Article Text
PDF
Abstract
The robust design and conduct of pragmatic cluster randomised trials may be in tension with the ethical requirement to obtain written informed consent from prospective research participants. In our experience, researchers tend to focus on whether a waiver of consent is appropriate for their studies. However, pragmatic cluster randomised trials raise other important questions that have direct implications for determining when an alteration or waiver of consent is permissible. To assist those involved in the design, conduct and review of pragmatic cluster randomised trials, we outline four critical questions to consider: (1) What is the nature of the intervention being evaluated? (2) Is the choice to use cluster randomisation justified? (3) Can the risk of recruitment bias be addressed? and (4) Is an alteration or waiver of consent appropriately justified? We recommend that researchers and research ethics committees conduct a stepwise analysis of a planned cluster randomised trial using these questions. To illustrate the application of this stepwise analysis, we use three pragmatic cluster randomised trials in the haemodialysis setting as case studies.
- Clinical Trial
- Ethics- Research
- Informed Consent
- Risk Assessment
Data availability statement
Data sharing not applicable as no datasets generated and/or analysed for this study.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
Introduction
The conduct of pragmatic trials has been recognised as an urgent priority.1 As opposed to explanatory trials that evaluate the efficacy of interventions in optimal conditions, pragmatic trials evaluate the effectiveness of interventions in routine conditions to inform the decisions of patients, health providers and systems managers.2 3 Pragmatic trials commonly have few exclusion criteria, recruit participants at the time of clinical presentation, have clinically meaningful and patient-centred outcomes, use routinely collected data for outcome measures and include all randomised participants in the analysis. And while most pragmatic trials use individual (eg, patient) randomisation, approximately 35% use cluster randomisation that involves randomly allocating groups (eg, clinics) to study interventions.4
Pragmatic cluster randomised trials (CRTs) are ethically and methodologically complex. While international ethics guidance for CRTs exists,5 6 these documents do not specifically discuss issues raised by pragmatism. For example, one prominent issue in pragmatic trials is the recruitment process. Pragmatic trials aim to mirror routine clinical settings and ideally include all patients who will receive the interventions under investigation as a part of routine care. Soliciting prospective participants’ consent, however, can disrupt the workflow of busy clinics to the extent that the trial no longer mirrors routine practice. Prospective participants may also refuse to participate and, if recruitment and retention are too low, this can result in treatment-relevant differences between the study population and target clinical population thereby undermining the fundamental goal of pragmatic trials—to directly inform patient care.
CRTs also require larger sample sizes than individually randomised trials because outcomes from multiple participants in the same cluster are usually positively correlated. If research personnel are required to recruit participants, CRTs will consequently be more costly than individually randomised trials. Furthermore, prospective participants are often recruited after clusters have been randomised. If participants and those recruiting them cannot be blinded to their cluster’s allocation, there is a high risk of identification and recruitment bias7 8; in other words, ‘the potential for knowledge of the assigned intervention to influence whether individuals are recruited or selected into the analysis’.9 Essentially, CRTs are less likely to produce reliable results if consent is required.
This creates a point of tension: removing consent requirements in pragmatic CRTs reduces trial costs, eliminates recruitment bias and permits researchers to include all eligible patients without disrupting routine practice, but requiring informed consent is a central ethical protection in research involving human participants. As Capron explains, “informed consent for research is a great lake, fed by three historical rivers,” where the first and oldest river is the ethical principle of respect for autonomy, and the second and third rivers are human rights and research regulations.10 Both international human rights documents and research regulations articulate strict consent requirements in part due to historic examples of unlawful medical experimentation. Respect for autonomy means that informed consent is about protecting people ‘against the wrong that occurs when they, as moral agents with the right and responsibility to chart their own lives and actions, are not given an opportunity to decide whether or not to accept an intervention involving their person or things intimately associated with their being’.10 Any infringement on this great lake, this ethical and legal requirement, needs to be carefully considered and adequately justified.
Based on our experience, researchers focus primarily on whether their pragmatic CRT meets regulatory criteria for an alteration or waiver of consent. Indeed, ‘the trend has been to find reasons for dispensing with individual, voluntary, informed consent’.10 However, a shallow analysis of the tension that considers only regulatory criteria for a waiver of consent is inadequate. For pragmatic CRTs, a deeper dive into the great lake requires a stepwise analysis of issues that are antecedent to the permissibility of a waiver of consent. Building on our previous work,11–15 we outline four questions to assist researchers and research ethics committees (RECs) with navigating the consent river in pragmatic CRTs (see figure 1). We recommend those involved in the design and review of these trials consider the following: (1) What is the nature of the intervention being evaluated? (2) Is the choice to use cluster randomisation justified? (3) Can the risk of recruitment bias be addressed? and (4) Is an alteration or waiver of consent appropriately justified? Although not addressed in this paper, consideration of other issues such as those raised by privacy and confidentiality procedures, the benefits and harms of a study, and participant selection procedures, is required for a comprehensive ethical analysis.
Stepwise analysis to assist researchers and research ethics committees with the design and conduct of pragmatic cluster randomised trials.
To illustrate our proposed stepwise analysis, we use three pragmatic CRTs as case studies (see boxes 1–3). We deliberately chose studies from the haemodialysis setting as the nephrology community has been leading the way in the conduct of pragmatic CRTs, thereby establishing a precedent for how future studies are designed and implemented. Many pragmatic CRTs in this setting have also been ‘facilitated and encouraged through specific funding mechanisms from sponsors critical to nephrology research, including the National Institutes of Health, Department of Veterans Affairs and the Patient Centered Outcomes Research Institute’.16 Accordingly, there are several large-scale pragmatic CRTs in the haemodialysis setting that have recently been completed,17–19 terminated20 or are in progress.21 22 All information about the exemplar trials was retrieved from publicly accessible websites or published material.17 18 20 23–35
Higher versus lower serum phosphate targets in patients undergoing dialysis (HiLo) trial
The HiLo trial aimed to evaluate whether a higher (>6.5 mg/dL) vs lower (<5.5 mg/dL) target phosphate titration protocol alters the rates of all-cause mortality and hospitalisations in patients receiving maintenance haemodialysis.20 31 Titration protocols were implemented by in-centre dietitians within the framework of the dialysis providers’ current procedures. They expected to randomise 120–150 haemodialysis centres in the USA providing treatment to approximately 4400 patients. Monthly data transfers from DaVita and the University of Utah electronic medical records were deidentified and combined into a single database for outcome assessment. The Duke University Institutional Review Board approved the study. To obtain informed consent from patients without on-site research personnel, the HiLo investigators distributed tablet devices that connect to secure web-based electronic consent modules on the study website.31 The electronic consent modules are accompanied by concise educational videos explaining clinical research participation, phosphate and its management in kidney failure, and details of the HiLo trial. A centralised team of nephrologists was also available to answer potential participants’ questions by telephone. Patient enrolment began in 2020. According to the investigators, ‘For the study’s primary objective, HiLo will be a success if it yields evidence that the Hi or Lo target is superior to the other… HiLo will also be a success if it enhances the clinical trial culture in nephrology by ushering in wider use of stakeholder engagement, eConsent, electronic health record data, remote study monitoring, and innovations in primary outcomes’.20 However, after recruiting about 500 participants into the trial, imbalances in baseline characteristics and enrolment rates were detected.35 As a result, the trial was changed to individual randomisation.35 The trial was subsequently terminated by the Data and Safety Monitoring Board due to futility of enrolment and futility of separation between study arms.22
Major outcomes with personalised dialysate temperature (MyTEMP) trial
The MyTEMP trial evaluated the effectiveness of adopting personalised cooler dialysate (0.5°C and 0.9°C below the patient’s predialysis body temperature, to a minimum dialysate temperature of 35.5°C) as a centre-wide policy to reduce cardiovascular-related death and hospitalisations compared with standard temperature dialysate (36.5°C).18 84 dialysis centres in Ontario, Canada were randomly allocated to the study interventions, which resulted in 15 413 patients over 4 years receiving maintenance haemodialysis at the temperature assigned to their centre. Most primary outcome data were obtained from provincial databases. Only patients’ deidentified temperatures and blood pressure were obtained from a random sample of 15 haemodialysis sessions per month. Western University’s Health Science Research Ethics Board, on behalf of 13 institutions overseeing 45 haemodialysis centres, approved the study with a waiver of consent.23 The remaining institutions received ethics approval and were granted a waiver of consent from their local research ethics committees. Participating centres were provided with written information to give to prevalent patients at the start of the trial informing them of their centres’ allocated temperature and their right to be dialyzed at a different temperature should they or their physician choose.23 The prespecified goal of achieving 80% of haemodialysis treatments in participating centres to be adherent to the random assignment was achieved, although reasons for non-adherence were not recorded.18 Posters indicating that a trial was ongoing and that patients could speak with their health provider to learn more about it were placed in waiting areas or near weighing scales at participating centres. According to the investigators, ‘When applied as a centre-wide policy, personalised cooler dialysate did not reduce the risk of cardiovascular mortality or hospital admissions for major adverse cardiovascular events compared with a dialysate temperature of 36·5°C’.18 However, patients in the personalised cooler dialysate group were more likely to report feeling uncomfortably cold on dialysis than patients in the standard-temperature group,18 25 26 and commentators believe that the results should not be interpreted as ‘definitive evidence that cooling the dialysate is ineffective for individual patients’.28
Time to reduce mortality in end-stage renal disease (TiME) trial
The TiME trial evaluated the effects of implementing a minimum haemodialysis session duration of ≥4.25 hours on mortality, hospitalisations and health-related quality of life compared with usual care (no trial-driven approach to session duration).17 266 dialysis centres in the USA were randomly allocated to the study interventions, which resulted in 7035 incident patients over 4 years receiving maintenance haemodialysis for the duration assigned to their centre. All data were collected routinely and obtained from electronic health and administrative records of the dialysis provider organisations. Under authorisation agreements with the dialysis provider organisations overseeing the 266 dialysis centres, the trial was approved and granted a waiver of consent by the University of Pennsylvania Institutional Review Board. Patients were given written information about the trial, including their facility’s randomised assignment and the ability to modify their session duration if a ≥4.25-hour session was believed to be inappropriate by their physician. Patients were also given telephone access to contact members of the research team for additional information or to opt out of having their data included in the trial dataset. An external Data and Safety Monitoring Board terminated the trial ‘because of a lower than anticipated difference in session duration between the intervention and usual care groups, and an interim analysis indicating a conditional power close to 0% to detect a hazard ratio of 0.85 for mortality’.17 The use of a waiver of consent contributed to the lack of compliance with study interventions and, consequently, a lack of an experiment since, according to the investigators, ‘among the reasons identified by facility personnel for low adherence to the intervention, reluctance by patients to undergo treatments longer than those of many other patients at the same dialysis unit seemed to be the most important’.17 The investigators found no difference in mortality or hospitalisation rates for the intervention compared with usual care.
Stepwise analysis
Question 1: what is the nature of the intervention being evaluated?
Study interventions in a CRT can be delivered at several different levels,36 and the ability to distinguish between them is crucial since it has direct implications on the appropriateness of using cluster randomisation. For the purposes of this paper, we focus on two levels: cluster-level and individual-level interventions.
Cluster-level interventions can only be delivered to an entire cluster. For example, researchers may evaluate the cost-effectiveness and energy efficiency of a novel water filtration method compared with standard reverse osmosis water filtration in a haemodialysis clinic’s centralised water treatment system. An inherent feature of these interventions is that they are not divisible at the individual level, meaning that exceptions cannot be made for any individual within a cluster. It also means that it is difficult, if not impossible, for individuals to avoid the intervention. Generally, individuals would have to leave their cluster to avoid the intervention which, in this example, means patients would need to seek care at a clinic that is not enrolled in the study. When researchers want to evaluate a cluster-level intervention, individual randomisation cannot be used and, in these cases, the justification for using cluster randomisation is that it must be adopted out of necessity.
Individual-level interventions are divisible at the individual level, meaning that individuals can avoid the intervention by refusing its administration. For example, patients receiving haemodialysis treatments at clinics enrolled in the experimental arm of the HiLo trial would ‘experience liberalised diets and less intensive phosphate-binder regimens’.20 Patients can avoid the intervention by requesting typical phosphate-binder regimens and maintaining strict diets. Similarly, in the MyTEMP trial18 and TiME trial,17 treatment temperatures and duration were set as a local centre’s policy, but it would be possible for a patient to have their treatment temperature or duration modified. When researchers want to evaluate an individual-level intervention, either individual or cluster randomisation can—in theory—be used. However, given the statistical inefficiency and methodological complexities with CRTs (discussed below) as well as the manifold ethical issues they raise,5 the use of cluster as opposed to individual randomisation must be justified.
It is common for CRTs to evaluate complex interventions involving both cluster-level and individual-level components. For instance, to reduce complications associated with haemodialysis, a CRT may involve the reorganisation of nursing scheduling and workflow, a novel protocol for nursing management of complications, and a self-directed patient education tool on best practices for preventing complications during treatment. When researchers want to evaluate a complex intervention involving a cluster-level component, individual randomisation cannot be used to evaluate the cluster-level component of the intervention. But the use of cluster randomisation to evaluate the individual-level component of the intervention requires justification since an alternative, statistically efficient design may be used (eg, split-plot designs which involve randomly allocating clusters to the cluster-level intervention components and then randomly allocating individuals within clusters to the individual-level intervention components).37
Question 2: is the choice to use cluster randomisation justified?
If researchers aim to evaluate an individual-level intervention or a complex intervention involving an individual-level component, using cluster randomisation should be justified. The investigators of the HiLo, MyTEMP and TiME trials provided similar reasons: the HiLo trial used cluster randomisation ‘to simplify trial operations and to ensure that only a single study-specific phosphate target is implemented in each facility’20; the MyTEMP trial used cluster randomisation ‘to enhance intervention uptake and adherence (logistical convenience) and to minimise cross-group contamination’23 and the TiME trial used cluster randomisation ‘to facilitate implementation of the intervention and minimise contamination’.17
The most common reason for using cluster randomisation to evaluate an individual-level intervention is to avoid or minimise contamination.13 Other reasons commonly provided include scientific considerations (eg, when the treatment effect of interest is a cluster-level effect, ie, how effective the intervention is for the average cluster rather than the average participant), logistical or administrative convenience, to enhance compliance, to reduce costs or to increase the pragmatism or external validity of the trial by mirroring how interventions are delivered in routine practice. But whether any of these justifications are appropriate requires more context: researchers and RECs should consider how the unit of randomisation impacts the study from an ethical, practical and statistical standpoint.
From an ethical standpoint, exposing research participants to the risks of research is justified, in part, by the prospect of generating socially valuable knowledge. Yet scientific rigour is a prerequisite to social value: ‘unless research generates reliable and valid data that can be interpreted and used by the specific beneficiaries of the research, it will have no social value and participants may be exposed to risks for no benefits’.38 Researchers should, therefore, choose the most scientifically rigorous design to answer their research question.
From a practical standpoint, using cluster randomisation may avoid some of the logistical challenges associated with individual randomisation. Consider the MyTEMP trial. If researchers were to randomise patients rather than haemodialysis centres, they would have to contemplate the logistics of randomisation and intervention delivery for 15 413 patients. Logistics are clearly simplified if each of the 84 centres is randomised to either a standard dialysate temperature policy or a personalised dialysate temperature policy as all patients within a centre are managed with identical temperatures.
From a statistical standpoint, cluster randomisation is statistically inefficient with respect to sample size compared with individual randomisation. For example, in the MyTEMP trial, 84 clusters assuming a total of 6846 person-years per arm (ie, using the harmonic cluster mean (163) multiplied by half the clusters (42) as a proxy for the total person-years) were required to achieve 80% power to detect a 20% hazard rate reduction.23 An individually randomised trial would have required only 3956 person-years per arm. Thus, the MyTEMP trial required a 73% inflation in the sample size to achieve the same power as an individually randomised trial. This means that more research participants are exposed to the risks of research in a CRT compared with an individually randomised trial.
The increased number of participants required in CRTs may be balanced against potential cost reductions due to pragmatic design features. For example, according to the HiLo trial investigators, “Unlike a traditional explanatory trial that might incur $50-$100 million in costs… HiLo will be executed for 5%-10% of the cost”.20 But these cost savings result from pragmatic design choices; for example, the use of electronic health records for outcome assessment instead of incurring the costs of additional data collection procedures. Pragmatic design features that make CRTs less expensive can also make individually randomised trials less expensive. Moreover, it is important to compare the costs of a CRT and an individually randomised trial with consent processes rather than comparing the costs of a CRT without a consent process to an individually randomised trial with a consent process. Whether consent is required from research participants depends on the study interventions and data collection procedures, irrespective of the unit of randomisation.39
Even if costs can be reduced by using cluster over individual randomisation, considerations of scientific validity will likely override any cost savings considerations. Unlike individually randomised trials, researchers will commonly perform randomisation in a CRT before individuals are identified and recruited. Furthermore, blinding participants to the allocated study interventions can be challenging. This means there is an increased risk of identification bias, recruitment bias or both with CRTs compared with individually randomised trials. In the HiLo, MyTEMP and TiME trials, randomisation occurred before it was possible to identify research participants. The HiLo trial had major imbalances in baseline characteristics and enrolment rates between the study arms and, as a result, switched to individual randomisation after 10% of the participants had been enrolled.32 35 The investigators of the MyTEMP and TiME trials were able to avoid these biases with a waiver of consent—if there is no recruitment process, there is no risk of recruitment bias. This means that CRTs are more prone to bias and thus less likely to produce reliable, scientifically valid results than individually randomised trials when consent is required.7
To summarise, using cluster over individual randomisation is difficult to justify for individual-level interventions. Even the most commonly provided reason—the need to avoid or minimise contamination—requires careful scrutiny as individually randomised design can account for a substantial amount of contamination.40 CRTs may have some practical advantages, but they require more participants and will often be more expensive and less likely to produce scientifically valid results compared with individually randomised trials when informed consent is required. The solution, however, is not to dispense with consent requirements to address risks of bias. The solution is to use individual randomisation when possible because researchers can simultaneously uphold ethical standards while increasing the likelihood of producing scientifically valid results since identification and recruitment bias are not introduced when consent is obtained prior to randomisation.
Question 3: can risks of bias be addressed?
If using individual randomisation is not possible, or the use of cluster randomisation has been appropriately justified, steps must be taken to address risks of bias. There are various steps researchers can take to circumvent or mitigate bias in CRTs. When feasible, researchers should recruit participants before clusters are randomised and implement the intervention with minimal delay. This would avoid identification and recruitment biases. However, for many CRTs, randomisation occurs before it is possible to identify and recruit participants. In these cases, researchers should identify and recruit participants as soon as possible after clusters have been randomised and ensure blinded recruiters provide both intervention and control groups with similar information about the trial.8 This means that prospective participants can be fully informed about the purpose and other details of the trial while concealing allocation at the time of enrolment.
In some cases, participants may be able to determine their allocation status after enrolment, which can lead to performance bias, non-adherence or attrition. Considering the TiME trial, if participants are informed of the study objectives, then they could directly infer their centre’s allocation from the treatment they receive. This may, for example, lead participants in the intervention group to refuse longer dialysis sessions. Steps to mitigate these biases (eg, using objective outcomes as opposed to patient-reported outcomes) and to minimise or address non-adherence and attrition (eg, establishing effective communication channels between researchers and participants, using intention-to-treat analysis) should be considered when designing the trial.
If there is a demonstrably high risk of identification and recruitment bias and the various steps to mitigate these biases are not practicable, researchers may seek an alteration or waiver of consent requirements for the study interventions to circumvent the risks of bias.
Question 4: is an alteration or waiver of consent requirements justifiable?
Recent trends show that pragmatic CRTs are increasingly conducted without consent,4 exemplified by the MyTEMP18 and TiME trials17 and two other ongoing pragmatic CRTs in the haemodialysis setting.21 22 33 34 According to international ethics guidelines for CRTs, an alteration or waiver of consent requirements is permissible when (1) the research would not be feasible if informed consent is required, (2) the research has important social value and (3) the research poses no more than minimal risk to participants.5 6
To determine the level of risk in research, RECs are tasked with ‘comparing the probability and magnitude of anticipated harms with the probability and magnitude of harms ordinarily encountered in daily life or during the performance of routine physical or psychological examinations or tests’.6 Exactly how to operationalise this definition is challenging. One reason why the MyTEMP trial was considered minimal risk research was that the intervention was ‘similar to a quality-control measure that could be implemented by a dialysis centre director’.23 Likewise, the TiME trial was considered minimal risk research because, in part, ‘the intervention consists of a dialysis session that is within the range of usual care administered in the USA as well as in many other countries’.17 The HiLo trial was not considered minimal risk research because the experimental intervention ‘differs from the current opinion-based standard of care’.20
These claims reflect a belief that pragmatic CRTs inherently meet the definition of minimal risk research when they evaluate usual care interventions. There are two problems with this belief. First, when study interventions are a protocolised subset of a continuous usual care range, the interventions no longer constitute usual care.41 For example, typical haemodialysis treatment duration is between 3.5 and 4.5 hours, thrice weekly. When the range of care was dichotomised in the TiME trial, patients whose cluster was allocated to implementing a minimum session duration of 4.25 hours were no longer receiving usual care. The dichotomisation of a continuous usual care range introduces potential risks (and benefits) that are distinct from those associated with the range of care in routine practice. In the case of the TiME trial, the risk determination should be made by comparing the known benefits and harms of receiving longer treatment sessions to the known and distinct benefits and harms of shorter treatment sessions.
Second, interventions within the standard of care may lack evidence, may have divergent profiles of benefits and harms and may pose serious risks to certain individuals.42 Therefore, if the interventions are within the standard of care, researchers and RECs should nonetheless evaluate the known or expected differences in burdens associated with the interventions, and ‘if there are sufficiently large differences in the expected burdens between the two arms of a (pragmatic trial), a study should be deemed to exceed minimal risk’.43 If we consider the MyTEMP trial, it is irrelevant to the determination of the risk level that the intervention could be implemented by a dialysis centre director. The comparison should be the expected burdens of receiving cooler dialysate compared with standard dialysate temperature.
An assessment of research risks also requires consideration of how the care trajectory of an individual may be altered as a result of trial participation. This means that research and RECs should evaluate the ‘source of risk that is attributable to participating in a study compared with not participating in it’.43 To make this comparison, Kane et al provide a useful, although controversial, heuristic: “Where usual care trials are aimed at measuring outcomes that entail large gains or losses of welfare, they cannot generally be presented as minimal risk”.44
Social value ‘refers to the importance of the information that a study is likely to produce’,6 where the importance of the information is determined either from its relevance to ‘understanding or intervening on a significant health problem or because of its expected contribution to research likely to promote individual or public health’.6 We do not wish to contend that pragmatic CRTs per se have limited social value because of the risks of bias. Of course, there are contexts in which pragmatic CRTs may be preferable to other designs. But even well-designed pragmatic CRTs can lack social value; for example, if ‘its endpoints are unrelated to clinical decision-making so that clinicians and policy-makers are unlikely to alter their practices based on the study’s findings’.6 Therefore, RECs should carefully consider whether the anticipated social value of a pragmatic CRT is important enough to justify an alteration or waiver of consent requirements.
Even if it can be argued that a pragmatic CRT constitutes minimal risk research and has important social value, researchers must also demonstrate that the trial would be infeasible to conduct with informed consent to be granted an alteration or waiver of consent. The argument for the infeasibility of consent is straightforward for pragmatic CRTs evaluating cluster-level interventions: when the intervention is delivered to the cluster, rather than cluster members, it is difficult or even impossible for cluster members to avoid the intervention and, as a result, their refusal of informed consent is meaningless.45 In other words, requiring informed consent would create expense and inefficiency without materially furthering the goal of showing respect for participants’ autonomy.46
Researchers and RECs must be wary of similar arguments applied to pragmatic CRTs evaluating individual-level interventions. For many pragmatic CRTs such as the HiLo, MyTEMP and TiME trials, a consent process for individual-level interventions will create additional expenses and inefficiencies but, importantly, would further the goal of respecting the autonomy of participants whose care is modified for research purposes. It may be that in some pragmatic CRTs, a consent process for an individual-level intervention such as which antiseptic solution to use during preparation for surgery, might be outside the scope of patient autonomy. But, since individual-level interventions are, by definition, directly delivered to cluster members, it is theoretically feasible to obtain their consent at the time of intervention delivery. Moreover, when asked, patients prefer that their consent be obtained in pragmatic trials, even those evaluating interventions used routinely in clinical practice.47 Therefore, appealing to increased costs or risks of bias introduced by a recruitment process are not, in and of themselves, sufficient for justifying the infeasibility criterion. If cost and bias were sufficient reasons, the implication is absurd: informed consent requirements should be dispensed for all research given that recruitment procedures always require additional resources and, for research involving postrandomisation recruitment, increase risks of bias.
The perceived infeasibility of consent in pragmatic CRTs may be caused by the belief that lengthy discussions and written documents are necessary features of informed consent. Researchers and RECs may not be considering alternative models of consent that are both consistent with pragmatism and the goal of respecting patient autonomy. For example, integrated consent—wherein health providers discuss the study with their patients and document verbal consent in their health records—can facilitate an efficient recruitment process without disturbing the workflow of busy clinics.48 When standard written informed consent administered by a research coordinator is demonstrably infeasible (for reasons other than, or in addition to, costs and risks of bias), the research has important social value, and research participation poses no more than minimal risk, researchers and RECs should first consider a simplified consent approach while keeping in mind that autonomy-promoting opt-in processes are preferable to opt-out processes.
Conclusion
We acknowledge the imperative to conduct high-quality randomised controlled trials. Patients are exposed to burdens and harms in clinical practice because of the paucity of evidence to support the decisions of health providers and systems managers in many settings. It, therefore, seems unjust not to conduct pragmatic, patient-centred research that has the potential to improve the quality and duration of patients’ lives. However, the enthusiasm for using CRTs to promote pragmatic aims merits pause and reflection. Following our stepwise analysis, researchers and RECs should carefully consider whether cluster randomisation is necessary, as individual randomisation can be designed to promote pragmatism while circumventing the complex ethical and methodological issues in CRTs. If RECs do not have the expertise to judge whether cluster randomisation is appropriate, they should ‘verify that a competent expert body has determined the research to be scientifically sound or consult with competent experts to ensure that the research design and methods are appropriate’.6
This is not to say that CRTs should never be used; rather, researchers and RECs should consider the ethical, practical and statistical implications of using cluster randomisation. It can be very difficult to design a CRT free of bias and, if these risks cannot be adequately addressed, it may be better to use individual randomisation. It should also not be assumed that pragmatic CRTs are low-risk studies. A careful assessment of the risks posed by research participation is required—even if interventions are used routinely in clinical practice. If the risks are minimal, using a waiver of consent for individual-level interventions is often imprudent. Obtaining consent is likely feasible at the time of intervention delivery or, if using research personnel to obtain written informed consent proves to be infeasible, simplified consent that mirrors routine clinical consent practices should be prioritised over waiving consent requirements entirely.
Data availability statement
Data sharing not applicable as no datasets generated and/or analysed for this study.
Ethics statements
Patient consent for publication
References
Footnotes
X @coryegoldstein, @charlesweijer
Contributors CEG, CW, MT and SND conceived of the manuscript. CEG wrote the first draft and revised based on feedback from CW, MT and SND. CEG is the guarantor of the manuscript and accepts full responsibility for the work.
Funding This work was supported by a Canadian Institutes of Health Research (CIHR) Project Grant (PJT-479757). CEG is supported by a CIHR Postdoctoral Fellowship Award.
Competing interests CW receives consulting income from Cardialen and Eli Lilly & Company. Other authors declare no competing interests.
Provenance and peer review Not commissioned; externally peer reviewed.