Alzaforum comments on Biogen's results
The last full day of the Alzheimer's Association International Conference, which ended today in Chicago, saw scientists pack a room the size of an aircraft hangar in anticipation of a late addition to the scientific program. They came to see the data behind a tantalizing press release issued earlier this month, which had claimed that BAN2401, the anti-Aβ protofibril immunotherapy being developed by BioArctic, Eisai, and Biogen, reduced amyloid in early Alzheimer's disease and also slowed cognitive decline. The upshot? According to results presented in Chicago by Eisai's Lynn Kramer, the antibody appears to have done what it was designed to do. Over 18 months, fibrillar amyloid fell in all treatment groups compared with placebo; indeed, plaques melted by a whopping 93 percent in participants on the highest dose. This dose was reported to have reduced cognitive decline by 47 percent as measured by the ADAS-Cog, and by 30 percent on the ADCOMS, a new composite measure to detect early cognitive decline. At 856 participants with MCI due to AD or mild AD, this trial is the largest one yet to post both amyloid reduction and a downstream benefit on symptoms.
But wait. There's a catch. The data set is complicated. That's in part because this Phase 2B trial started out with adaptive randomization, which means that as participants accumulated, new enrollees were more apt to be put on doses that looked likelier to be effective. The trial featured a Bayesian statistical analysis at 12 months, and ended with a more conventional, aka “frequentist,” analysis at 18 months. That was complex, but the scientists had planned it that way.
Unplanned, however, was a curve ball regulators threw their way in July 2014, well over a year after the trial had started enrolling. Driven by safety worries about ARIA-E, the European Medicines Agency insisted that APOE4 carriers be excluded from the highest dose, 10 mg/kg infused biweekly. By that point in time, 74 carriers had been enrolled at that dose. What's more, the EMA demanded that all APOE4 carriers who had been on that dose for less than six months stop the trial, effectively kicking out 26 people who had no ARIA-E. This left 48 in the highest dose group. ARIA-E is a temporary inflammatory reaction typically seen within the first three months of Aβ immunotherapy; it is being studied intensely by most companies developing anti-Aβ antibodies. Most instances of ARIA-E occur within the first three months of dosing.
Kramer said that the trial sponsors argued against this restriction, to no avail. Paul Aisen from the University of Southern California, San Diego, said this regulatory request was odd, and difficult to understand. The EMA’s constraint likely compromised the adaptive randomization, said Andy Satlin, who initially led the BAN2401 program at Eisai and is now at Intra-Cellular Therapies in New York. This may have contributed to an outcome whereby the dose that proved to be the best received a lower number of participants than the second-best dose. In effect, the adaptive algorithm, which was increasingly leaning toward the highest dose as the trial went on, bumped up against the EMA’s APOE4 constraint, and a disproportionate number of participants ended up in the nearest dose. Suzanne Hendrix, a biostatistician and president of Pentara Corp., Salt Lake City, told Alzforum the restriction in effect biased comparisons with the placebo group. "The trial ended up with a different population in those two arms. That is not a true randomization anymore," Hendrix said.
Hendrix noted that among people with MCI and early AD, APOE4 carriers tend to decline faster on the ADCOMS, a cognitive composite she co-developed in hopes of picking up subtle changes at early disease stages (Wang et al., 2016). Outside this ADCOMS data, the idea of APOE4 speeding up progression is not established in the field, however. AD scientists do agree that APOE4 hastens amyloid deposition and brings on symptoms at a younger age, but they not agree on an effect on progression. “APOE4 carriers have an earlier onset, but their expected rate of decline is not different from patients with other APOE genotypes. However, there are conflicting reports in the literature on this topic,” wrote Lars Lannfelt of Uppsala University in Sweden, who made the mouse antibody that led to BAN2410 (see full comment below).
Despite the skewed randomization, statisticians and clinicians who gathered in the hallways after the presentation were cautiously upbeat. "Overall the results are positive and the amyloid effect is impressive," Hendrix said. "I believe this antibody works," Aisen agreed.
So do others. “In summary, there is dramatic amyloid lowering, with some apparent slowing in decline at the highest dose,” wrote Stephen Salloway of Butler Hospital, Providence, Rhode Island. David Holtzman of Washington University, St. Louis, considers the overall data “very promising” (full comment below), and Jeffrey Cummings of the Cleveland Clinic in Las Vegas thinks the results support target engagement and suggest a dosing strategy for phase 3. Randy Bateman of WashU wrote, “The field is clearly moving forward with the ability of a fourth drug to remove amyloid to a normal level, as measured by PET. Now with aducanumab, gantenerumab, and n3pg, BAN2401 has demonstrated reversal of amyloid plaques to normal levels, representing a milestone in the history of Alzheimer’s disease” (full comment below). Other commentators quibbled that the sponsors could have anticipated scrutiny over the APOE4 distribution and included that subgroup analysis in their AAIC presentation.
As the trial was enrolling, scientists conducted frequent interim analyses to steer new enrollees toward the doses that appeared early on to be most likely to work. This was done in hopes of making the trial more efficient (Satlin et al., 2016). It soon turned out that the action would be at the higher doses, and by the end of enrollment, 161 people were on 10 mg/kg biweekly, 253 people on 10 mg/monthly, and 247 people were on placebo. Only 52 people were in the 2.5 mg/kg biweekly dosing group, 51 in the 5 mg/kg monthly, and 91 people in the 5 mg/kg biweekly groups. Following the EMA request, the second-highest dose group ended up chock-full of APOE4 carriers, at 89 percent, compared with 30 percent in the highest dose group, and 70 percent in the placebo group.
At 12 months, a Bayesian analysis of the ADCOMS results estimated that the highest dose was 98 percent likely to slow decline. Separately, the trial's designers had previously set the criteria by which they would declare success at that time point to be an 80 percent likelihood of a clinically significant difference of 25 percent slowing of decline from baseline; the 12-month analysis calculated this likelihood to be 64 percent. The 80 percent was an awfully high bar, Hendrix told Alzforum, and may have been influenced by a push in the field at the time to reach for large effect sizes.
Bayesian trials had been debated in the Alzheimer’s field, but rarely put into practice. At AAIC, several commentators wondered if a simpler design could have worked as well. Hendrix expressed similar thoughts. "It's possible that with fewer dose arms, they might have met the primary endpoint at 12 months,” she said. On the other hand, Satlin said, the strong p values of the 10 mg/kg dose results may only have been achieved because the adaptive algorithm placed disproportionately more patients into this group than a conventional trial of the same size would have done.
So what were the results? Across the board, the antibody reduced amyloid in the brain. A time- and dose-dependent reduction saw PET SUVRs fall by up to 0.3 units in those on the highest dose, a 93 percent drop. Kramer said that on manual reads of the scans, 81 percent of treated patients went from amyloid-positive to amyloid-negative.
For the cognitive analysis, Kramer focused on the two highest doses. The 10 mg/kg biweekly group had a 47 percent reduction in cognitive decline as judged by the ADAS-Cog, and a 30 percent reduction on the ADCOMS. A 26 percent slowing on the CDR-SB was not significant. The 10 mg/kg monthly dose group—the one containing mostly APOE4 carriers—performed about midway in between placebo and the highest dose, showing a trend toward slower cognitive decline on the ADAS-Cog, ADCOMS, and CDR-SB but no statistical significance. The placebo group declined at a similar rate to placebo groups in recent large AD studies.
In a subgroup of patients who underwent spinal taps, CSF Aβ42 rose dose-dependently in all treatment arms, to more than 300-fold the level of the placebo group. Kramer said this was expected as the antibody pulled the peptide from plaques into the soluble fraction. Scientists generally believe that in the run-up to Alzheimer's dementia, CSF Aβ42 decreases as amyloid plaques deposit in the brain, sequestering it there. Kramer said studies were ongoing to determine how much of this Aβ was free versus bound to antibody. For CSF total tau, there appeared to be a slight reduction in the top two doses combined, though there was considerable scatter in the data.
And how about ARIA-E? As with other antibodies, it occurred in a dose- and APOE-dependent way, 48 times total across all groups, including two in placebo. Just shy of 10 percent of participants in the highest dose had an episode of ARIA-E, as did seven of that group’s 48 APOE carriers. Most ARIA-E was detected only on MRI, though five instances caused headache, visual disturbances, or confusion; two of those were in the highest dose, Kramer reported. In toto, this amounts to less ARIA-E than seen with gantenerumab or aducanumab.
Where does BAN2401 go next? "We view this as robust enough to approach regulatory authorities to discuss next steps in terms of additional trials or even breakthrough status," Kramer said in a press briefing before the main presentation.
In the general media, on social media, and among analysts, the results received the full range of responses, from an enthusiastic thumbs-up to a merciless drubbing over the uneven APOE4 carrier allocation. Both Biogen and Eisai’s stock prices dropped about 10 percent, but are starting to recover.—Tom Fagan and Gabrielle Strobel