Ethical issues posed by cluster randomized trials in health research

1. Who is a research subject?

To determine whether the ethical principles and regulations governing research apply, one must first conclude that a study is human subjects research and then identify the research subjects. Indeed, much of the debate on the Keystone study focused on whether the study was in fact human subjects research [14, 26–28]. For instance, Baily argued that the Keystone study is not human subjects research:

"The project was not designed to use ICU patients as human subjects to test a new, possibly risky method of preventing infections; rather, it was designed to promote clinicians' use of procedures already known to be safe and effective for the purpose. Each hospital engaged in a classic quality-improvement activity in which team members worked together to introduce best practices and make them routine, with quantitative feedback on outcomes being intrinsic to the process. Such activities should not require IRB review." [26].

But neither novelty nor risk is at the core of what constitutes human subjects research. Rather, recalling Levine's definition of research above, research is a systematic intervention designed to produce generalizable knowledge. Miller and Emanuel argue that "the project was...[human subjects research] since it prospectively implemented a protocol of infection control interventions and tested hypotheses regarding its effectiveness. Publication of the study results suggests that a goal was to produce generalizable results" [14].

While the line between quality improvement activities and human subjects research can be very difficult to draw [29], we believe the distinction is rarely an issue for cluster randomized trials. The difficult issue for cluster trials is to establish who counts as a research subject. The question is of considerable importance, as only research subjects properly fall under the aegis of research ethics committees and protections such as informed consent. Knowledge translation cluster trials commonly intervene on healthcare providers but measure outcomes on patients. Are the healthcare providers research subjects in this case? What about the patients? Other cluster trials, such as the COMMIT study, involve community level interventions and collect data on a subset of community members. Are just those who were sampled for data research subjects, or are all members of the community research subjects?

U.S. regulations define a research subject as a "living individual about whom an investigator...conducting research obtains (1) Data through intervention or interaction with the individual, or (2) Identifiable private information". It goes on to define 'interventions' as "both physical procedures by which data are gathered...and manipulations of the subject or the subject's environment that are performed for research purposes" [30]. With respect to knowledge translation trials, when healthcare workers are the target of the study intervention it might be argued that they are, as a result, research subjects. But in such studies, if patients are only indirectly impacted by the study intervention and if no identifiable private information is collected, should we consider them research subjects? (This, we take it, might be a better way of understanding Baily's point about the Keystone study.) What about community members in a cluster trial in which the intervention is applied at the level of the community? The answer depends on how we understand the phrase "manipulations of...the subject's environment" [31]. To identify who is a research subject in cluster trials, we need a clear understanding of what sorts of environmental manipulations properly invoke the protective apparatus of research ethics and regulation. Each of these issues is explored in a subsequent article in the series.

2. From whom, how, and when must informed consent be obtained?

The ethical principle of respect for persons generally requires that researchers obtain the informed consent of research subjects. In the cluster trials literature it is commonly claimed that the need for informed consent from individuals depends on whether the study intervention is delivered at the level of the cluster or the individual [32, 33]. With a cluster-level intervention, individual refusal of informed consent may be, in effect, rendered meaningless. If an individual within a cluster refuses study participation he or she will, in many cases, be unable to avoid exposure to the study intervention and this undermines the very purpose of consent [33]. As Edwards and colleagues put it, in such studies, "the autonomy principle is lost except insofar as the individual has any democratic choice of who the guardian is and some right to consultation by the guardian" [32]. When the intervention targets individual research subjects, generally informed consent can and should be obtained. In these cases, "it is only trial entry that takes place without individual consent, as the individual treatments offered can be declined or accepted by each participant. This resembles a conventional trial where consultation over consent implies that available alternatives are offered and that these always include routine care" [32].

While this approach seems broadly correct, further work will need to justify in terms familiar to research ethics committees and regulators why individual consent may not be required in cluster trials when the intervention targets the cluster. We see two possible justifications. First, ethical and regulatory requirements for informed consent apply only to research subjects. If it turns out that, for instance, patients or community members who are only indirectly impacted by the study intervention are not research subjects, then informed consent is ipso facto not required. Second, requirements for informed consent may be waived if four conditions obtain: the research poses no more than minimal risk; the rights and welfare of subjects are not adversely affected; the research could not be carried out practicably otherwise; and, when appropriate, subjects will be debriefed [34]. The applicability of these criteria to cluster trials requires further analysis to provide researchers and research ethics committees with practical guidance. When does a cluster trial pose only minimal risk to subjects? When does a waiver of consent not adversely affect the rights and welfare of research subjects? How rigorously are we to understand the requirement that the research could not practicably be carried out? Must subjects be debriefed and, if so, how should this be done?

When must informed consent be obtained from healthcare workers in cluster trials? There are at least three dimensions of this issue that require further consideration. First, when the study intervention targets an entire hospital or primary care practice it may be difficult for a healthcare worker who refuses consent to avoid the study intervention. Second, health care workers are commonly believed to have an obligation to engage in quality improvement. Third, as Hutton and colleagues point out, "if a health care professional chooses not to participate in a study, they [sic] are in effect denying their patients the potential benefits of participation. Healthcare providers ought to do the best for their patients..." [10].

When the study intervention is administered at the individual level, it is generally agreed that the informed consent of the research subject must be obtained [33]. But when a cluster trial involves a behavioral intervention, the informed consent process may lead to treatment contamination [35, 36]. Edwards and colleagues explain that

"[i]nforming the controls fully about the experimental arm(s) is likely to produce the very effect that randomizing by cluster was designed to avoid -- that is, prompting controls to adopt the treatment(s) under investigation. One option is to withhold information about the novel treatment from controls, on the grounds that they are getting conventional care and are therefore in the same position as people outside the experiment" [32].

But can information about the details of the study intervention be withheld from research subjects in the control arm consistent with the principle of respect for persons?

Finally, Klar and Donner raise a difficult question regarding the timing of informed consent that requires further exploration. To illustrate their concern they point to two studies examining the impact of vitamin A administration on early childhood mortality. In the first study, the unit of randomization was the household and informed consent was obtained from study participants prior to randomization [37]. In the second study, the unit of randomization was the community and informed consent was only obtained after randomization [38]. The authors worry that

"[t]he relative absence of ethical guidelines for cluster randomized trials appears to have created a research environment in which the choice of randomization unit may determine whether informed consent is deemed necessary before random assignment...It seems questionable, on both an ethical level and a methodological level, whether the unit of randomization should play such a critical role in deciding whether informed consent is required [before randomization]" [39].

A subsequent paper in the series examines each of these questions in detail.

3. Does clinical equipoise apply to CRTs?

The ethical principle of beneficence obliges researchers to not harm needlessly and, where possible, to promote the good of research subjects. The application of beneficence to cluster trials raises two broad questions.

First, do researchers have an ethical obligation to research subjects in the control arm to provide more than usual care? The question arises out of the belief that, while subjects in the experimental arm may benefit as a result of study participation, those in the control arm are exposed to risks and burdens without the prospect of such benefit. Glanz and colleagues state:

"Meeting [ethical] requirements...is particularly challenging when individuals or communities are assigned to control or comparison groups that do not receive the intervention hypothesized to be most effective. The control subjects may be burdened disproportionately by data collection requirements without receiving the benefit of services or resources" [36].

According to Klar and Donner, "some investigators have attempted to ensure that these individuals can still benefit from participation by offering a minimal level of intervention or, alternatively, by offering all individuals the intervention by the technique of delaying its intervention in the control group" [39]. While intuitively appealing, these approaches require further reflection. If denying research subjects in the control arm access to the hoped for benefits of the experimental intervention is ethically impermissible, then why is it permitted to give them only "minimal" benefits or to delay their access to these benefits?

Second, as data accumulate in a cluster trial, is there an obligation to modify or stop the study if one of the interventions appears unsafe or unexpectedly effective? For a variety of reasons, data monitoring committees are not commonly used in cluster randomized trials. When data monitoring committees are employed, they require clear guidance as to their ethical obligations. Glanz and colleagues have argued that concerns of safety or unexpected efficacy may require a data monitoring committee to modify or stop a study prematurely [36]. They point out that "interim analysis could show a clear improvement in psychological or medical outcomes associated with an intervention. It would then be reasonable to offer the more effective strategy to all communities or participants" [36]. It is well recognized that early differences between interventions may be the result of chance or bias rather than a true intervention effect. How much evidence of a "clear improvement" ought there be before a data monitoring committee recommends that a study ought to be modified or stopped?

In the literature on individually randomized trials, the concept of clinical equipoise helpfully frames questions regarding researcher obligations to subjects in the control group and when data monitoring committees ought to recommend modifying or stopping a clinical trial. As described above, clinical equipoise permits a trial to be started when there exists a state of honest, professional disagreement in the community of expert practitioners as to the preferred treatment [19]. By implication, a trial ought to be stopped when the moral warrant for its conduct no longer obtains [40]. It is unclear, however, whether clinical equipoise can be applied to cluster trials. The concept is commonly understood as emerging from the fiduciary relationship between physician-researcher and patient-subject [41]. Cluster trials may involve neither physician-researchers nor patient-subjects. For instance, in both the NEXUS trial and the Keystone study, the targets of the study intervention were health care workers themselves. In the COMMIT study, the targets of the study intervention were communities and community members. If clinical equipoise is to be used to address issues posed by cluster trials, a moral foundation relevant to cluster trials will have to be articulated for it. The applicability of clinical equipoise to CRTs is considered in detail in a subsequent paper in the series.

4. How do we determine if the benefits outweigh the risks of CRTs?

The principle of beneficence requires that the benefits of study participation stand in reasonable relation to its risks. Numerous publications describe the variability in review from one research ethics committee to the next. For instance, Hearnshaw documents wide discrepancies in requirements for ethics review and time to approval in 11 European countries for a study involving an information pamphlet and questionnaire for elderly patients and their physicians [42]. While part of the variation in ethics review is a result of regulatory differences among countries, the lack of a structured approach to the ethical analysis of risk is thought to be an important contributing factor. Described in detail above, component analysis provides research ethics committees with a systematic approach to the ethical analysis of benefits and harms in research [18]. The applicability of component analysis to cluster randomized trials is, unfortunately, unclear. If component analysis is to be applied to cluster trials, a number of conceptual hurdles will first have to be cleared.

Does the distinction between therapeutic and nontherapeutic procedures hold in cluster trials? The first step in component analysis is the demarcation of therapeutic and nontherapeutic procedures. The distinction between therapeutic and nontherapeutic procedures is generally unproblematic in cluster trials with an individual level intervention. These individual level interventions commonly are drug, surgical, or behavioral interventions that aim to benefit research subjects, and, thus, they are straightforwardly therapeutic interventions. The difficulty is posed by cluster trials involving a cluster level intervention. Public health trials commonly involve a cluster level intervention designed to improve the health of a community and its members. For instance, the COMMIT trial employed a multimedia campaign to increase quit rates in heavy smokers and reduce the prevalence of smoking in the community. Ought we to understand these interventions as therapeutic? Even more difficult to classify are complex interventions that aim to modify healthcare worker behavior in knowledge translation trials. The NEXUS trial used audit and feedback and educational messages to attempt to reduce physician orders for needless radiographs. Should we classify these procedures as therapeutic or nontherapeutic interventions?

The second step in component analysis is to ask whether therapeutic procedures meet the ethical standard of clinical equipoise [18]. Question #3 above considers in detail the applicability of clinical equipoise to cluster trials. The third step in component analysis is to ask whether the risks of nontherapeutic procedures are minimized consistent with sound scientific design, stand in reasonable relation to the knowledge to be gained, and, if the study involves a vulnerable population, pose no more than a minor increase above minimal risk [18]. The applicability of each of these standards to cluster trials deserves exploration. Of particular interest is the meaning of minimal risk in the context of a cluster trial. Minimal risk is commonly defined as the risks of daily life of a healthy person [43]. When cluster trials target households, neighborhoods, or communities, it is unclear whether an individualistic understanding of minimal risk remains appropriate. Might minimal risk refer to the quotidian risks faced by clusters rather than individuals? What impact would such an understanding have on the review of cluster trials? A subsequent paper in the series analyzes these questions in detail.

5. How ought vulnerable groups be protected in CRTs?

The principle of justice requires that vulnerable groups in research both be protected adequately and not unduly denied access to research benefits. Vulnerable groups are commonly understood to include pregnant women, prisoners, children, and incompetent adults, and cluster trials have studied all of these groups. Althabe and colleagues describe a cluster trial of a multifaceted behavioral intervention to improve obstetrical care in Argentina and Uruguay [44]. Hickman and colleagues randomized specialist drug clinics and prisons to test whether the use of dried blood spots to test for hepatitis C would increase uptake of diagnostic testing by injection drug users [45]. Kipping and colleagues describe a pilot cluster trial in which schools with children 9 and 10 years of age were randomized to receive an obesity prevention intervention or no intervention [46]. De Smet and colleagues randomized 13 intensive care units in the Netherlands to receive digestive tract decontamination with oral and intravenous antibiotics, digestive tract decontamination with oral antibiotics only, and usual care in an attempt to reduce 28-day patient mortality [47].

A variety of additional protections apply when clinical research involves a vulnerable group. The inclusion of the vulnerable group in research must be required to answer the study hypothesis; a vulnerable group cannot be used merely as a population of convenience. When prospective research subjects are incapable of providing informed consent, a surrogate decision maker must provide consent on their behalf. Finally, the risks of nontherapeutic procedures must not exceed a minor increase above minimal risk. Cluster trials, particularly those involving interventions applied at the level of the cluster, may further restrict the ability of vulnerable groups (or their surrogate decision makers) to choose research participation freely. Does this imply that greater protections for vulnerable groups in cluster trials are required? How might one meaningfully enhance protections without impeding research that may benefit the health of vulnerable groups?

Research conducted in developing countries raises a host of ethical issues [17]. Consider Bolton and colleagues' description of the first cluster randomized trial of psychotherapy in sub-Saharan Africa [48]. Depression is a common and serious health problem in sub-Saharan Africa, with a prevalence estimated at 21% [48]. Unfortunately, few treatments are available for those suffering from depression in impoverished countries. Antidepressant drugs are too expensive and psychotherapy, developed for use in industrialized countries, has not been tested for efficacy. In the trial, 30 villages in rural Uganda were randomized to receive psychotherapy or usual care. Study subjects were identified with the help of community leaders, healers, and other knowledgeable persons and, after they provided verbal informed consent, the diagnosis of depression was confirmed with a culturally appropriate questionnaire. In villages allocated to the intervention arm, subjects received group-based interpersonal psychotherapy for 90 minutes each week for 16 weeks. In control villages, research subjects were free to seek out whatever interventions they wished. Symptoms were again assessed after the intervention period. The study intervention proved highly effective in treating depression: after the intervention, 6.5% of subjects in the intervention group met the criteria for severe depression, compared with 54.7% of subjects in the control group. Upon completion of the study, group psychotherapy was made available to the control communities.

The trial of interpersonal psychotherapy for depression illustrates both the potential and challenges of cluster trials in developing countries. Researchers conducting cluster trials face ethical issues that flow from cultural differences and disparities in access to health care between host and sponsor countries. What ethical standards, including those for informed consent, ought to apply: those of the host or sponsor country? What treatment should research subjects in the control arm receive? Clinical equipoise requires patients enrolled in a trial not be exposed to treatment known to be inferior to treatments available in clinical practice. But, one might ask, available where? In developed countries, standard treatment for major depression includes antidepressant drugs and psychotherapy. Must subjects in the control arm receive the best, proven therapy, even if it is locally unavailable? In communities with substandard access to healthcare, do researchers have an obligation to provide research subjects with treatment for medical conditions not related to the study condition? What obligations do researchers and study sponsors have to research subjects and host communities after completion of the study? Do they have a moral obligation to provide participating communities with access to the study intervention (if it proves effective), and, if so, for how long? These important justice issues are explored in a subsequent paper in the series.

6. Who are gatekeepers and what are their responsibilities?

There is a growing consensus in the research ethics literature that researchers have obligations to communities participating in research. The ethical principle of respect for communities flows from the recognition that the community has moral worth and, as a result, researchers have a duty to protect and promote its interests [24]. The community-researcher relationship has been described as a partnership, in which community consultation and negotiated agreement are key features [25]. When a community has a legitimate political authority empowered to speak on behalf of its members, researchers may additionally be required to seek community consent to research participation. Importantly, community consent does not supplant the requirement for individual informed consent to study participation. While protections for communities in research may straightforwardly apply to cluster trials in which the unit of randomization is the community, their applicability across the scope of cluster trials is uncertain. Cluster trials randomize diverse groups that are not communities -- households, primary care practices, hospital wards, classrooms, and neighborhoods -- and whose moral status is not well characterized. In the cluster trials literature, the gatekeeper has emerged as a key player in protecting the interests of these diverse groups and their members [32, 33, 49]. However, a variety of questions regarding the role, function, and authority of gatekeepers have yet to be explored adequately.

Who are gatekeepers? When cluster trials involve an intervention that is administered at the cluster level, difficulties in obtaining meaningful individual informed consent have led to the practice of using gatekeepers [49], guardians [32], and cluster representation mechanisms [33] to protect group and individual interests. Edwards and colleagues define a gatekeeper as "an agent...who has the power to 'deliver' [a] cluster," and who acts as an advocate on behalf of cluster interests [32]. Hutton, defines gatekeepers as "people in either political or administrative positions who are able to give consent for those within a cluster to be randomized" and whose consent may occur on multiple "levels" [49]. Current descriptions of gatekeepers, however, do not give a clear account of who can act as a gatekeeper when there are no clear administrative or political structures in place. The diversity of groups studied in cluster trials poses a challenge to how we identify gatekeepers, and how group characteristics influence who may serve as representatives.

What are the functions of gatekeepers? Gatekeepers are described as being able to "deliver" [32] or "give consent for" [49] a cluster. The U.K. Medical Research Council guidelines describe the role of a gatekeeper as "analogous...to that of individuals for individual decisions" and says the gatekeeper must act "in the interests of the cluster/individuals in the cluster" [33]. Further, the gatekeeper must document that he or she "considers the cluster's participation in the trial to be in the interests of the cluster as a whole/in the interests of each member of the cluster (as appropriate...)" [33]. The potential for conflict among the various sets of interests protected by the gatekeeper requires careful examination. Acknowledging that community or cluster and individual interests are separable and may be conflicting [24, 50], how should a gatekeeper balance individual and cluster interests if they conflict? Gatekeepers, who may be in administrative positions (e.g., practice managers, hospital chief executive officers), will also have to balance cluster and institutional interests and consider the impact of the research on the organization for which they are responsible.

What are the sources of a gatekeeper's authority? One of the outcomes of the debate on community consent is the recognition of the importance of the issue of authority [24]. Only communities that possess a legitimate political authority empowered to speak on behalf of its members may provide community consent. The issue of authority for the variety of functions ascribed to gatekeepers requires careful scrutiny. When does a gatekeeper possess the authority to consent on behalf of the cluster? When individual consent cannot be obtained, does a gatekeeper have the authority to consent on behalf of the individuals in the cluster? A subsequent paper in the series critically appraises the role, function, and authority of gatekeepers in CRTs.