In April, a meta-analysis published in the Cochrane Database of Systematic Reviews pooled 17 randomized controlled trials involving more than 20,000 patients with mild cognitive impairment or mild dementia due to Alzheimer’s disease (AD) and concluded that antiamyloid monoclonal antibodies clear amyloid-beta from the brain but produce little-to-no effect on cognition and dementia severity.

For a field that had just delivered its first two FDA-approved disease-modifying drugs after decades of failure, the timing stung. Lecanemab and donanemab clear amyloid plaques from the brain with remarkable efficiency, and both slowed cognitive decline in pivotal trials by statistically significant margins.

Reaction was swift and irreconcilable.

Paul Aisen, MD, who directs the Alzheimer’s Therapeutic Research Institute at the University of Southern California in San Diego and helped build the trial infrastructure the field runs on, was among several researchers who asked the journal to withdraw the review. Cochrane declined.

“You cannot combine different drugs with different targets, different doses, and different populations and then average out all the results,” Aisen said. He believes the authors diluted the apparent benefit of the newer drugs by pooling them with first-generation antibodies that had limited plaque-clearing ability.

Alberto J. Espay, MD, a neurologist at the University of Cincinnati in Cincinnati who has published a series of reanalyses of the trial data, considered the review rigorous and unwelcome in equal measure. “It’s the truth we don’t want to hear,” he said. “Cochrane is among the most conservative review chambers there is. Of course, many of us are pissed off.”

David Knopman, MD, has spent his career between those poles. A Mayo Clinic neurologist, he resigned from an FDA expert advisory committee after the agency approved the first antiamyloid drug, aducanumab, over the panel’s opposition. Knopman called the Cochrane review a corruption of the process — the equivalent, he said, of reviewing modern antidepressants by pooling them with first-generation drugs abandoned for their side effects and concluding the drug class doesn’t work. “It was fundamentally, analytically, and practically flawed,” he said.

The drugs themselves hit every endpoint — primary and secondary — and held up under survival analysis. But that wasn’t the sticking point for Knopman. 

“I have some serious doubts about the clinical benefits of lecanemab and donanemab, but the Cochrane review didn’t address that,” he said. Prescribing the drugs, he added, is “a little bit of an act of faith.”

That phrase — “act of faith” — captures the field’s strange position. The drugs do something real. Whether that something is enough for patients, for the health system, and for a hypothesis that has set the AD research agenda for 30 years depends on who you ask.

To some, lecanemab and donanemab mark the beginning of a medical revolution. To others, they are a rounding error dressed up as progress.

What the ‘Amyloid Hypothesis’ Got Right…

Beginning in the early 1990s, researchers proposed that AD originates when amyloid-beta, a peptide fragment, aggregates into plaques between neurons. The accumulation triggers a downstream cascade: tau tangles inside neurons, synaptic dysfunction, neuronal loss, and then the erosion of memory. Clear the plaques, the reasoning went, and the cascade stalls at its source. Researchers called it the “amyloid hypothesis.”

Autopsy studies showing thick clumps of amyloid in the brains of patients with AD bolstered the theory. The genetics made the case look airtight. Every mutation known to cause autosomal-dominant early-onset AD increases amyloid production or shifts it toward the aggregation-prone form. In Down syndrome, those who carry a third copy of the amyloid precursor gene develop the pathology almost universally. A rare Icelandic variant that reduces amyloid production confers protection. The logic follows: More amyloid leads to more disease; less amyloid leads to less disease.

Amyloid is also the only pathologic protein exclusive to AD, explained Steven T. DeKosky, MD, a neurologist at the University of Florida in Gainesville who helped establish amyloid imaging as a research tool. In the early 2000s, his team contributed to the clinical validation of Pittsburgh Compound-B, a PET tracer that allowed researchers to see amyloid plaques in a living brain for the first time.

Early transgenic mouse models of AD were driven by amyloid biology, limiting researchers’ ability to study other disease mechanisms, DeKosky said. Mutating the amyloid gene or the enzyme that processes it produced an AD-like phenotype in mice; tau mutations led to a mouse model for frontotemporal dementia.

Everything — biology, genetics, imaging, and animal models — seemed to steer researchers toward amyloid as the linchpin to stopping AD.

…and What It Got Wrong

What the research did not show, even then, was a clean relationship between plaques and cognitive decline.

“There are a ton of pathology studies that showed there was no significant relationship between amyloid plaque number and cognition,” DeKosky said. “It was always tau that was the correlate.”

His own team published synapse counts in the early 1990s showing a correlation of 0.8 between synapse number and cognition, far stronger than anything amyloid produced.

The closest correlates of decline were tau tangles and synaptic loss, not plaques. The field knew this from the beginning, DeKosky said, and pursued amyloid anyway. The available tools left no other option.

The focus on amyloid that shaped a generation of AD research may have been shortsighted, said Karl Herrup, PhD, a neurobiologist at the University of Pittsburgh in Pittsburgh.

In his 2021 book How Not to Study a Disease: The Story of Alzheimer’s, he argued that despite 30 years and tens of billions of dollars, no one has established the cellular mechanism by which amyloid kills neurons. The plaques are present when neurons die, but presence is not mechanism.

“We have no cell biological model for how amyloid, in any of its states of aggregation, actually kills neurons,” he said. “Amyloid is indeed neurotoxic. If you dump it onto cells in a dish, they’ll die. If you inject it into the brains of experimental animals, neurons will die. But you can do the same thing with soy sauce, and it’s not telling us much about [AD].”

Here, the field’s sharpest opponents found unexpected common ground. Asked about the missing mechanism, Aisen did not dispute it.

“I don’t think we have a good understanding of the mechanisms by which amyloid damages neurons.” Instead, he argued that genetics carry the burden of proof on their own.

The first drug to emerge from this hypothesis arrived under a cloud. Aducanumab won FDA approval in 2021 over the near-unanimous objection of the agency’s own advisory committee, which found the efficacy data unconvincing. Knopman and two other members resigned in protest. The controversy never fully dissipated.

Whatever their merits, lecanemab and donanemab arrived under the cloud of suspicion aducanumab left behind.

How to Interpret the Data

Everything turns on a single number, and the number is small.

In the phase 3 Clarity AD trial evaluating lecanemab, the drug slowed decline on the 18-point Clinical Dementia Rating-Sum of Boxes (CDR-SB) scale by just 0.45 points over 18 months. Donanemab, in the phase 3 TRAILBLAZER-ALZ 2 trial, slowed decline by 0.67 points in the low-to-moderate tau population over roughly the same amount of time.

Still, cognition declined in all patients in both trials, although the treated groups declined a little less. Lecanemab’s maker, Eisai, framed that gap as a 27% slowing of disease progression.

Espay frames it differently. On an 18-point scale, 0.45 points is 2.5% of the range. Both figures are mathematically correct but leave entirely different impressions, a point he and his colleagues, including Herrup, made in a 2024 analysis.

The question is whether the effect is large enough for anyone to notice. The smallest change a patient or clinician can perceive has been estimated at roughly 1 point on the CDR-SB scale for mild cognitive impairment and 2 points for mild dementia, according to a 2024 review. Both drugs fell below those thresholds.

By that yardstick, the benefit is real in the statistical sense but invisible in the experiential one.

The problem may have been the framing, said Jason Karlawish, MD, a geriatrician and bioethicist at the University of Pennsylvania in Philadelphia who co-directs the Penn Memory Center and is the author of the 2021 book The Problem of Alzheimer’s. Trial patients were clustered between roughly 1 and 5 on the CDR-SB scale, not spread across the full 18-point range. Calling a half-point shift trivial against an 18-point ruler inflated the denominator.

“The 18-point scale rhetoric is just that. It’s a rhetorical point,” he said. “This is a fascinating academic debate where the majority of the wagons return solidly in a circle and the guns are pointing inward.”

It was not, he added, a debate that served patients or families. But if that debate were to be had, Karlawish noted that the scale is not evenly spaced because movement from one descriptor to the next can mean the difference between mild forgetfulness and loss of independence. The rate of change over time is slower in the treated group. Karlawish considers that a real effect, though he noted the field lacks randomized follow-up beyond 18 months to know how the gap evolves.

DeKosky drew a distinction the defenders rarely make in public. “Statistical significance is not the same as clinical significance,” he said. “Since the amyloid always comes first and appears to stimulate the tau to come, I was hoping if we slowed down the amyloid or stopped it, we would slow down the process of tau.”

The CDR-SB thresholds and the small half-point differences did not persuade Knopman. Those thresholds, he said, were derived for symptomatic drugs in more advanced disease and were never validated for disease-modifying therapy given early, such as lecanemab and donanemab. “Before these drugs,” he said, “there was no success on which to judge what was significant.”

Espay disagreed, saying the drugs’ effect sat below the threshold of perception, which left clinicians in an awkward position. Patients “have to trust us because they cannot tell,” he said.

Calling the benefit clinically trivial, Herrup noted that donepezil, a cholinesterase inhibitor that temporarily boosts memory by raising acetylcholine levels in the brain, produced a larger separation from placebo in one third of the time.

Aisen rejected the framing entirely. An 18-month trial is too short to judge a drug that works by slowing the rate of decline, he said. The benefit is small early on, but it accumulates. A patient who declines more slowly holds onto everyday abilities longer — managing finances or living without full-time care. That could mean a 5- to 6-month delay in progression after 18 months of treatment. Six years of treatment could lead to a 2-year delay in progression, he said.

There are some data suggesting Aisen may be right. Four-year extension results for lecanemab show the treated and untreated curves continuing to separate, though slowly. But the comparison relies on historic controls rather than a true placebo arm — the original placebo patients were switched to the drug after the trial ended.

“[Historic] controls are interesting,” DeKosky said, “but they can’t prove the point.”

Measuring Risk

Amyloid-related imaging abnormalities (ARIA) are a significant complication of antiamyloid therapies that appears in two forms: vasogenic edema, fluid leaking into brain tissue, and microhemorrhage, small bleeds in the vessels the drug is clearing.

In the Clarity AD trial, ARIA with edema occurred in 12.6% of treated patients vs 1.7% of those on placebo. Among patients carrying two copies of the APOE4 gene, the population at highest genetic risk for the disease itself, the rate climbed toward one third on imaging and higher still for donanemab.

This complication is built into treatment rather than incidental to it, said Jeffrey Petrella, MD, a neuroradiologist at Duke University in Durham, North Carolina. Amyloid does not collect only between neurons; it also lines the walls of cerebral blood vessels, he noted.

“When you take out the amyloid in the vessel walls, it leaves a little gap, and you can have leakage of fluid and eventually leakage of blood cells,” he said. Because the antibodies travel to the brain through the bloodstream, the vessels are the first thing they reach. “ARIA is intrinsic to antibody therapies that enter the brain through the bloodstream.”

Most ARIA is mild, and most of it does resolve, said DeKosky, who chaired the data and safety monitoring board (DSMB) that reviewed every scan in the lecanemab trial.

Knopman’s experience at Mayo Clinic supported the same point. With conservative screening, he put the real-world rate of new ARIA at around 8%. The safety concerns raised before the drugs launched, he said, were “grossly overstated by the opponents.”

Petrella’s worry runs in a different direction, one the trials did not address. The edema clears, but the microhemorrhages leave something behind.

“There’s always iron deposition in areas where there’s been hemorrhage,” he said, because the iron in blood does not wash away. Over years of treatment, small bleeds can accumulate, and the literature links their number to later cognitive decline.

“We know that microhemorrhages, even if they don’t turn into a catastrophic hemorrhage, can cause cognitive decline, which is the very thing we’re trying to stop here,” he said. Whether years of antiamyloid therapy could produce something similar is unknown, Petrella added.

ARIA is not the only concern these drugs carry. A 2023 meta-analysis showed that antiamyloid antibodies accelerate the loss of brain volume compared with placebo. However, the meaning of that finding is unclear. To critics, the shrinkage suggests injury; to defenders, it may reflect plaque clearance rather than tissue loss.

Infusion-related reactions — chills, fever, headache, nausea, and drops in blood pressure — occurred in roughly one quarter of lecanemab patients. Petrella noted that the biweekly infusions require visits to the same centers used for chemotherapy, adding a significant logistical burden for patients and clinics.

Several deaths were reported during lecanemab’s open-label extension, each involving cerebral macrohemorrhages. At least two cases involved blood thinners: one in a patient on anticoagulant therapy and in another who received a thrombolytic for a suspected stroke.

The interaction has led to warnings against using the drug in patients on anticoagulants, excluding a large population of older patients with atrial fibrillation or cardiovascular disease and narrowing the already thin pool of eligible patients.

A Global Verdict

While US researchers argued over effect sizes, much of the rest of the developed world weighed in. Britain’s National Institute for Health and Care Excellence (NICE) reviewed and rejected both drugs three times, judging the benefits real but too small to justify the cost. Both manufacturers have since appealed, and NICE has reopened its review.

Italy rejected reimbursement for both drugs and Netherlands for lecanemab. Both hold European marketing authorization, but only Germany has made them broadly available, launching lecanemab in September 2025 and donanemab in November under a controlled access program — though its health technology body later found no proven added benefit for either drug, and price negotiations are ongoing.

That conservative regulators in country after country looked at the identical data and declined approval is evidence that caution is not a fringe position, Herrup argued.

Although an estimated 10%-20% of the 7 million people with AD in the US meet the criteria for antiamyloid therapy, uptake has been strikingly thin. Roughly 13,000 patients had received lecanemab as of late 2025.

The drug carries a list price of $26,500 per year, with total treatment costs, including amyloid testing, infusion visits, and MRI monitoring for ARIA, running substantially higher. Medicare Part B covers the drug, but patients without supplemental insurance may have to pay about 20% of the costs out of pocket.

A subcutaneous formulation of lecanemab approved in 2025 for maintenance dosing may ease the burden of biweekly infusions, but a second bottleneck remains. No single blood test yet confirms the diagnosis with enough certainty to justify the risk and expense.

Knopman attributed the thin uptake to caution in the prescribing community, while Espay framed it more sharply.

“If people look at the data, they’d say this is not much of a win for patients,” Espay said.

Even if the cost per patient falls, Herrup noted that the US medical infrastructure cannot expand to meet the size of the disease. “There’s just no way our healthcare system can incorporate this and remain solvent,” he said.

What Comes Next

If there is a point of agreement in this debate, it is that the current drugs are not the final answer and that targeting amyloid alone is not enough.

Even Aisen, the most vocal of the defenders, frames the current drugs as a starting point rather than an arrival, with the future running through earlier treatment and combination therapy.

Researchers generally agree that addressing tau accumulation is critical. Indeed, two anti-tau antibodies are now entering phase 3 trials.

On May 14, Biogen reported that diranersen, an antisense oligonucleotide designed to reduce tau production in the brain, showed both reductions in tau pathology and signs of cognitive benefit in a phase 2 trial of patients with early AD — the first time any tau-directed therapy has achieved both in a randomized study.

Though the trial missed its primary statistical endpoint and the effect was modest, Biogen is advancing the drug to phase 3.

Separately, Eisai is testing etalanetug, an anti-tau antibody, in combination with lecanemab — combating amyloid and tau together — in two ongoing trials.

Other drugs in the AD pipeline target a range of other mechanisms affected by the disease, including neuroinflammation, metabolism and insulin signaling, synaptic dysfunction, and vascular biology.

Karlawish, who prescribes the drugs and wants better ones, cautioned against the assumption that any single target would be enough, noting that brains in patients with AD carry amyloid, tau, alpha-synuclein, TDP-43, and vascular disease in varying combinations.

“I just don’t think any rational person would say our strategy is 100% in drug development,” he said.

Is Amyloid the Wrong Target After All?

So is it time to move on from amyloid? For Aisen, that answer is an emphatic “no,” and he’s certain the hypothesis has been vindicated.

“I am thrilled that we’ve proven the hypothesis correct,” he said.

Herrup hoped the trials would change his mind. “I really hoped I was wrong,” he said. But his position remains unchanged.

DeKosky, who helped validate the imaging tool that made the entire enterprise possible, offered the simplest summary. “All we’ve shown at this point is that this disease can be pushed around,” he said.

Early in DeKosky’s career, a leading neuropathologist told him flatly, “Steve, you will never find a drug that will treat this disease. I’ve looked at the brains of thousands of people and there’s nothing in there.”

Against that backdrop, pushing the disease around is a significant step forward, the first crack in a wall that held for decades. But it is not a cure. Karlawish framed the question the way he believed it should be asked.

“The difference between drug and placebo is big enough to say that it’s up to a patient to decide, ‘Is this benefit worth my time and effort and risk?,’” he said, “as opposed to the system saying, ‘it is not worth your time; therefore, we’re not going to make it available to you.’”

Aisen has received research grants from Lilly, Eisai, and the National Institutes of Health (NIH) and has served as a consultant for Merck, Roche, BMS, Genentech, AbbVie, Biogen, ImmunoBrain Checkpoint, Arrowhead, AltPep, and Neurimmune. Knopman has served on DSMBs for the Dominantly Inherited Alzheimer Network study and for a tau therapeutic agent for Biogen, for no personal compensation. He has been an investigator in clinical trials sponsored by Biogen and Lilly and has served as a consultant for Samus Therapeutics, Third Rock, Roche, and Alzeca Biosciences, for no personal compensation. DeKosky chaired the DSMB for the Eisai-sponsored Clarity AD trial of lecanemab and reported serving on the medical advisory board of Acumen Pharmaceuticals, chairing DSMBs for Prevail Pharmaceuticals and Vaccinex, and being a consultant for Novo Nordisk. Petrella has received research support from the NIH and Sanofi and a gift from Gates Ventures and has served on advisory boards for Cortechs.ai, icometrix, and Biogen. Karlawish, Herrup, and Espay reported having no relevant financial disclosures.