Summary: A critical research letter warns that a novel statistical method used to support a new class of Alzheimer’s drugs can severely overstate clinical efficacy. The study evaluated quantile aggregation, a technique that groups patients, averages their outcomes, and tracks trends across those clusters.
Originally deployed in an influential reanalysis of Eli Lilly’s Alzheimer’s drug donanemab, the researchers discovered via computational simulations that this mathematical approach hides patient variability and can inflate the perceived relationship between amyloid clearance and cognitive improvement by a staggering 29 times its actual magnitude.
Key Facts
- The Method in Question: Quantile aggregation divides a patient cohort into specific groupings, averages their individual clinical results together, and looks for patterns across those aggregated blocks.
- The 29x Inflation Trap: In simulations tailored to mirror recent clinical trials, the quantile aggregation method exaggerated the strength of the link between clearing amyloid plaques and slowing cognitive decline by 29 times.
- Masking Patient Variability: By blending large groups of patients and averaging their outcomes, the technique effectively masks real-world variability in cognitive changes among individual patients, manufacturing a false sense of predictability.
- Breaking the Randomization: The method strips away vital clinical trial protocols by pooling together patients who received the actual drug with those who received a placebo, rendering it unable to determine if amyloid removal is causing cognitive benefits.
- Resurrecting Failed Trials: To prove the method’s flaws, researchers ran data from a completely failed 2014–2023 trial of the drug solanezumab through the quantile aggregation model. The method fabricated a strong, entirely false link between amyloid reduction and better cognitive scores where none existed.
- Academic Independence: Senior author Sarah Ackley highlighted that working outside of pharmaceutical industry financial incentives allowed independent academic researchers the freedom to rigorously audit consequential drug evaluation methodologies.
Source: Brown University
A statistical approach being used to support a new class of Alzheimer’s drugs may lead to overstated claims about how the drugs work, according to a new study led by researchers at the Brown University School of Public Health.
Published in JAMA Neurology, the research letter focused on quantile aggregation, a new statistical technique that divides people into groups, averages their results together and then looks for patterns across those groupings.
The letter examined how the approach works when applied to cognition and amyloid, a protein that builds up in the brains of people with Alzheimer’s disease. The approach was originally published in an analysis of Eli Lilly and Company’s Alzheimer’s drug donanemab.
“Many researchers believe reducing amyloid buildup could slow memory loss and cognitive decline associated with the disease, making it a major target for newer Alzheimer’s drugs,” said the study’s senior author Sarah Ackley, who is an assistant professor of epidemiology at Brown’s School of Public Health and runs the Computational Epidemiology Lab
“The problem is that using this method to assess the effect of amyloid removal on cognition can produce misleading results.”
The researcher’s concern is that the approach can make the link between amyloid reduction and cognitive improvement appear much stronger than it is, according to the analysis. The study the researchers looked at was a reanalysis of the original data from the randomized control trial on donanemab. It was led by scientists affiliated with the drug maker.
“When we did simulations, we found that you could basically take a very weak relationship between amyloid and cognition and make it appear as something that looked really strong and important,” Ackley said.
The team expected there might be problems with the method but were struck by how large the effects were.
In simulations that were designed to reflect the conditions from recent trials, the team found the method showed the relationship between amyloid and cognition to be 29 times higher than its actual magnitude.
The researchers said this happens because by combining large groups of patients and averaging their results together, the process hides variability in cognitive change between patients. That can make it look like reducing amyloid is more predictive of cognitive benefit than it is.
The method also combines patients who received the drug with those who received a placebo. Without that randomization, the analysis cannot reliably determine whether amyloid reduction is actually causing cognitive benefit or whether other factors are at play, according to the study.
To illustrate this, the team also tested the method using data from the Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease Study that ran from 2014 to 2023. That trial tested if the drug solanezumab could slow cognitive decline in older adults with elevated amyloid levels in their brains, an early sign associated with Alzheimer’s disease.
The trial showed solanezumab did not slow cognitive decline, yet when the team ran the data from that trial through the analysis of donanemab using the quantile aggregation method, it came back showing a strong link between lower amyloid and better cognitive outcomes.
“We basically built a case that this method is going to give you misleading results,” Ackley said. “It made a failed trial look like it had successfully removed amyloid and that the removal of amyloid had reduced cognitive decline. In reality, the drug did neither of these things.”
Ackley emphasized that the findings do not settle the broader question of how the new Alzheimer’s disease drugs work. Instead, she says, the work highlights a need for more rigorous statistical methods. She also emphasized the need for more data sharing in Alzheimer’s research, especially as new treatments become more widely used and covered by public programs like Medicare.
“Our study was simple, but a great demonstration of the value of academic research,” she said. “Working outside of industry incentives gave us the freedom to closely examine a methodological issue affecting how some of the most consequential new drugs are understood.”
Key Questions Answered:
A: Not necessarily. Many neuroscientists operate on the hypothesis that clearing amyloid buildup will slow down memory loss, making it the primary target for a new wave of expensive Alzheimer’s medications. However, the actual, raw relationship between removing plaques and keeping a patient’s mind sharp can be quite weak and unpredictable in practice. This study proves that a flawed math equation can take a very weak clinical relationship and make it look incredibly powerful on paper.
A: Imagine if a school averaged together the test scores of whole classrooms instead of looking at individual students. The extreme failures and unique successes get completely erased, leaving behind a perfectly smooth, deceptive average. Quantile aggregation groups different patients together and averages their results. By smoothing out the messy, real-world differences between how individual human beings decline, it hides patient variability and falsely forces a line that looks like clearing amyloid is perfectly saving memory.
A: To thoroughly pressure-test this math framework, the Brown team took the historical data from a drug called solanezumab, which clinical trials proved completely failed to clear amyloid or stop cognitive decline. When they fed those failed results into the quantile aggregation model used for newer drugs like donanemab, the math completely distorted reality, spitting out an analysis that falsely claimed the drug successfully removed amyloid and cured cognitive decline.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this Alzheimer’s disease and neuropharmacology research news
Author: Juan Siliezar
Source: Brown University
Contact: Juan Siliezar – Brown University
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Methodological Considerations for Quantile Aggregation in Alzheimer Disease Trials” by Michael D. Flanders, Michelle Caunca, Renaud La Joie, Lon S. Schneider, and Sarah F. Ackley. JAMA Neurology
DOI:10.1001/jamaneurol.2026.1240
Abstract
Methodological Considerations for Quantile Aggregation in Alzheimer Disease Trials
Is amyloid reduction an appropriate surrogate outcome for clinical benefit in anti-amyloid trials? This question is central to interpretation of current results and to the design of future trials. Inherently noisy cognitive measures complicate efforts to answer this question, prompting the use of alternative analytic approaches.
Recently, quantile-aggregation methods that regroup trial participants by posttreatment amyloid burden rather than by random assignment have found clear amyloid-cognition relationships in data from TRAILBLAZER-ALZ 2. The statistical properties and limitations of this approach for interpreting trial data have not been characterized.
