Neuroprotection Is Unlikely to Be Effective in Humans Using Current Trial Designs: An Opposing View
To summarize recent neuroprotective trials, there has been no useful balance of benefit over risk.1 It is premature to shoot the messenger without evidence either that the results have been misleading or that we have missed trends. Confirmatory trials and/or meta-analysis support the validity of the results, and similar designs have successfully revealed the risks and benefits of thrombolysis. Trial designs have been imperfect, but the message is accurate: the drugs or doses we tried were inadequate. We need safer and better drugs, we need to know when and how to use them, and we need to maximize the efficiency of our trials. All are possible with current designs.2
Drug and Dose Selection Have Been Flawed
Originally, limited efficacy data from a single laboratory and an acceptable toxicity profile were sufficient to proceed to clinical development. Choice of dose range, the optimal duration of therapy, patient subgroup selection, and the likely extent of benefit were all reserved for testing in clinical trials. The misplaced optimism of such an approach is now recognized, however, and guidance on preclinical testing has been elaborated.3 Rigorous external scrutiny of preclinical data now occurs in the more informed companies,4 and clinical development is deferred pending resolution of any deficiencies. This is the first role of the (now mandatory) trial steering committee.
The drug classes under investigation are evolving, and new approaches are still being discovered. Glutamate antagonists have been widely studied.1 While effective in small animal models, their short time window in the laboratory, failure to protect white matter or oligodendrocytes, and dose-limiting psychoactive effects are notable and perhaps fatal disadvantages. Lubeluzole and eliprodil suffered dose-limiting QTc prolongation. Gavestinel doses were restricted by liver function disturbances. Perhaps the safety and multimodal effects of magnesium, or novel approaches, eg, with proteasome inhibitors, may prevail.
A difficulty with better tolerated drugs that may have a bell-shaped dose-response relationship, such as neutrophil inhibitory factor (ASTIN trial) or BMS-204,352 (POST trials), is that dose selection relying on the maximum tolerated dose may be flawed: instead, some measure of efficacy needs to be incorporated. Here, modern adaptive dose selection designs are going to be profoundly useful. The ASTIN trial uses a design first applied in cancer trials: patient responses are monitored “on line” and used to inform a computer algorithm, which “decides” on the most informative dose to employ at the next randomization opportunity (for discussion, see http://lib.stat.cmu.edu/bayesworkshop/Bayes99.html). This design should revolutionize neuroprotective drug development: not only will the optimal dose be selected, the trial will also decide when enough confidence in the drug is attained to justify progression to phase III; alternatively, it can be used to determine futility with scientific rigor, rather than risking abandoning a potentially useful compound on management whim. Such new designs place exacting demands on the steering and data safety monitoring committees, however, since new skills are required.
Entry Criteria Have Tightened
Patients may have been poorly selected for trials: the CLASS trials optimistically treated patients up to 12 hours from stroke onset, and the “definitive” lubeluzole phase III trial extended the time window from 6 to 8 hours.5 At worst, this has limited trial power. Patient selection has been improved for current trials, however. The ASTIN trial has a 6-hour window, but vigorous efforts are keeping the average time to treatment nearer to 4 hours and also discouraging recruitment of patients who narrowly miss exclusion on several criteria. Interactions among entry criteria are increasingly considered. Selection according to a combination of time window and persisting evidence of salvageable tissue (diffusion/perfusion mismatch on MRI) is being used for a sipatrigine trial and in the DIAS trial of desmoteplase. Even in the long-running IMAGES trial with magnesium, in which late treatment is very unlikely to cause harm and so marginal benefits may be worthwhile, incentives are offered to keep the proportion treated early above 40%, and the MR IMAGES substudy selects on the basis of MR mismatch. Higher risk of adverse effects effectively shortens the window of utility: a fixed risk offset against a benefit that diminishes with time translates into overall harm after a finite time delay; in contrast, a completely safe treatment could in theory be given extremely late without offsetting any minimal benefit.2
Power Can Be Drained by Misplaced Optimism and Suboptimal Endpoint Selection
If early reperfusion can achieve 12% absolute improvement in independence, later neuroprotection will surely be less effective: to power a study for a 10% to 12% absolute benefit is unrealistic. Current trials are targeting smaller benefits: GAIN recruited more than 3000 patients between 2 trials; IMAGES seeks a 5.5% improvement in 2700 patients (though even that may be optimistic), but more important, sequential or adaptive sample size designs are now accepted,6 as evidenced by ASTIN and DIAS. Through stratification, GAIN achieved almost perfect balance in prognostic factors between groups.7,8 Methods to limit baseline imbalance in prognostic variables are becoming more sophisticated: adaptive randomization in conjunction with interactive telephone response systems can achieve balance on up to 12 variables without implications for drug supplies and without leading to impractical block sizes.9 Unfortunately, the most sophisticated systems are still viewed with suspicion and combinations of techniques have yet to be implemented.
Trial endpoints are being better selected: with unselected populations showing a U-shaped distribution of Barthel outcomes, even a marked improvement across the board leads to few patients crossing any arbitrary boundary set in the range of 60 to 90 Barthel points. Thus, recent trials like GAIN7,8 have identified trichotomous outcomes, have combined measures on various scales using the global statistical test, or have set variable outcome thresholds according to initial stroke severity. Crucially, through the agency of steering committees, invaluable data from trials and registers involving many thousands of patients such as GAIN are creating a resource that is informing the design of current and future trials using designs such as ASTIN.
Designs Should Change but Must Remain Realistic
Past approaches precluded prior treatment with thrombolysis, yet several putative neuroprotectants will likely be effective only in the presence of reperfusion. Current trials allow thrombolysis according to local guidelines, even where alteplase is used off-label: the POST, IMAGES, and ASTIN trials are examples. Institutional Review Board approval may even be dependent on allowing such treatment. Subgroup analysis of completed trials that allowed thrombolysis has not hinted at improved outcomes with combination treatment.8 Unfortunately, although protocols encourage early initiation of the putative neuroprotectant, practicalities often dictate that thrombolysis is started or even completed before randomization. These issues still need to be tackled.
Quality of Data Is Paramount
Previously, incentives were offered to accelerate patient recruitment; in recent trials, data quality has taken priority. It is now routine to expect training of staff in trial procedures (NIHSS, SSS, and Barthel), feedback on performance (ASTIN, POST), central reading of images (GAIN, DIAS, POST), and external monitoring—even in academic trials such as IMAGES.
Current trials are applying techniques new to stroke but which have proven worth in other fields. Unfortunately, no trial has yet employed all of the available strategies to maximize its chances of success. Nevertheless, designs are available and in current use that will allow a truly neuroprotective compound to demonstrate its value. We have the technology; we must demand it is properly used.
The opinions expressed in this editorial are not necessarily those of the editors or of the American Stroke Association.
Kennedy Lees holds grants to investigate stroke trial design, is principal investigator of the IMAGES (sponsored by the UK Medical Research Council) and GAIN International (sponsored by GlaxoWellcome) trials, serves on the data safety monitoring committees for the ASTIN (Pfizer) and DIAS (PAION) trials, and was a member of the STAIR group.
Lees KR. Neuroprotection. Br Med Bull. 2000; 56: 401–412.
Lees KR. Prospects for improved neuroprotection trials in stroke.In: Chan P, ed. Cerebrovascular Disease. Cambridge: Cambridge University Press; 2001: 430–443.
Stroke Therapy Academic Industry Roundtable (STAIR). Recommendations for standards regarding preclinical neuroprotective and restorative drug development. Stroke. 1999; 30: 2752–2758.
Lees KR, Green AR, Odergren T. Comparison of neuroprotective data for NXY-059 in animal models with the STAIR criteria. Cerebrovasc Dis. 2001; 11 (suppl 4): 77.
Diener HC. Lubeluzole in acute ischemic stroke treatment: lack of efficacy in a large phase III study with an 8-hour window. Stroke. 1999; 30: 234.
Lees KR, Asplund K, Carolei A, Davis SM, Diener H-C, Kaste M, Orgogozo J-M, Whitehead J, for the GAIN International Investigators. Glycine antagonist gavestinel in neuroprotection (GAIN International) in patients with acute stroke: a randomised controlled trial. Lancet. 2000; 355: 1949–1954.
Weir CJ, Lees KR. Value of minimisation for treatment allocation in acute stroke trials. Cerebrovasc Dis. 1998; 8 (suppl 40): 32.