Design of Stroke-Related Clinical Trials

Phase 1

Phase 1 studies establish doses in humans that do not have high toxicity, establish a route of administration, and determine the clinical pharmacology. The general bound for dose exploration is determined by the severity of the condition being treated. In studies of cancer chemotherapy, where the underlying disease is almost universally fatal, relatively high levels of sub-lethal side effects may be acceptable in therapeutic agents. In that setting, the intent of a phase 1 trial is to discover the maximum tolerated dose (MTD). The occurrence of toxicity that is unacceptable is termed dose-limiting toxicity (DLT); hence, the MTD is determined by the absence of a DLT, with the MTD being one dose level below a DLT.

For most stroke indications, the cerebrovascular disease produces a range of morbidity and mortality, rather than a uniformly fatal outcome, making drug doses with substantial toxicity less acceptable. Accordingly, there is no consensus regarding the need to escalate the dose to the MTD. The choice of the dose range to be explored for toxicity in initial phase 1 trials is, instead, often based either on doses found to show efficacy for stroke in animal experiments or efficacy for other diseases in humans. The upper bound of escalation chosen for a phase 1 study would be treatment specific and may depend on such factors as whether the agent is to be used for long-term prevention among mild stroke patients (in which case the occurrence of even infrequent toxicity events might be unacceptable), or whether the agent is to be used short term to prevent major disability or death amongst severe stroke patients (in which case a higher toxicity rate could be acceptable). Also less well defined in phase 1 studies for stroke are standard definitions for varying levels of toxicity below DLT. A phase 1 study funded by NINDS aimed to establish the MTD of human serum albumin as a neuroprotective agent for participants with recent ischemic stroke. A Safety Evaluation Committee of neurologists and cardiologists evaluated participants’ records at each dose level according to pre-specified guidelines to determine whether severe and serious adverse events occurred during the 72 hours after ictus. The existence of DLT was determined by a consensus after each member of the committee reviewed the charts of all the participants at each dose level.

The selection of initial, lowest dose to be tested in phase 1 dose-escalation trials varies. In cancer trials, a common approach starts from one-tenth of the dose that causes 10% mortality in rodents (LD ₁₀ ). In stroke, when toxicities are less acceptable, a starting dose might be one-tenth of the dose that causes maximal efficacy in animals. The dose is increased by a smaller percentage each time. A classic escalation sequence is a “3+ 3” modified Fibonacci scheme, in which 3 patients are treated at each dose, escalating to the next dose if no toxicities are observed or testing an additional 3 patients if a toxicity is observed. Continual reassessment methods (CRMs) are an alternative design for a phase 1 study and use information from all prior-tested patients to choose the most informative dose to test on the next patient, via either frequentist or Bayesian methods. Simulation studies show that the MTD may be reached sooner with some CRM methods than with the Fibonacci method, putting fewer participants at risk from higher doses. For example, the trial evaluating the iron chelator, deferoxamine mesylate (DFO) treatment was a phase 1 dose-finding trial using an algorithm for continual reassessment every third patient.

Phase 2

Phase 2 studies are important bridges between early pharmokinetic and overt toxicity studies in phase 1 and large, pivotal efficacy trials in phase 3. Phase 2 studies do not seek to draw definitive conclusions about treatment efficacy but do often seek evidence to justify proceeding to an expensive and time-consuming phase 3 study. Phase 2 studies primarily rule out clearly ineffective treatments; they assess futility, side effects, toxicity, logistics of treatment administration, and project trial costs. Palesch et al. explained the futility design aspects in therapeutic stroke trials. In phase 2 stroke-prevention trials, a physiologic or surrogate outcome, such as a risk factor reduction, may be used as the primary outcome, when the lead outcome for a future phase 3 study, new stroke or death, would take years to ascertain. In phase 2 therapeutic trials, biomarker surrogate or auxiliary outcomes, such as reduction of infarct growth or hemorrhage growth on brain imaging, ^, may be used as the primary outcome, when it is judged that the biomarker has less random variability and will allow a smaller sample size than more variable clinical endpoints. If no such biomarker exists for a particular agent’s mechanism of action, therapeutic phase 2 stroke trials will use the same clinical outcome measures as planned for phase 3 trials. Phase 2 trials are informally categorized further as phase 2A and phase 2B, with phase 2A trials very preliminarily and phase 2B trials more powerfully determining readiness of particular doses and approaches to move on to definitive phase 3 testing.

Using a one-sided test and inflated type-I-error level enables phase 2 studies to proceed with smaller sample sizes, less time, and fewer resources than phase 3 studies. Single-arm phase 2 studies using historical data as a reference have the advantage of requiring even smaller sample sizes than a study with randomized, internal controls, with the same error and effect size parameters. A critical assumption of a single-arm design is that the concomitant clinical care and outcomes of the disease have not changed over time, so that historical controls are an accurate representation of current patients. Another critical requirement is that the same outcome measure is collected in the same way in the new trial as in the historical comparator. For example, in the two NINDS rt-PA Study trials, a common source of historical control data, the modified Rankin Scale, was scored assessing disability from all diseases present in the patient, not just those inferred by the examiner to be due to stroke, and the National Institutes of Health stroke scale (NIHSS) was collected under the instructions “score what you see,” rather than encouraging repetition of the exam if the examiner had expected a different response based on disease knowledge. Some newer trials changed the instructions for the Rankin and/or the NIHSS to make the measures more specific to the current stroke or pathophysiologic expectations, rendering the scores not directly comparable to the prior trial. If there is uncertainty about the validity of historical controls, a concurrent comparison group must be included in the trial, but the single-sided hypothesis and an inflated type-I-error level would still lead to a reduction in sample size over a phase 3 design. ^, A completed phase 2 stroke treatment trial of combined intravenous and intra-arterial tissue plasminogen activator (t-PA) with historical controls was determined to be worthwhile and moved forward to phase 3. ^,

Phase 2 studies may also be used to choose a dose from a set of tolerable doses or to select the “best” among a set of treatments. In a study of amyotrophic lateral sclerosis (ALS), investigators combined a dose-selection study with a phase 2 futility study, a design that may be applicable to dose-finding studies in stroke. Bayesian dose-finding studies may also be applicable and have been used in several phase 2 trials, including the Acute Stroke Therapy by Inhibition of Neutrophils (ASTIN) trial and the Argatroban with Recombinant Tissue Plasminogen Activator for Acute Stroke (ARTSS-2) trial. Bayesian dose-finding trials require (i) consensus agreement regarding the level of prior knowledge or, more frequently, the use of neutral priors, (ii) confirmation as to whether outcomes can be assessed rapidly, and (ii) evaluation of the investigators’ ability to reduce bias. ^,

A phase 2 design is especially attractive when the availability of participants is limited in relation to the rate of the development of new treatments. A successful phase 2 study does not guarantee a successful phase 3 trial. ^, Nevertheless, using a phase 2 design can reduce the number of futile phase 3 studies and can provide an overall reduction in the long-term cost of trials. ^{, ,}