Galasko’s presentation offered a glimmer of hope when he described results of a follow-up examination, conducted after treatment was suspended. That analysis hinted at a possible belated clinical benefit for a low dose of the study drug; however, that only became evident after most patients had already been taken off the treatment. Although far from conclusive, the findings may serve as a cautionary tale for adaptive-style trials that rely on interim analyses, said Galasko.

RAGE is a cell-surface receptor of the immunoglobulin superfamily. It binds advanced glycation endproducts (AGEs); these are modified forms of lipids and proteins that become glycated when exposed to sugars. When bound to their receptor, AGEs, which form during normal aging and in higher amounts in patients with diabetes, cause inflammation and oxidative damage to cells. Recent work has shown that RAGE also binds amyloid-β (Aβ) and mediates toxic effects of Aβ oligomers in neurons (see ARF related news story) The PF-04494700 RAGE inhibitor blocks this interaction, so the hope was that it would provide a combined treatment effect across inflammatory and amyloid-related processes. In preclinical studies, the compound decreased brain Aβ load in transgenic mice and improved their performance on behavioral assays. A prior 10-week-long Phase 2a safety trial showed a good safety profile of this drug in patients with AD, but no changes on clinical measures (Sabbagh et al., 2011).

The trial Galasko described at CTAD (see on ClinicalTrials.gov) was a proof-of-concept trial designed to test safety, tolerability, and efficacy of two doses of PF-04494700 compared to placebo over 18 months. It enrolled 399 AD patients randomized to one of three groups: The first (135 patients) received 20 mg per day of the drug, the second (132 patients) received a 5 mg dose, and the third (132 patients) were on placebo. All patients were over age 50, with Mini-Mental State Examination (MMSE) scores between 14 and 26. Many patients were on AD medications, such as acetylcholinesterase inhibitors or memantine, but not on diabetic or immunosuppressive drugs, as these might interfere with RAGE’s mechanism of action.

The researchers calculated ADAS-cog (Alzheimer's Disease Assessment Scale-cognitive subscale) scores as the primary outcome measure, along with various safety indications. As secondary outcomes, they conducted additional cognitive tests and collected data on structural magnetic resonance imaging (MRI) measurements, Aβ imaging using positron emission tomography (PET), and levels of cerebrospinal fluid (CSF) biomarkers Aβ and tau on subsets of patients.

According to the trial’s design, researchers were to take measurements at baseline and every three months thereafter up to 18 months. The trial was not Bayesian, but it did include some adaptive features—mainly, it could change course based on the results of two interim analyses conducted by the Data and Safety Monitoring Board (DSMB), an independent group of experts that advises trial investigators.

The first interim analysis, which was conducted six months after half of the intended subjects had been randomized, was purely a safety analysis. It showed that patients receiving the higher dose of the drug had more serious side effects, and their ADAS-cog scores worsened more quickly compared to the other two groups. As a result, that arm of the trial was stopped. The low dose was safe. The second interim analysis, conducted a year later, was a futility analysis to assess whether the low-dose treatment appeared to be providing a prespecified level of benefit. It narrowly missed that level; hence, all patients were taken off the medication and the trial was essentially over. But researchers continued to follow participants for one more visit, up to three months later, to learn more; about half of the original patients completed a last visit at 18 months.

When Galasko and colleagues analyzed the complete dataset including all 18-month measurements, things got interesting (or confusing, depending on one’s point of view). They found that the rate of cognitive decline in the high-dose group had slowed down, and their ADAS-cog scores were now similar to those of the placebo group, at least for those patients who continued to return for follow-up visits. Galasko could not explain the faster cognitive worsening in this group or their subsequent stabilization of cognitive symptoms. “We don’t know the mechanism of toxicity. We did not find any evidence of vasogenic edema or any other abnormality on MRI,” he said.

When the researchers analyzed the results of the 69 patients in the lower-dose group who completed the 18-month analysis, they found post-hoc evidence of improvement in their ADAS-cog scores compared to the 68 patients in the placebo group. There were, however, no differences in other clinical outcome measures or in the rates of hippocampal shrinking or CSF biomarker measures. This improvement in ADAS-cog scores was not detected at the 12-month timepoint, and it disappeared if the researchers only analyzed test scores obtained from patients while they were receiving the study drug.

Although the clinical trial did not provide evidence to support continued development of PF-04494700, Galasko said ending the trial early may have prevented the scientists from seeing any positive effects that required more than 12 months to become apparent. “We need to be careful about adaptive designs and stopping rules if we want to get the maximum amount of information about a drug,” he said. Other researchers at the meeting pointed out that the findings are difficult to interpret, given that the trial was stopped early and so many patients dropped out. This is the second investigational AD treatment Pfizer discontinued recently (see ARF related news story).—Laura Bonetta.