This is Part 1 of a two-part story. See also Part 2.
5 August 2008. On the last day of the International Conference on Alzheimer’s Disease (ICAD), held 26-31 July in Chicago, Paul Aisen of University of California, San Diego, used his plenary lecture before a still-respectable audience to put recent disappointing clinical trial results in Alzheimer disease into a broader perspective and to paint a way forward. Aisen heads the Alzheimer’s Disease Study Group, an NIA-funded cooperative involving some 50 U.S. and Canadian centers that was originally set up by the late Leon Thal and colleagues. He consults extensively for companies developing AD drugs.
In the past year, Neurochem’s Alzhemed and Myriad’s Flurizan failed Phase 3 trials, and Elan’s Phase 2 results of its passive immunotherapy bapineuzumab were clearly positive only in a subgroup of patients. These results have revived debate about the validity of the amyloid hypothesis, with some scientists arguing the hypothesis is wrong and others insisting it has still not been truly tested in the clinic. A recent paper about how the Phase 1 participants of the original AN1792 vaccine fared years later has fueled the dispute further (see Holmes et al., 2008). Suddenly, the terms “Baptists” and “Tauists,” which the field had retired years ago to most everyone’s relief, were back in currency again. Rather than enter into the heat of this controversy, Aisen took a step back and explained broadly how clinical trials in AD have evolved over the past 15 years, why current experimental treatments are facing a particular technical hurdle in Phase 2, and what could be done to overcome it. This story is largely a summary of Aisen’s plenary lecture.
On a timeline of drug development, the 1980s were the decade of acetyl cholinesterase development, the 1990s the decade of their clinical trials, 1997 to 2001 saw the approval of the three such drugs that are in wide use (donepezil, rivastigmine, galantamine), then memantine came along in 2003. All those are symptomatic. Currently, the field focuses intensely on generating and testing disease-modifying drugs. By Aisen’s estimate, the years 2010 to 2015 are when the first ones might reasonably be expected to come on line; however, to make their regulatory approval possible, methods of trial design and drug testing have to change.
In the 1990s, the FDA and leaders in the field jointly developed approval guidelines for the first wave of drugs (see also ACT-AD workshop). Trials had to have twin primary outcomes—a memory/cognitive test plus a global or functional measure. For the former, researchers largely settled on using the ADAS-Cog; it worked well for the mild to moderate population that enrolled in the trials, Aisen said. The CIBIC-plus scale worked as a global measure, as did the CDR-SB, the ADCS-ADL, and the DAD. For these symptomatic drugs, these measures enabled a separation between placebo and treatment curves on both cognition and clinical outcome. This was true even in short trials of 12 weeks, in which placebo groups generally do not decline. The effects were small—about two points on the ADAS-Cog. But they were statistically significant, and the FDA never mandated a minimum effect size.
“This was a straightforward pathway for symptomatic AD,” Aisen said. It led to approval of five drugs (including tacrine, which sees little use because it can damage the liver), and the FDA’s guidelines are still the same. But many scientists feel that the guidelines are not adequate for current science anymore. To this day, Aisen said, there has been neither a successful MCI trial nor prevention trial. Perhaps most urgently, no disease-modifying drug has garnered regulatory approval. One part of the problem is that the ADAS-Cog is insensitive to changes in very mild AD, a population that by a strong consensus of AD researchers stands a better chance of responding to drugs than people who have sustained damage to their brain for several years longer.
What can be done? For one, the ADAS-Cog can be made more sensitive for earlier stages of AD, i.e., MCI or CDR 0.5, for example, by adding executive function and delayed recall tests. Moreover, there is room to improve its analysis as well as its standardization, Aisen said. Alternative measures such as the NTB deserve attention, and computerized batteries are being developed, though in Aisen's view they are not quite ready for widespread use in ADCS trials. All these areas were topics of presentations at ICAD in Chicago. These are nuts-and-bolts kinds of tweaks that are entirely doable in the near future, Aisen said. In addition, it is hugely important to make an effort to recruit and especially to train more clinical sites, that is, principal investigators, study coordinators, raters, trial monitors. Equally important are efforts to recruit patients and design the practical experience of trials such that more patients and their caregivers stay in the trial to the end. All those efforts are necessary as a greater number and range of experimental therapies are leaving preclinical research and getting ready for human tests.
The goal of obtaining a disease modification label from the FDA poses a trickier challenge. From the regulatory point of view, the FDA at present demands a complex trial design of randomized start and withdrawal, whereas the European counterpart EMEA is open to simpler designs. Both agencies are open to the field’s increasing use of biomarker data, though not yet in lieu of cognitive clinical outcomes. The randomized start design means that one half of patients takes drug from the beginning of the trial, whereas a second set is first randomized to placebo and at 18 months switches over to active drug. This is meant to distinguish between symptomatic and disease-modifying drugs in this way: with a symptomatic drug, the second group of patients should catch up cognitively with the first, because a symptomatic drug merely produces an initial hump of benefit but then patients decline at the same rate, or parallel slope, as placebo-treated patients. With a disease-modifying drug, the second group of patients would never catch up to the first, because the earlier group’s rate of decline has slowed already in the previous treatment period and the curves of the two sets of patients diverge.
While this is nice in theory, it’s fair to say researchers and drug companies do not love this design. The second randomization doubles the length of a typical Phase 3 trial to 36 months; given present dropout rates of 15 to 35 percent, that is simply impractical, Aisen said. Aisen showed two recent examples of trial designs. The Alzhemed Phase 3 design of parallel 18-month placebo/treatment arms followed by a joint open-label extension is a fine design that should work. With an effective drug, that is. In this case, the failure was due to the drug itself, as well as to variations between the clinical sites. In the Dimebon Phase 2 trial, the data at 12 months looked like the combined curves of a disease-modifying and a symptomatic effect. That is, treated patients rapidly improved above baseline in the first 12 weeks and then declined more slowly than the placebo group (Doody et al., 2008). In this case, the slopes behaved in this mixed way in a standard trial design that also did not employ randomized start.
The message to take away at this point, Aisen said, is that getting the disease modification label is really not critical. It’s not the label companies should worry about. The bigger and more treacherous methodological challenge is how to optimize trial design in Phase 2 for drugs that are believed to work by a disease-modifying mechanism. This is the Phase 2 problem.
Where exactly is the dilemma? It is that with a disease-modifying drug, scientists cannot expect to see proof of efficacy in a Phase 2 trial. With the symptomatic drugs, they could. Why is that? Aisen explained that the underlying biology suggests that disease-modifying drugs will produce no short-term benefit but rather lessen the slope of a person’s decline curve over time. They still get worse, but less so. Over the long run, that would serve the patient well, but in the relatively short time frame accessible to Phase 2 trials, it’s a disadvantage over a quick symptomatic boost. What’s more, people with very mild AD or with MCI/CDR 0.5 do not decline at all in six months and only minimally by 12 months on the existing standard measures. To see the slope of decline curves diverge, a trial needs to enroll many hundreds or thousands of people and follow them over 18 months—but that is basically the recipe for a Phase 3, not a Phase 2. So companies are faced with the choice of running hugely expensive Phase 2 trials or moving into Phase 3 without proof of efficacy. “There currently is no rational progression from a small Phase 2 to a large Phase 3,” Aisen said. This means, in essence, that Phase 3s can be as risky as Phase 2s. In the absence of Phase 2 efficacy data, Phase 3 trials of disease-modifying drugs can be expected to fail eight out of 10 times, Aisen said. A Phase 3 program costs $200 million. The annual cost of AD to the nation currently is estimated at $100 billion. “The potential gains justify even this high risk,” Aisen said. To see how companies deal with the Phase 2 problem and how it can be circumnavigated, see Part 2 of this story, to be posted tomorrow.—Gabrielle Strobel.
This is Part 1 of a two-part story. See also Part 2.