Real-world ARIA rates from registries are landing close to trial estimates

The pivotal trials of lecanemab and donanemab established the ARIA frequencies that defined the labels - symptomatic ARIA-E in the low single-digit percentages overall, materially higher in APOE4 homozygotes, and a small subset of serious events. The open question through 2025 was whether routine clinical use, with broader patient selection and less-rigorous monitoring than a trial, would produce meaningfully worse numbers.

Through the first half of 2026, the answer from the registry and academic-cohort literature - principally ALZ-NET and parallel academic datasets - is that the real-world rates are landing within the same broad range as the trials. Specifics:

• Overall incidence of any ARIA on protocol-defined MRI is consistent with trial-era estimates, with the expected dominance of asymptomatic findings.
• APOE4 homozygotes continue to carry the highest absolute risk, replicating the genotype-stratified trial picture. This is the empirical basis for routine APOE genotyping before initiation.
• Serious ARIA events remain rare. The reported case-level outcomes are consistent with the safety language in the labels, including the small number of fatal cases reported in pre-marketing and post-marketing experience.

There are real caveats. Registries oversample academic and large-system practices that resemble trial sites. Community-practice signal is sparser. Non-protocol MRI cadence in some practices may underdetect asymptomatic findings, biasing reported incidence downward. And dose-titration practice for donanemab - a key safety lever - varies across sites.

The implication for the field is that the ARIA conversation is shifting. The question is less "do these drugs behave the same outside the trial" and more "is the right patient being treated in the right setting" - see our Signal on the ARIA monitoring bottleneck and our Insight on the rate-limited anti-amyloid rollout.