Glycine Antagonist in Neuroprotection for Patients With Acute Stroke: Title and subTitle BreakGAIN Americas: A Randomized Controlled Trial FREE

In recent years, considerable efforts have been devoted to developing treatments aimed at restoring cerebral perfusion or limiting neuronal damage following acute ischemic stroke. Recognition of the "ischemic penumbra," a region of reduced cerebral blood flow in which cell death might be prevented,¹ has focused attention on treatments that might minimize or reverse brain damage when administered soon after stroke onset. Enlargement of infarct volume is determined by changes in metabolism caused by initiation of the ischemic cascade. This cascade involves energy supply failure, membrane depolarization, release of neurotransmitters (including glutamate in large amounts), accumulation of intracellular calcium, increased production of nitric oxide and free radicals, development of cellular edema, and finally, cell death.² Each step along the ischemic cascade offers a potential target for therapeutic intervention. To date, several classes of neuroprotective compounds have been investigated in phase 3 trials for acute stroke. They have included calcium channel antagonists, N-methyl-D-aspartate (NMDA) receptor antagonists, free radical scavengers, anti–intercellular adhesion molecule 1 antibody, GM-1 ganglioside, γ-aminobutyric acid agonists, and sodium channel antagonists, among others.³ All of the trials have yielded disappointing efficacy results and some showed safety problems, including increased mortality or psychotic effects, which resulted in their early termination.^{4 - 6}

The compound gavestinel (GV150526) was shown to have high affinity and high selectivity for the glycine site of the NMDA receptor complex, and inhibited NMDA-induced depolarization at low doses in preclinical studies.⁷ When administered up to 6 hours from stroke onset, gavestinel reduced infarct size by 50% in a rat middle cerebral artery occlusion model of ischemic stroke.⁸ In phase 2 studies, gavestinel was well tolerated when administered at projected neuroprotective doses.^{9 - 11} The phase 3 Glycine Antagonist in Neuroprotection (GAIN) trials, GAIN Americas and GAIN International, were undertaken using nearly identical protocols to determine whether patients treated with gavestinel within 6 hours of stroke onset have improved functional outcome (independence) at 3 months, and to further assess the compound's safety in the clinical setting of acute stroke. Here we report the final results of GAIN Americas, which was conducted in the United States and Canada.

The GAIN Americas trial, conducted at 132 medical centers in the United States and Canada, was a randomized, double-blind, placebo-controlled, parallel-group study designed to evaluate the efficacy and safety of gavestinel in patients with a clinical diagnosis of acute stroke.

The eligibility criteria for the trial are outlined in Table 1. Stroke severity was assessed using the National Institutes of Health Stroke Scale (NIHSS).¹² Total baseline NIHSS scores could range from 2 to 41, due to the limb weakness inclusion and level of consciousness exclusion criteria. Eligible patients with hemorrhagic stroke were included in order to provide information about the safety and preliminary efficacy of gavestinel among patients with intracerebral hemorrhage (ICH). Patients who received intravenous recombinant tissue-type plasminogen activator (rt-PA) within 3 hours of stroke onset, and who met the eligibility criteria after rt-PA infusion, were enrolled. All patients, or their legally authorized representatives (in accordance with local regulations), provided written informed consent to enter the trial. All participating centers obtained approval for the consent form and the study protocol from their local ethics or institutional research review committees prior to randomizing patients.

The treatment assignment schedule was generated by the statistician on the independent safety and efficacy data monitoring committee (SEDMC) and was given to the centralized randomization service. Patients were randomly assigned (1:1) to receive either gavestinel or placebo, with stratification by age (≤75 or >75 years) and stroke severity (NIHSS scores categorized as 2-5, 6-13, ≥14) yielding 6 strata. These NIHSS categories were selected based on their associations with 6-month functional outcome in a large, population-based study.¹³ Treatment allocations were obtained by pharmacists at the study sites by telephoning the randomization service, which used an interactive voice-response system.

The study drug was provided in prenumbered packs of amber vials containing gavestinel sodium or vehicle placebo (Glaxo Wellcome Inc, Research Triangle Park, NC). The contents of the vials were diluted in 5% dextrose (D5W) due to the poor solubility of gavestinel in saline. Since gavestinel is light sensitive, the infusion bag and tubing were covered with opaque green plastic sleeves during preparation and administration. The sleeves also helped prevent unblinding of study personnel by keeping them from seeing the pale yellow color of the diluted gavestinel solution. Accidental unblindings of personnel involved in patient care were recorded in the case report form.

The study drug was given intravenously over 3 days. Patients received either placebo or a total of 1800 mg of gavestinel administered as a loading dose of 800 mg (400 mg in 250 mL of D5W over 30 minutes, then 400 mg in 250 mL of D5W over 3.5 hours), followed by 5 maintenance doses of 200 mg each in 150 mL of D5W over 15 minutes at 12, 24, 36, 48, and 60 hours after the start of the loading dose.

Baseline assessments included demographics, medical history, cerebrovascular event history, prestroke level of independence by the Modified Rankin Scale (MRS),¹⁴ stroke severity by the NIHSS, and stroke subtype by the Oxfordshire Community Survey Project classification.¹⁵ Study personnel were trained and certified in using the NIHSS through video training procedures.¹² The presence of ICH was determined from a brain computed tomography (CT) scan taken within 18 hours of stroke onset, or a magnetic resonance imaging (MRI) scan taken within 6 hours. Digitized images were read by a blinded committee of 3 independent neuroradiologists. At day 7 or at hospital discharge, whichever came sooner, stroke subtype was classified by the criteria used in the Trial of ORG10172 in Acute Stroke Treatment (TOAST).¹⁶ Routine hematologic and clinical chemistry parameters were measured at baseline and after treatment completion. A central laboratory performed the analyses. Study treatment was terminated if creatinine values exceeded 2.0 mg/dL (178.8 µmol/L), or if bilirubin, aspartate aminotransferase, or alanine aminotransferase values exceeded 4 times the upper limit of reference range on these or locally obtained laboratory tests. Electrocardiograms were obtained at baseline and during the treatment phase and were read at a central blinded reading center.

The primary outcome was functional capability in activities of daily living at 3 months as measured by the Barthel Index (BI).¹⁷ Barthel Index scores were trichotomized as follows: 95-100 ="independence" (little or no help required), 60-90 = "assisted independence" (some help required), and 0-55 = "dependence" (help required with most or all activities). Deaths were analyzed along with the 0-55 group. The BI cut points were chosen based on prior studies that established that scores of 60 and 95 defined meaningful clinical subgroups.^{18 - 19} Guidelines for administering the BI were reviewed at prestudy training meetings. To minimize bias in the assessment of the primary outcome, the person performing the BI at 3 months could not be a person involved in caring for the patient during the initial hospitalization.

Secondary outcomes included level of independence determined by BI at 7 days or hospital discharge (whichever came sooner) and at 1 month, and both stroke severity determined by NIHSS and level of handicap determined by MRS at 1 and 3 months. Other outcomes included survival through 3 months and a global test, composed of dichotomies of the BI (95-100, 0-90/dead), NIHSS (0-1, 2-42/dead), and MRS (0-1, 2-5/dead) measured at 3 months. This was a modified version of the global test used in the National Institute of Neurological Disorders and Stroke (NINDS) trial of rt-PA for acute stroke in that it omitted the Glasgow Outcome Scale.¹⁹ Reduction in ischemic lesion volume growth from baseline to 3 months by diffusion-weighted MRI,²⁰ health care resource use through 3 months, and functional outcome and quality of life using the Stroke Impact Scale²¹ at 1 and 3 months were additional prespecified secondary outcomes that will be reported separately.

Adverse events, both serious and nonserious, were recorded. Certain adverse events were designated as "events common to stroke patients" in the protocol. These included progression of stroke, hemorrhagic transformation, brain herniation, aspiration pneumonia, deep vein thrombosis, and recurrent stroke. For such events that also met the standard definition of a serious adverse event, the requirement for immediate reporting by investigators was waived (with prior regulatory approval) unless the event proved fatal or the investigator otherwise felt the event required urgent notification. Other types of serious adverse events followed usual reporting requirements. The SEDMC reviewed all safety data after the first 150 patients were enrolled, and then after each subsequent 250 patients were enrolled. Fatal events were reported to the SEDMC as they occurred.

The primary efficacy analysis included all ischemic stroke patients who received at least 1 dose of study drug. Randomized patients who did not go on to receive any study treatment were excluded. Following the intent-to-treat principle, patients who received any medication were analyzed in the treatment group to which they were randomized, regardless of treatment received. Treated patients who were lost to follow-up or withdrew consent prior to their 3-month assessments were assigned their last posttreatment BI score.

Barthel Index score distributions in active and placebo groups were compared using the extended Mantel-Haenszel χ² test (1 df)²² stratified by baseline stroke severity and age group, at the 2-sided, 5% level of significance. The scores used for the extended Mantel-Haenszel test were 2 for BI 95-100, 1 for BI 60-90, and 0 for BI 0-55 or dead. These scores were optimal under a proportional adjacent odds model. The null hypothesis was that distributions of the trichotomized outcome would be the same in both treatment groups. Given the scoring, positive z values were interpretable as gavestinel superior to placebo, and negative z values as placebo superior to gavestinel. The trial was powered at 90% at 2-tailed α = .05 for an (unstratified) outcome distribution for the trichotomized BI, as shown in Table 2. This allowed for a 10 percentage-point difference between treatment groups in the best outcome category in favor of gavestinel and a 6 percentage-point difference in the worst category. The outcome distribution in the placebo group was based on the Clomethiazole Acute Stroke Study.²³ A sample size of 673 evaluable, nonhemorrhagic patients per treatment group was required to detect the stated differences. Assuming 10% of randomized patients would have an ICH, and no more than 5% would be unevaluable, a total sample size of 1580 acute stroke patients was required.

Two interim analyses and a final analysis were scheduled. These used a 2-sided test with equal allocation of type I error (.025 in either direction), but with asymmetrical stopping boundaries. Due to the conservative nature of the stopping boundaries for the interim analyses, only a slight adjustment (at the fourth decimal place) was needed in the P values required by the group-sequential design. For the routine reporting of P values below, nominal P values were used, ie, no further adjustments for group-sequential testing were applied.

Prestated secondary analyses examined the secondary outcome measures described above using the extended Mantel-Haenszel test for the BI outcomes and the Mantel-Haenszel test²⁴ for the NIHSS and MRS outcomes. The global test used the same model as the NINDS trial but our method of estimation used multinomial theory and weighted least squares to estimate and test the common odds ratio.²⁵ Prespecified subgroup analyses of BI distributions at 3 months stratified by time to treatment (≤4, >4 hours), use of rt-PA, and baseline age and NIHSS score also used the extended Mantel-Haenszel test. Safety analyses included all treated patients (hemorrhagic and ischemic stroke) grouped according to treatment actually received, rather than as assigned at randomization. As per study protocol, efficacy for ICH will be determined in a pooled analysis combining ICH patients from both GAIN trials.

Enrollment began in April 1998 and ended in October 1999 with 132 sites (100 in the United States, 32 in Canada) contributing patients. Follow-up was completed in January 2000. The trial profile is summarized in Figure 1. Of the 1646 patients randomized, 41 did not receive study treatment. The most common reasons for failing to treat were consent obtained too late or withdrawn (n = 11), rapid improvement of symptoms (n = 7), drug administration delays (n = 6), or other eligibility criteria violations (n = 17). Of the 1605 patients treated, 237 were classified as having hemorrhagic strokes (236 ICH, 1 other) by the image adjudication committee. The remaining 1368 patients constituted the ischemic stroke group. Follow-up was complete on 1343 (98.2%) of these patients; 25 patients withdrew consent or were lost to follow-up before 3 months. None were withdrawn from the study by investigators for adverse events. Earlier posttreatment assessments were carried forward for the missing 3-month data in 24 cases. One patient was excluded from the analysis because consent was withdrawn prior to any follow-up assessment. The primary efficacy population (N = 1367) therefore included 701 ischemic stroke patients randomized to gavestinel and 666 randomized to placebo.

In this population, 77 violations of inclusion or exclusion criteria were reported in 74 patients (36 randomized to gavestinel, 38 to placebo). The most frequent violations were late (>6 hours) study drug administration (53 instances, 3.9% of patients) and use of an investigational drug or device in the past 3 months (13 instances, 0.9% of patients). Forty-five patients (6.4%) in the gavestinel group and 41 patients (6.1%) in the placebo group stopped treatment prematurely for reasons other than death. Of these, 37 were discontinued due to adverse events (18 gavestinel, 19 placebo) and 10 due to elevated laboratory values (6 gavestinel, 4 placebo). Thirty patients died before treatment was completed: 21 (3.0%) in the gavestinel group and 9 (1.4%) in the placebo group. Clinical sites reported 5 cases where the treatment blinding was accidentally broken during the course of treatment.

Table 3 summarizes the baseline characteristics of the 2 treatment groups. There were no major imbalances, but fewer patients in the gavestinel group had a history of previous transient ischemic attack. The proportion of patients older than age 75 years was 39.1% in the gavestinel group and 38.0% in the placebo group. Median time to treatment was 5.2 hours in both the gavestinel and placebo groups, with 17% and 19%, respectively, treated within 4 hours of stroke onset.

The primary outcome analysis is summarized in Figure 2. Patients treated with gavestinel did not show a statistically significant improvement in independence at 3 months compared with patients treated with placebo, after adjustment for age and baseline NIHSS score. No treatment effect was detected, even after adjusting for the imbalance in history of transient ischemic attack. No significant differences were found on any of the secondary end points (Table 4), including survival through 3 months (Figure 3). In addition, no significant difference was found using the global statistic (odds ratio [OR], 1.05; 95% confidence interval [CI], 0.85-1.31; P = .64).

Results for prespecified subgroups of interest are shown in Table 5. The distributions of the BI at 3 months were compared within each of the 6 age and NIHSS strata. A significant difference was found in only 1 stratum among the younger patients with milder strokes (age ≤75 years, NIHSS 2-5). After adjustment for age, baseline NIHSS, sex, time to treatment, use of rt-PA, and stroke subtype (lacunar vs nonlacunar), the difference for this stratum persisted (P<.001, Wald test statistic for logistic regression with an assumed common log OR). Among those treated within 4 hours of stroke onset, no significant differences were demonstrated for those receiving gavestinel vs placebo.

Thrombolytic therapy with intravenous rt-PA was administered to 333 patients: 166 (24%) in the gavestinel group and 167 (25%) in the placebo group. Time to treatment with the study drug was somewhat shorter among patients given rt-PA (median hours: 4.9 gavestinel, 4.8 placebo) compared with those not given rt-PA (median hours: 5.4 gavestinel, 5.3 placebo). Within the rt-PA subgroup, gavestinel-treated and placebo-treated patients were similar with respect to age, baseline NIHSS score, time to treatment, sex, and stroke subtype (TOAST classification). No significant improvement in BI at 3 months was found for those who received rt-PA plus gavestinel compared with those who received rt-PA plus placebo.

The final safety analysis was performed on all treated patients (n = 1605). The 3-month case fatality rate was similar in the gavestinel (23%) and placebo (20%) groups. The specific causes of death as well as the types and frequencies of both serious and nonserious adverse events experienced by the 2 treatment groups were very similar. Adverse events of relevance to this study population that occurred in 5% or more of patients within the first 7 days are shown in Table 6. Events prespecified as "events common to stroke patients" were reported at similar overall frequencies through 3 months among patients treated with gavestinel (81%) and placebo (80%). As expected from the phase 2 studies, transient rises in bilirubin levels (≥1.67 times the upper limit of reference range) were seen more often in those treated with gavestinel (19%) than in those treated with placebo (7%) due to a pharmacokinetic interaction with bilirubin metabolism. Although follow-up CT scans were not required by protocol, investigator reporting of symptomatic or asymptomatic hemorrhagic transformation (shown in Table 6) or parenchymatous hemorrhage (3% in each treatment group) was similar regardless of study treatment. As expected, reports of either of these bleeding events were more frequent in the ischemic stroke subgroup treated with rt-PA (54/333, 16%) than in the subgroup not treated with rt-PA (63/1034, 6%).

To date, the GAIN Americas trial (N = 1367) and the GAIN International trial (N = 1455) are the largest trials of a neuroprotectant for acute stroke. Both were adequately powered to detect a difference in the primary end point and to provide definitive conclusions about the study medication's efficacy for ischemic stroke. Despite promising preclinical studies with gavestinel, neither GAIN trial demonstrated any treatment benefit for this compound.²⁶ In GAIN International, 34% of gavestinel-treated patients and 35% of placebo-treated patients were independent at 3 months, compared with 39% and 37%, respectively, in GAIN Americas. Mortality at 3 months was 20% in GAIN International and 21% in GAIN Americas. The serious adverse event experience was also similar in the 2 trials, with neither showing the drug to cause harm. Unlike other types of NMDA receptor antagonists, gavestinel had no significant neurologic or cardiovascular adverse effects.

In GAIN Americas, the only subgroup analysis that showed a favorable 3-month outcome for gavestinel was among younger patients with mild strokes (age ≤75 years, NIHSS 2-5). After adjustment for possible baseline confounders, this finding persisted. However, a benefit for this subgroup was not found in GAIN International. A treatment effect was reported for patients younger than 70 years in the nalmefene phase 2 study,²⁷ but was not confirmed in its phase 3 study.²⁸ The favorable outcome for gavestinel in this 1 subgroup in our study remains unexplained.

The other prespecified secondary analyses revealed no other subgroups of benefit. Even among patients treated within 4 hours of stroke onset (n = 241), no beneficial effect for gavestinel was suggested. Patients who received combined therapy of gavestinel and rt-PA did not show a significant improvement compared with those who received placebo and rt-PA. These 333 patients, with 166 receiving both gavestinel and rt-PA, represent the largest experience with a neuroprotectant in combination with thrombolytic therapy to date. However, since study procedures called for completing the rt-PA infusion before randomization into GAIN, some patients who responded rapidly to rt-PA might not have been enrolled in GAIN. The only other study combining a neuroprotectant (lubeluzole) with rt-PA was terminated early by the sponsor.²⁹ The results among 89 patients enrolled in that trial (randomized 1:1 to rt-PA, plus lubeluzole or placebo) showed that the combination was feasible, safe, and might increase efficacy over rt-PA alone. Other neuroprotective agents need to be assessed for potential benefit in combination with rt-PA.

There were some important and unique design features included in the GAIN studies. Randomization was centralized and stratified into 6 strata by age and stroke severity, thus providing well-balanced treatment groups, which have been lacking in some trials.^{23 ,30 - 31} Exclusion of strokes that did not include an NIHSS motor component score of at least 2 decreased the chances of having too many mild strokes with a high spontaneous recovery rate, thus causing a type II error. Another unique feature was the use of the trichotomized BI as a primary outcome measure. In previous stroke trials, various cutoff scores have been used for the BI, but less than 60 has been the time-honored categorization for definite dependence.³² The additional BI cutoff score of 95 was chosen on the basis of the NINDS study as a measure of definite independence.¹⁹

Stroke forums, such as the Stroke Therapy Academic Industry Roundtable II, have emphasized the importance of a highly involved steering committee working in partnership with the industry sponsor. In the GAIN Americas trial, the protocol was developed by an independent academic steering committee working with the pharmaceutical sponsor. Moreover, this academic-industry collaborative group remained active in the ongoing supervision of the trial, while the GAIN Americas database was managed and analyzed at an academic center.

The efficacy of neuroprotectants in preclinical animal studies has not been corroborated by human acute stroke trials to date.^{23 ,28 ,30 - 31 ,33 - 35} Part of the reason may lie in fundamental differences between stroke in the animal models (mainly rodents) and in humans. The experimental conditions of animal models do not accurately reflect the variability found in clinical stroke patients with comorbid conditions. The failures of neuroprotective compounds in humans to date has led to the formulation of recommendations for minimum standards for the preclinical testing of putative neuroprotective drugs.³⁶ These recommendations include requirements for collecting data on the adequate penetration of drug across the blood-brain barrier, efficacy in multiple species including primates, and reproducibility by independent laboratories.

We considered several factors that might explain our neutral results. First, did we misjudge the therapeutic window of opportunity? In animal models of middle cerebral artery occlusion, gavestinel reduced infarct volume by 50% when given up to 6 hours from the onset of ischemia.⁸ In our data, there was no suggestion of benefit even among patients treated within 4 hours. Inadequate penetration of the drug into the brain was also considered. In a preliminary study of central nervous system penetration in 1 acute stroke patient, gavestinel was detected in the cerebrospinal fluid 2 hours after the start of intravenous administration of the drug and gradually increased.³⁷ However, no testing has been done to establish whether the compound reaches the infarct core or penumbra. We also considered whether the study population included too many mild and lacunar strokes. The majority of patients in the GAIN study had larger infarcts with some involvement of the cortex, where NMDA receptors are concentrated and where the drug was expected to have the most beneficial effect. We also considered whether our failure to detect a treatment effect was due to insensitivity of the outcome scales. Although the scales used in this trial are easy to use, have good reliability, and have been used in many acute stroke trials, there is still debate on how they should be used and interpreted in clinical trials.³² These scales may not be sensitive enough to detect small treatment differences, and have "ceiling" effects in detecting differences at higher levels of functioning. Moreover, the BI does not assess cognitive or perceptive functions, which have a marked influence on global outcome following stroke. Given the concordant findings using multiple outcome measures in the GAIN trials, it is unlikely that choosing another outcome scale would have altered our conclusions.

The GAIN Americas trial, the largest phase 3 acute stroke trial conducted in North America, had some unique design features and was well executed. The study failed to show a neuroprotective effect for gavestinel when administered within 6 hours of acute ischemic stroke. This glycine antagonist joins the growing list of neuroprotectants that have not shown improved outcomes for patients with acute stroke, despite promising preclinical results. We still believe neuroprotection remains a viable strategy for acute stroke treatment and should continue to be studied.