The asterisk on the anti-amyloid Alzheimer drugs

Two anti-amyloid monoclonal antibodies — lecanemab (Leqembi, FDA-approved 2023) and donanemab (Kisunla, FDA-approved 2024) — clear amyloid plaque from the brain and reduce the rate of cognitive decline in early Alzheimer’s disease by roughly 25–33% over an 18-month trial window, versus placebo. Those are real results, in adequately powered phase-3 trials, with cognitive endpoints both companies and the regulator agreed to in advance.

There is also a number both trials report and neither company foregrounds: patients on the active drug lose more brain volume than patients on placebo. On the order of half a percent of whole-brain volume over 18 months, dose- dependent, and consistent across the pivotal trials.

What that number means is the unsettled question of the field. Two hypotheses explain it. They predict the same 18-month observation. Distinguishing them requires data that does not yet exist.

Hypothesis one: pseudoatrophy

The sponsors and a substantial fraction of the senior Alzheimer-research community argue the volume loss is an artefact of removing physical material that was previously occupying space in the brain. Amyloid plaque has volume. Clearing it makes the brain measurably smaller on MRI without any neuronal loss having occurred. By this account, the cognitive benefit is the signal and the volume change is the cleared-plaque shadow. It is, in this story, harmless.

The argument is internally coherent. Anti-amyloid antibodies do clear plaque. Plaque does occupy space. The volume change is in roughly the right magnitude to be consistent with plaque clearance plus secondary fluid shifts.

Hypothesis two: real neurotoxicity

The competing account is that anti-amyloid antibodies recruit microglia, and microglia, once recruited, do not stop at the plaque. They prune synapses, strip dendrites, and contribute to a low-grade neuroinflammatory state in the surrounding parenchyma. ARIA — amyloid-related imaging abnormalities, the trial-defined adverse event — is the visible end of this same process; the sub-clinical version may be the volume signal. By this account, the cognitive benefit and the genuine neuronal loss are both real, and the trial window is simply not long enough to see whether the neuronal loss matters.

The shape of the problem

Both hypotheses predict the same correlation in the 18-month data: more plaque cleared, more volume lost. The trials confirm exactly that correlation. Neither trial reports the trajectory of the volume change beyond the 18-month primary endpoint. The question that distinguishes the two hypotheses — does the volume loss plateau (consistent with pseudoatrophy) or continue (consistent with ongoing damage) — is exactly what the published pivotal data cannot answer.

The open-label extension studies will, eventually, produce that trajectory. They are running now. The numbers have not been published.

The other gap: anticoagulated patients

Both trials excluded patients on therapeutic anticoagulation. The real-world Alzheimer population includes a substantial fraction of patients on warfarin, DOACs, or antiplatelet therapy — atrial fibrillation, mechanical valves, post-stroke prophylaxis — and that fraction climbs with age.

Post-marketing case reports now document fatal cerebral haemorrhages in patients on lecanemab or donanemab while also on therapeutic anticoagulation, particularly in carriers of two copies of the ApoE4 allele. The current FDA and prescriber guidance recommends caution rather than contraindication; in practice, the line between caution and contraindication is interpreted by individual prescribers under real prescribing pressure, and the recipients of the drug already include patients the pivotal trials would have excluded.

What we would want to see

Three things would meaningfully reduce the uncertainty.

• Five-year volumetric MRI follow-up from the open-label extension cohorts, with the trajectory plotted past month 18. Plateau favours pseudoatrophy. Continued decline favours neurotoxicity.
• A prospective registry that does not exclude anticoagulated or ApoE4-homozygous patients, with hard endpoint capture (cerebral haemorrhage, mortality) and pre-specified subgroup analysis.
• Cognitive trajectory past 36 months. The 25–33% reduction in decline is an 18-month figure. If the volume loss is real damage, the curves should re-converge or invert. If it is pseudoatrophy, the curves should continue separating.

None of this is hostile to the drugs. The 18-month cognitive benefit is real and in the magnitudes the trials reported. The asterisk is that the 18-month window is short relative to both the disease and the mechanism of the drug, and the data needed to convert the asterisk into a footnote has not yet been collected.

How we would use a body model here

A composed model of central-nervous-system pharmacology, blood-brain-barrier transport, anti-amyloid antibody binding kinetics, and microglial activation can in principle simulate both hypotheses and report the divergence point. We have built the components and are running the scenarios against the published Clarity-AD and TRAILBLAZER-ALZ-2 anchors. The honest answer at this stage is that the model, like the trial, sees the 18-month observations consistent with both stories — and the divergence between them shows up only when the simulation is run past month 30. Which is, of course, exactly the data the field is missing.

This is where mechanistic modelling earns its keep. Where the trial has run out, the model has not. The output is not a substitute for the long-term data — it is a prior on what the long-term data is likely to look like, and a way of asking which sub-population is most worth following first.