Challenges in conducting clinical trials in nephrology: conclusions from a Kidney Disease—Improving Global Outcomes (KDIGO) Controversies Conference

1Chronic kidney disease (CKD) affects ∼10% of adults in high-income countries,, and its treatment is burdensome and costly, accounting for substantial proportions of health budgets., It is now recognized as an important cause of death, particularly in low- and middle-income countries where diabetes is becoming common and resources for treatment are limited., Other conditions affecting the kidney, such as acute kidney injury, and kidney stone disease, also contribute a substantial burden of morbidity and mortality. Despite the high individual patient and societal burden of kidney diseases, the amount of reliable information available to guide kidney patient care is very limited. Although some treatments have been assessed in randomized trials, most of these have been too small to detect treatment effects of a magnitude that can realistically be achieved with a single intervention (e.g., reductions of 15%–20% in major outcomes such as death or disability). Experience to date indicates that conducting trials in patients with kidney disease on the sort of scale that has led to major therapeutic advances in other specialties (e.g., oncology, diabetes, and cardiology) is challenging.

2In order to discuss the most significant barriers to conducting trials in patients with kidney disease and to propose potential solutions, KDIGO (Kidney Disease: Improving Global Outcomes) convened an international multidisciplinary Controversies Conference in Paris, France, titled “Challenges in the Conduct of Clinical Trials in Nephrology” in September 2016. The meeting began with plenary talks that aimed to identify the key themes for discussion related to trial design (including the specific topics of how to measure kidney and nonkidney outcomes) and trial execution before 4 breakout groups considered the key issues in detail. After each of the breakout sessions, ongoing deliberations were reported and discussed in plenary sessions. This paper synthesizes the main areas of discussion, agreement, and remaining controversies addressed at the conference. The conference agenda and selected presentations from the meeting are available on the KDIGO website (http://kdigo.org/conferences/clinical-trials/).

3The discussion was based on the general principle that, in order to conduct a successful randomized trial, there are 4 main requirements: randomization of a sufficient number of patients (to ensure sufficient numbers of outcomes); assurance of adherence to the allocated treatment; reliable ascertainment of relevant study outcomes; and appropriate statistical analysis. These requirements can be met through a combination of sound trial design and efficient conduct. Much of the discussion at the conference centered on the concept of “streamlined” trial design, which focuses on the main determinants of trial quality and avoids unnecessary elements of design or conduct that increase cost and complexity.

4Trial design (analogous to experimental design) and trial conduct were considered separately. However, it should be noted that many aspects of a trial’s design influence its conduct and can therefore be decisive in determining whether a trial is successful. Because of their large scope, discussions about trial design were organized into 3 breakout groups that considered the design of trials in kidney disease, how to measure different kidney-specific outcomes, and assessment of other outcomes. The fourth breakout group focused on optimizing trial conduct. Table 1 brings together key findings from the breakout groups with regard to challenges in the design and conduct of randomized trials in patients with kidney disease. More complete details are provided in the following.

5The discussions identified important trial design principles that are generic but inconsistently applied in trials among patients with kidney disease, as well as issues that are specific to such trials.

6A key first step when planning a trial is to formulate the trial question. This should be both clinically relevant (i.e., addressing a major area of clinical uncertainty) and relevant to patients (i.e., aiming to avoid an outcome or condition that patients themselves consider significant). Patients should be involved in discussions when planning trials.

7An important underlying principle in trials testing superiority of an intervention is that the aim should always be to compare treatment arms that differ substantially in their expected effects on the primary outcome and to preserve this separation for the duration of the trial. The likely magnitude of this difference can sometimes be usefully tracked through assessing differences in a surrogate variable between randomized treatment groups (e.g., plasma low-density lipoprotein cholesterol in the SHARP trial), provided that blinding at an individual participant level is strictly enforced. Failure to maintain adequate adherence to study treatment leads to a loss of separation between groups and loss of statistical power. This may be a particular problem in trials among patients with CKD who are typically receiving multiple treatments, and are often required to attend clinics frequently, and are particularly prone to drug toxicity and intercurrent illnesses that may require the trial treatment to be modified (as observed in the EVOLVE trial in which study drug discontinuation was much higher than anticipated).,

8As a general principle, eligibility criteria should be practical and broad. This ensures a widely generalizable result and facilitates efficient recruitment. Inclusion criteria should identify a suitable population of patients who are likely to have the type of outcome that the treatment is anticipated to prevent. Guideline committees may help to maximize the size of eligible populations by highlighting areas of uncertainty, by avoiding making recommendations where evidence is weak, and by stating where placebo-controlled trials are needed.

9Exclusion criteria should be constructed that exclude patients who have a definite indication or contraindication for at least 1 of the study treatments, who are likely to be nonadherent to the trial protocol, or who are not expected to survive for the duration of the trial. Therefore, patients who would have obvious safety issues (from relevant comorbidity) or who are at high risk of a potential pharmacologic interaction related to the intervention should generally be excluded (although this might not always be necessary if a Data and Safety Monitoring Board is charged with monitoring safety in specific patient groups). An important group to exclude is those who are likely to stop allocated treatment because of intolerance or nonadherence (“drop-out”) or who are likely to start the study drug (or a drug with a similar mode of action) outside of the trial (“drop-in”). It may be possible to identify patients who are likely to be nonadherent during a prerandomization “run-in” period so that they can be withdrawn before being randomized. This helps to preserve study power. Run-in periods may be particularly beneficial in trials of dialysis patients, a complex group for which maintaining adherence to generally well-tolerated medications can be difficult.,

10Kidney disease has a wide variety of causes, and many different pathophysiologic mechanisms are responsible for disease initiation and progression, as well as for its complications. This makes it unlikely that a single treatment would have a large relative effect on major outcomes (particularly on outcomes such as death or progression to end-stage kidney disease). In determining sample size, therefore, particularly for large Phase 3 outcome trials, it is important to avoid overly optimistic assumptions about treatment effect size, even when there are apparently large effects on disease biomarkers. In practice, this means that relative risk reductions in major outcomes of more than ∼20% are unlikely to be observed. Nonetheless, in high-risk patients such as those with kidney disease, small relative risk reductions may still correspond to large reductions in absolute risk and therefore be clinically and economically worthwhile.

11It is inappropriate to estimate relative risk reductions (for major outcomes) from systematic reviews composed solely of small randomized trials; such trials can only achieve statistically significant results (and hence publication) if—by the play of chance—their effect estimates are larger than the truth, which means that such reviews will tend to yield inflated effect size estimates. When the evidence from previous small trials is limited in this way, a trial that is capable of detecting a realistically moderate effect (e.g., a 15%–20% reduction in a major outcome) is more rational than one that aims to detect an effect estimated in a systematic review. When designing kidney trials, there may be considerable uncertainty when estimating event rates owing to a lack of contemporary data from population-based studies. In addition, event rates derived from population-based registries may not reflect those of an enrolled trial population. The potential for high levels of nonadherence to study treatment, which can have a particularly detrimental effect on study power, should also be considered. Adequate drug exposure and study power may therefore be achieved by planning on a predefined total number of primary outcomes and a minimum duration of follow-up in a study population of approximately the correct size. A Data and Safety Monitoring Board can advise on early termination of a trial if convincing evidence of harm or efficacy emerges before trial completion or if there are other reasons (e.g., irremediable failure to recruit or futility) that make it inappropriate to continue.

12The principles of statistical analysis for randomized trials are well documented and should be applied in trials among patients with kidney disease. For example, intention-to-treat analyses should be specified for the primary analysis. Several particular problems are commonly encountered. First, where sample sizes are limited, there has been a temptation to overinterpret subgroup findings or where nonadherence is a problem, to conduct potentially biased on-treatment analyses. In superiority trials, such analyses should be considered exploratory rather than confirmatory, except where the analysis plan provides a clear justification for them before database lock and consideration has been given to the number of comparisons being made.

13In reporting the results of a clinical trial, the primary outcome should be emphasized, even if the results do not support the intervention. However, the full interpretation of a trial’s results should relate to the totality of evidence, meaning the primary outcome plus secondary and safety outcomes. If evidence is indeed inconclusive, this is an acceptable and valuable conclusion, and further randomized trials may still be useful if uncertainty remains.

14The choice of outcome measures in trials designed to assess kidney disease status (i.e., function, damage, or disease activity) should depend on the disease setting and phase of clinical development. Consideration should be given both to the stage of the disease and how rapidly it is progressing. Table 2 provides a matrix of general approaches to and strategies for selecting endpoints. The measurement of activity of disease and kidney structure, if available, may add useful information to measures of kidney function. Disease-specific markers (or outcomes) that reflect the underlying pathophysiology or molecular pathways of disease may be helpful in the setting of primary kidney diseases such as systemic lupus erythematosus, IgA nephropathy, and polycystic kidney disease. However, in settings where progression of kidney disease is a multifactorial process, such as with hypertension or diabetes mellitus, a kidney-specific outcome may be best assessed using the estimated glomerular filtration rate (eGFR).

15When outcome measures reflect the underlying pathophysiology of a disease, the effects of an intervention may be large enough to detect reliably in small to medium sized trials. Therefore, surrogate outcomes are suited to smaller Phase 2 clinical trials used to establish proof-of-concept, optimal drug dose, and information on tolerability. Markers of glomerular or tubular damage, inflammation, fibrosis, etc., or a combination of markers may be considered at this phase. A demonstrable difference in the average eGFR (or change in the eGFR) between treatment groups may also be possible in small- to medium-size trials.

16In larger Phase 3 trials, it may also occasionally be appropriate for outcome measures in support of the primary outcome to include measures of structural damage or disease-specific markers. Acceptable surrogates include measuring total kidney volume in autosomal dominant polycystic kidney disease., However, such trials must be sufficiently large to also provide reliable information on the safety of the intervention. Changes in eGFR over time may remain a more practical and acceptable method for assessing progression of kidney disease in many trials (Table 2).

17Directly measured GFR may occasionally be necessary if a treatment might influence variables used to measure eGFR through mechanisms other than effects on glomerular filtration (e.g., muscle mass changes or tubular secretion of creatinine). In the pediatric population, most clinical research is conducted in the setting of rare diseases, which, by definition, have small disease populations. Even so, studies in pediatric populations are subject to the same principles for evaluating progression of CKD as in adults. However, using creatinine as a marker of GFR is problematic in children younger than 2 years of age because creatinine levels rise rapidly during infancy. The European Medicines Agency provides guidance on extrapolating efficacy and safety data from adults to children to inform pediatric investigation plans.

18An important current issue is whether change in albuminuria is acceptable as a surrogate marker of CKD progression. In the context of nephrotic syndrome, large changes in albuminuria are an acceptable marker of kidney disease activity. Albuminuria may also be appropriate in the setting of structural damage, but it may not be in the setting of hemodynamic dysfunction or acute reversible disease. Albuminuria may also be an appropriate surrogate if there is evidence that the effects of treatment are durable. Possible endpoints for evaluating treatments include prevention of incident macroalbuminuria, remission from macroalbuminuria to normoalbuminuria, and a predetermined decrease, such as a set quantitative change. More data are needed to better understand whether and how changes in albuminuria correspond to disease progression and how changes can be meaningfully applied to designing trial endpoints. The US National Kidney Foundation, the US Food and Drug Administration, and the European Medicines Agency will convene a meeting on March 15 to 16, 2018, to discuss these issues.

19Outcomes that measure aspects of disease status other than kidney structure or function may be disease-specific morbidity or mortality and can include assessments of quality of life or everyday functioning or economic impact. Outcomes should capture the expected, plausible treatment effects (benefits and harms), be relevant to patients and health care providers, and be appropriate for the phase of clinical development. Excess burden to patients, their families, health care providers, and research staff should be avoided. When available, measurement instruments that have operating characteristics known to be within acceptable limits should be used.

20In determining outcome measures, patient and caregiver perspectives should be sought and considered for all clinical trials in nephrology. Standard definitions for key core outcomes identified as being important to patients on dialysis or with kidney transplants are being developed by the Standardised Outcomes in Nephrology initiative., For example, outcomes that the Standardised Outcomes in Nephrology initiative has identified to be important to patients on hemodialysis include mortality, functioning of vascular access, and cardiovascular disease, but also include symptoms such as fatigue, pruritus, cognitive function, and functional limitations. The US Food and Drug Administration has provided specific guidance on how Patient Reported Outcomes, which include assessments of quality of life, can be developed and validated to assess these symptoms to support labeling claims. Examples for which streamlined Patient Reported Outcomes may be particularly important include trials of treatments for anemia (alongside outcomes related to clinical safety and efficacy).

21Composite outcomes should combine components that make sense for the specific intervention, patient population, and disease state and should be of approximately comparable clinical importance. All-cause mortality is rarely an appropriate primary outcome in kidney trials because it is neither sensitive to any real effects on particular causes of death nor generalizable to different types of patients. It is preferable to create composite outcomes comprising related events that are all likely to be influenced favorably by treatment (and assess the effects on safety outcomes separately). If kidney and cardiovascular outcomes are to be combined in a single composite primary outcome, it is important to ensure that sufficient information will be available on both disease components to be able to guide treatment decisions. Similarly, the use of co-primary endpoints (for which an effect has only to be demonstrated on one of them) is not generally appropriate unless each is in some way relevant to patients and analyses of all such endpoints are adequately powered.

22When available and appropriate for the particular clinical context under study, standardized disease outcome definitions that are feasible to apply at scale should be considered. However, if needed, new definitions should be considered in special situations such as heart failure in the context of dialysis. When a composite outcome is used, it is important to assess the effects of treatment on its components and related outcomes. Analyses of events that recur (e.g., vascular access procedures or hospitalization for heart failure) may be important to patients and payers, and appropriate statistical methodology to analyze recurrent events is available. Where continuous measures are used, clinically important differences should be defined and justified.

23Instruments assessing health-related quality of life can assess the economic impact of a treatment to inform payers. Application of such instruments should follow the same principles as those for other patient populations without kidney disease, and appropriately streamlined methods for gathering such data should be included in trial designs.

24Increasing the number of large, important, and relevant clinical trials will require a culture shift within the nephrology community. A multipronged approach is required to improve study conduct and help all stakeholders to better understand the ways in which high-quality clinical trials improve patient care. One practical solution for streamlining the conduct of trials is integrating research processes and procedures into routine care (as has been done so successfully in oncology, diabetes, and cardiology), and success in this depends on improved community awareness.

25Inefficiencies in clinical trial conduct jeopardize the ability to address important clinical questions and are a disservice to trial participants. Table 3 lists potential strategies to improve the efficiency and effectiveness of clinical trials in nephrology and in other disciplines. Such strategies would be expected to enhance the rights of participants (e.g., by ensuring that consent procedures provide information in an accessible form) as well as their safety and well-being (e.g., by minimizing the requirements for study visits and invasive tests or by more effective methods of pharmacovigilance). This work builds on the Quality by Design approaches developed by the Clinical Trials Transformation Initiative.,

26Increasing participation in clinical trials is a major goal and requires a range of strategies (Table 4). Patients and clinicians should be made aware of the value of research participation and the importance of randomization. This could be done through education via videos, webinars, targeted advertising strategies, or peer group discussions. Educational efforts, both formal and informal, need to be dedicated, consistent, and constant. Patient advocacy groups for rare diseases have been successful at engaging patients and providers about the meaning and value of research and are a potential resource for ideas and collaboration.

27Processes for informing patients and health care team members about specific clinical trials should be systemically embedded in health care communities. At individual centers, repositories of current and planned studies should be accessible via various internet portals, and information about studies should be displayed in the waiting areas and offices of medical facilities. For recruiting into specific trials, it may be helpful to institute a process of “prescreening” whereby research coordinators develop lists of potentially eligible patients and, where permissible, provide those patients with information about the trial and “preconsent” them. This then enables recruitment to proceed rapidly once the trial receives full ethical and regulatory approval (and also helps to identify sites that will not have sufficient patients to contribute). Electronic health records may provide an opportunity to identify large numbers of potential participants who may be eligible and should be invited. Widespread invitation ensures that patients are empowered to decide whether they want to participate in a trial rather than waiting for their doctor to hand-select them.

28Research champions within countries and regions should be recognized and identifiable. Clinical research organizations, academic research organizations, and networks of trialists should be encouraged to share information about the enrollment performance of individual study sites. This could speed completion of enrollment and reduce wasting of resources on poor performing sites. Excellence in research should also be recognized by national audits and by payers, with rewards given both for randomization and success in achieving quality data and completeness of follow-up.

29The amount and type of data collected in a clinical trial affect the ability to recruit patients, follow their progress, and complete the trial. Inefficiencies in data collection increase trial costs, labor, and burden of participation (for both participants and the research team). To date, trials in patients with kidney disease have tended to collect too many data fields, most of which do not contribute to answering the main clinical question and lead to unnecessary complexity and difficulty in recruiting., Researchers should identify small core datasets required for each study, minimize the frequency of measurements, and simplify the collection of data. The specific processes developed for an individual study will vary depending on the study’s purpose, type of intervention, available resources, and stage of development.

30Well-established national dialysis and transplant registries in many countries provide an opportunity to streamline data collection in trials. Focused kidney-specific templates that use consistent terminology, definitions, and sets of variables could be of value if these templates were integrated into registries or electronic health records.

31Depending on the circumstances, expensive central laboratory analyses may not be essential where an outcome is a measure of difference in a biomarker between randomized arms. Variation in calibration for a biomarker between laboratories has little impact on the magnitude of differences between randomized groups. For example, in the SHARP trial, analyses of differences in routine plasma (or serum) creatinine measured every 6 months at local hospital laboratories allowed low-cost, but reliable, assessment of the effects of ezetimibe/simvastatin on the progression of CKD. When using local laboratories, however, it is important to know their reporting units, reference ranges, and analytic methods to ensure that summary analyses are meaningful.

32There is increasing evidence that verification of clinical outcome data (usually referred to as outcome adjudication) may have little effect on the relative risk reductions reported by trials and that this process could also be streamlined in certain situations.

33Approaches such as using “run-in” periods to identify patients who are unlikely to adhere to study treatment and/or attend clinics, minimizing unnecessary data collection, limiting excessive numbers of study visits (perhaps by arranging follow-up by telephone or electronic health records where possible), and expediting in-person visits (e.g., by avoiding lengthy waits in the hospital pharmacy) may all help maintain adherence and follow-up (Table 1). Adherence should be monitored centrally, and each study treatment “dropout” should prompt investigation into the reason and discussion as to whether study treatment can be restarted.

34In clinical studies, the term withdrawal of consent is problematic because of its lack of specificity and because it is frequently confused with a participant’s wish to stop study treatment or not undergo a certain trial procedure. Specific levels of withdrawal from the protocol-specified follow-up, which range from a patient not attending clinic visits but perhaps agreeing to clinical note review to an absolute withdrawal with no further data being provided, should be embedded in case report forms. To better capture follow-up data for patients who are no longer participating in a study, patients could be asked to agree that electronic health records can be accessed. This would allow for capturing information about their outcomes, adverse events, and concomitant medications during and after study participation. Engaging general practitioners and primary care providers could also help in establishing streamlined methods for complete study participant follow-up.

35Trial monitoring can be time-consuming and resource intensive, and therefore simplifying these processes can have a profound impact on reducing the burden of the trial conduct. When using direct electronic data entry, processes for trial monitoring and source data verification can become much more efficient. For example, risk-based approaches to monitoring (focusing on those data that are critical to trial quality) and central statistical monitoring (using study data to identify unusual patterns of performance) can reduce and prioritize site visits.

36Safety reporting should be tailored for each trial protocol. In early phase development, rigorous detailed adverse event ascertainment is necessary, but this level of event recording may not be necessary when the safety profile of a treatment is well-known. During protocol development, regulators such as the US Food and Drug Administration and European Medicines Agency can advise on which specific adverse events need to be collected and which do not. Regulators can also advise on the level of information that needs to be collected. Clinical narratives are burdensome, may reduce trial participation, and should be focused only on those serious adverse events where such data may be informative, such as suspected unexpected serious adverse reactions. Reliable review of safety during a trial is best achieved by examining randomized comparisons of aggregated data by the unblinded Data and Safety Monitoring Boards.

37The lack of adequately powered randomized trials in nephrology has led to a problematic imbalance between the clinical need of patients with kidney disease and the amount of reliable evidence to inform practice. This KDIGO conference highlighted some of the key challenges faced by those trying to perform large trials. These include a lack of uncertainty among nephrologists who have often adopted treatments before adequate evidence of efficacy or safety is available, smaller treatment effects than were predicted from effects on surrogate biomarkers, inappropriate selection of outcomes including those with little relevance to patients or unlikely to be affected by treatment (e.g., all-cause mortality), difficulty in identifying large numbers of eligible patients, high levels of nonadherence to study treatment by overburdened patients, and overcomplicated trial conduct (Table 1). Adoption of the approaches outlined in this report has the potential to dramatically improve the quality of clinical trials in nephrology and substantially enhance the evidence base for the safe and effective treatment of patients with kidney disease.

38This conference was sponsored by KDIGO and was in part supported by unrestricted educational grants from Abbvie, Achillion, Akebia Therapeutics, Alexion, Amgen, AstraZeneca, Bayer HealthCare, Fresenius Medical Care, KBP Biosciences, Keryx Biopharmaceuticals, Merck, Omeros, Relypsa, Roche, and Vifor Fresenius Medical Care – Renal Pharma.