Study Questions Methods Used in Alzheimer’s Drug Analysis

A statistical approach used to support amyloid-targeting treatment for Alzheimer’s disease may lead to overstated claims about amyloid-cognition relationships, an analysis suggested.

The study focused on quantile aggregation, a statistical technique that divides trial data into quantiles, averages the results of each quantile, and looks for patterns across groupings.

Weak relationships between amyloid and cognition in individual-level analyses were much stronger when quantile aggregation was used, reported Sarah Ackley, PhD, of the Brown University School of Public Health in Providence, Rhode Island, and co-authors.

“In our simulations, aggregation inflated the strength of the apparent association 29-fold,” they wrote in a JAMA Neurology research letter.

Quantile aggregation recently was used in a secondary analysis that evaluated post-treatment amyloid and clinical outcomes in people who received placebo or donanemab (Kisunla) for early symptomatic Alzheimer’s disease in TRAILBLAZER-ALZ 2. The Eli Lilly-sponsored analysis reported that lower post-treatment amyloid levels correlated with slower clinical progression.

By hiding variability in cognition, quantile aggregation can produce “misleading results that overstate amyloid-cognition relationships,” Ackley pointed out.

The technique combines participants who received donanemab with those who received placebo, she told MedPage Today. “When we do randomized controlled trials, randomization is the essential ingredient for causal interpretation. If we ignore this critical piece of information, we can’t conclude that amyloid reduction is responsible for cognitive benefit,” Ackley said.

“In trials of amyloid-targeting drugs for Alzheimer’s disease, cognitive benefits have been modest during the short follow-up, raising the question of whether benefits may increase over longer follow-up,” she noted.

In their study, Ackley and co-authors evaluated quantile aggregation using both simulated data and publicly available trial information. They followed the same steps used in recent work and compared the results against individual-level analyses.

The simulated trial data included 1,600 participants with conditions reflecting recent findings. Publicly available information was from the Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease (A4) study, a phase III trial that showed no cognitive benefit with solanezumab.

Individual-level analyses in the simulated trial data showed a weak association between post-treatment amyloid levels and cognitive change (R² = 0.03, 95% CI 0.02-0.05), Ackley and colleagues reported. Under quantile aggregation, the association increased substantially (R² = 0.87, 95% CI 0.63-0.97).

Individual-level analyses in the A4 study also showed a weak association between post-treatment amyloid and cognitive change (R² = 0.04, 95% CI 0.00-0.07). Quantile aggregation, on the other hand, yielded a near-perfect association (R² = 0.99, 95% CI 0.99-1.00), the researchers stated.

The findings do not affect the outcomes of the trials that led to donanemab’s FDA approval. “Donanemab has demonstrated meaningful clinical benefit,” said Eli Lilly Group Vice President Mark Mintun, MD.

“In TRAILBLAZER-ALZ 2, the primary registration trial, patients showed statistically significant slowing of decline on both iADRS [integrated Alzheimer’s Disease Rating Scale] and CDR-SB [the sum of boxes of the Clinical Dementia Rating scale]. Those results stand on their own, independent of any secondary analysis or methodological debate,” he told MedPage Today.

“The quantile aggregation approach was used as an exploratory tool to examine whether the degree of amyloid plaque clearance tracked with the degree of clinical benefit, grouping participants by post-treatment amyloid levels to visualize a dose-response relationship across the full range of outcomes,” Mintun noted. “We agree that quantile aggregation should not be interpreted in isolation and multiple types of analyses are useful.”

A limitation of the study is that it demonstrated shortcomings of quantile aggregation without exploring alternatives, Ackley and colleagues acknowledged.

“There exist established approaches for evaluating mechanism and surrogacy in trials: causal mediation, instrumental variable techniques, and within-arm analyses would be appropriate in this setting,” they wrote. “These methods explicitly preserve randomization, retain informative patient-level variability, and permit causal interpretation.”