Controlled clinical trials are a cornerstone in the search for medical therapeutic advances. Since the introduction of this concept, there has been substantial progress in the design, conduct, and analysis of such studies, and many have become quite complex. A growing awareness of ethical issues emerging from such trials has led to increased public scrutiny and dialogue between patients, investigators, institutions, and funding agencies. The recent broader role of industrial/financial sponsorship of such endeavors has led to new issues and challenges for all interested parties.
The Data and Safety Monitoring Board (DSMB) is now considered a critical component of the administration and conduct of large multicenter clinical trials. Early cessation of large trials because of major advantages or untoward effects of the treatment under study has recently underscored the pivotal internal peer review role of the DSMB. Although the general need for treatment monitoring has been recognized by the National Institutes of Health since at least 1979, surprisingly little information exists concerning terms of reference, authority, composition, and operating procedures by which such groups function.1 It is the purpose of this essay to explore these issues further, with the intent of fostering a constructive dialogue and framework from which to construct DSMB guidelines.
Who Composes the DSMB?
Those charged with the responsibility of designing, leading, conducting, and sponsoring clinical trials must give special care to the structure and role of the DSMB.2 It should provide informed and independent advice to the principal investigator, steering committee, and sponsor. For the purposes of this discussion, the principal investigator of a trial will be presumed to be the chairman of the steering committee, and the word “sponsor” will refer to either a peer-reviewed funding agency or an industrial source. A DSMB chairman should be appointed jointly by the steering committee and the sponsor: this individual is independent and knowledgeable in the clinical trial’s area of study. In collaboration with the steering committee, he or she should be empowered to select an appropriately constituted, multidisciplinary group to monitor the trial’s progress. The group selected should be composed of physicians with extensive clinical investigative experience in the area of study, a biostatistician with substantial experience in trial monitoring who is independent of the study, an ethicist or an individual schooled in the social sciences or humanities to provide an additional societal dimension and perspective, and a biostatistician (nonvoting) representing the Data Coordinating Center in charge of compiling, analyzing, and preparing interim data summaries for the DSMB’s consideration. In anticipation of arriving at resolution of difficult issues, having an odd number of voting members may be prudent. The members of the DSMB should be free of conflict of interest as it relates to the sponsor and intervention, should have declared this in writing, and should receive only expense reimbursement and reasonable remuneration for time spent fulfilling their tasks.
What Is the Relationship Between the DSMB and Other Groups?
It is important to define from the outset the roles and relationships of the steering committee, principal investigator, sponsor, and Data Coordinating Center.3 Appropriate channels of communication, lines of responsibility, accountability, and authority are essential prerequisites to smooth and effective clinical trial conduct. Although some have argued that DSMB positions for both the principal investigator and sponsor are appropriate, compelling reasons suggest that their presence during DSMB deliberations may diminish the frank and open discussion that must ensue if all DSMB members are to adequately fulfill their responsibilities. Moreover, it is impossible to preclude that the thinking and actions of both principal investigator and sponsor would be influenced by access to all trial information and discussions developed at periodic DSMB meetings. A compromise that usually meets the needs of all parties is selective representation and participation by the principal investigator at open sessions of DSMB meetings. This ensures that matters of concern, questions and developments as they relate to planning and policy of the trial, and unforeseen evolving issues can be brought forward appropriately. Similarly, the DSMB, through its Chairman, may have reason to engage the principal investigator in a dialogue on a periodic basis or after interim meetings. Whether the DSMB relates to the sponsor through the principal investigator and Steering Committee or directly (particularly in non–industry-sponsored trials) is a moot point that should be subject to negotiation at the outset of each trial, recognizing that there may be a need for some individual tailoring on this point. These matters notwithstanding, autonomy of the DSMB and dissociation from steering committee and sponsor should be a clear and mandatory guiding principle that is appreciated to be in the best interests of all parties.
What Is the Function of the DSMB?
The primary responsibility of the DSMB is to ensure patient safety. In addition, as the clinical study unfolds, the DSMB must be satisfied that patients enrolled in the study are kept adequately appraised of the risks associated with entry into the investigation by means of informed consent. Having been satisfied before its initiation that the study addresses a meaningful question and that the protocol is designed in a scientifically and ethically valid manner, the DSMB should ensure that the anticipated rate of recruitment to the study is adequate to achieve the necessary frequency of end points of interest, thereby satisfying the ultimate study objectives. These techniques, often involving group sequential design methodology, must strike an appropriate balance between the capacity to detect major harm while at the same time avoiding substantial attenuation in the statistical power of the primary analysis.4 5 6 Guidelines for early termination of the study should be formulated and published by the DSMB at its first meeting before any data review. A group sequential design that accounts for interim analyses with a use function as proposed by Lan and DeMets,5 taken in conjunction with the totality of evidence available, provides the necessary framework from which the DSMB can make informed and prudent decisions. Particular care should also be taken to establish confidence in the Data Coordinating Center, the quality, accuracy, and timeliness of the data flow, and the confidentiality of access to the data, which are key elements of this process. Emerging from this assessment would be a plan for the DSMB to meet at appropriate time intervals to assess study progress. At these junctures, appropriate statistical techniques to define intergroup differences as they relate to the end points of interest should be employed. During interim analysis by the DSMB, attention should be given to ensuring that the data presentation highlights not only study end points but all known and potential unfavorable side effects. Such meetings also permit review of any suggested protocol amendments before their implementation.
What Are the Considerations Associated With DSMB Decision Making?
Perhaps the most critical decision the Data and Safety Monitoring Board may face is whether or not to recommend that the trial cease because of excess harm or benefit in a treatment group before its planned conclusion.7 It has been argued that a trial that suggests that a new treatment is either equally or less effective than conventional care should not be continued, because the distinction between whether it is ineffective or harmful is not required.8 This has led to the development of a “futility index,” defined as the conditional probability that the trial will fail to demonstrate a benefit of therapy, given the observed results. Use of such an approach can be achieved with minimal loss of statistical power. If the index increases during the evolution of a randomized trial, then the likelihood of achieving a positive result progressively declines. The purported advantage of this approach is that continuation of a trial deemed to be futile is wasteful of resources and potentially unethical. This view must be contrasted with one that suggests that if evidence for harm from an innovative therapy is not strong, dedicated advocates of the therapy may challenge the results arising from a trial that has been stopped earlier than planned, especially when the confidence limits associated with definition of the treatment effect are broad. Hence, still other patients could be exposed to unnecessary risk if another clinical trial were launched to investigate the same issue, thereby subjecting more patients to potentially unnecessary hazards. Clarity and precision on the issue of harm are especially important when the therapy has already been approved for general use, as was recently the case for flosequinan in heart failure.9
There are several examples of cardiovascular trials that were terminated earlier than planned. Three treatment arms of the Coronary Drug Project were terminated early in the 1970s because of adverse experiences10 11 ; these decisions to terminate were not controversial. In 1989, the Beta-Blocker Heart Attack Trial was stopped early because of strong evidence of a beneficial effect of propranolol on all-cause mortality.12 A favorable effect had been more or less expected, given the prior published report of benefit in the Norwegian Timolol Study.13 By contrast, the very early termination of the Cardiac Arrhythmia Suppression Trial (CAST) because of increased risk of arrhythmic death attributed to encainide and flecainide created a lengthy debate.14 The ethics of continuing this trial were questioned; however, the DSMB felt that overwhelming evidence was necessary to offset strongly held prior assertions concerning the clinical value of antiarrhythmics. Importantly, the CAST trial was also a caveat for proponents of surrogate end points, since suppression of ventricular arrhythmias did not accurately predict the effect on mortality. Monitoring of cardiovascular clinical trials over the past quarter century has provided important contributions to clinical research. Given the tenuous nature of many assumptions that modulate sample size calculations (such as the magnitude of the treatment effect), it is not surprising that some trials are terminated earlier than planned because of a greater-than-projected benefit, whereas others conclude in the face of unexpected major adverse events. The multiplicity of mechanisms of new drug actions sometimes conspires to offset the anticipated benefit of agents undergoing evaluation. It should also be appreciated that the patient monitoring associated with entry into a clinical trial and the scrutiny with which adverse effects are evaluated are more intense and comprehensive than those that normally occur in general medical care. Accordingly, unfavorable side effects have the potential for greater harm when applied to a broader patient population for which the therapy is ultimately intended and in whom surveillance of unfavorable side effects is less detailed.
DSMB members participating in such decision making must also take into account the consistency of the data and evidence emerging over the course of the study and be cognizant of random variations in event rates. Another key element for the DSMB to consider is the currency of the data under review and whether they typify an adequately representative cross section of the trial cohort to be studied. Moreover, the diversity of actions of a therapeutic agent under study and the nature of the disease process may conspire to produce an early clustering of harmful effects before emergence of late, long-term beneficial effects, ie, the treatment impact may vary throughout the planned period of observation. This highlights the need for caution and patience lest a decision to discontinue a trial be too precipitous. Complex therapy, such as certain surgical procedures, may be associated with a learning curve during implementation, with resultant diminished effectiveness and increased complication rates early in the experience of a new center in a multicenter trial. This phenomenon may also be manifest in complication rates associated with small- as opposed to large-volume centers. Finally, differing margins of benefit and harm in a variety of patient groups are a common challenge in large clinical trials; unless these groups are predefined from the outset, such analyses are unlikely to be useful in influencing the decision for early termination of the study. When comparing novel therapy with established treatment, it is important to realize that small absolute differences may be of major clinical significance in a disease that is widely prevalent.15
The nature, feasibility, cost-effectiveness, and general applicability of the treatment strategy inevitably arise as factors in DSMB deliberations. Hence, a therapy administered through an implantable infusion pump requiring frequent supervision and home care presents a degree of difficulty that would be offset only by a substantial improvement in symptoms or survival. These issues need to be framed in the context of the prevalence of the disease under study as well as its natural history, morbidity, and mortality. In diseases such as advanced cancer and heart failure, which cause substantial suffering as well as major mortality, one is ideally seeking a therapeutic strategy that would both relieve suffering and prolong useful life. Paradoxically, however, DSMBs may find themselves evaluating a trial in which suffering may be ameliorated but mortality not affected favorably but actually increased. Acquisition of such knowledge may be of importance and potentially clinically relevant, thereby underlining a need for a balance between statistical stopping guidelines, ethical perspicacity, common sense, and good clinical judgment.
If the terms of reference and composition of the board are vital to its successful collaboration in the conduct of clinical trials, so too is its operational strategy. Whereas a component of the DSMB meetings may appropriately be open to the principal investigator and sponsor, the major work of the DSMB should be conducted in strict confidence. It cannot be overemphasized that breaches of this confidence by any member of the DSMB can seriously undermine the successful conduct and overall integrity of the trial as well as its ultimate interpretation. Preparation of the data according to treatment groups with appropriate confidence limits should be undertaken in a comprehensible and user-friendly fashion according to a format discussed at the outset and probably analogous to figures and tables to be published in the final report. An adequate number of end points and length of time should have elapsed between meetings to provide enough information from which to make informed judgments. Although information presented to DSMBs has traditionally been identified as arising from treatment or control groups, a more balanced approach to data interpretation and trial monitoring may be achieved by blinding (where possible) treatment assignment. When this approach is used, DSMB members are disciplined to assess similarity or differences between two groups of unknown identity and then required to consider how their judgment would be affected if the strategy under investigation were assigned to one or another group. At any point in the progress of the trial, the DSMB may choose to unblind itself if separation of the groups is sufficient to contemplate a recommendation or special analysis, change, or cessation of the trial. In the event that a harmful effect is evident, the threshold for earlier termination than planned may be lower than when benefit occurred. If effective therapy for the condition under investigation already exists, then a more stringent threshold for harm can be incorporated into the trial by use of an asymmetrical statistical design as opposed to the more conventional parallel analysis.16
It is difficult to overestimate the importance of careful recording of the DSMB discussions and conclusions, since they may be of vital importance and heavily scrutinized at the conclusion of the trial by all interested parties, including regulatory agencies. Moreover, each DSMB meeting is greeted with anticipation and concern by the investigators, the sponsor, the steering committee, and local ethical review boards who are charged with surveillance of events at individual sites. Special care must be taken not to raise undue confidence or anxiety resulting from the DSMB deliberations, especially as they relate to the likelihood of early termination of the trial or emerging differences or lack of differences between the study groups. Ideally, unless there are concerns about the conduct of the trial, ie, the accuracy or timeliness with which information is conveyed from investigative sites to the data monitoring center and on to the DSMB or the rate of entry of patients and development of events, a terse written statement from the DSMB chairman to the principal investigator about the desirability of continuing the trial provides the best communication. In the event that the DSMB concludes that the trial should be terminated, it must communicate this recommendation rapidly and effectively to the steering committee and sponsor. In this circumstance, extensive discussion will likely be required among these three groups regarding the nature, weight, and consistency of the evidence that led to the recommendation. Authority for implementation or recommendation to temporarily suspend, stop, or modify a trial rests with the steering committee, which would be expected to act on it in concert with the sponsor with all deliberate speed.
The DSMB is in a unique position to provide the steering committee and sponsor with informed and incisive commentary about the trial results and their interpretation at the conclusion of the study. The collaborative role of the DSMB in this regard should not be underestimated, since it may enhance the overall interpretation, synthesis, and publication of the final report as written by the principal investigator and the steering committee.
Careful definition of the terms of reference, composition, reporting structure, relationships, and operational strategy of Data and Safety Monitoring Boards can further enhance the already crucial role they play in the monitoring and successful conduct of clinical trials.
It is a pleasure to acknowledge the many professional colleagues from whom we have learned concerning the issues discussed here and the editorial assistance of Sharon Campbell.
The opinions expressed in this editorial comment are not necessarily those of the editors or of the American Heart Association.
- Received September 29, 1994.
- Accepted December 12, 1994.
- Copyright © 1995 by American Heart Association
Gordon RS Jr, Chairman, National Institutes of Health Clinical Trials Committee: Clinical trial activity. NIH Guide for Grants and Contracts. 1979;8:29.
Colton T, Freedman LS, Johnson AL, Machin D. Practical issues in data monitoring of clinical trials. Stat Med. 1993;12:5-6.
Topol EJ, Armstrong PW, Van de Werf F, Kleiman N, Lee K, Morris D, Simoons ML, Barbash G, White H, Califf RM, on behalf of the Global Utilization of Streptokinase and Tissue Plasminogen Activator for Occluded Coronary Arteries (GUSTO) Steering Committee. Confronting the issues of patient safety and investigator conflict of interest in an international clinical trial of myocardial reperfusion. J Am Coll Cardiol. 1992;19:1123-1128.
O’Brien PC, Fleming TR. Multiple testing procedures for clinical trials. Biometrika. 1979;35:549-556.
Lan KK, DeMets DL. Discrete sequential boundaries for clinical trials. Biometrika. 1983;70:659-663.
Task Force of the Working Group on Arrhythmias of the European Society of Cardiology. The early termination of clinical trials: causes, consequences, and control. Circulation. 1994;89:2892-2907.
Packer M. Profile study: main results. Presented at the American College of Cardiology Annual Meeting; March 19, 1994; Atlanta, Ga.
Friedman LM, Bristow JD, Hallstrom A, Schron E, Proschan M, Verter J, DeMets D, Fisch C, Niew AS, Ruskin J, Strauss H, Walters L. Data monitoring in the Cardiac Arrhythmia Suppression Trial. Online J Curr Clin Trials. 1993;(79)930731.
Whitehead J. On the basis of maximum likelihood estimation following a sequential test. Biometrika. 1986;73:573-581.