Background The recent remarkable proliferation of multicenter clinical trials investigating novel treatments for acute ischemic stroke has created a new dilemma for stroke investigators. Promising agents and trials are many, competent clinical centers few. May a single clinical center, without scientific bias, participate in multiple, concurrent clinical trials recruiting overlapping populations of subjects with acute stroke?
Summary of Comment Many clinical stroke centers already participate in multiple overlapping trials. I delineate a variety of patient-allotment strategies that are presently used, although often not explicitly stated. Methods include assignment by time window, unique criteria, site, belief in agent efficacy, remuneration, and quota urgency. All have drawbacks, chiefly the introduction of bias into the selection of subjects for clinical trials, compromising the generalizability of trial findings to broader clinical populations. Several solutions are proposed that permit single centers to participate in multiple studies while minimizing or eliminating selection bias, including within-center use of a permuted-block enrollment strategy. The issue of trial allocation among multiple available studies has received insufficient attention from clinical trial designers, epidemiologists, and statisticians. Currently an inescapable problem for clinical stroke investigators, it arises in any clinical medical research field with rapid basic-to-clinical scientific advances.
Conclusions It is recommended that centers participating in multiple, concurrent clinical trials (1) use an articulated, formal method of patient allocation that is blinded or indifferent to clinical variables; (2) avoid assignment solely according to time since onset or unique patient criteria; and (3) avoid any influence of belief in agent efficacy on allocation.
I wish to draw attention to and explore a new methodological dilemma that the recent profusion of stroke treatment trials has created for active stroke investigative centers: how to avoid or minimize study bias when assigning patients among competing trial protocols. It is our privilege to work in the fourth great epoch of stroke investigation, the age of treatment innovation.1 One of the characteristic features of this era that surely would have astonished and pleased our predecessors is the abundance of promising therapies under active investigation for the amelioration of acute ischemic stroke. Phase II or phase III clinical trials of novel agents are legion and include standard antithrombotics, novel anticoagulants, thrombolytics, fibrinogenolytics, calcium channel blockers, excitatory amino acid receptor blockers, free radical scavengers, anti-inflammatory monoclonals, and neuronal membrane regeneration enhancers. A partial registry of US clinical trials listed 63 stroke clinical trials underway at the end of 1994 (personal communication with G.R. Houser, National Stroke Association, November 1994), many of which overlap to some extent in drawing on the same pool of patients who present to investigating sites within 3 to 48 hours of onset of acute cerebral ischemia. Individual centers presently participate in up to eight acute ischemic stroke studies concurrently.2 3 4 However, the issues raised by simultaneous participation in multiple, overlapping trials have been underrecognized and underexplored. A thorough search of MEDLINE 1964 through 1994 and of major texts and monographs on clinical trials revealed no directly relevant analysis of this issue.5 6 7 8 9 10 11 12 13
It is important to distinguish the problem of overlapping protocols from other multiple trial issues to which it bears passing resemblance and which have received greater analysis. The general prohibition against enrolling one patient simultaneously in separate trials of two differing interventions (eg, a thrombolytic and a neuroprotective agent) is widely acknowledged. Deciding which agent is responsible for any resulting beneficial or adverse effect poses insuperable difficulties. A related issue that has drawn less, but still some, attention from statisticians and epidemiologists concerns trials in which one patient is allocated sequentially to more than one type of therapy.14 This issue arises in studies that require, for example, randomization among two “induction” regimens and among two subsequent “maintenance” regimens.
In contrast, the question addressed here is how best to assign a patient to one among several concurrently available trials for which the patient is eligible. For example, a stroke trialist at stroke center X is called to the emergency department at midnight to see a 50-year-old woman with new-onset atrial fibrillation and symptoms of right middle cerebral artery ischemia noted 30 minutes earlier. On shuffling through an ever-ready card stack of clinical-trial inclusion and exclusion criteria, the physician/investigator realizes that the patient is potentially eligible for several ongoing randomized trials in which stroke center X is currently participating, including a 3-hour-window trial of intra-arterial thrombolysis, a 6-hour-window trial of an excitatory amino acid blocker, and a 24-hour-window trial of a low-molecular-weight heparinoid. How should the investigator decide which trial to discuss with the patient and family?
The major problem facing investigators participating in multiple studies is the introduction of bias into clinical trials. Conscious or unconscious bias may distort the selection of subjects, producing a subset of enrolled patients unrepresentative of the general population. This distortion may limit the generalizability of trial findings to broad clinical communities and, in the extreme case, cause a trial to produce false-positive or false-negative results. Currently, assignment of patients among multiple candidate protocols is carried out by various schemes, often not fully articulated, that harbor underrecognized potential sources of bias. These methods include assignment by time window, unique criteria, site, belief in agent efficacy, remuneration, and quota urgency.
Current Allocation Strategies
One common strategy among trial centers is to participate in several acute stroke trials, each with a different time target for initiation of therapy after symptom onset. In accord with their presumed mechanism of beneficial action, thrombolytic and neuroprotective agents generally have briefer windows for entry (90 minutes, 3 hours, 6 hours), while antithrombotic and regeneration agents allow longer intervals (24 hours, 48 hours). A center may choose to assign all patients presenting within 3 hours to a hyperacute thrombolytic trial, those between 3 and 6 hours to an acute neuroprotective trial, and those between 6 and 24 hours to an urgent antithrombotic trial.
There are overt and covert sources of bias in this assignment strategy. Clearly, centers following this strategy systematically underrecruit patients presenting very early into lesser urgency trials by steering all patients presenting immediately after stroke onset to the most hyperacute trial. Underrepresentation of early epochs in longer window trials may prevent identification of a subgroup with distinctive benefits, or hazards, from the tested intervention. Less obvious potential sources of bias in this strategy include time shifting. An investigator who feels a particular patient is a better candidate for the therapy of a less urgent rather than urgent trial, consciously or unconsciously, may not pursue the rapid evaluation required for hyperacute trials with customary vigor; the investigator thus allows a patient to become eligible for randomization only after the deadline for the more urgent therapy has elapsed, which gives the misleading appearance that the patient was a candidate only for a less urgent trial. Additionally, investigators may defer evaluating patients who present at an inconvenient hour for a hyperacute trial but proceed later in the day with enrolling them in a less urgent trial. This may cause a hyperacute trial to fail to identify particular risks or advantages of late-night or early-morning therapy.
Acute stroke trials frequently differ in the breadth of their inclusion and exclusion criteria, with varying age cutoffs, coexisting conditions allowed, concurrent medications permissible, and stroke subtypes sought (large-vessel middle cerebral artery only, anterior circulation only, nonlacunar only, all ischemic strokes). Some stroke centers participate in concurrent studies with narrow and broad criteria, and they enroll all patients meeting narrow criteria in the more demanding trial and patients ineligible for the stricter study in a more broadly inclusive trial. The perils of this approach are similar to those of allocation by time window. The chief problem is that patients with certain traits, such as younger age range or fewer concomitant medical conditions, are preferentially diverted from the broader to the narrower eligibility trial, possibly preventing the broader trial from fully analyzing a subgroup with differential benefit or risk.
Several stroke centers comprise a number of component hospitals and place different trials at different facilities. All patients presenting to a Veterans Administration hospital may be entered in one trial, and all patients presenting to an affiliated community hospital may be enrolled into another. This provides a center with the scientific benefits of participating in multiple protocols and the flexibility of coping with trial terminations by extending remaining trials to additional hospitals, without operating overlapping protocols at the same institution. While this strategy has many attractive aspects, it is not free of potential sources of bias. One major source of difficulty is the handling of non–emergency department patients referred to the stroke center. An investigator learning of an acute stroke patient by telephone from a local primary-care physician, outside emergency department, or direct patient call must suggest to which hospital the patient be transferred, with choices among contending protocols available at different sites again raising opportunities for selection bias. Moreover, referring physicians (or informed patients themselves) may skew enrollment by preferentially directing acute stroke patients to a hospital with an intervention they favor. Community neurologists not infrequently prefer to transfer young acute stroke patients to a hospital performing a thrombolysis trial rather than one engaged in a neuroprotective study. As an outside-referral population is likely to differ in many important demographic and clinical aspects from the local catchment-area population, skewed referral patterns may weight trials disproportionately with subjects who have distinctive responses to trial agents.
Belief in Agent Efficacy
From among several available trials, investigators may choose to offer a patient one that tests an agent they feel is especially promising for that patient’s clinical situation. A patient with elevated blood pressure may be offered a trial of a neuroprotective rather than an anticoagulant agent; a patient with a history of minor cardiac disease may be offered a trial of a neuroprotective agent without known cardiac side effects rather than a study of a neuroprotective agent that prolongs the QT interval; a young patient may be offered a thrombolytic trial perceived to be riskier but having a greater potential impact on outcome rather than an anticoagulant or neuroprotective study. The obvious drawback of this allocation method is that assignment reflects an investigator’s a priori biases about the efficacy or appropriate clinical role of agents being tested, the precise issue the trial is designed to define. Selective steering of one class of patients into one particular trial enriches that trial and depletes other trials of a subgroup who may actually benefit or experience harm from the tested agents in ways unforeseen by the investigator, leading to false-positive or false-negative findings. While few stroke centers deliberately set a policy of allocation by belief in agent efficacy, many fail to specify a formal means of allocation that automatically determines for the investigator to which trial a patient should be assigned. The resulting vacuum, in which assignment is left up to the enrolling investigator, may allow beliefs about agent efficacy to imperceptibly or subtly influence an investigator’s trial allocation decision. Any allocation in which trial assignment is not formally specified in a manner blinded or indifferent to the particulars of a patient’s clinical condition must be considered prima facie suspect.
Remuneration and Quota Urgency
Clinical centers may choose among concurrently available trials by assigning patients to the study that provides greater reimbursement of center costs or the study for which the center is less close to filling its minimum enrollment commitment to remain an active site. Paradoxically, while these methods of trial allotment seem to spring from baser motives than a belief in differential agent efficacy for a particular patient, they actually have the advantage over the latter method of at least not selecting patients in accordance with an a priori belief about the answer to the very hypotheses the trials are designed to test. The more minor defects of allocation by remuneration or quota urgency reside in providing incentives for centers to enroll patients of borderline eligibility to receive more substantive remuneration or meet pressing quota deadlines, encouraging overliberal interpretation of inclusion and exclusion criteria. These methods also offer no guidance in choosing among trials of equal reimbursement of costs or equal quota urgency.
Toward a Solution
Although present methods of assignment among competing trials all have deficiencies, several acceptable and easily instituted approaches to the concurrent trial dilemma may be delineated.
The purest approach is for stroke centers simply not to participate in more than one trial at a time. Some recent trials have indeed placed a ban on participating sites taking part in any overlapping studies. The drawbacks of this approach are several. First, the United States presently suffers from an acute shortage of qualified stroke trial sites, that is, experienced centers with the expertise, manpower, and commitment to carry out often time-consuming and ponderous clinical studies. Laudable efforts to recruit and organize nontraditional trial centers, run by emergency physicians, primary-care physicians, or nonacademic neurologists, have met with only limited success. Efficient testing of the many promising agents now entering the clinical study arena can occur only if stroke centers participate in overlapping trials. Also, single trials rarely provide reimbursement sufficient to sustain the costs of a clinical research service and leave a center vulnerable to sudden loss of funds and dispersal of acquired expertise should the particular trial be unexpectedly halted. A more enlightened national funding mechanism for stroke trial sites, which would provide sufficient reimbursement of costs within each trial and monies to sustain centers between trials, is to be wished for devoutly but is unlikely to transpire in the current funding environment. Center viability consequently often requires participation in multiple trials.
Fortunately, a variety of methods of avoiding or reducing selection bias among multiple trials are available, once it is recognized that the problem of assigning patients at a single stroke center among concurrently available trials possesses fundamental structural homologies to the problem of allocating patients to different arms of a small clinical trial, a problem that has received much attention in the clinical trial and statistics literature. While simple randomization generally produces balanced assignments among treatment arms in large trials, randomization may produce serious imbalance in trials with a small number of subjects. To overcome this difficulty, a variety of strategies have been developed15 16 17 that may also be applied to the analogous problem of a stroke center seeking to assign its relatively small number of trial candidates among competing stroke trials.
Assignment by fixed schedule, such as enrolling alternating patients in alternating trials, has several attractive features, including assignment by a simple, formal method and assurance of balanced allocations. The theoretical drawback of this method is that investigators are aware beforehand of which trial the next patient is to be offered. If an investigator or referring physician has a priori beliefs about patient traits that are optimal for each agent, they may more aggressively enroll or refer patients with those traits when they know a particular trial is next in the sequential enrollment schedule. This potential source of bias is minimized but not fully eradicated by the awareness of investigators and referring physicians that, within each trial, patients are randomized among active and control arms, so receipt of a desired agent is not guaranteed by enrollment in a particular trial.
An individual center could choose to randomize eligible patients among concurrently available trials. After initial evaluation of a patient is completed, which trial to discuss with that patient could be determined by standard random allocation methods. This strategy ensures that investigators and referring physicians making decisions about how to approach patients remain unaware of trial allocation. However, with a small group of patients simple randomization may result in significant imbalance of assignment. A stroke center enrolling 20 acute ischemic stroke patients per year, for example, would like to minimize the possibility of having 15 patients assigned to one trial and only 5 to another.
This technique uses weighted randomization to provide balanced assignments while maintaining uncertainty regarding next allocation. If one trial receives a disproportionate initial share of patients through simple randomization, subsequent patients can be allotted by the equivalent of a biased coin that is more likely, but not certain, to yield assignment to the underpopulated trial. For example, a random digit between 1 and 6 might assign patients to one trial and between 7 and 9 to another, a two-thirds bias. More complicated schemes can give weight not only to numbers of patients in each group but to any other variable the investigators would like to affect the proportion of trial assignments, such as remuneration or quota urgency. The drawbacks of this technique include its relative complexity of administration and lack of intuitive appeal to clinicians.
This method combines the desirable features of fixed schedule and simple randomization assignment. Within blocks of 4, 8, or other intervals, patients would be assigned in balanced pseudorandom fashion to different trials (eg, ABBA, AABB, or BABA). Investigators would be blind to the underlying sequences and block length. This strategy ensures balanced assignment to competing trials while maintaining investigator and referring physician uncertainty regarding the next trial allocation (a small number of guesses at the end of blocks may be accurate). Ease of administration, balanced assignment, and maintenance of physician uncertainty make permuted block pseudorandomization an extremely attractive method for adoption at single centers.
Implementation of automatic allocation schemes such as the above at single-hospital stroke centers would not be difficult. Arrangements would have to be made to avoid allocating patients to studies for which they are ineligible, either by restricting allotment categories to enrollable studies or automatic reassignment. For centers with multiple-component hospitals and multiple studies, it is logistically burdensome to obtain bureaucratic approval for every study at every hospital. A unified, cross-hospital approval process would reduce but not completely resolve this obstacle.
An entirely different approach to resolving the multiple, overlapping trial dilemma is to rely on correction of single-site bias at the level of the entire multicenter trial population. A tendency among some sites to enroll or avoid enrolling patients with particular traits may be balanced by countervailing tendencies at other sites, producing a representative study population overall. Also, the statistical impact of recruitment bias in a single site may simply be overwhelmed by the preponderance of unbiased subjects recruited at other centers. In this approach, it is argued that judgments regarding the generalizability of trial findings to broad clinical populations need not rest solely on complete freedom from bias in recruiting individual trial subjects; such judgments may also be based on the demographic and clinical characteristics of all actually enrolled patients and their similarity to general clinical groups. A major deficiency of this approach is that it has encountered only lukewarm acceptance among wary clinicians. While undergirding the successful European Carotid Surgery Trial and the ongoing International Stroke Trial, this type of design met severe criticism in interpretation of the Extracranial/Intracranial Bypass Trial.18 19 20 21 In addition, trial designers must be wary of biases likely to be encountered at a preponderance of participating sites that will tend to skew the entire study population, such as siphoning off very early presenting patients from acute and urgent trials to hyperacute trials.
The allocation of patients among concurrently available trials is a substantial issue that has hitherto received inadequate attention. It is currently an epidemic problem for stroke investigative centers, reflecting the burgeoning of basic and clinical stroke research over the last decade, but arises in any discipline in which rapid advances in physiology and pharmacology make available simultaneously multiple promising agents for clinical testing. The issue of trial allocation thus has broad applications in clinical medicine generally and is pertinent to many other active clinical trial research fields including epilepsy, multiple sclerosis, affective disorder, acute myocardial ischemia, and oncology. Further analyses by clinical investigators and formal statistical modeling by statisticians would help to clarify the best approaches to minimizing bias in trial assignment among concurrently available trials. Provisionally, it may be recommended that centers participating in multiple, overlapping trials should (1) use an articulated, formal method of patient allocation blinded or indifferent to clinical variables, (2) avoid assignment solely according to time window and unique criteria, and (3) avoid any influence of belief in agent efficacy on allocation. Clinical trial coordinating centers should recognize that participating sites meeting these criteria may participate in overlapping trials without undue risk of introducing patient selection bias.
- Received January 6, 1995.
- Revision received April 13, 1995.
- Accepted April 25, 1995.
- Copyright © 1995 by American Heart Association
Pullicino PM, Caplan LR, Hommel M. Cerebral Small Artery Disease. New York, NY: Raven Press Publishers; 1993.
Feldmann E, Easton JD. Clinical Trial Roster. Brown University Division of Cerebrovascular Diseases and Sonology; 1992.
Lyden PD, Rapp K, Babcock T, Rothrock J. Ultra-rapid identification, triage, and enrollment of stroke patients into clinical trials. J Stroke Cerebrovasc Dis. 1994;4:106-113.
Biller J, Saver JL. Clinical Trial Roster. Northwestern University Stroke Program; 1994.
Pocock SJ. Clinical Trials: A Practical Approach. New York, NY: John Wiley & Sons; 1983.
Shapiro SH, Louis TA, eds. Clinical Trials: Issues and Approaches. New York, NY: Marcel Dekker Inc; 1983.
Meinert CL, Tonascia S. Clinical Trials: Design, Conduct and Analysis. New York, NY: Oxford University Press; 1986.
Iber FL, Riley WA, Murray PJ. Conducting Clinical Trials. New York, NY: Plenum Medical Book Co; 1987.
Capildeo R, Orgogozo JM, eds. Methods in Clinical Trials in Neurology. New York, NY: Stockton Press; 1988.
Cato AE, ed. Clinical Drug Trials and Tribulations. New York, NY: Marcel Dekker Inc; 1988.
Porter RJ, Schoenberg BS, eds. Controlled Clinical Trials in Neurological Disease. Boston, Mass: Kluwer Academic Publishers; 1990.
Spilker B. Guide to Clinical Trials. New York, NY: Raven Press Publishers; 1991.
Whisnant JP, ed. Stroke: Populations, Cohorts, and Clinical Trials. Boston, Mass: Butterworth Heineman; 1993.
Pocock SJ. Clinical Trials: A Practical Approach. New York, NY: John Wiley & Sons; 1983:72.
Mosteller F, Gilbert JP, Mcpeek B. Controversies in design and analysis of clinical trials. In: Shapiro SH, Louis TA, eds. Clinical Trials: Issues and Approaches. New York, NY: Marcel Dekker Inc; 1983:13-64.
Chalmers TC. The control of bias in clinical trials. In: Shapiro SH, Louis TA, eds. Clinical Trials: Issues and Approaches. New York, NY: Marcel Dekker Inc; 1983:115-127.