This is a two-part story. See also Part 2.
28 July 2008. It is an observation noted so often that it’s taken as gospel: in Alzheimer disease clinical trials, placebo groups seem to be worsening at a more leisurely rate than they once did. If so, AD drug makers face serious problems. The newest generation of AD therapeutics tend to slow disease progression rather than improve symptoms, so if placebo groups do not deteriorate at predictable rates, perfectly effective drugs could slip under the statistical radar.
Much of the discussion on waning placebo decline has centered around the ADAS-Cog, an 11-item portion of the Alzheimer’s Disease Assessment Scale that measures cognition and serves as a primary outcome measure in many AD drug trials. “What was bandied around back in the early 1990s is that decline on the ADAS-Cog would be six to nine points per year,” said Rachelle Doody of Baylor College of Medicine in Houston, Texas, noting an earlier analysis (Stern et al., 1994) that produced an equation for predicting progressive ADAS-Cog decline from baseline scores. “We never see this much decline now. Four points at one year is typical.”
Several studies presented this week at International Conference on Alzheimer's Disease (ICAD) use different approaches to address this conundrum. Tomorrow afternoon, Lon Schneider of the University of Southern California Keck School of Medicine in Los Angeles, will present results from a meta-analysis that essentially debunks the notion that placebo participants have declined more slowly in recent AD trials. Roy Jones, director of the Research Institute for the Care of Older People (RICE) in Bath, U.K., and colleagues presented a poster yesterday showing that placebo patients in AD trials of donepezil initiated between 1995 and 1999 have declined more slowly than placebo groups of earlier (1990-1994) donepezil trials. And in a poster presentation Wednesday (see ARF companion story), Michael Irizarry and other GlaxoSmithKline colleagues in Research Triangle Park, North Carolina, and several U.K. sites will share their analysis identifying baseline participant characteristics that predict subsequent ADAS-Cog decline in placebo groups of AD trials. Despite differing design, data sets, and conclusions, these studies underscore the field’s intensifying need to gauge placebo decline more effectively in future AD trials.
A Test of Urban Legend
After “sitting in meetings and listening to experts say the placebo groups are not responding as they’re supposed to,” Schneider decided to apply some rigorous number-crunching toward a test of this claim. He and colleagues reasoned that an analysis of placebo data pooled from many trials could most effectively evaluate this widespread perception, based largely on unfavorable results from individual trials. The researchers combed the literature and hit up company officials for raw data from randomized, double-blind, placebo-controlled AD trials lasting at least six months. They found 103 such trials between 1991 and 2005 and were able to extract the following information from 87: trial size, countries, number of sites, treatment allocation ratios, enrollment dates, age, gender, ADAS-Cog, and Mini-Mental State Examination (MMSE) scores. More than a third of the studies in his meta-analysis were unpublished, Schneider told ARF.
Consolidating the placebo data from these 87 studies, Schneider and colleagues found no changes in baseline characteristics or cognitive change scores across the 15-year period of the trials. Collectively, the placebo subjects showed mean MMSE declines of -0.73 (SD = 0.94) and -2.77 (SD = 0.94) at six and 12 months, respectively. ADAS-Cog changes at those time points were 1.44 (SD = 1.58) and 4.13 (SD = 2.39).
"When you look at this, it becomes a numbers thing,” Schneider told Alzforum. “Sampling and play of chance alone may lead to not showing the decline.” Consider a six-month, European Phase 3 trial of phenserine for probable AD, whose lack of efficacy in cognition, global function, or activities of daily living was chalked up to an unusually slow rate of placebo decline. Those unpublished data—presented by Bengt Winblad (Karolinska Institutet, Stockholm) at the 2005 International Conference on Alzheimer’s Disease and Parkinson’s Disease in Sorrento, Italy—marked a turning point for the field, causing many to cast their first suspicious glare on placebo decline patterns. And rightly so—at 12 weeks, the placebo group had improved more than a point on the ADAS-Cog and returned to baseline by six months.
Although the phenserine study was relatively large (375 patients), the placebo group was small (76 patients). Therein lies a problem, Schneider said. In his analysis to be presented tomorrow at ICAD, the size of the placebo arm was a critical factor in subsequent cognitive decline. Smaller studies show greater variability in mean placebo progression—some improve, some worsen, Schneider said. In a recent phone conversation, he drew an analogy to a coin toss experiment in which laws of probability predict 50 percent of the flips to turn up heads. However, in smaller samples—say, 10 flips—the number of heads could swing widely, maybe from one to nine. If the coin were tossed 1,000 times, the outcomes would fall within a range much closer to the expected 50 percent.
Similarly, Schneider’s meta-analysis of AD trials showed that at six months, placebo sample sizes less than 100 had a 37 percent chance of not showing significant change, whereas placebo groups of more than 200 all showed significant worsening. In 12-month studies, 95 percent of placebo groups larger than 100 declined significantly. “This is just measurement and sampling theory that I am demonstrating is a major problem in AD study design,” Schneider said.
In his meta-analysis, placebo groups from studies that allowed use of cholinesterase inhibitors showed about the same ADAS-Cog decline as did placebo subjects in studies from an earlier time when cholinesterase inhibitors were not allowed. This finding would seem to discredit the idea that use of cholinesterase inhibitors has led to slower placebo decline in AD trials of newer, disease-modifying treatments. “What this practically means is that you don't have to exclude patients on cholinesterase inhibitors from clinical trials, all other things being equal,” Schneider noted. Less deterioration was also linked with trials having non-English speaking sites, perhaps reflecting the challenge of using clinical instruments in demographics for which they were not originally designed. In addition, the researchers found less variance associated with more frequent assessments. While this point suggests that boosting the number of evaluations in clinical trials may improve reliability, it should be considered in context with a recent study (Gold, 2007 and ARF companion story) that found frequent assessments associated with practice effects.
Sid Gilman, a neurologist at the University of Michigan, Ann Arbor, who has broad experience in AD drug trials, studied Schneider’s data when the two met informally during a meeting several weeks ago. In a phone interview with ARF, Gilman said he was struck by the similarity between the mean ADAS-Cog changes for the pooled placebo patients in Schneider’s analysis and for the placebo group in an 18-month trial of folate and vitamins B12 and B6 to lower homocysteine in AD. “A single trial comes very close to what Lon was showing overall,” Gilman said of the homocysteine lowering study, with which he was not involved.
In that trial, the mean ADAS-Cog decline in 169 placebo patients was 1.12 points at six months and 4.46 at 12 months, compared with 1.44 and 4.13 points, respectively, in Schneider’s analysis. Data from the homocysteine lowering study were first presented last year at the Alzheimer’s Association International Conference on Prevention in Washington, DC, by lead investigator Paul Aisen, then at Georgetown University and now at the University of California, San Diego. Aisen is director of the Alzheimer’s Disease Cooperative Study (ADCS).
Further support for Schneider’s ADAS-Cog figures comes from placebo data of several other large, 18-month AD trials—of simvastatin and tarenflurbil (aka Flurizan)—presented at ICAD this week. In the statin and Flurizan trials, involving placebo groups of around 200 and 800, respectively, placebo decline on the ADAS-Cog hovered around 1.5 points at six months and four points at 12 months—mirroring the figures from Schneider’s meta-analysis. Disappointing results from the highly anticipated tarenflurbil trial, in which Schneider was a co-lead investigator, were reported in brief last month by Myriad Genetics, Salt Lake City, Utah, which has decided to discontinue the drug (see ARF related news story). None of these three 18-month trials were included in Schneider’s meta-analysis of placebo data.
Doody told ARF in a phone interview that she found “no surprises” in Schneider’s findings. But “you can’t take this and apply it as a critique of specific studies,” she said. As lead investigator in a recently published Russian trial of Dimebon, Doody noted that though the small placebo group of 77 mild to moderate AD patients worsened significantly (more than five points on the ADAS-Cog at six months), the drug effect did not depend on that decline. In that trial, patients on Dimebon actually had better ADAS-Cog scores at six and 12 months than at baseline (see ARF related news story).
Placebo Behavior in a Decade of Donepezil Studies
Jones, director of the Research Institute for the Care of Older People (RICE) in Bath, U.K., and colleagues took a closer look at the placebo decline dilemma using a different approach: a meta-analysis of placebo data from AD trials of the acetylcholinesterase inhibitor, donepezil, the world’s most prescribed Alzheimer’s medication. The researchers collected data on 3,403 patients who participated in 13 randomized, double-blind, placebo-controlled AD trials of donepezil initiated between 1990 and 1999. All pooled data came from mild to moderate AD patients (MMSE scores 10-26) who took part in AD trials that used the MMSE and/or ADAS-Cog to measure post-baseline cognition and were within a donepezil clinical development program run by Eisai Inc. and Pfizer Inc., the drug’s developers, wrote Jones in an e-mail to ARF. He said that data from trials run after 1999 were not included because double-blind studies of donepezil after this date focused on MCI, severe AD, and vascular dementia.
Jones and colleagues found that patients in more recent (1995-1999) trials had significantly lower mean MMSE decline (-0.56 points) from baseline to week 24, compared with those enrolling in earlier (1990-1994) trials (-1.28 points). Over the same time frame, placebo subjects of more recent donepezil trials also worsened less on the ADAS-Cog (1.03 points versus 1.82 points in placebo participants of earlier trials), but this difference was not significant. Jones noted that their data included more information on MMSE changes and less on ADAS-Cog changes, reflecting availability of information from the studies used in their analysis.
As disclosed in the ICAD meeting abstract, this study resulted from discussions and analyses by an expert working group initiated and funded by Eisai and Pfizer. In the U.S., donepezil is co-promoted by these two companies and distributed by Eisai. Jones received honoraria for his participation in the analysis, which was presented in an ICAD poster yesterday.
Jeff Cummings, director of the Alzheimer’s Disease Research Center at the University of California, Los Angeles, said in a recent ARF interview that he does not find Jones’s and Schneider’s findings incompatible. “Lon’s point is that larger, longer trials are more likely to be associated with placebo decline. That probably is also true of Roy’s data,” said Cummings, who has consulted and served on speakers’ bureaus for Eisai and Pfizer. “But Roy’s data also alert us to the fact that at least within some sample sets, there seems to be a trend toward less placebo decline in more recent trials.”
As they were done by the same companies, the trials in Jones’s study had relatively similar design. In addition, the researchers observed a comparable donepezil-placebo treatment difference in the earlier and later trial sets, suggesting a consistent treatment effect across the decade of studies. Such similarities presumably made their trials more homogeneous than those in Schneider’s analysis and allowed more in-depth analysis of design details, Jones noted. He did mention that the donepezil studies post-1995 tended to have somewhat smaller sample sizes than did the pre-1995 trials, lending support to Schneider’s suggestion that sample size is important for demonstrating placebo decline.
Strangely enough, the apparent “smarting up” of placebo subjects in more recent trials did not correlate with patient characteristics that would predict this effect. Jones and colleagues found that individuals enrolling in later (post-1995) trials had, in fact, lower baseline MMSE scores. Low screening MMSE was identified in the study by Irizarry and other GlaxoSmithKline colleagues as a significant independent predictor of 24-week ADAS-Cog decline in AD trials. However, that same study found that patients whose ADAS-Cog performance deteriorated markedly between screening and baseline were less likely to show significant 24-week cognitive decline. This would seem consistent with the finding by Jones and colleagues that patients with lower baseline MMSE tended to decline less at 24 weeks.
Also somewhat unexpected was the finding by Jones’s team that placebo subjects in later (post-1995) donepezil studies—who seemed to decline at slower rates—were older, had other medical conditions, and used more medications than did patients in the earlier group of trials. Jones speculated that the increased comorbidity and medication use could have stemmed from less restrictive entry criteria in later studies, presumably because more experience had been gained about the use of donepezil. That patients in the later studies appeared to deteriorate less could also result from advances in care and social support for AD patients, improvements in the management of AD risk factors, and general advances in medical care for older people, Jones noted.—Esther Landhuis.
This is a two-part story. See also Part 2.