Bulletin of the World Health Organization

Ethical challenges in cluster randomized controlled trials: experiences from public health interventions in Africa and Asia

David Osrin a, Kishwar Azad b, Armida Fernandez c, Dharma S Manandhar d, Charles W Mwansambo e, Prasanta Tripathy f & Anthony M Costello a

a. University College London, Institute of Child Health, 30 Guilford Street, London WC1N 1EH, England.
b. Perinatal Care Project, Diabetic Association of Bangladesh, Dhaka, Bangladesh.
c. Society for Nutrition, Education and Health Action (SNEHA), Dharavi, Mumbai, India.
d. Mother and Infant Research Activities (MIRA), Kathmandu, Nepal.
e. Department of Paediatrics, Kamuzu Central Hospital, Lilongwe, Malawi.
f. Ekjut Project, Chakradharpur, Jharkhand, India.

Correspondence to David Osrin (e-mail: dosrin@ich.ucl.ac.uk).

(Submitted: 14 January 2008 – Revised version received: 10 October 2008 – Accepted: 06 January 2009 – Published online: 24 July 2009.)

Bulletin of the World Health Organization 2009;87:772-779. doi: 10.2471/BLT.08.051060


“I am because we are and because we are I am. A person is not a separated and isolated individual but a community of related individuals. Despite all this, an individual has the right to self determination and authorization.” Joseph Mfutso-Bengo, Malawian bioethicist.1

“I am because we are and because we are I am. A person is not a separated and isolated individual but a community of related individuals. Despite all this, an individual has the right to self determination and authorization.”

Joseph Mfutso-Bengo, Malawian bioethicist.1

This paper describes ethical discussions arising from our attempts to improve public health evidence in low-income settings. It considers two overlapping issues: the ethics of testing social interventions and the ethics of cluster randomized controlled trials (RCTs). We are not ethicists but public health practitioners who work on maternal and newborn survival, and our interest in ethics arises from operational challenges. We started our first cluster RCT a decade ago, and currently work on six: one in Bangladesh, two in India, one in Malawi, and two in Nepal (Table 1).

Ethics of public health research in low-income countries

Recent papers on public health ethics in the Bulletin of the World Health Organization identify a need to reconcile the claims of individual and mass approaches,24 a need encapsulated in the quotation with which this paper begins. The current emphasis of research guidelines on individual autonomy – the protection of the vulnerable – is primarily a reaction to historical tragedies.59 It draws on the ethical tradition of deontological morality: the question of whether research treats participants as they ought to be treated or as they have a right to be treated.5,10 This accords with the Hippocratic obligation to benefit the individual patient and its hallmarks are consent to participation and the right to non-interference.1114 There are other possible perspectives. Public health research lends itself to utilitarian morals within which it should maximize health or happiness for the greatest number of people:5 an action may be justified by its overall results rather than by its effects on the individual participant. This perspective, known as consequentialism,10 is attractive but difficult because it reminds us of the very abuses that led to the drafting of the existing guidelines. Nevertheless, a focus restricted to individual self-determination does not necessarily resonate with our experience of social life, the connectivity between people and the public health agenda. We hesitate to generalize but it is worth floating the idea that the individualistic perspective is at least partly a product of European enlightenment thinking and that this may not always coincide with traditions of thought in other societies.

Discussions about research ethics in low-income countries have been dominated recently by concerns about trials of drug regimens to prevent mother-to-child transmission of HIV.15 Contentious issues include the use of placebo control groups, the testing of interventions that are likely to be less effective than current best practice (with substantial discussion on how far best practice implies a global standard of care) and post-trial access to therapy.1619 Although the resulting controversy has been beneficial – airing concerns about “North–South” relationships and inequity – it gives a limited impression of public health research partnerships in low-income countries.

It is easy to see that trials of public health interventions involve a tension between individual autonomy and potential public benefit.20 Some interventions leave little option for voluntary participation, others more. Our work, for example, aims to improve the experience and outcomes of maternity. Much of it involves testing the effects of community women’s groups on morbidity, care seeking and mortality (Fig. 1). The intervention model is a cycle in which the groups discuss maternal and child health, identify problems and successes they and their neighbours have faced, prioritize issues for action, design strategies to address them, enact the strategies (either as groups or within the wider community) and evaluate their success (Fig. 2). The intervention is made on a background of sub-optimal health-care provision and usage, the point being that we are testing strategies to improve quality and uptake of appropriate care in a situation where health system best practice is an aspiration rather than a reality. The intervention requires participants to opt in: individuals have to make an effort to access it by joining and attending a community group and involving themselves in its activities.

Fig. 1. A women’s group in Dhanusha, Nepal
Fig. 1. A women’s group in Dhanusha, Nepal
David Osrin
Fig. 2. Example of an intervention model: sequence of community group activities in the City Initiative for Newborn Health, Mumbai, India
Fig. 2. Example of an intervention model: sequence of community group activities in the City Initiative for Newborn Health, Mumbai, India

Cluster randomized controlled trials

The use of RCTs to test public health interventions is increasing, a predictable development given their importance as a source of evidence. The defining feature of a cluster RCT is that the unit of allocation is a group rather than an individual.21 Cluster trials are important to public health for four reasons. Many public health interventions are delivered to groups, areas, institutions or systems collectively rather than individually; testing the delivery of an intervention to an individual might raise concerns if others are not included; individually-delivered interventions might spread among family, friends or the wider community; and we are often interested in the mass effectiveness of interventions that will be rolled out to the public with varying degrees of quality, uptake, adherence and response, even if they have been shown to be efficacious in individual cases.22 We will not discuss the issues of design,23 analysis21,24,25 and reporting of cluster RCTs,2628 but we think that researchers have an ethical responsibility to be aware of them, to seek advice and to report on the basis of appropriate analyses.27,2931

Challenges and responses

Group consent

Cluster RCTs involve two levels of consent:28 for the involvement of the group and the individual. Commonly, local guardians or representatives – elected leaders, community elders or group heads – act as cluster guardians who consent to participation. In communities in which collective decision-making is customary, communal leaders may express the collective will. To do this, they have to decide that participation is in the best interest of the community, a utilitarian judgement that may also be contested given that communities are usually amalgams of smaller communities.29,32 In a complex society, how do we identify individuals – or groups – who speak for the many? How far can we say their guardianship extends to such a decision? We need to bear in mind that representatives will almost never have been appointed for the context in which we seek their assent and so their fitness for this role may be questionable.

We have found group consent challenging. At the minimum, we need to document the choice of representatives and the reasons for approaching them.33 In all our trials, we have tried to make the process of cluster consent as open as possible. Since it is unlikely that one type of person can represent a community, we have sought agreement from a range of stakeholders. In most cases we have held community meetings at which the idea of the trial has been discussed and consent for it sought (Table 1 and Box 1). This has not always been written consent, which tends to have been reserved for elected representatives and village heads, but it has at least been inclusive.

Box 1. Examples of ethical challenges faced

Consent from cluster guardians
In all trials, the design of the cluster RCT was explained to cluster guardians before initiation.
Bangladesh: we held meetings and took verbal consent from community leaders, religious leaders, local chairmen, and elected administrative union heads.
India (Jharkhand, Orissa): we held open community meetings with village elders, opinion leaders and headmen. Written permission was granted by village elders. Clusters roughly corresponded with panchayats, although the project is spread across two states. In Orissa, panchayats have elected sarpanchs who acted as cluster guardians. In Jharkhand there had not been panchayat elections for some years and the debate about whether to hold them or to continue with the traditional system of governance involving village headmen is sub-judice: we obtained consent from traditional village headmen.
India (Mumbai): we held community meetings and took verbal consent from general practitioners, community-based organizations, non-government organizations, municipal representatives, health-post personnel, representatives of the Integrated Child Development Services, political officers of major parties and self-declared social workers. We explained to representatives what the community might contribute to the work: assistance with mapping, help to identify households, births and deaths, nominating community members to join the team and attending periodic meetings. In one area, we were advised to seek permission from a political leader (bhai), but were unable to meet him over 8 visits. A local general practitioner kindly discussed the trial with him and permission was granted.
Malawi: we held community meetings with district assembly members. We took signed consent from the District Commissioner, traditional authorities, group village headmen and village headmen.
Nepal (Makwanpur): we held open community meetings and took signed consent from heads of village development committees. Initially, one chairmanasked for a donation to each village development committee before approval could be given. Chairpersons also applied pressure for us to employpersonally recommended staff. Both these issues were resolved by a combination of resilience, a clear position and the support of the chairman ofthe district development committee, who emphasized the importance of the trial and our lack of financial wherewithal.
Nepal (Dhanusha): we held open community meetings and took signed consent from secretaries of village development committees.
Consent by individuals
In all trials, attendance at community group meetings is voluntary and group members have been told that the findings of their work will be compared with other areas in which groups are not active.
In all trials, verbal consent to data collection is taken after reading out an explanation. Participants are told that we hope that the data will contribute to the improvement of maternal and child health. The Jharkhand and Orissa trial Data Safety Monitoring Board recommended signed or thumbprint consent for individual data collection, and this has been introduced. We have disseminated baseline study findings to government and local stakeholders through programmes of meetings.
In Nepal, women in intervention clusters were told about the formation of community groups to identify the problems of women and children and to find local solutions. This information was conveyed both in women’s groups and in households when interviewers visited to collect data. In control clusters, women were told that they happened to be in the control group by chance, and that only information on pregnancy, delivery, births and deaths was being collected in their area, while women’s groups would be formed in other villages. They were also told that once the first phase of the trial was over, a similar intervention would be started in their villages. This was the earliest of our trials and the intervention was introduced in control areas in 2005.
We face some problems in clarifying the nature of the trial for individual women in control clusters. Since some of the trials involve a single visit for postnatal interview, interviewers explain that data are being collected to find out about maternal and child health to benefit local communities. Field staff have said that they find it difficult to put across the idea that somewhere else women are attending voluntary groups, but not in the respondent’s area, and that the areas will be compared. In reality, the conceptual notion of denial of an intervention to the control area is manifest only in the lack of a (novel and controversial) local women’s group, which the participant might not want to attend if it existed. Field staff have said that explaining this is challenging.
In Mumbai, although the trial involves a partnership with government, being identified too closely with government programmes can be a liability for participation. For example, family planning programmes do not have a good reputation among our participants, and we have found it necessary to assure them of our independence.
Benefits to control areas
Bangladesh: we undertook training in maternal and newborn care for heath service providers and traditional birth attendants. Pregnant women were encouraged to use health facilities. Women’s group members and trained traditional birth attendants accompany women to the facility as a result of which women get medical attention more readily.
India (Jharkhand, Orissa): we engaged with area health committees and conducted appreciative inquiry workshops for auxiliary nurse midwives. We helped to set up health committees in all clusters, to improve the links between communities and frontline public sector health providers.
India (Mumbai): as part of the wider City Initiative for Maternal and Newborn Health, we helped to upgrade health post structure and equipment, conducted appreciative inquiry workshops at maternity homes and undertook a programme of quality improvement at public sector hospitals in partnership with the municipal corporation.
Malawi: we undertook training in newborn care for health service providers and strengthening of the programme for prevention of mother-to-childtransmission of HIV, through multiplier funding.
Nepal (Makwanpur and Dhanusha): we undertook training in newborn care for all district health workers, community volunteers and traditional birth attendants. We provided equipment for all district health facilities and essential drugs for primary health centres and health posts.
Requests by participants
Rural women’s groups requested funds for travel and refreshments, common practice in non-government organization work (in Bangladesh and Nepal). We argued that the aim was for groups to be sustainable after withdrawal of support, and that such a provision was unlikely to be sustainable. In Nepal, since a small block grant was available for each group, we suggested that group members should decide on expenditure themselves.
Rural women’s groups requested ambulances, building of roads and health facilities (in all rural projects to some degree). We emphasized that solutions had to come from the community, and that it was not within our capacity to offer these services. We developed linkages with providers and were able to advise participants on how and where to make a case for provision.
Urban women’s groups requested medicines and fertility treatment (Mumbai, India). We developed guidelines and contact details for sources of care, which were provided by group facilitators.

RCT, randomized controlled trial.

Three examples provide food for thought. When the MIRA Makwanpur trial began in rural Nepal, we were able to engage the district development committee and village development committees in the process of cluster selection, randomization and allocation. However, as the crisis of governance associated with Nepal’s Maoist insurrection developed, the legitimacy and responsibilities of these institutions were contested. By degrees, the realities of running a trial in an area with two governments – each considered illegitimate by the other – became clear. We speculate that politics embodies the idea of cluster guardianship writ large, since it involves claims that individuals, parties and movements speak for larger groups. Indeed, the suggestion that a given leader might not be able to give consent for the involvement of his or her constituency could be seen as an affront to such claims.

Difficult though this issue is, it becomes more confusing if the definition of a “cluster” is not based on existing geopolitical demarcations. In Mumbai, India, we work within vulnerable urban slum areas. This means that the boundaries of clusters are not obvious and they do not necessarily represent existing political or sociocultural spaces. Who gives consent for the inclusion in a trial of part of an urban agglomeration of densely packed but diverse humanity? Certainly, we have consent from municipal health service providers but, since one of the intrinsic problems is the lack of access to their services, it would be hard to say that they speak for a cluster. We have to fall back for permission on political and cultural leaders, elected members and community organizations, bearing in mind that it is always best to avoid associating public health interventions with specific political incumbents.

And what if the intervention under test aims to change the status quo? The Ekjut intervention in Jharkhand, India, tries to help women in underserved communities to take charge of their health needs. The process itself raises (at least conceptual) notions of empowerment and agency that could be seen as socio-politically destabilizing. How would potential cluster guardians deal with this possibility? More provocatively, how ethical is it to take consent from (mostly male) guardians who embody the social structure within which the participants will be manoeuvring? On balance, we feel that the opt-in nature of the interventions under test relieves us of at least some of the burden of anxiety. Cluster guardianship is less critical if individuals are able to decide for themselves whether to be physically involved in the trial; but perhaps this allows us to skirt the issue.

Individual consent

Group consent is not a substitute for individual consent, which we think should follow similar lines to individualized studies. Participation may be related to the intervention itself and to data collection. Community members should be made aware of the trial and asked if they would like to participate. An individual’s right to refuse to participate should be respected, despite consent at the representative level,32 and trial documentation should clarify the opportunities for cluster members to avoid the risks associated with an intervention.33

All our trials require verbal consent for data collection. Participants are informed of the reasons for the exercise, the scope of the questions involved and the likely time required for an interview, as well as the fact that the decision not to participate will not be penalized. We have not so far been concerned about verbal consent in this context. From the intervention perspective, things are less concrete. In the Ekjut trial, for example, the agenda of the first women’s group meeting is to seek consent for future meetings. This is obtained after explaining the approach that will be followed, that participation in meetings is voluntary and groups can decide not to meet. In the MaiMwana trial, Malawi, women’s informed voluntary participation in intervention activities is considered to imply consent. We confess to some variation in the degree to which participants in women’s group activities are aware of the study design or the existence of control groups. They tend to be invited to participate in the unmasked intervention and are informed that they are involved in testing a new approach which will be compared with other areas that do not have women’s groups.

Timing of consent

Consent after randomization is common in cluster RCTs. Because of the scale of work, it is often easier to choose a population, define and draw a random sample of clusters in which the trial will be done, randomize the allocation, then seek consent from cluster guardians.34 This conforms in general terms with Zelen’s approach to individually randomized designs, in which consent is taken after randomization.35,36 In an individualized trial, we seek consent for participation and explain the process of randomization so that the participant is aware that she may or may not receive the intervention under test. Presumably a similar approach should apply to clusters: we should seek consent from cluster guardians for inclusion in the trial and explain that data collection will be done across all study clusters, but that allocation will be random.

Table 1 shows that we have sought cluster guardian consent pre-allocation in several cases, but not in all. In our experience, community representatives have been able to grasp the idea of randomization easily: stakeholders in Makwanpur rapidly conceived the trial design as a lottery. One way of looking at – and presenting – the situation is as a pilot test of an intervention in the first phase of roll-out. It is undoubtedly true that some representatives of control areas have been unenthused by the fact that their areas are not to receive the intervention under test until a later date, and it would be disingenuous of us to claim that the process has been smooth sailing. However, if the idea of the RCT is rendered overt, consensus can usually be reached.

Ethical approval

We have faced two particular challenges to ethical review.13,15 First, it may not be easy to find an ethical committee to which to put the case for an RCT. This is not the usual business of governmental public health. For example, although the City Initiative for Newborn Health was undertaken in partnership with the Municipal Corporation of Greater Mumbai, we struggled to find an ethical committee with sufficient mandate to be able to cope with review of a trial that involved operational research and social interventions: most ethical committees deal with clinical trials. A second problem is the blurred line between public health interventions and trials, particularly when trials are conducted as part of incremental changes in health services or community action. When a trial is conducted in a sector with few precedents for ethical review, individuals within the system may not see it as necessary to apply for approval from an ethics committee or to seek group or individual consent. On the other hand, the Nepal Health Research Council, which reviewed the Makwanpur and Dhanusha trials, has a national role across the spectrum of health and has developed a great deal of experience. Members of the Council and the Ministry of Health also sit on the trial Data Safety Monitoring Boards. The same was true in Malawi, which has an active National Health Sciences Review and Ethics Committee.

Benefits to control groups

In the fall-out from the HIV research debate, advisory bodies have struggled with the question of what constitutes the appropriate standard of care for control groups.7,37 In addition, because public health interventions are usually not masked, the possibility of “resentful demoralization” in members of control groups is high.38 It seems to us, however, that the current debate has been limited to a technological frame of reference: testing a new protocol of care for a specific disease against optimal implementation of an existing protocol. Public health interventions may be conceptually different if what we are trying to do is improve the general quality and uptake of care in a situation of vulnerability, limited resources, inequity and system weaknesses. It makes little sense to test an improvement in a system against the best possible version of the existing system, since that is precisely where the problem lies.

The issue turns, we think, on equipoise – a state of uncertainty about the pros and cons of either therapeutic arm. If we genuinely do not know if a change in health services or community action will lead to better outcomes than the status quo, a trial is a sensible option, and it is reasonable for the individuals in control groups to experience existing health-care norms.39 Failure to test interventions against current realities could deprive disadvantaged populations of incremental improvements in health. This is not an exemption clause: we believe that control groups should receive benefits for participation, although the scope of reasonable benefits is uncertain.14,40 Provision of care that is not directly linked to the research question (ancillary care) is particularly important in low-income settings, and is a current topic of debate for the Malawi National Health Sciences Review and Ethics Committee. As a research team, we hold to the maxim “no survey without service” and, in all of our trials, control groups receive benefits, summarized in Box 1. These extend to a duty of care for people in control clusters when data collection teams identify risks to health. We are happy to break protocol if individuals are at risk, assisting them, for example, with transport and negotiations for emergency medical care.

Post-trial adoption

As we have said, we see a cluster RCT as a first step in the roll-out of an intervention that may benefit public health. The intervention is introduced in a limited number of groups, its effectiveness is evaluated and, if the trial suggests that it is effective, it is rolled out to the control groups, with modifications based on experience. If the trial suggests that it is ineffective, control group individuals are protected from wasteful intervention. Participants in RCTs should have the opportunity to access superior care if the trial shows that one intervention is more effective than another, and communities involved in studies should benefit in the long run. If only things were this simple. First, we have found the cluster RCT to be less of a gold standard than it might appear. Trials often require replication to deal with questions about generalizability or simply to develop a groundswell of conviction. Health care involves politics, and so does research, and trial findings are exposed to a wide range of interpretation. Second, research funding bodies are generally not in the business of roll-out. This puts the onus on researchers to try to leverage uptake by (usually) the public sector. Again, if there is buy-in at a high level in the initial stages the process may be easier, but there is a difference between notional agreement and the commitment of resources. Involving many partners in identifying, developing, conducting and disseminating research is one way to leverage such commitment. One of the issues for all our trials is the sustainability of the community women’s groups after the trial ends. Withdrawal demands extensive discussions with group members. In all cases, we have agreed to give the control clusters the benefits of the intervention if it is shown to be effective. In some cases (Bangladesh, Jharkhand and Orissa, Mumbai), we have assurances of financial support for this from existing funders. In others, we are trying to negotiate a roll-out plan with government representatives.


We support the use of cluster RCTs of public health interventions in low-income countries. The current move towards more rigorous evaluation has occurred because it is increasingly agreed that public health research has not delivered credible evidence as often as it could have. We should also consider the ethics of not doing research. If less than 10% of current global funding for research goes to diseases that afflict more than 90% of the population (the “10/90 gap”), this gap is itself an ethical issue.13 Good research tells us if things work – or if they do not – and ethics may be served equally by protecting people from exposure to costly and ineffective interventions.14,41


We thank all the community groups and stakeholders in the trials discussed, and Glyn Alcock, Sarah Barnett, Sonia Lewycka, Joanna Morrison, Mikey Rosato and Naomi Saville. We thank the coordinators of the projects involved in our collaboration. At the Ekjut project, Nirmala Nair; at the City Initiative for Newborn Health, Neena Shah More; at the MaiMwana project, Tambosi Phiri; at the MIRA Makwanpur project, Kirti Tumbahangphe; and at the MIRA Dhanusha project, Bhim Shrestha.

Funding: Most of the authors are members of Towards 4+5, a Research Programme Consortium funded by the United Kingdom (UK) Department for International Development. David Osrin is funded by a Career Development Fellowship from The Wellcome Trust. In Bangladesh, the Perinatal Care Project is funded by the UK Community Fund and the UK Department for International Development. In India, the Ekjut study is funded by the Health Foundation and the City Initiative for Newborn Health by the ICICI Centre for Child Health and Nutrition and The Wellcome Trust. In Malawi, the MaiMwana project is funded by Saving Newborn Lives (Save the Children USA), Effective Interventions, and the UK Department for International Development. In Nepal, the MIRA Makwanpur trial is funded by the UK Department for International Development, the World Health Organization, the United Nations Fund for Population Activities, and the United Nations Children’s Fund; and the MIRA Dhanusha trial by a Swiss foundation and the United States Agency for International Development. The funding bodies played no part in the development of the article or the decision to submit it for publication.

Competing interests: None declared.