Ethics of Adaptive Designs for Randomized Controlled Trials

Abstract

Over recent decades, adaptive trial designs have been used increasingly, accompanied by debates about whether such trials offer ethical and methodological advantages over traditional randomized controlled trials (RCTs). This study examined how methods and ethics experts believe adaptive trials affect the ethical character of clinical research, compared to traditional RCTs. We conducted in-depth interviews with 17 researchers from bioethics, epidemiology, biostatistics and/or medical backgrounds. While about half believed adaptive trials are more complex and may thus threaten autonomy, these respondents also believed this challenge is not insurmountable. Most respondents felt efficiency and potential for participant benefit were the main justifications for adaptive trials. There was tension on whether adaptive randomization in response to increasing information disrupts clinical equipoise, with some insisting that uncertainty still exists and therefore clinical equipoise is not disrupted. These findings suggest that further discussion is needed to increase the awareness and utility of these study designs.

INTRODUCTION

Traditional ways of conducting RCTs have generally involved a strict adherence to a 50:50 allocation scheme, use of an unbiased control group with the option to stop the trial early due to futility, danger to participants, or if one of the interventions shows superiority.¹ However, RCTs have been criticized for being inefficient, expensive, impractical in emergencies and other situations, and presenting some ethical tensions when large numbers of people are exposed to a potentially inferior treatment through randomization,² particularly as the trial continues into its later stages. Adaptive clinical trials, which allow planned modifications to the clinical trial based on accumulating data, have been used as an alternative to traditional clinical trials. This is because they are considered to be more efficient, more practical and more ethically acceptable option when compared with traditional or non-adaptive clinical trials.

There has been an increase in the use of adaptive clinical trial designs around the world. Recently, the US congress directed the FDA to facilitate increased use of adaptive trials designs in order to speed up bringing of new drugs to market,³ while others have recommended adaptive trials because of a view that they are more ethically acceptable in emergency situations such as the recent Ebola outbreaks in West Africa and the Democratic Republic of the Congo (DRC)⁴ where concerns were raised about having a placebo arm for the duration of a potential trial. Because of their flexibility, others have also recommended the use of adaptive designs as being more suited to learning health systems,⁵ which offer ways of testing new interventions closer to natural conditions. However, debates exist about whether adaptive trials relieve the ethical tension in RCTS or create new ethics concerns; this debate is still ongoing.

It is generally believed that adaptive trials, under certain conditions,⁶ reduce study duration and eliminate delays between study phases.⁷ In addition, adaptive trials are said to decrease the risks of a trial to participants as the seemingly better intervention is increasingly likely to be provided to participants as the allocation ratios change.⁸ However, it has been argued by some that adaptive trial designs violate ethical principles such as equipoise and justice,⁹ when compared to traditional randomised trials. It has also been argued that adaptive trials, by virtue of their complexity, may violate patient autonomy because it may be more difficult to achieve informed consent.¹⁰ While some proponents of adaptive trial designs claim that the designs provide more benefits to study participants by exposing them to a potentially more effective arm,¹¹ others argue that in fact this may not be a benefit for the participants, given that very few clinical trials end up being successful.¹² This study was conducted to gather the perceptions of researchers and ethicists on the challenges and opportunities presented by adaptive trial designs, their views on the ways in which the trials affect validity, efficiency, risks, benefits, autonomy, and fairness and to offer some insights into how to resolve some identified challenges.

METHODS

The study participants were researchers involved in research in adaptive clinical trials and were recruited from academic institutions in the United States (US), Canada, Asia and Europe, as well as US Federal government agencies. Participants were interviewed either in person or via Zoom. Participants were included in the study if they had conducted an adaptive trial, had published a paper on adaptive designs or had served on an ethics committee where adaptive trials were reviewed. Snowball sampling method was used to identify additional participants, and interviews were conducted using an interview guide (supplementary material 1) with open-ended questions. The interview guide consisted of general statements to guide the discussion, including participants’ interest and attitude towards adaptive designs and ethical issues such as respect for persons, equipoise and justice. Interviews typically took 30–60 minutes; all interviews were recorded. After the interviews, the recordings were transcribed, checked for errors and then analysed to identify codes, nodes and finally themes through reading and re-reading of the transcripts, and the themes were refined through independent reading by a second person. This process was informed by the literature and the emerging patterns in the data. The themes in this report have been reported through the use of summaries of themes as well as verbatim quotations, edited for grammar (supplementary material 2). This study received ethical approval from both the Johns Hopkins Bloomberg School of Public Health Institutional Review Board and the University of Zambia Biomedical Research Ethics Committee.

RESULTS

We conducted in-depth interviews with 17 researchers, who came from different backgrounds including bioethics (5), epidemiology (6), biostatistics (8) and medicine (4) (Some respondents had expertise in more than one field). All were involved in conducting or studying adaptive clinical trials at the time of the interview. The respondents were from different types of institutions, such as international non-governmental organizations, universities and government agencies. Themes that emerged in interviews are grouped into key topic areas.

RESPECT FOR PERSONS

Complexity of trials may negatively affect the autonomy of participants

There was a general agreement that adaptive trials are more complex than ordinary clinical trials. Eight of the 17 respondents generally felt that this complexity can constitute a threat to the autonomy of the participants that need to be addressed. This was expressed by several respondents, one of whom is quoted in the following excerpt:

R2: I think that adaptive trials have a greater potential for confusion regarding informed consent due to their complexity and additional decision points. While they may seem attractive on a superficial level, there are also risks that need to be discussed with both patients and the scientific community working on the study.

One of the respondents felt that the complexity as well as the fact that the trial is tailored to the patient’s disease may cause some participants to enter the trial with some sort of therapeutic misconception, thinking that they are more likely to get a positive outcome, when in actual fact such outcomes—even in adaptive trials, this respondent believes, are relatively rare or similar to those of fixed randomization trials.

Another respondent expressed discomfort that the changes that are likely to be made in the course of an adaptive trial may make it difficult for the participants to be fully informed about the trial, as described here:

R9: I suppose that’s useful, but it doesn’t address the real issue. For example, if I’m involved in a trial and the treatment or dosage is changed without my knowledge. Maybe I was warned that this could happen when I enrolled and gave my consent. But does that exonerate the researchers? Have they fulfilled their commitment to the study participants by informing them upfront? It’s like agreeing to terms and conditions on a website before using it. It may make me a little more comfortable, but not much.

There was also an observation that it was important to strike a balance between making trials more complex while still having them be understandable:

R9: my position would be that one needs to basically try to maintain a balance between making trials more and more complex by adding new and more complicated features versus trying to add as many features as possible without making the trial design totally incomprehensible to anybody but an expert.

Although adaptive trials are complex, the threat to autonomy is not insurmountable

There was a general consensus among these respondents (8/17) that the complexity of adaptive trial designs was not necessarily an insurmountable problem as there were ways to address this through communication.

R6: Find a way to explain the study design in layman’s terms to minimize the risk of misunderstanding. I think you would have to find a good way to communicate. I don’t think it’s impossible. Most people would understand if you said something like, ‘If it looks like it’s not working, we might stop early and we have some things on standby that we might try if we have a good reason. You have to trust us to make the right decision.’”

Some felt that the lack of understanding might also arise in traditional trials, and may be no worse in adaptive trials. For example, one of the respondents argued that even a fixed randomization trial can sometimes be complex, while another said that researchers already do not do a good job explaining trials, adding that participants usually do not fully understand when they consent.

R8: … Adaptations may make the trial more mysterious and harder to understand, which could make participants uncomfortable. I don’t think it’s any different from a fixed design that can be complicated and lead to confusion. An adaptive design may be more complicated than a non-adaptive design, it may make it a little harder but it’s not impossible to overcome in terms of explaining it and helping people to understand it.

For some respondents, it was felt that there may be room for developing standards for disclosing information to participants as the trial continued to accrue participants.

Trial participants do not need to understand everything for consent to be valid

Two of the participants felt that although the complexity of adaptive trials may make it difficult for patients to understand them, this does not invalidate the consent that patients give before a trial. The following excerpts illustrate this point.

R15: I like to look at the whole study as the object of the intervention, not the issue of what the probability of randomization is going to be. It can get very complicated to even discuss whether or not to tell participants what the randomization probability is. I don’t think you should. Instead, you should say that it’s a study that will try to treat as many people as possible in the right way given our current knowledge and leave it at that. I don’t believe that’s a violation of patient autonomy because they don’t know all the logistics used in any trial to make it successful.

R16: if the goal of informed consent is to ensure that participants get a full understanding of what they are consenting to and what is going on, then we should just shut down the entire research enterprise. Because one thing we know empirically is that this is not what we’re doing and not what happens, and people don’t have a full understanding even under optimal circumstances. I don’t think that’s what informed consent is about.

Adaptive trials may serve to enhance autonomy

One respondent had a different view, positing that by not ignoring developing information, adaptive trials may in fact enhance autonomy because patients will be informed as the trial progresses and the adaptations are being made.

R13: So, I think the idea is that an adaptive trial would try to enhance autonomy because it allows for learning and evolution based on earlier participants. The autonomy of each person is upheld because we’re not ignoring growing information about a treatment when considering whether or not to enroll. So, I think that’s one way that adaptive trials are supposed to enhance autonomy.

BENEFICENCE

Generally, several themes emerged under the broad heading of beneficence. These ranged from those related to whether adaptive trials were more efficient to whether adaptive trials allowed participants a greater chance of individual benefit. Each theme will be discussed further below.

Adaptive trials are more efficient than fixed randomized trials

The majority of respondents (11/17) felt that adaptive trials provide the researcher a chance to get to the answer more quickly than in an ordinary fixed randomization trial. It was also felt that this efficiency is particularly helpful when emerging data suggest that arms should be dropped or when a trial has to be stopped early because of accumulating evidence.

R12: Adaptive trials have the potential to increase efficiency, make decisions earlier with fewer patients and resources, and accelerate the drug development timeline for the benefit of companies and patients. They also have the potential to treat patients in a more personalized manner, for example by randomizing with a higher probability to a drug that seems to be working better in their particular cohort, as in response-adaptive randomization designs. Additionally, they have the general ability to remain flexible enough to answer scientific questions of interest in a setting where a traditional design would be inefficient or impractical.

Adaptive trials may recruit more people than ordinary trials

There were four respondents who felt that it was not necessarily true that adaptive trials recruit fewer participants than ordinary trials. In actual fact, they stated that where adaptive trials had only two arms, they were in fact less efficient than ordinary trials and would have the effect of exposing more participants to unproven therapy or harm.

R17: …. You might want to say that this is not an ethical issue, but I think it is. The efficiency of the trial, how long it takes, and how many participants are involved are all important factors. Using an adaptive design means that you may need to enroll 30% more participants to achieve the same level of statistical confidence. Would it be better to do a shorter trial and get the answer more quickly?

Adaptive trials allow fewer people to get ineffective or unsafe interventions

Three out of 17 participants felt that one advantage of adaptive trials is that they offer the chance to participants to get the less harmful intervention compared to fixed randomisation trials. Other terms used instead of “less harmful intervention” were “beneficial” and “favourable”. By adapting and dropping arms as the trial progresses, the chances of getting the effective or less harmful arm was higher.

R6: The benefit of an adaptive design for participants is that if there is an underperforming arm, it can be dropped, and trial participants are less likely to be enrolled in an arm that is ineffective.

However, one respondent reported that he wasn’t particularly worried about patients being exposed to a therapy that doesn’t work unless it was actually toxic.

R15: All these factors go into the mix, but if there’s a therapy that doesn’t work, I believe the ethical pressure is a lot less. Yes, more people are exposed to a therapy that doesn’t work, but unless it actually kills or is overly toxic, I don’t get as worked up about a situation where more people are in the trial and nothing is working. That’s the same as the situation outside the trial. So, my main concern is for situations where the therapy does work, not where it doesn’t, unless you can convince me otherwise. If it doesn’t work and isn’t overly toxic, it doesn’t matter whether you’re in or out of the trial.

Therapeutic misconception, that any adaptive experiment is necessarily better or more ethical than nothing

While some respondents agreed that adaptive trials increase the chance that individual participants get a more favourable intervention, two out of 17 respondents disagreed with this notion. This minority subset did not share the view that adaptive trials are generally more ethical than ordinary clinical trials. Another respondent stated that there was a therapeutic misconception arising from the notion that any experimental therapy is better than no experimental therapy, and it was just a matter of figuring out which therapy was better than nothing.

R10: I’ve seen statisticians give talks where they say that adaptive trials are more ethical because more people end up in a better-performing arm. Maybe in some circumstances, it makes sense. But we have to be careful about trying to sell trials as therapy. There are contexts like clinical equipoise where I think it’s reasonable to say things like that but then again, we have to be careful about how we communicate what exactly we’re doing in trials. … I’m skeptical that they are more ethical, but I can see why there’s interest in them because of that perception that is out there.

Concerns about adapting before there is sufficient evidence

Three of the respondents raised concerns about the possibility that adaptations may be done without sufficient evidence, potentially exposing participants to harm. It was also pointed out that with appropriate expertise, this may be resolved.

R14: I would rather say that if decisions about adaptations are made by someone who is smart enough not to make changes to the design based on insufficient data, then I don’t see how we could harm participants. We need to think more about this, except if the design is changed in the wrong way. But this would happen if you have people who are not skilled enough and make changes when it’s not relevant.

It was also pointed out that having an arm that is doing better may not necessarily be equivalent to a meaningful clinical benefit for the participant.

Possible to game the trial

One of the respondents had concerns about the possibility that some participants may disadvantage other participants who are in a more desperate situation by waiting until the trial is near the end to join, giving themselves a higher chance of getting in a more favourable arm.

R10: … There is a basic concern in adaptive trials that it could potentially disadvantage patients who are in a more desperate situation. If you’re self-interested and thinking about enrolling in an adaptive trial, your first question should be how close the study is to finishing. And you basically want to be the last patient who enrolled if what you’re after is benefit.

Complexity of adaptive designs may affect interpretation

Two participants felt that the complexity of adaptive trials may make them inefficient, as they require a certain level of expertise and special statistical methods, making them harder to sell to other investigators and regulatory authorities.

R12: Adaptive trials usually require much more intensive development of design features and operating characteristics, such as power and type I error. They may only be useful in a limited setting and may not do what they’re supposed to do. The results of the trial may even show that an adaptive trial and all the work that went into setting it up really wasn’t necessary. They can also be a harder sell to investigators, regulatory authorities, and other stakeholders who are more accustomed to thinking about traditional designs. In some cases, they may actually perform worse than traditional designs or preclude us from answering questions that could have been answered by a traditional design.

EQUIPOISE

Two opposing themes emerged concerning equipoise among respondents. One was that during an adaptive trial, the moment allocation probabilities are changed, the state of clinical equipoise with which a trial normally is started is disrupted, and this raises ethical issues. Conversely, other respondents felt that even after changing allocation probabilities, you still have uncertainty and therefore equipoise is not disrupted. Continuing means you are still trying to resolve the uncertainty. We now present the results of these themes in turn.

Equipoise is disrupted after changing trial allocation probabilities

Five respondents felt that equipoise is disrupted by the moment you decide to change allocation probabilities. One respondent felt that this is especially so when the study is un-blinded, because the accumulating data may alter the state of equipoise for the investigators.

R1: … I was giving an example where, if you’re adapting by changing the allocation ratio in a non-blind trial, you need to be very conscious that you may be introducing… investigators need to be aware that this might alter the state of equipoise.

Two respondents stated that the accumulating data itself causes equipoise to be disrupted, for example when you drop an arm during an adaptive trial.

R6: I’m going to assume, and maybe this is a bad assumption, that if we go into a trial with an understanding that there’s equipoise at the outset, before any adaptations are made, that equipoise may be disrupted by the information gained early on. This is why underperforming arms are thrown overboard. If you lose that position of equipoise, you take action and replace the underperforming arms with ones for which there is a similar amount of uncertainty.

R12: Most response-adaptive designs start out with that total uncertainty about that relative efficacy or safety that is, with equal randomization. Once evidence emerges that for a given patient, one arm is showing more benefit than another, it is true that equipoise is disrupted. The response adaptive randomisation algorithm attempts to address this by assigning a patient to the “better” arm with higher probability, which seems ethically attractive, but may not always function as intended. On the other hand, traditional RCTs simply ignore or blind themselves to accruing data that might reach the same conclusion regarding relative outcomes, or choose not to act on it by changing the randomization probabilities, and in doing these things, appear to maintain equipoise. I think it is an open discussion what the ethical implications are here, and whether they are truly features of the designs or of the fact that we are or are not looking at existing data during the trial in that way.

One respondent, while agreeing with the idea that accumulating data disrupts equipoise, raised a question about when one would be satisfied that you have sufficient information to stop the trial:

R17: If equipoise means we haven’t yet met the standard of evidence to close the trial and make a decision about which treatment to use for future patients, then we can deviate from 50/50 pretty far before reaching that point. The question is: what’s your breaking point for saying we shouldn’t continue the trial because we’re already getting a pretty good sense of the answer? People have different values and views about that.

In contrast, one respondent held the view that equipoise does not hold in a situation involving a highly lethal disease:

R11: This is to say that those who represent equipoise as a balanced system ignore or see an ideal world where all the “experts” can be gathered to have a say. When fatality is very high, it’s extremely difficult to compute the potential risk-benefits of a given intervention.

Adapting the trial does not mean equipoise is disrupted

Five respondents felt that adapting a trial does not mean that uncertainty no longer exists. Continuing the trial after an adaptation is still a way of resolving the uncertainty. One respondent gave the following example:

R10: When we start the study, we have sufficient uncertainty that all of these arms (A, B, C, and D) can be considered rivals for each other. As the study progresses, that starts to look less and less likely in theory. Accordingly, there would be fewer experts willing to assign patients to arms B, C, and D.

There was also a realization by the respondents that in an adaptive randomization trial as the adaptations continue, eventually the researcher is faced with the question about what would be an acceptable ratio at which most experts would agree that continuing the allocations would no longer be acceptable or ethical.

R10: The challenging part comes when we think about the threshold where we say it’s no longer acceptable to randomize people. What exactly is that threshold?

This respondent went further to look at this from the participant’s perspective, stating that when it is obvious that one arm is doing better than the others all the people would choose that arm, when given a choice.

R10: … But are those people who are being allocated to arms B, C, and D going to be happy with this? Surely if we gave them the choice, they would all pick A. So why is it okay? I think that’s really interesting. It forces those of us who are in favor of equipoise to take seriously the idea that we still don’t know if nobody should be getting B, C, or D. Even though A looks much better in the context of this study, if we set our thresholds right and design the study correctly, we’re not harming people by giving them B, C, or D. It’s not inconsistent with competent care to do that because we haven’t yet answered the question to the level of certainty that we agreed on from the onset.

To address issues of justice that may arise as the adaptations continue, one respondent countered that this is because there is an assumption that continuing to adapt means someone is being denied something.

R3: I would say the adaptation itself doesn’t raise problems of justice. What raises problems of justice is whether you’re attending to the underlying uncertainty and ensuring that you’re not allocating people to forms of care that no one would recommend for them. I think those who object to continuing adaptations believe that if you’re adapting, you’re depriving someone of something. That presumes there’s a single agent whose uncertainty you care about. But maybe there is no such single agent; there’s just a diversity of opinions from lots of experts who may disagree with each other. The point of the study is to resolve that disagreement.

Another respondent made the point that adaptation is a way of maintaining equipoise, until the trial is done:

R13: … Even if you’re changing the randomization frequency or adding another arm at some point down the road, you’re still doing that with a general sense of equipoise about the outcome. So, I think that equipoise still has to be a guiding ethical principle around adaptive trials. Given the changes to the trial and the complexities of adaptive trials, you need to be extra certain that there are still contributions being made to a general feeling of equipoise until the trial is complete.

Another respondent felt that there is a tendency to assume that equipoise is only when the uncertainty was 50:50. This raises a question about how people define equipoise and how they interpret its requirements.

R14: Equipoise is a complex concept. For me, it requires that we have uncertainty. Whenever we’re sure that one option is better than the other, it’s not equipoise and we need to give the better option to patients. But when we’re uncertain and need more evidence to get out of the uncertainty zone, we’re in equipoise. So equipoise doesn’t mean 50/50.

Discussion

There has been increased interest in the use of adaptive trial designs. With it has arisen a debate about the ethical acceptability of some aspects of these designs. This study set out to explore these ethical issues by collecting the views of bioethicists and those involved in conducting these trials. Individuals interviewed in this study generally agreed that adaptive clinical trials are more complex than ordinary clinical trials, and about half believed that this complexity challenged how well participants would understand studies through informed consent, thus constituting a potential threat to the autonomy of participants. Several of these same participants, however, believed that this threat to autonomy is not insurmountable and that both communication can be improved and simplified. Several participants also stated that participants need not understand everything about the study for consent to be valid, noting that understanding in traditional trials is often incomplete. Respondents generally agreed that adaptive trials are more efficient because they provide the researcher with a chance to get to the answer more quickly and with fewer participants than in an ordinary fixed randomization trial. However, a few others felt that efficiency only applied to studies with more than two arms, because, according to at least one respondent, two-arm adaptive trials were believed to be less efficient than ordinary trials and would have the effect of exposing more participants to unproven therapy or harm. There were also concerns about adaptive trials leading to therapeutic misconception, and the possibility of adaptations being done without sufficient evidence, being difficult to interpret and harder to sell to other investigators and regulators. Ethical concerns were also raised about the disruption of clinical equipoise after changing allocation probabilities in response adaptive randomization, but others insisted that despite changing the probabilities, uncertainty still exists and therefore clinical equipoise is not disrupted. We will now discuss these issues in turn.

Respect for persons

A key theme raised by respondents was whether or not adaptive trials threaten the ability to get valid informed consent due to their complexity. It has been argued that adaptive trials, by virtue of their complexity, may violate patient autonomy because it may be more difficult to achieve informed consent.¹³ While a number of respondents agreed with this position, others asserted that trial complexity is not unique to adaptive clinical trials, but also exists in fixed randomization trials. Respondents generally agreed that there are often ways to overcome this threat to autonomy, including through better and more engaging consenting procedures. Furthermore, some respondents felt that participants need not understand everything in the study for consent to be valid. On the contrary, one participant felt that compared to many fixed randomization trials, adaptive trials do not ignore emerging information from the trial, and thus the adaptations are one way of enhancing the autonomy because participants are better informed and respected at every stage of the trial.

It is accepted in research ethics that participants should understand key elements of a study’s purpose, procedures, risks and benefits when they choose to participate in any trial or research project. This stems from the principle of respect for persons, and is a way of ensuring that participants with agency have sufficient information to make choices that are consistent with their wellbeing, values, and rights. Critics of adaptive designs argue that the complexity of these studies make it difficult for participants and members of ethics committees to understand the trials, making it difficult to have a robust ethical review process¹⁴ as well as genuine informed consent.¹⁵ As an example, it is argued that a concept such as randomization in an ordinary clinical trial is not easily understood by a significant portion of research participants,¹⁶ and adaptive randomization, which is much more complex, makes it more challenging.¹⁷ Some of our respondents agreed with this position. However, arguments about the complexity of adaptive trials threatening the autonomy of participants assume that most research participants would still not understand the adaptive trial if sufficient time and effort was given to the consenting process. We posit that this is not correct, and that given sufficient time, sufficient understanding can be achieved. Informed consent should ideally be an iterative process, and for complex designs investigators should make use of simple strategies aimed at improving understanding of consent. Numerous studies have shown ways in which study explanations can be made simpler to improve understanding, and include the use of multimedia platforms, animated videos, slideshows with voice overs, simplification of consent documents (such as lower reading level, easier phrasing and styling), allocating more time for discussions and having participants take a quiz to assess understanding and feedback.¹⁸ These and other interventions, either singly or in combination with others have been shown to improve understanding. It is therefore our considered view that the complexity of adaptive clinical trials is not a barrier to achieving valid informed consent.

The view, voiced by some respondents, that understanding of the adaptive trial need not be absolute or entirely comprehensive for informed consent to be valid raises a question, often raised in consent for standard trials, about the extent to which participants must grasp elements of consent before it should be considered valid. Both Wendler¹⁹ and Wendler and Grady²⁰ have argued that while core elements of informed consent are essential and must be understood by participants before they can take part in a study, a full understanding of randomization itself is not necessary, as one only needs to understand that allocation to a study arm will not be based on investigator judgment about what is individually best for a patient. It is also generally agreed that during the consenting process, comprehension of study procedures is an ethical imperative, but there is continuing debate about to what extent this comprehension should go, and at what threshold enrolment of a given participant should proceed or not proceed.²¹ Clearly, the question of how much information to disclose to participants is an ethical question, and involves making value judgments about what is important from the perspective of individual researchers and participants, sometimes balancing a forest versus trees comparison.²² While some researchers and participants may favor providing (or receiving) a detailed rendering of the study information (the trees), others may be satisfied with – or even achieve higher overall understanding -- getting the big picture (the forest).

For instance, for blinded trials it has been suggested that it would still be ethical to withhold information on their actual chance of being randomized to each arm, as long as such information about varied randomization probabilities as data accumulates is discussed during the consenting process.²³ However, disclosing the actual chance of randomization at the time of recruitment for many participants might serve as an incentive to participate and thus might be relevant information to include.

Some respondents felt that the complexity of adaptive trials may lead to therapeutic misconception. Hey and Kimmelman²⁴have argued that at least for early phase adaptive trials that typically rely on surrogate and short-term outcomes, adaptive allocation creates confusion among participants about risks and benefits of participation because they lead to incorrect estimates of the chance of a research trial benefit (therapeutic misestimation) as well as not understanding the difference between the purpose of research versus individualized care (therapeutic misconception). This, they argue, is because results from early phase trials with surrogate endpoints are a poor guide for clinical benefit given that there is often a discordance between these results and those of a phase 3 study. Hey also argues because these unrealistic expectations work to the advantage of investigators who use them to boost recruitment, investigators have no incentive to dispel these unrealistic expectations.²⁵

Buyse has stated that while outcome adaptive trials hold much promise, they should not be sold to participants as beneficial because this may not in fact be the case, suggesting a troubling lack of transparency.²⁶

Clinical Equipoise

Clinical equipoise is consistently a core ethical issue discussed in relation to adaptive clinical trials. According to Freedman, a state of clinical equipoise exists when there is genuine professional uncertainty within a community of experts about whether a treatment is preferred.²⁷ In this study, two opposing themes emerged with respect to whether equipoise is disrupted when an adaptive trial changes its randomization scheme in response to emerging data. On the one hand, respondents felt changing allocation probabilities disrupts the state of clinical equipoise with which a trial is started, while others felt that even after changing allocation probabilities, there is still uncertainty and therefore equipoise is not disrupted, and continuing is a means resolving the uncertainty.

It is generally agreed that equipoise is an ethically necessary, though not sufficient requirement before a randomized clinical trial can take place. In a typical 1:1 fixed randomization clinical trial, participants will normally have an equal chance of being assigned to either the intervention or the control arm, a situation which stays the same until the trial comes to an end or is terminated. In outcome adaptive randomization, the probability of assignment to each trial arm increases or decreases according to accruing information, depending on which arm appears to be superior or inferior. That is, if upon interim analysis, arm A seems superior, then more participants would be randomized to arm A than arm B until the next interim analysis. These probabilities have to be calculated using extensive computer simulations, and must be outlined before the trial starts, in order to protect trial integrity as well as prevent type I error. The main ethical argument advanced by proponents of outcome adaptive randomization is that it reduces the number of participants in a trial that would be assigned to an inferior treatment while still achieving valid scientific results.²⁸ Some respondents felt that the accumulating data, especially in an un-blinded adaptive trial, or in an outcome adaptive trial where study arms are dropped, will alter the state of equipoise for the investigators. They argue that unlike in a traditional trial, which appear to maintain equipoise by ignoring accumulating data until the end of the trial, adaptive trials lead to the disruption of equipoise as data accumulates and study characteristics are adjusted. This is consistent with several other researchers, who have argued that equipoise is disrupted the moment data becomes available suggesting that one arm is potentially superior, making it ethically problematic to continue with a 1:1 randomization scheme.²⁹ Some have called the continuation of randomization when the individual clinician is not in a state of equipoise “a violation of trust” between the clinician and the patient.³⁰ This view of equipoise is what has come to be called theoretical (individual) equipoise and was originally espoused by Fried.³¹

However, others have argued that the correct view of equipoise should be clinical (or collective) equipoise, which is instead judging uncertainty among a community of experts.³² As long as there is no agreement among a group of well-informed experts – that is, where genuine uncertainty exists among experts given existing evidence, then equipoise is not necessarily disrupted, and it would still be ethical to continue with the trial.³³

To counter the view that continuing to recruit when equipoise if disrupted is unethical, we can use the example of a fixed randomization design, where recruitment continues until sufficient data accumulates to make meaningful conclusions, even considering the inclusion of interim analyses and stopping rules. Unless sufficient evidence has accumulated to stop the trial, the recruitment must continue to the end in order to generate information that provides enough confidence to the majority of experts in the relevant field, generally realized through a predetermined level of statistical significance given the initial 1:1 randomization scheme. We argue that a similar situation applies to response adaptive randomization. When at interim analysis the randomization scheme is changed based on preliminary trends toward superiority of one of the arms, it does not mean that sufficient evidence has accumulated for the expert community that renders continuation of the trial unethical. The interim results merely suggest that an arm is currently working better than another arm, and, as such, that more participants should receive that arm’s intervention; but the trial continues as evidence is not yet sufficient to convince a community of experts. A number of respondents were in agreement with this view.

One may also take the view that the accumulating data does in fact alter equipoise because more information has become available, which means the researcher (or a data monitoring committee) has acquired more knowledge about whether one treatment is superior or inferior to the other. By ignoring information in a traditional clinical trial, one is violating equipoise, and by accounting for accumulating information in an adaptive trial, equipoise is still respected, if not enhanced.

The question still remains, as Bothwell and Kesselheim³⁴ put it, as to what constitutes sufficient evidence to convince which experts given that medical communities often do not agree about what level of evidence is sufficiently conclusive. This question was also posed by some of the respondents in this study. Importantly, in both cases, the trial continues until prespecified levels of confidence in outcomes are achieved.

Beneficence

Perhaps the most common justification for the use of adaptive clinical trial designs is that they are more efficient than traditional fixed randomization clinical trials.³⁵ A majority of the respondents in this study felt that these studies are more efficient because they provide the researcher a chance to more quickly get to their answer than an ordinary fixed randomization trial. A few respondents also felt that adaptive clinical trials were more beneficial or favorable because they offer participants a higher chance to get the less harmful (or more beneficial) intervention. The efficiency and beneficence were described in different ways, including the capacity to recruit fewer patients with fewer resources, the potential to randomize a patient to a drug with a higher probability of working better, as well the flexibility that allows changes to the study as it proceeds. This efficiency, coupled with the potential benefits for participants are often given as the reasons why adaptive trials are seen to be more ethical than fixed randomization trials.³⁶ In addition,

However, a small group of respondents felt that the efficiency of adaptive clinical trials is only applicable for trials that have more than two arms, because, in their view, two arm adaptive trials may end up recruiting more participants than fixed randomization studies, exposing participants to unnecessary risk of harm. This view is similar to what has previously been expressed by others in the field with respect to adaptive allocation trials.³⁷ It has previously been reported that the methodological literature may have overstated the advantages or understated the disadvantages of adaptive designs.³⁸ They observe that these trials are often complex, may need adequate infrastructure that requires additional resources, may be unsuitable for studies with long term outcomes, while trials with early stopping rules may miss secondary outcomes, safety and subgroup effects. It is clear that while these trials may be efficient and offer many benefits to both researchers and participants in some circumstances, in other cases they may be unsuitable and it is better to rely on non-adaptive designs.

The results of this study are from a relatively small group of bioethicists and researchers with experience in adaptive trial designs. These findings may therefore not be representative of the entire community of researchers involved in clinical trial designs. In addition, given that this study was done among participants who actively conduct adaptive clinical trials, it is likely that some interviewees are more likely to speak in favour of these trials compared to independent observers. In order to get a balanced view, this study also included those who are generally skeptical about these trials. Despite these observations, the findings provide a summary of some of the common ethical issues that are associated with these designs.

Conclusion

Despite their increasing popularity in the medical literature, adaptive trials continue to generate debate around ethical challenges with which they are associated. We found that among ethicists and trialists, it was generally felt that despite adaptive trials being more complex than ordinary clinical trials, and therefore constituting a potential threat to the autonomy of participants, this threat to autonomy is not insurmountable and that participants need not understand everything about the study for consent to be valid. It was also found that efficiency and potential for participant benefit were the main justification for their use. While some had ethical concerns about the disruption of clinical equipoise after changing allocation probabilities in response adaptive randomization, others insisted that despite the changing the probabilities, uncertainty still exists and therefore clinical equipoise is not disrupted. Adaptive designs have raised persistent and important ethical questions that continue to be the subject of debate among researchers and ethicists.