Systematic Review of Nimodipine
To the Editor:
I noted with interest the systematic review of nimodipine from Dr. Horn et al (Stroke. 2001;32;2433–2438). The authors are to be complimented for a thorough review of difficult literature; their summary brings clarity to a previously confusing issue. I concur with their conclusions that preclinical data suggested nimodipine NOT be pursued in clinical trials.
The authors emphasized stroke infarct volume as an outcome measure. They meta-analyzed the effect of nimodipine on infarct size (their Figure 1), in which an overall favorable effect is shown for nimodipine. The “methodological score” they used to rate the quality of the reviewed articles gave a point if the article included both behavioral and morphometric outcomes.
It seems to me that this review could be taken as further indictment of morphometry as a valuable outcome measurement: infarct volume in the rodent brain seems not to predict effects either on functional outcome or in human clinical trials. This controversy has been bubbling for a while now, and this article serves to crystallize it. The only remaining arguments in favor of morphometry as an end point are (1) it is simple and (2) the data are parametric so standard statistical analysis can be used. Arguments against morphometry include (1) the variance is so huge (if it is reported honestly) that sample sizes must be increased beyond what is typically reported and (2) it has limited relevance to functional outcome. Horn et al have buttressed this latter point, perhaps unintentionally.
As we struggle to resolve the paradox of positive animal/negative human trials, we would be well served to keep this review in mind. Unless a putative neuroprotectant shows effects other than reducing rodent infarct volume, it is very unlikely to prove useful in human stroke victims. I would go further and suggest that rodent infarct volumetry is useless, but I would bow to the wise and articulate rebuttals from my colleagues in this area. Nevertheless, we must now require functional improvement in animals before proceeding to clinical trials, as the authors suggest.
Systematic Review of Nimodipine
To the Editor:
The second publication from the Stroke Therapy Academic Industry Roundtable (STAIR)1 is a welcome addition to the burgeoning literature on the failure of acute stroke trials. Although most of the STAIR consensus proposals are sensible and can be enacted, one is of concern, namely the suggestion that the primary outcome of acute trials should be based on a global outcome statistic, much as that used by the National Institute of Neurological Disorders and Stroke (NINDS).2 The global outcome approach combines 2 or more clinical measures (which are chosen a priori by the trialist) and is statistically efficient, ie, it can improve the power of a trial and therefore the probability of a definitive result.3 However, the statistic has no “units” and, therefore, no obvious meaning to trialists, clinicians, patient, carers, funding bodies, or society in general, except to say that a study is positive, neutral, or negative. The purpose of trials is to improve outcome through the development of new interventions, and therefore the outcome used in a study should be meaningful to those who will use or be exposed to the new treatment. The global test fails on this issue. On a related point, it is questionable whether it is ethical to consent a patient into a trial where the primary outcome is based on a global outcome statistic. When consenting a patient into a trial, we are primarily interested in conveying the potential benefit and hazard of the experimental intervention when compared with the control treatment. With respect to benefit, we should center our explanation around the trial’s primary outcome, which is possible if it is impairment, disability, or dependency, and impossible if it is the global outcome statistic because there is no way to describe its meaning and significance to the patient. Interestingly, those discussing use of a global test in NINDS raised a related issue, namely how drug labels should be worded “if statistical significance was achieved for the global test but not for the individual outcomes.”3
If we did not have a clinical outcome that was sensitive to intervention-related change, then maybe we would have to rely on a “global” statistical outcome. However, we know from the NINDS PROACT II and STAT trials2,4,5⇓⇓ that measures of dependency (modified Rankin Scale) and disability (Barthel Index) are sensitive to absolute changes of 10% or more. Indeed, given enough patients, absolute shifts of 1% in dependency can be detected, as in the acute aspirin trials.6,7⇓
An argument given by the STAIR II panel supporting the global outcome statistic is that several outcomes of interest are analyzed together. However, conventional analysis designs can achieve the same with important dimensions of recovery being included as secondary outcomes, including impairment, quality of life, cognitive impairment, and mood.
With this reasoning, we do not need an abstract outcome. The STAIR II panel needs to review their recommendation on the global outcome statistic and instead support the use of existing and proven clinical scales. My own vote8 goes to using a measure of dependency (modified Rankin Scale) because it is easy to assess and its meaning is easy to communicate, particularly to patients during the consent procedure; this does not of course imply that this scale is perfect or that we should stop looking for better clinical measures of outcome.
- ↵Stroke Therapy Academic Industry Roundtable II (STAIR-II): recommendations for clinical trial evaluation of acute stroke therapies. Stroke. 2001; 32: 1598–1606.
- ↵Tilley BC, Marler J, Geller NL, Lu M, Legler J, Brott T, Lyden P, Grotta J, for the National Institute of Neurological Disorders and Stroke (NINDS) rt-PA Stroke Trial Study Group. Use of a global test for multiple outcomes in stroke trials with application to the National Institute of Neurological Disorders and Stroke t-PA stroke trial. Stroke. 1996; 27: 2136–2142.
- ↵Bath PMW, Lindenstrom E, Boysen G, De Deyn P, Friis P, Leys D, Marttila R, Olsson J-E, O’Neill D, Orgogozo J-M, Ringelstein B, van der Sande J-J, Turpie AGG, for the TAIST Investigators. Tinzaparin in acute ischaemic stroke (TAIST): a randomised aspirin-controlled trial. Lancet. 2001; 358: 702–710
We appreciate Dr. Bath’s careful and thoughtful analysis of the proposals raised in the STAIR-II article. His main concern seems to be in relationship to the recommendation that a global test be used in most phase III clinical trials as the primary outcome measure. We believe, based on the results of the NINDS rt-PA trial and the experience from many unsuccessful acute stroke trials, that using a global test is a reasonable approach for assessing therapeutic response in acute stroke therapy trials.
In the case of binary outcomes (as used in the NINDS) where patients are considered a success or failure on each outcome, the global approach produces an odds ratio that gives a patient or physician an indication of how the odds of a favorable outcome on treatment compares to the odds of a favorable outcome on placebo. A favorable outcome is defined as minimal to no post-stroke disability. Odds ratios are commonly used to describe treatment benefit in clinical trials when there are single outcomes, often adjusted for covariates. This is not considered unethical. If preferred, the global test can also be formulated to provide a relative risk so that the result can be explained in terms of likelihood of a favorable outcome.1 Again, relative risks are commonly used to describe treatment benefit in clinical trials with single outcomes.
The main reason to use a global approach is the concern that there is no one measure of success in stroke. A single outcome may measure only one dimension. For example, a patient may be functioning at a high level on the Barthel Index but have aphasia. We would not consider this a great treatment success. Similarly, a marked improvement may occur on a neurological assessment scale, such as the NIH Stroke Scale, but the patient might still have substantial disability, such as impaired gait.
We have been working with regulatory agencies to help them understand and accept the global approach, but we acknowledged in our article that we have not fully achieved that goal. Hopefully, in the near future, regulatory agencies will more readily acknowledge the difficulties of defining and identifying therapeutic effects with purported acute stroke therapies and agree that a global test incorporating several relevant outcome measures is a reasonable and acceptable primary outcome measure.
As an analogy to the use of a global test in clinical trials, the scientific community accepts analysis of variance where one tests for differences in means among groups and is satisfied to reject the null hypothesis, even if the individual pairwise tests are not statistically significant. With both ANOVA and in the global test, the individual tests are useful in helping to interpret the overall test. The scientific community also accepts nonparametric rank tests and their associated probability value as evidence of treatment benefit. If we used variables without categorization and used a global test that did not require binary outcomes, we would still be able to compare groups in terms of favorable outcome, although indeed we would quantify the difference in rank scores.2 The same would be true if the primary outcome was only an individual measure, such the Rankin scale score. We generally would report a P value and rank score.
Therefore, for all of these reasons, plus those discussed in a previous article in Stroke,3 we believe that a global test derived from multiple appropriate individual outcome measures that assess a variety of outcome measures such as neurological status, disability, handicap and perhaps imaging determination of lesion size, is the best method to evaluate treatment effects of a lack of effect in acute stroke treatment trials. (Marc Fisher, MD, and Barbara C. Tilley, PhD, for the STAIR Group.)