Applying causal mediation analysis to the A4 trial — which tested solanezumab in cognitively unimpaired adults with elevated brain amyloid — this methodological investigation found that indirect effects of solanezumab on cognitive decline operating through cerebral amyloid were estimated near zero across multiple analytic specifications. Critically, confidence interval widths varied by a factor of up to 17 depending on how exposure, mediator, outcome, and covariate adjustment were defined, signaling extreme sensitivity to analytic choices.
The broader implication cuts to the heart of Alzheimer's drug development. Amyloid PET clearance has been increasingly accepted as a surrogate endpoint by regulators — most prominently in the approvals of lecanemab and donanemab — yet its validity as a true surrogate for cognitive benefit remains contested. The 1989 Prentice Criteria demand that a surrogate fully capture a treatment's effect on the true outcome; solanezumab's flat cognitive results make it a sobering test case, as zero benefit leaves little mediation to detect. This analysis cannot adjudicate whether amyloid is a valid surrogate for effective treatments; it only confirms the method struggles when the drug itself doesn't work. The 17-fold variance in confidence intervals is a methodological red flag: surrogate evaluation via mediation is only credible when hypotheses are pre-registered and biologically grounded. As a preprint not yet peer-reviewed, these conclusions warrant caution. The finding is incremental but timely, offering a procedural warning to trial designers who treat amyloid clearance as a regulatory shortcut.