12 June 2007. For most people in AD research, a conference is the best place to pick up the latest and greatest on treatment trials, because those data are either published with long delays or not at all. The second Alzheimer’s Association International Conference on Prevention, held 9-12 June in Washington, D.C., had a fair amount of trial data to offer. The research comes not a minute too soon. Worldwide, 26.6 million people have Alzheimer disease today and projections put the number of cases at 100 million by the year 2050, according to estimates presented at the conference by Ron Brookmeyer of Johns Hopkins Bloomberg School of Public Health in Baltimore, Maryland. Here are selected tidbits of trial data.

Alzhemed
If you were on the edge of your seat about the much-anticipated results of the first phase 3 trial of Alzhemed, which is, after all, the first anti-amyloid and purportedly disease-modifying drug for AD that has gotten this far, you can safely slide back, exhale, and prepare to wait a little longer. Paul Aisen of Georgetown University yesterday faced a standing-room-only press room packed with reporters and even some curious senior scientists from competing companies who did not want to await the formal scientific presentation scheduled for 2 hours later. But Aisen had to tell them that he had no data to show even though the company Neurochem of Laval, Quebec, had previously said that they would present the results of this pivotal trial at this conference. The problem? The 67 centers participating in this U.S.-Canadian trial had sent raw data that varied so greatly from one center to the next that the variability essentially rendered the statistical model by which the data were going to be analyzed unworkable. “The data is a poor fit to the statistical model,” Aisen said. Hence, the statisticians are now trying to modify the statistical model to better fit the data, and only then will they be able to analyze the endpoints. Aisen said that he expected this process to take another month.

It is important to realize that this does not mean that the trial is known to have failed to reach its primary endpoints and that its scientists are conducing a post-hoc analysis to find some promising hints amid the data. The changes to the statistical method happen while the data remains blinded. It simply means that there is no efficacy or safety data at this point.

This trial was a huge, expensive effort, and problems with it will harbor lessons on what to avoid in the future as other experimental drugs approach phase 3 and will have to be tested in equally large studies to have adequate power for make-or-break trials. This 18-month trial enrolled 1,052 people 50 years and older with mild to moderate AD at 50 U.S. and 17 Canadian centers. The people were randomized to take either placebo or 100 or 150 milligrams of drug twice a day of Alzhemed (aka tramiprosate), a patented variant of the amino sulfonic acid taurine that is thought to bind to the Aβ peptide and interfere with its fibrillization. All took an acetylcholinesterase inhibitor; memantine use was optional. The trial used psychometric test batteries as well as structural MRI and CSF Aβ and tau biomarkers as outcome measures. Seventy-five percent of patients completed the trial.

But when it came time to analyze, the researchers found that there was a highly significant center effect. The reasons Aisen was able to give at this point for the diverging results included that some patients were on memantine while others were not, the patients took memantine for differing lengths of time, some patients went on or off antidepressants during the time, some people changed their acetylcholinesterase inhibitor dose, and others took vitamin E (though vitamin E has been shown in a large clinical trial to have little effect, at least in amnestic MCI; see Petersen et al., 2005). The variation among sites was so great that it violated the assumptions underlying the statistical model the study set out to use. Consequently, the scientists, with input from the FDA, are currently working to identify the confounding factors and include them in the statistical model, Aisen said. Then they will enter the actual data for the treatment groups into the adjusted model.

Aisen presented no numbers. He said that preliminary analysis suggested a treatment effect on the cognitive endpoints and on the MRI, i.e., the disease-modification endpoint. The trial’s woes highlight the difficulties of conducting large, long trials. Apparently, people changed the doses and combination of their background medications during the trial. In theory, they should not do that, but of course they are patients first and trial subjects second, so their doctor will act in their best interest. These patients can be made to leave the trial, but that weakens the data even further because regulatory rules require that the analysis include all patients who are randomized into the trial, not only those who took all trial drug to the end (called intent-to-treat analysis). A large trial magnifies these problems, Aisen said. Whatever the outcome of this particular trial, Aisen said the experience with it will teach the field about how to conduct such large trials.

Neurochem scientists also presented a series of posters. One showed tramiprosate, or 3-amino-1-propanesulfonic acid, to simply be taurine elongated by a further peptide bond. (Taurine itself is a common ingredient in infant formula and energy drinks.) The posters claimed that tramiprosate is neuroprotective by interfering with various proposed pathways of Aβ toxicity in slice and cell-based assays.—Gabrielle Strobel.