticians may have been considering a broader variety of adaptations and other trial conduct aspects such as maintaining the blind, protocol violations, and implementation of the adaptations. Concerns about the potential biases introduced by unfamiliar types of adaptations may have led academic biostatisticians to generally be more cautious, whereas the consultant groups have more frequently simulated, designed, implemented and analyzed such designs and therefore have less concern about certain forms of potential statistical biases given past experience.
Second, the perspectives on ethical advantages/disadvantages may be highly correlated with experiences in the design and conduct of the more complex adaptive designs. As stated in the FDA guidance, the more complex designs are less understood which may impact the assessment of ethical perspectives. Professional groups, consultant biostatisticians, academic clinicians, academic biostatisticians, and other stakeholders, alike, will make ethical judgments based on their own experiences and professional lenses. For example, consultant biostatisticians bring to the table an approach relying heavily on Bayesian modeling and decision-making. The academic biostatisticians’ views may reflect their experience in the logistical and operational aspects of trials—for example, the complexity of informed consent, protocol deviations and risks for operator bias.
Study limitations
This study involved a relatively small, but experienced group of clinical trial experts actively developing rigorous clinical trials, although importantly many had little experience with ACT designs. Although this research elicited opinions from various stakeholders regarding the performance of ACT designs, caution should be heeded when extrapolating these opinions to others of the same constituencies or assuming their representativeness in the broader biomedical community. We did not survey institutional review board members, as the ethics review of trials typically occurs after the design phase we focused on, although this would be an interesting area for further study. In addition, medical ethicists would have been able to provide a broader consideration of the ethical issues involved in ACT planning and implementation and would be important to include in future studies. In retrospect, it may have been beneficial for the respondents to consider the ethical aspects of ACTs within the full Emanuel framework (value, scientific validity, fair subject selection, favorable benefit-risk ratio, independent review, informed consent, and respect for the research subjects) [32]. Designers of ACTs may do well to consider these seven elements when comparing a proposed ACT to a more traditional fixed trial in the planning process. In addition, greater involvement of patients or patient advocates is an important part of the trial process and should be included early in the planning of any type of clinical trial [33]. Finally, the expressed opinions on a range of ethical issues come from four constituent groups with widely variable experience and familiarity with ACTs.