OBJECTIVES: This research aims to evaluate the amyloid-beta biomarker, identified by positron emission tomography (PET) imaging, as a surrogate endpoint for cognitive function, measured by Clinical Dementia Rating-Sum of Boxes (CDR-SB), in Alzheimer’s disease (AD) trials of anti-amyloid-beta monoclonal antibodies (MABs).

METHODS: Data from randomised controlled trials (RCTs) were identified through literature review. A fixed-effects bivariate meta-analytic model was used to evaluate the surrogate relationship between treatment effects on amyloid-beta and CDR-SB across all RCTs and MABs, with intercept, slope and conditional variance as measures of association. The surrogacy for individual therapies was evaluated using subgroup analyses and a hierarchical model to borrow information across treatments.

RESULTS: The review identified twenty RCTs reporting treatment effects of seven MABs on amyloid-beta and CDR-SB. The overall surrogate relationship across all MABs was strong: slope 1.4 (95% CrI: 0.54, 2.2) and conditional variance 0.02 (95% CrI: 0, 0.05). The surrogate relationships for individual treatments showed large uncertainty around the surrogacy parameters. The use of the hierarchical model reduced the uncertainty around the key parameters. For example, the surrogate relationship for lecanemab was weak in the subgroup analysis: slope 2.15 (95% CrI: -1.49, 5.7) and conditional variance 0.06 (95% CrI: 0, 0.42), but stronger when using the hierarchical model: slope 1.66 (95% CrI: 0.11, 3.26) and conditional variance 0.03 (95% CrI: 0, 0.20). The reduction in the width of CrI was 74% (56%-95%) for slope and 34% (4%-68%) for conditional variance, when comparing results from the hierarchical model with subgroup analyses.

CONCLUSIONS: Our results showed that amyloid-beta was a good surrogate endpoint for CDR-SB when assuming a common surrogate relationship for all included treatments. The surrogate relationships were weak for individual treatments. They were improved when borrowing information from other treatments in the hierarchical model, albeit with notable uncertainty around the surrogacy parameters.