On April 15 the Cochrane Collaboration published a meta-analysis of seventeen randomized trials of anti-amyloid drugs, pooling data from more than twenty thousand patients and three decades of work. [1] The review's verdict was a single adjective and a single qualifier: the effects on cognition and function were, in its language, "absent or trivial." The lead author was Francesco Nonino of Bologna; the senior author, Edo Richard of Radboud. [2] Between them they had no financial relationship with any of the companies whose drugs they were reviewing. Between them they had just assessed the evidence behind Leqembi and Kisunla — two drugs being infused, on the afternoon of publication, into patients at memory clinics across the United States. [1]

The meta-analysis is the kind of document that reaches a very specific audience with a very specific claim. Cochrane is not a commentator. It is the consensus instrument of evidence-based medicine — the thing clinicians and regulators reach for when they want to know what a large body of trials, looked at together, actually shows. [2] What this one showed was that three decades of intervention designed around a single biomarker — amyloid plaques, visible on PET scan — has produced drugs that clear the biomarker and leave the patient roughly where she was.

The effect size the pooled data surfaced was, in one of the review's more understated lines, smaller than what most clinicians can detect at the bedside. [1] The authors did not phrase it as a condemnation. They phrased it as a finding.

Eisai, which co-markets Leqembi with Biogen, called the review "scientifically deeply flawed" within hours. [3] The Alzheimer's Association's chief scientific officer, Maria Carrillo, offered the formulation that has become the field's moral shorthand: "behind every data point is a person." [3] The New York Times, reporting the story Thursday, gave those voices most of the room. [3] The version of the story that reached most readers was "Stirs Debate" — two plausible camps, each with arguments, readers invited to pick.

But Cochrane is not one of the two camps. That is what Cochrane is for.

A field falls in love with a biomarker the way a map-maker falls in love with a landmark: once it is drawn, everything else is measured against it. Amyloid plaques are visible. They can be counted. They can be reduced. When Leqembi and Kisunla (Eli Lilly's donanemab, approved in 2024) went to the FDA, the case they made was that the plaques came down. [4] That was not in dispute. What was never quite settled, and what Cochrane has now restated at scale, is the other thing — whether the mind came back with them. The scan improved. The patient did not.

The improvement was not zero. Cochrane's pooled estimates show a whisper of cognitive benefit — small, possibly real, certainly below the threshold a family would notice at the dinner table. [1] Set against that whisper is another column on the same page. Amyloid-Related Imaging Abnormalities, known as ARIA, are brain bleeds and swellings that show up on MRI after infusion. In the lecanemab trials, a substantial fraction of treated patients developed ARIA; a smaller but non-trivial number were symptomatic, and several died. [5] The drugs' labels acknowledge this. Infusion centers screen for the APOE ε4 gene variants that multiply the risk. [5]

Medicare, which approved reimbursement for Leqembi in 2023 conditional on patient registries, now pays roughly $26,500 a year per patient for the drug alone, before the MRIs, the infusion center fees, and the neurologist visits that Medicare also pays. [4] The register of patients is now many tens of thousands wide. A multibillion-dollar annual spend is funding the intervention that a pooled reading of every adequately sized trial now calls trivial. [1]

It is the sort of contradiction that could only arise in a field where the biomarker came to stand in for the disease. Amyloid is not Alzheimer's. Amyloid is a correlate of Alzheimer's — one of several, and perhaps not the primary one. For thirty years the field was organized around the proposition that if you cleared the plaques, you treated the condition. The proposition was testable. It has been tested. It failed, and the failure has been measured.

What happens next is a question of organization rather than science. The Alzheimer's Association's statement has the register of a patient-advocacy group defending a therapeutic win at the moment of regulatory approval — which is what it is. [3] Eisai's statement has the register of a manufacturer protecting a revenue line — which is what it is. Cochrane's statement has the register of a review that has nothing to sell, and that register is what is most often missing from what the reader sees.

Matthew Herper, writing on X on the afternoon of publication, offered a summary that made the point without asserting it: "This new analysis of Alzheimer's drugs is such a good example of how the field has been distorted by the amyloid hypothesis." [6] The observation is not that the drugs do nothing. The observation is that the drugs' case was built on a distortion, and that the distortion has now been measured against twenty thousand patients.

Infusion schedules will not change this week. The FDA approval still stands. The Medicare reimbursement still stands. The families who began treatment this spring will keep arriving at clinics. What has changed is only the thing a meta-analysis can change — the weight of what is known. Cochrane has put a number on a whisper. Whether the clinic lets the number travel from the journal to the waiting room is now the field's responsibility, and not the review's.

-- KENJI NAKAMURA, Tokyo