new pathways for evaluating potential acute stroke therapies

View point

New pathways for evaluating potential acutestroke therapies

Marc Fisher�1, Kenneth Cheung2, George Howard3, and Steven Warach4

Abstract Pharmacological therapy for acute ischemic stroke

remains limited to one successful, approved treatment: tissue

plasminogen activator within 3 h of stroke onset. Many

neuroprotective drugs and a few other thrombolytics were

evaluated in clinical trials, but none demonstrated unequi-

vocal success and were approved by regulatory agencies. The

development paradigm for such therapies needs to provide

convincing evidence of efficacy and safety to obtain approval

by the Food and Drug Administration (FDA). The FDA moder-

nization act of 1997 stated that such evidence could be

derived from one large phase III trial with a clinical endpoint

and supportive evidence. Drugs being developed for acute

ischemic stroke can potentially be approved under this act by

coupling a major phase III trial with supportive evidence

provided by a phase IIB trial demonstrating an effect on a

relevant biomarker such as magnetic resonance imaging or

computed tomography assessment of ischemic lesion

growth. Statistical approaches have been developed to opti-

mize the design of such an imaging-based phase IIB study, for

example approaches that modify randomization probabilities

to assign larger proportions of patients to the ‘winning’

strategy (i.e. ‘pick the winner’ strategies) with an interim

assessment to reduce the sample size requirement. Demon-

strating a treatment effect on a relevant imaging-based

biomarker should provide supportive evidence for a new

drug application, if a subsequent phase III trial with a clinical

outcome demonstrates a significant treatment effect.

Introduction

Currently, the only therapy for acute ischemic stroke with

proven efficacy and regulatory approval is tissue plasminogen

activator (t-PA) initiated within 3 h after stroke onset (1). A

recent combined analysis of several t-PA trials demonstrated

that the greatest efficacy of this treatment occurred within

90 min of stroke onset and beneficial effect that likely persists

until 4�5 h after stroke onset (2). The lack of additional

approved therapies is not for lack of effort, as many other

potential pharmacological stroke therapies were evaluated in

clinical trials (3). The only other acute stroke therapies to

demonstrate a statistically significant treatment effect in phase

III trials were the defibrinogenating agent, Ancrod, in a 3-h

window enrollment trial and the thrombolytic, ProUrokinase,

in a 6-h window trial (4, 5). Neither of these therapies is

currently approved by regulatory agencies because additional

confirmatory evidence of efficacy is lacking. Many other

pivotal, efficacy trials assessing thrombolytic and neuroprotec-

tive agents in acute stroke patients were either terminated for

futility or failed to achieve evidence of a statistically significant

treatment effect (6). These many acute stroke therapy trials

failed to achieve significance for a variety of reasons, as outlined

in Table 1, but provided valuable lessons for the future.

The performance of phase III acute stroke therapy trials is

predicated on the concept of demonstrating that the therapy

will show a statistically significant, clinically meaningful

treatment effect in an appropriate group of patients (7). The

design and performance of such trials must also conform with

regulatory requirements to allow for drug approval. Therefore,

trial designers and sponsors must be aware of changing trial

design concepts and the regulatory environment that will

decide whether adequate proof of treatment efficacy was

demonstrated. State-of-the-art design for acute stroke therapy

trials and regulatory requirements are both evolving, as the

lessons from prior trials are digested and interpreted and newer

statistical and patient assessment methodologies become

available (8).

Correspondence: Dr Marc Fisher�, UMASS/Memorial Healthcare,

119 Belmont St., Worcester, MA 01605, USA. Tel: 144 508 334 6641;

Fax: 144 508 334 6695; email: [email protected] of Neurology, University of Massachusetts Medical School,

Worcester, MA, USA2Department of Biostatistics, Columbia University Mailman School of

Public Health, New York, NY, USA3UAB School of Public Health, Birmingham. AL, USA4National Institute of Neurological Disorders and Stroke, Bethesda, MD,

USA

& 2006 The Authors.52 Journal compilation & 2006 International Journal of Stroke Vol 1, May 2006, 52–58

Development process for acute strokepharmacological therapies

The development of acute stroke therapies is primarily based

upon the hypothesis that reducing ultimate infarct size will

translate into improved clinical outcome months and years

later (9). The basic concept is that if the extent of infarction is

on average reduced by a pharmacological intervention then

stroke patients should demonstrate improved function and in

the National Institute for Neurological Disorders and Stroke

(NINDS) t-PA trial the treated patients had an improved

outcome and also demonstrated a reduction in computed

tomography (CT)-confirmed infarct volume (10).

The reduction of injury approach for developing acute

stroke therapies serves as the primary basis for animal studies

to assess novel pharmacological therapies. Thrombolytic or

neuroprotective drugs targeted at appropriate aspects of the

ischemic cascade are initially assessed in animal stroke models

(typically in rats), with the primary measure of efficacy being a

significant reduction in infarct volume in placebo-controlled,

blinded, randomized experiments with careful control of

physiological variables (11). Dose-ranging, delayed treatment

effect, and extensive toxicology studies are also necessary

before filing for an Investigational New Drug (IND) license.

Some experts also advocate additional preclinical experiments

in gyrencephalic species such as cats or primates, and many

pharmaceutical companies are now heeding this recommen-

dation (11, 12).

Drugs showing promise with preclinical testing and ap-

proved for an IND can then be tested in humans. The initial

human studies, phase I and phase IIA, are directed at establish-

ing safety. Phase IIA studies focus on demonstrating safety in

acute stroke patients with the general characteristics of the

overall stroke population (13). Exclusion criteria for this phase

will reflect particular concerns for the drug being evaluated.

Typically, phase IIA studies are performed in a dose-escalation

fashion with a 2:1 or 3:1 randomization paradigm of active

drug to placebo in relatively small patient numbers. Safety is

assessed by an independent Data and Safety Monitoring Board,

as each dose nears or is completed. The phase IIA study

continues until the maximally tolerated dose (MTD) is

identified or a dose producing a plasma concentration well

above that shown to be effective in animals is achieved without

serious side effects.

Promising drugs showing a reasonable initial safety profile

are advanced to phase IIB studies, focusing on acquiring

additional safety data and potentially demonstrating hints of

therapeutic efficacy. In stroke drug development, demonstrat-

ing possible efficacy in phase IIB studies has been difficult. In

some studies, such as the recently completed phase IIB

abciximab study, a signal of apparent efficacy on clinical

outcome measures was detected (14), but this is rare. This is

expected, as studies designed to provide adequate power to

show an association with clinical outcome measures are by

definition phase III trials. However, sponsors would like to

observe some signal of efficacy at this stage before making the

large commitment of financial and personnel resources needed

to perform phase III studies. In other therapeutic areas, such as

multiple sclerosis, detecting an efficacy signal in a modest

sample size is easier in phase IIB with a relevant biomarker

(15). Finding relevant biomarkers for acute ischemic stroke

drug development and using them in phase IIB studies could

dramatically impact upon the drug development process by

enhancing development of promising drugs, by stopping

development of unpromising drugs, and by providing the

confirmatory evidence required for FDA approval.

Regulatory process and stroke drugdevelopment

In the United States, the marketing of drugs by pharmaceutical

companies requires that the relevant division of the FDA grant

approval. The drug seeking approval must show evidence of

both safety and efficacy (16). The evidence of safety and

effectiveness is derived from ‘adequate and well-controlled

investigations, including clinical investigations, by experts

qualified by scientific training and experience to evaluate the

effectiveness of the drug involved, on the basis of which it could

be fairly and responsibly concluded by such experts that the

drug will have the effect it purports or is represented to have

under conditions of use prescribed, recommended or sug-

gested in the labeling or proposed labeling thereof ’ (16). In the

past, the FDA has typically required two phase III clinical trials

with clear objectives, methods, controls, and protocols for a

drug to be approved. Two phase III trials were required to

provide independent confirmation of treatment efficacy and to

avoid a type I error of presuming efficacy when it may have

occurred by chance.

Table 1 Potential explanations for unsuccessful pharmacological acute stroke trials

(1) The drug tested did not have robust treatment effects and underwent insufficient preclinical testing to determine the true lack of efficacy

(2) Drug toxicity precluded achieving adequate patient plasma or central nervous system levels necessary to produce a reasonable treatment effect

(3) The trial did not have sufficient power to detect a modest, but clinically meaningful treatment effect

(4) Patients were included in the trial that were unlikely to benefit from the treatment under investigation

(5) Too many mildly affected patients were included in the trial and these patients had a high placebo response rate reducing the power for detecting

treatment effects

(6) Too many patients were included at delayed time points after stroke onset when potential treatment effects were less likely to occur

(7) The outcome measure used to assess a treatment effect was insensitive

& 2006 The Authors.Journal compilation & 2006 International Journal of Stroke Vol 1, May 2006, 52–58 53

M. Fisher et al. View point

In 1997, an amendment to the 1962 Federal Food, Drug and

Cosmetic Act was passed, known as the Food and Drug

Administration Modernization Act (FDAMA). In FDAMA,

the requirement for substantial evidence of effectiveness was

reduced to a single phase III trial and confirmatory evidence

(17). How to best approach the FDAMA requirements for

stroke drug development leading to an approvable package of

information remains contentious. Peck et al. (18) in a recent

paper suggest that a phase III clinical endpoint trial is in fact

empirical because the clinical outcome measure used as the

primary endpoint in such a trial measures the effect of

treatment on a clinically relevant measure of patient outcome.

However, such a trial is not likely to assess how the drug

induces such an effect by its pharmacological activity. Another

assessment approach is to perform a trial that provides causal

confirmation of a drug’s pharmacological activity. Demon-

strating a drug’s significant effect on relevant biomarkers

defined as ‘a laboratory measurement that reflects disease

activity’ could provide causal evidence of drug activity (19).

Although it must be acknowledged that the development of

such biomarkers can be difficult and may be associated with

problems such as measurement variability, reliability of the

measurement, overinterpretation of the significance of the

biomarker, and expense of obtaining the biomarker (20),

development of relevant biomarkers for acute ischemic stroke

that are related to its pathophysiology and evolution is critical.

Relevant biomarkers could be used to provide causal evidence

of a drug treatment effect in a trial with a reasonable sample

size and should be used along with the results from one phase

III clinical trial with an empiric but clinically relevant endpoint

for potential FDA approval based upon FDAMA. When a

relevant biomarker demonstrates that it is correlated with

improved clinical outcome, it may then serve as a surrogate

marker for clinical outcomes. No validated biomarkers or

surrogate markers are currently available for assessing pur-

ported acute stroke therapies.

The most advanced potential biomarker for acute ischemic

stroke is to assess changes in ischemic lesion volume with MRI

from a baseline, pretreatment time point to a delayed time

point 30–90 days later (21). Assessing treatment effects on

ischemic lesion growth in human stroke trials is a potentially

relevant biomarker because the target of acute stroke therapy is

salvage of the ischemic penumbra and the preclinical evalua-

tion of potential stroke therapies is primarily directed at

reducing infarct size. Diffusion-weighted MRI (DWI) can

detect regions of high-energy metabolism failure and loss of

ion homeostasis within minutes of stroke onset (22). These

ischemic regions on DWI may be partially reversible with early

intervention such as thrombolysis, but their natural history is

to proceed to infarction without treatment (23). At a delayed

30–90 days time point, standard T2-weighted MRI can be used

to determine the extent of infarcted tissue, and this volume can

be contrasted with the initial volume of ischemic tissue on

baseline DWI scans. With early DWI studies obtained within

6 h after stroke onset then compared with 30–90 days T2 scans,

natural history studies in untreated patients demonstrated a

50–100% increase in mean ischemic lesion volume and 10–

15% show no lesion growth or shrinkage (24). Preliminary

studies also suggest that a significant correlation exists between

increasing lesion volume from early to delayed MRI studies

and clinical outcome. Patients with no lesion growth or

shrinkage have a statistically significant better outcome than

patients demonstrating lesion growth (21). Apparent relevant

biomarkers for phase IIB acute stroke trials would be to

compare the percentage of patients with no ischemic lesion

growth in a treated group versus a placebo group or assessing

the treatment effect on mean ischemic lesion volume change

over time.

MRI can also be used to choose patients more likely to

respond to interventions. Perfusion MRI (PWI) evaluates

blood flow in the brain’s microcirculation and can identify

readily hypoperfused areas (22). The combination of DWI/

PWI can provide a crude approximation of potentially salvage-

able ischemic tissue or the ischemic penumbra. A commonly

observed initial DWI and PWI pattern is that the PWI volume

is larger than the DWI volume, so-called ‘DWI/PWI mis-

match’, as seen in Fig. 1 (25). The region of PWI abnormality

without DWI abnormality may represent an approximation of

the ischemic penumbra, the presumed target of acute stroke

therapies (26). Uncontrolled intravenous t-PA studies suggest

that stroke patients with a diffusion/perfusion mismatch are

more likely to have a favorable delayed clinical outcome than

patients without a mismatch (26, 27). Currently, in selected

centers, patients are chosen for intravenous t-PA treatment

beyond 3 h after onset, if they demonstrate a clear DWI/PWI

mismatch such as that observed in Fig. 1. It is, however,

appreciated that the diffusion/perfusion mismatch concept

Fig. 1 The diffusion-weighted image (DWI) on the left demonstrates bright

areas in the left parietal region, indicative of acute ischemic injury. The mean

transit time (MTT) image on the right demonstrates a larger area of

hypoperfusion in that region. This mismatch – area of hypoperfusion that

appears normal on DWI – approximates the ischemic penumbra in acute

stroke and therefore identifies the optimal patient for enrollment in clinical

trials.


View point M. Fisher et al.

only approximates the ischemic penumbra and further refine-

ments of penumbral characterization will occur (28). Perfu-

sion CT can also provide information about the status of brain

perfusion and qualitative measurements of cerebral blood flow

and cerebral blood volume (29). Regions with collapsed CBV

on perfusion CT likely represent irreversible ischemic injury

and a mismatch between such regions and those with reduced

CBF and normal CBV can be identified. Further quantification

and validation of these CBF and CBV values on perfusion CT is

needed. When this occurs, perfusion CT may also be a useful

biomarker.

Statistical designs for detecting signals ofefficacy in early phase trials

Phase I and IIA safety trials

For a promising therapy to move quickly to a pivotal phase III

trial, it is imperative that a potentially effective and safe dose

regimen be identified. The primary objective of phase I and

phase IIA trials is to establish safety, and identify an MTD, or

the largest dose that can be safely given (30). Traditionally, the

MTD is approached from lower doses and follows the principle

that three patients are treated at each nontoxic dose level and

up to six at each level showing toxicities. Then, the MTD is

estimated by the dose immediately below the level where safety

issues become apparent (i.e., the ‘toxic’ level). This paradigm,

first applied in cancer trials, may not be the appropriate

methodology for stroke trials. In stroke trials, the expression

of a toxic level is frequently a substantial clinical event such as a

cerebral hemorrhage. Because such events carry such a sub-

stantial morbidity and mortality, the acceptable proportion of

patients undergoing such events is quite small. For example, it

would be difficult to conceive of a ‘safe’ treatment for acute

stroke that would be associated with a symptomatic hemor-

rhage rate above 10%. This is in contrast to cancer trials where

an anticipated negative patient outcome absent dramatic and

successful therapy makes a much higher level of toxic responses

tolerable. Because the acceptable proportion of patients with

toxic responses is so small for stroke studies, it is likely that the

true safe dose will be substantially overshot by the step-up

approach with three patients in a stratum. For example,

consider a dose where the true toxic rate (i.e., true hemorrhage

rate) is 20%. The algorithm would ‘step up’ to the next higher

dose if either: (1) no toxic responses were observed in the first

three patients, or (2) one toxic response was observed in

the first three patients and three additional patients were

observed with no toxic events. The likelihood of observ-

ing no toxic responses among the first three patients would

be (1–0�20)3 5 0�512, or 51�2%. The likelihood of one toxic

event among the first three patients is (3!/(2!� 1!) (0�201)

(0�802) 5 0�384, and the probability of subsequently observing

no toxic effects is 0�512; hence, the likelihood of observing one

toxic effect and then observing three additional patients with

no toxic effects is 19�7% (0�384� 0�512 5 0�197). As such,

even when the true rate of toxic events is twice the acceptable

level (20% in comparison with 10%), there is still a 70�9%

(51�2119�7%) of assuming that the dose level is safe and

stepping up to the higher level. Using this approach, the true

hemorrhage rate for an individual patient has to be approxi-

mately 63% before there is less than a 5% chance of ‘stepping

up’ to a higher dose and assuming that the current dose is ‘safe.’

Clearly, no drug dose with a 63% chance of causing a

hemorrhage would be considered safe.

Recent statistical methods have been proposed to determine

the MTD in the case where the acceptable level of toxicity is at a

low level (e.g., 10%). At the heart of the statistical challenge is

that the step-up approach described above does not allocate

many patients at levels close to the toxic level. With either three

or six patients evaluated at the toxic level, the proportion of

patients with toxic responses are measured with crude percen-

tages that are multiples of 16�7% (i.e., 1/6th), and the widths of

the confidence limits about these estimates on small sample

sizes are disappointingly large.

Perhaps the most promising detection method to address

this challenge is the ‘continual reassessment method,’ which

addresses this concern by using a strategy using information

accrued during the trial, so that the next patient will be treated

at the most currently estimated MTD (31–33)–in essence, the

approach keeps allocating individuals at the ‘best guess’ of the

MTD. This approach then has the advantage of moving more

quickly to dose levels close to the true MTD, resulting in a

larger number of patients allocated to doses that are close to the

true MTD through avoiding processes that ‘waste’ patients by

allocating them to doses that are substantially below the MTD.

In contrast to the commonly used 313 step-up procedure, a

much greater level of precision in the estimated toxicity is

provided because a larger number of patients are assessed in

the true neighborhood of the MTD. In turn, this greater level of

precision allows reliable estimation in the neighborhood of low

levels of toxicity (i.e., 10%). This approach can be optimized by

analytic techniques and computer simulations when applied to

actual trials (34).

Dose-finding phase IIB trials

The identification of a potentially effective dose(s) should be

the primary objective of phase IIB trials. Relatively little is

known about the dose–efficacy relationship of an agent prior

to conducting a phase IIB trial. Typically, the MTD is used in

phase IIB trials, but lower doses may be safer and equally

effective. A better strategy may be to evaluate several doses and

select the ‘best’ dose for the phase III study.

Two-stage designs

To have adequate power and control of type I error rates, the

sample size required for multiple-dose comparison trials

would be infeasible if conventional one-stage, balanced ran-



domization designs that use a clinical outcome endpoint are

used. A phase IIB trial should be efficient regarding dose

finding with a realistic sample size and the use of outcome-

adaptive methods may advance trial conduct. One of the first

approaches proposed was to adjust the allocation scheme to

assign a larger number of patients to the treatment that appears

most promising at the current time. The practicality of this

‘play-the-winner rule’ (35, 36) is limited as the outcome of

assigned treatment must be observed before establishing the

‘winner’ in order to make future assignments. At the extreme,

even having a 90-day outcome measure as the primary out-

come would limit enrollment of patients to approximately

a 3-month cycle. These limitations have led to a wide-

spread recognition of timeliness issues for sequential methods

(37, 38).

Among the responses has been the development of ‘two-

stage designs’ that provide a good balance between ethical,

economical, and timeliness considerations (39). Briefly, these

approaches use a first stage where patients are randomized to a

large number of doses in an attempt to ‘seek’ the most effective

doses. Following the first stage, the interim data can be used to

eliminate ineffective doses: additional patients will then be

randomized to potentially effective doses. Eliminating ineffec-

tive doses in the first stage again avoids ‘wasting’ patients by

assigning them to doses that are unlikely to show benefit

during the second stage. Such designs are ethically and

economically appealing because the average sample size will

be reduced while maintaining comparable statistical power

and type I error rates compared with one-stage designs. For

example, Table 2 compares a two-stage design with the

conventional one-stage, balanced randomization design in a

three-armed trial with two active doses versus the placebo. In

this example, the maximum number of patients required by

the two-stage design is 126, whereas the one-stage design needs

a total of 216 patients. In addition, since the two-stage design

will stop the trial after the first stage if neither dose is superior

to the placebo in terms of the observed number of responses,

the average number of patients will be 87 if in reality neither

dose is better than the placebo (40).

Impact on the phase III trials and regulatory issues

There is an extensive statistical literature of outcome-adaptive

designs, but their trial application is limited due to perceived

logistical challenges to implementation. A more controversial

approach is the use of Bayesian methods that promise the

possibility of providing a flexible and unified framework for

inference (41). Bayesian analysis has the ability to incorporate

information available at the initiation of the study, and this

information holds the possibility of substantial reductions in

the sample size. This is in contrast to the standard approach

(frequentist approach) where the interpretation of each trial is

limited to only the data provided by the study.

Bayesian methods are viewed by many investigators as being

subjective because of the necessity of specifying the investiga-

tor’s belief before data collection (i.e., the ‘prior distribution’),

making the results of the study both a function of the study

results and the prior assumptions. This joint dependence on

the assumptions and the resulting study data creates the

possibility of differing interpretations of the data resulting

from the same clinical trial (42). Finally, the process of

conducting phase III trials involves the selection of promising

treatments from the phase I/II studies. A study can appear

promising both for its true effect, but also because the

estimated effect appears inappropriately large in the specific

phase I/II studies that were conducted. As such, it is possible

that the prior information is biased, and the use Bayesian

analysis will ‘carry forward’ these biases through their use in

the prior distributions (42). Conversely, advocates of Bayesian

analysis have argued that the approach is rather robust and that

the potential gains from a Bayesian approach more than offset

the potential risks. If Bayesian approaches are considered,

caution should be exercised in light of FDAMA. If a phase IIB

trial outcome is to serve as confirmatory evidence under

FDAMA, the conventional frequentist method that provides

a ‘P value’ with respect to a predefined hypothesis concerning

relevant biomarkers or may be more convincing to regulatory

agencies than Bayesian inferences. The use of novel statistical

methods and the regulatory process need to maintain a

Table 2 Comparison of a two-stage design and the conventional one-stage design in a three-armed study (two active doses and one placebo)

One-stage design Two-stage design

Description 1. Enroll 72 patients per arm 1. Enroll 22 patients per arm

2. Choose a dose if a one-sided, two-sample binomial

test between the dose and the placebo is significant

at 0�025 level according to the Bonferroni’s adjustment

2. If any dose has at least two more responses than the placebo,

then take the dose with the largest number of responses

(called the winning arm) to stage 2 with the placebo.

3. Enroll 30 patients each to the placebo arm and the winning arm.

4. Declare a dose effective if it has at least 6 responses more

than the placebo.

Type I error� 0�05 0�05

Power 80% 80%

Sample size 216 126

�The type I error is defined as the probability of selecting any ineffective.



dynamic relationship to allow for a clinical trial program that

will demonstrate convincing proof of efficacy and lead to

regulatory approval.

Conclusions

The lessons learned from the many prior unsuccessful acute

stroke trials and the recent availability of likely relevant

imaging biomarkers of drug activity using DWI/PWI and

perfusion CT provide an opportunity to enhance trial design

for acute ischemic stroke pharmacological interventions that

should accelerate the evaluation of new, promising drugs. The

use of a relevant biomarker in phase IIB studies could provide

causal evidence of drug activity and presumably be used with

one, large phase III with an empirical, clinical outcome

measure to seek FDA approval based upon the requirements

of FDAMA. Newer statistical methodology can be used in

phase IIB trials to also provide valuable and timely information

about drug efficacy or lack thereof. The confluence of relevant

biomarkers, newer statistical methodology, and moderniza-

tion of FDA requirements for drug approval should lead to a

new era of more rapid assessment and hopefully approval of

additional acute stroke drugs.

References

1 The National Institute of Neurological Disorders and Stroke rt-PA

Stroke Study Group: Tissue plasminogen activator for acute ischemic

stroke. N Engl J Med 1995; 333:1581–7.

2 Cheng YD, Al-Khoury L, Zivin JA: Neuroprotection for ischemic

stroke: two decades of success and failure. NeuroRx 2004; 1:36–45.

3 Fisher M, Schabitz WR: An overview of acute stroke therapy: past,

present and future. Arch Int Med 2000; 160:3196–20.

4 Furlan A, Higashida R, Wechsler LA et al. Intraarterial prourokinase

for acute ischemic stroke. The PROACT-2 Study: a randomized

controlled trial. JAMA 1999; 282:2003–11.

5 Sherman DG, Atkinson RP, Chippendale T et al. Intravenous

ancrod for treatment of acute ischemic stroke. JAMA 2000; 283:

2395–403.

6 Kidwell CS, Liebeskind DS, Starkman S et al. Trends in acute ischemic

stroke trials through the 20th century. Stroke 2001; 32:1349–5.

7 Fisher Mfor the Stroke Therapy Academic Industry Roundtable:

Recommendations for advancing development of acute stroke thera-

pies. Stroke 2003; 34:1539–46.

8 Optimising Analysis of Stroke Trials (OAST) Collaborators, Bath

PMW: Optimising the statistical analysis of functional outcome in

stroke trials. Stroke 2003; 34:316.

9 Fisher M, Ratan R: New perspectives on developing acute stroke

therapy. Ann Neurol 2003; 53:10–20.

10 The National Institute of Neurological Disorders and Stroke (NINDS)

rt-PA Study Group: Effect of intravenous recombinant tissue plasmi-

nogen activator on ischemic stroke lesion size measured by computed

tomography. Stroke 2000; 31:2912–9.

11 Stroke Therapy Academic Industry Roundtable: Recommendations for

standards regarding preclinical neuroprotective and restorative drug

development. Stroke 1999; 30:2752–8.

12 Marshall JWB, Cummings RM, Bowes LJ et al. Functional and

histological evidence for the protective effect of NXY-059 in a

primate model of stroke when given 4 hours after occlusion. Stroke

2003; 34:2228–33.

13 Stroke Therapy Academic Industry Roundtable: Recommendations

for clinical trial evaluation of acute stroke therapies. Stroke 2001; 32:

1598–606.

14 Adams Hfor the AbESTT-II Investigators: Abciximab in emergent

stroke treatment trial – II. Presented at the 28th International Stroke

Conference (February 2004). Stroke 2004; 34:252.

15 Miller DH: Biomarkers and surrogate outcomes in neuro-

degenerative disease: lessons from multiple sclerosis. NeuroRx 2004;

1:284–9.

16 Katz R: Evidentiary standards for drug development and approval.

NeuroRx 2004; 1:307–16.

17 Food and Drug Adminstration Modernization Act of 1997, Pub L No.

105–115, 111 Stat 2295, 1997.

18 Peck C, Rubin DB, Sheiner LB: Hypothesis: a single clinical trial plus

causal evidence of effectiveness is sufficient for drug approval. Clin

Pharmacol Ther 2003; 73:481–90.

19 Katz R: Biomarkers and surrogate endpoints: an FDA perspective.

NeuroRx 2004; 1:189–95.

20 Mayeux R: Biomarkers: potential uses and limitations. NeuroRx 2004;

1:182–8.

21 Warach S, Pettigrew LC, Dashe JF et al. Effects of citicoline on

ischemic lesions as measured by diffusion-weighted magnetic

resonance imaging: citicoline 010 investigators. Ann Neurol 2000;

48:713–22.

22 Beaulieu C, Moseley ME: Diffusion-weighted and perfusion-weighted

magnetic resonance imaging in clinical stroke in current review of

cerebrovascular disease; in Fisher M, Bogousslavsky J (eds): Current

Medicine, 4th edn. Philadelphia, 2001: 59–68.

23 Kidwell CS, Saver JL, Mattiello J et al. Thrombolytic reversal of acute

human cerebral ischemic injury shown by diffusion/perfusion

magnetic resonance imaging. Ann Neurol 2000; 47:462–9.

24 Schellinger PD, Fieback JB, Jansen O et al. Strokemagnetic resonance

imagingwithin 6 hours after onset of hyperacute cerebral ischemia.

Ann Neurol 2001; 49:460–9.

25 Schlaug G, Benfield A, Baird AE et al. The ischemic penumbra:

operationally defined by diffusion perfusionMRI. Neurology 1999;

53:1528–37.

26 Parsons MW, Barber PA, Chalk J et al. Diffusion and perfusion-

weightedMRI response to thrombolysis in stroke.Ann Neurol 2002;

51:28–37.

27 Neumann-Haeflin T, de Rochemont M, Fiebach JB et al. Effect of

incomplete (spontaneous and postthrombolytic) recanalization

after middle artery occlusion. Stroke 2004; 35:109–15.

28 Kidwell CS, Alger JR, Saver JL: Beyond mismatch: evolving paradigms

in imaging the ischemic penumbra with multimodal magnetic reso-

nance imaging. Stroke 2003; 34:2729–35.

29 Wintermark M, Bogousslavsky J, Thiren JP et al. Prognostic accuracy

of admission cerebral blood flowmeasurements by perfusionCT in

acute stroke patients. Ann Neurol 2002; 51:417–32.

30 Storer B, DeMets D: Current phase I/II designs: are they adequate?

Journal Clin Res Drug Dev 1987; 1:121–30.

31 O’Quigley J, Pepe M, Fisher L: Continual reassessment method: a

practical design for phase I clinical trials in cancer. Biometrics 1990;

46:33–48.

32 Cheung YK, Chappell R: Sequential designs for phase I clinical trials

with late-onset toxicities. Biometrics 2000; 56:1177–82.

33 Cheung YK: On the use of nonparametric curves in phase I trials with

low toxicity tolerance. Biometrics 2002; 58:237–40.

34 Cheung YK, Chappell R: A simple technique to evaluate model

sensitivity in the continual reassessment method. Biometrics 2002;

58:671–4.

35 Zelen M: Play the winner rule and the controlled clinical trials. J Am

Statist Assoc 1969; 64:131–46.

36 Wei LJ, Durham S: The randomized play-the-winner rule in medical

trials. J Am Statist Assoc 1978; 73:840–3.



37 Rosenberger WF, Lachin JM: The use of response-adaptive designs in

clinical trials. Controlled Clin Trials 1993; 14:471–84.

38 Cheung YK, Thall PF: Monitoring the rates of composite events

with censored data in phase II clinical trials. Biometrics 2002; 58:

89–97.

39 Pocock SJ: Group sequential methods in the design and analysis of

clinical trials. Biometrika 1977; 64:191–9.

40 Thall PF, Simon R, Ellenberg SS: Two-stage selection and testing

designs for comparative clinical trials. Biometrika 1988; 75:303–10.

41 Diamond GA, Kaul S: Prior convictions: bayesian approaches to the

analysis and interpretation of clinical megatrials. J Am Col Cardiol

2004; 43:1929–3.

42 Howard G, Cutter CR, Coffey CS: Is bayesian analysis ready for use in phase

III randomized clinical trials? ‘‘Beware the Sound of the Sirens’’. Stroke.



new pathways for evaluating potential acute stroke therapies

Documents