Alzheimer's: 20 Years, 99% Failure, $42 Billion, Two Drugs

The disease

Alzheimer's disease kills neurons. It does so slowly, beginning a decade or more before any symptoms appear, and once the damage is done it cannot be reversed. By the time a patient forgets a name, millions of neurons are already gone.

Two proteins define the pathology. Amyloid beta accumulates into plaques between neurons. Tau protein forms tangles inside them. Plaques appear first, often years before symptoms. Tangles follow, and they correlate more closely with cognitive decline - where tau spreads, neurons die. The field spent decades debating which one to target. It bet almost entirely on amyloid.

7.4 million Americans over 65 are living with Alzheimer's in 2026. That number is projected to reach 13 million by 2050. The total cost of care - health services, long-term care, and the unpaid labor of family members - exceeds $400 billion per year in the United States alone. Seventy percent of the lifetime cost is borne by families.

For most of the disease's 120-year history, the only drugs available treated symptoms. Cholinesterase inhibitors like donepezil (Aricept, 1996) and memantine (Namenda, 2003) modestly improve memory and attention for months, sometimes a year. They do not slow the underlying destruction of neurons. They are palliative. No drug approved between 1906 and 2021 altered the course of the disease.

The bet

In 1984, researchers isolated amyloid beta - a sticky protein fragment that accumulates into plaques in the brains of Alzheimer's patients. By 1992, Hardy and Higgins had formalized the amyloid cascade hypothesis - the dominant theory of the disease: amyloid plaques are not just a marker of Alzheimer's. They cause it. Remove the plaques, stop the disease.

The evidence was compelling. Rare genetic mutations that increase amyloid production (like those in the APP and PSEN1 genes) guarantee early-onset Alzheimer's. A rare mutation in Iceland that reduces amyloid production protects against both Alzheimer's and age-related cognitive decline. The genetics pointed one way: amyloid drives the disease.

The pharmaceutical industry bet accordingly. For the next three decades, the vast majority of Alzheimer's R&D spending targeted amyloid - blocking its production, preventing its aggregation, or clearing it from the brain. Cumulative private R&D spending reached an estimated $42.5 billion. It was the most expensive bet in the history of drug development.

Why everything failed

The failures were not random. They taught hard lessons about where to focus - and where not to.

Lesson 1: Blocking amyloid production makes things worse. Three BACE inhibitors - verubecestat (Merck), atabecestat (J&J), and elenbecestat (Biogen/Eisai) - all reached Phase 3 trials. All three were stopped early. Not because they didn't work on their target - they successfully reduced amyloid production. But patients on the drugs declined faster than those on placebo. It turned out that BACE enzymes do more than make amyloid. They prune synapses and process other proteins the brain needs. Blocking them indiscriminately is like cutting a city's water supply to stop a flood.

Lesson 2: Most early trials enrolled the wrong patients. Before the era of amyloid PET scans and CSF biomarkers, Alzheimer's trials enrolled patients based on clinical diagnosis alone - cognitive tests and neurologist judgment. The problem: 20-36% of patients in the bapineuzumab and solanezumab trials turned out to be amyloid-negative on PET - they didn't have amyloid pathology at all. The drug was trying to clear plaques that weren't there.

Lesson 3: Too late is too late. Most failed trials enrolled patients with moderate-to-severe Alzheimer's. By that stage, neuronal damage is extensive and irreversible. Clearing amyloid from a brain that has already lost billions of neurons is like removing termites from a house that has already collapsed.

Lesson 4: Not all amyloid-clearing is equal. Bapineuzumab, solanezumab, and gantenerumab all targeted amyloid. None cleared enough of it. Bapineuzumab barely moved plaque levels - ARIA-related dose limits prevented higher concentrations. Solanezumab targeted soluble monomers but barely touched plaques. Gantenerumab didn't reach sufficient brain concentrations at the tested doses. The antibodies that eventually worked - lecanemab and donanemab - achieved something the earlier ones didn't: more than 50% reduction in brain amyloid, confirmed by PET imaging.

Lesson 5: The blood-brain barrier makes everything harder. The brain is protected by a tightly sealed vascular barrier that blocks most large molecules. Only about 0.1% of a typical antibody in the bloodstream crosses into the brain. This forces extremely high systemic doses to get enough drug to the target - which is why ARIA occurs. The blood vessels swell under the strain of clearing amyloid while saturated with antibody. Next-generation approaches like Roche's trontinemab use a "brain shuttle" - an engineered transferrin receptor binding domain - to cross the barrier more efficiently, potentially reaching therapeutic brain concentrations at lower doses with less vascular damage.

Aduhelm: the controversy that changed the rules

In March 2019, Biogen halted both Phase 3 trials of aducanumab for futility. An independent monitoring committee concluded the drug was unlikely to work. Biogen's stock dropped 29% in a day.

Seven months later, Biogen reversed course. A post-hoc analysis of the higher-dose arm of one trial (EMERGE) showed a small but statistically significant benefit. The other trial (ENGAGE) still showed nothing. Biogen filed for FDA approval.

On June 7, 2021, the FDA granted accelerated approval to aducanumab (branded Aduhelm) - overruling its own advisory committee, which had voted 10-0 against approval, with one member uncertain. Three committee members resigned in protest. The acting FDA statistician called the data "exceptionally uninformative." Biogen priced the drug at $56,000 per year.

In April 2022, the Centers for Medicare and Medicaid Services effectively killed the drug's market by restricting coverage to patients enrolled in approved clinical trials. Commercial revenue never materialized. In January 2024, Biogen discontinued Aduhelm and pivoted to lecanemab.

Aduhelm was a failure. But it was a useful failure. The controversy forced the field to define what "working" actually means for an Alzheimer's drug. It forced CMS to create a coverage framework for amyloid-clearing antibodies. And it forced the next generation of drugs to meet a higher bar.

What finally worked

In January 2023, the FDA granted accelerated approval to lecanemab (Leqembi), developed by Eisai and Biogen. Six months later, it received full traditional approval - the first disease-modifying Alzheimer's therapy to clear that bar. In July 2024, donanemab (Kisunla) from Eli Lilly followed.

Both drugs are monoclonal antibodies that clear amyloid from the brain. Both showed statistically significant slowing of cognitive decline. And both succeeded where earlier antibodies failed for specific, identifiable reasons.

Five things separated these drugs from the hundreds that failed:

• They cleared amyloid effectively - more than 50% reduction, confirmed by PET.
• They enrolled patients early, before extensive neuronal damage.
• They required biomarker confirmation, so every patient actually had amyloid pathology.
• They ran large trials with statistical power to detect modest effects.
• They learned from the BACE inhibitor disaster - that blocking amyloid production is different from clearing existing amyloid.

The question nobody agrees on

In April 2026, a Cochrane systematic review examined 17 clinical trials across 20,342 participants testing seven anti-amyloid antibodies. Its conclusion: the clinical effects of anti-amyloid drugs are "absent or trivial."

The review was immediately contested. Experts pointed out that five of the seven drugs in the analysis had failed - lumping them with the two that worked is like concluding antibiotics don't work because five out of seven tested compounds were ineffective. But the review surfaced a genuine tension.

Lecanemab slowed cognitive decline by 27% over 18 months. On the Clinical Dementia Rating scale, that translates to a difference of about 0.45 points on an 18-point scale. The effect is statistically significant, but its clinical meaningfulness - whether it changes the lived experience of the disease - is genuinely debated.

And the drugs are not free of harm. Amyloid-related imaging abnormalities (ARIA) - brain swelling and microbleeds - occur in 12-24% of patients on treatment. Most cases are asymptomatic and resolve. Some are not. Two patients carrying the APOE4 gene variant died from ARIA during routine clinical care - outside of controlled trial settings.

The access problem

Even for patients who would benefit, getting the drug is hard. Lecanemab and donanemab require confirmed amyloid pathology - either a PET scan (roughly $5,000) or a lumbar puncture. They require IV infusions every two to four weeks, with regular MRI monitoring for ARIA. They are approved only for early-stage disease.

An estimated 85% of Alzheimer's patients are diagnosed at moderate or advanced stages - too late for these drugs. The health system that failed to develop treatments for 20 years now has treatments it cannot deliver to most patients because it diagnoses them too late.

Blood-based biomarkers may change this. A test for phosphorylated tau (p-tau217) can detect amyloid pathology from a simple blood draw with accuracy approaching PET scans. If validated and deployed in primary care, it could shift diagnosis years earlier - into the window where these drugs actually work. The diagnostic bottleneck may break before the next generation of drugs arrives.

What comes next

The pipeline has not stopped. There are 32 Alzheimer's therapeutics in 48 Phase 3 trials. But the center of gravity is shifting. More than 70% of the pipeline now targets mechanisms beyond amyloid. And a new generation of research is looking past single-target hypotheses entirely: large cohort studies like the UK Biobank are profiling thousands of proteins in plasma and cerebrospinal fluid across tens of thousands of people, identifying signals - like GFAP and neurofilament light - that change a decade before symptoms appear. These studies may reveal new drug targets that the amyloid-first era never considered.

Next-generation amyloid: Roche's trontinemab uses a "brain shuttle" to cross the blood-brain barrier more efficiently, potentially requiring lower doses and causing less ARIA. Two Phase 3 trials (TRONTIER 1 and 2) are underway.

Tau: If amyloid is the match, tau is the fire. Tau tangles correlate more closely with cognitive decline than amyloid plaques do. But semorinemab (Roche) failed in Phase 3, and the tau field is earlier-stage. Phase 2 trials are running through 2026.

Neuroinflammation: TREM2, CD33, and other microglial targets aim to calm the brain's immune response, which may drive neurodegeneration independently of amyloid and tau.

GLP-1 agonists: The biggest recent disappointment. Novo Nordisk's semaglutide (the molecule behind Ozempic) failed both EVOKE Phase 3 trials in March 2026, showing no slowing of cognitive decline despite reducing neuroinflammation biomarkers by about 10%. The lesson: targeting inflammation outside the brain may not be enough to change what happens inside it.

Combination therapy: The most likely future is not a single drug but a cocktail - an amyloid-clearing antibody to remove the upstream trigger, a tau-targeting agent to address the downstream destruction, and perhaps an anti-inflammatory to reduce collateral damage. The DIAN-TU Tau NexGen trial is testing lecanemab plus an anti-tau antibody (E2814) in dominantly inherited AD - a small, special population, but the field's first real combination test. Sporadic-AD combination trials are harder, slower, and more expensive than monotherapy. The field is not there yet at scale.

Behind the $42 billion and two decades of failed trials are tens of thousands of patients who enrolled in Phase 3 studies that were ultimately terminated - people who gave years of their lives to advance the science, often knowing they might receive placebo. Their participation was not wasted. Each failure brought us closer to understanding the disease and finding new paths to slow it.

The question is no longer whether Alzheimer's can be slowed. It can. The question is whether we can slow it enough, safely enough, early enough, and cheaply enough to matter for the 7.4 million people who have it today.

For the full data - approved drugs, pipeline trials, mechanism breakdown, and Phase 3 readouts - see the Alzheimer's Disease Drug Landscape 2026 on TheraRadar.