Why 99% of Alzheimer's Drug Trials Fail

Everyone wants to invest in Alzheimer’s. Almost nobody should. Between 1998 and 2017, 146 experimental Alzheimer’s drugs entered clinical trials. Four were approved. That is a 2.7% success rate over two decades. When Jeffrey Cummings published his landmark analysis of Alzheimer’s drug development, the headline number was 99%: 99% of Alzheimer’s clinical trials showed no drug-placebo difference. For the decade 2002-2012, the failure rate was 99.6% (Cummings et al., 2019, Alzheimer’s Research & Therapy).

Across all therapeutic areas, the Phase 1-to-approval success rate runs about 13.8%. In Alzheimer’s, it is roughly 0.4%. That is not a difficult sector. That is a sector where the odds are stacked 35x worse than the rest of drug development. This pattern plays out repeatedly: a biotech announces Phase 2 Alzheimer’s data, the stock spikes 40% on biomarker engagement, Phase 3 fails, the stock drops 80%. The 99% rate exists for structural reasons. If you understand them, you can stop buying hype and start evaluating catalysts on their actual merits.

The Five Structural Reasons

1. The Blood-Brain Barrier

Most drugs cannot cross the blood-brain barrier. This is not a minor obstacle. It is the single most restrictive biological membrane in the human body. The barrier exists to keep pathogens and toxins out of the brain. It also keeps out most of the drugs designed to treat brain disease. Antibodies like lecanemab (Leqembi) and aducanumab can cross, but at very low concentrations. A peripherally administered antibody typically delivers less than 0.1% of its plasma concentration into the brain. That is why Biogen administers diranersen intrathecally via lumbar puncture: bypassing the blood-brain barrier entirely. Small molecules cross more easily, but most still fail to reach therapeutic concentrations in the brain.

2. Disease Heterogeneity

Alzheimer’s is not one disease. The pathology involves amyloid plaques, tau tangles, neuroinflammation, TDP-43, and vascular damage. Different patients have different dominant pathologies. A patient who is amyloid-positive and tau-positive responds differently from one who is tau-positive with significant neuroinflammation. The 2026 pipeline reflects this fragmentation: anti-amyloid antibodies (Leqembi, Kisunla), anti-tau antisense oligonucleotides (diranersen), GLP-1 agonists testing neuroinflammation pathways (semaglutide, which failed EVOKE completely), and 5-HT2A inverse agonists for psychiatric symptoms (remlifanserin). Each mechanism addresses one piece of a multi-pathology disease. A drug that works in the 30% of patients with a specific biomarker profile will show a diluted effect in a broad population trial.

3. The Surrogate Endpoint Problem

Amyloid plaque reduction is not the same as cognitive improvement. This is the trap that burned investors for 15 years before Leqembi. A drug clears plaques, the biomarker looks great, the stock jumps, and then Phase 3 shows zero clinical benefit. We covered this in our guide to surrogate endpoints: the FDA allowed accelerated approvals based on biomarkers for years, and multiple drugs were later withdrawn when confirmatory trials failed. The anti-amyloid antibody field was built on this premise. Aduhelm (aducanumab) received accelerated approval in 2021 based on amyloid reduction. Eisai pulled it from the market in 2024. Leqembi received traditional approval in 2023 because it showed both amyloid clearance and a modest cognitive benefit. The distinction matters.

4. Trial Duration and Dropout

Phase 3 Alzheimer’s trials run 18 to 24 months. That is a long time to track cognitive decline in a population that is, by definition, declining. Dropout rates of 20 to 30% are standard. When a trial loses a quarter of its patients over 18 months, the statistical power erodes. Leqembi’s CLARITY AD trial enrolled 1,795 patients and still had to contend with meaningful attrition. Semaglutide’s EVOKE and EVOKE+ enrolled 3,808 patients across two trials, and only about 85% completed the two-year primary endpoint. A longer trial with more dropouts means wider confidence intervals, which means a drug with a modest effect size can miss statistical significance purely from noise. This is exactly what happened with semaglutide: a 0.06-point CDR-SB difference at 104 weeks, buried in a confidence interval that spanned negative 0.48 to 0.36.

5. Animal Models Do Not Get Alzheimer’s

Mice do not develop Alzheimer’s disease. Every transgenic mouse model is an approximation. They develop amyloid plaques when you engineer them to overexpress mutant human APP. They develop tau tangles when you engineer them to overexpress mutant human tau. But they do not reproduce the full disease: the neurodegeneration, the cognitive decline, the complex interplay of multiple pathologies. A drug that clears plaques in a transgenic mouse can fail completely in humans. This is why Phase 1b biomarker data in humans matters more than any preclinical result. When Biogen reported 50 to 65% CSF tau reduction in the CELIA Phase 2 trial, that human biomarker data was more informative than any mouse study.

Leqembi Changed the Bar, Not the Odds

Lecanemab (Leqembi) was the first anti-amyloid antibody to show slowing of cognitive decline in a Phase 3 trial. CLARITY AD demonstrated a 27% reduction in CDR-SB decline over 18 months: a 0.45-point treatment difference on an 18-point scale (p=0.00005). That result, published in the New England Journal of Medicine, was the first time a drug showed a statistically significant clinical effect in Alzheimer’s.

But here is the number investors miss: 0.45 points on an 18-point scale is a 2.5% absolute difference. The drug does not stop the disease. It slows it by a fraction. Patients on Leqembi still declined by 1.66 points over 18 months. Placebo patients declined by 2.11 points. The drug buys roughly four to five months of delayed decline over an 18-month treatment period, at a cost of $26,000 per year and with a 12.6% incidence of amyloid-related imaging abnormalities (ARIA) on MRI.

Leqembi changed what the FDA considers an acceptable effect size. It did not change the biological reality that Alzheimer’s is a multi-pathology disease with no single target that addresses all of the damage. The 99% failure rate is a statement about the difficulty of the biology, not about regulatory standards.

The Current Catalysts, Ranked by Risk

Three live Alzheimer’s catalysts illustrate the framework. They are ranked by risk-reward based on mechanism, phase, and the structural reasons above.

Biogen diranersen (BIIB, $209.23, $30.8B market cap): Phase 2 CELIA data showed 26% slowing on CDR-SB at 18 months (0.54 points), 50 to 65% CSF tau reduction, and the first randomized evidence that reducing tau can slow cognitive decline. But the primary endpoint missed (dose response), the lowest dose worked best while higher doses produced more biomarker reduction but less clinical benefit, and Phase 3 has not started. This is the most compelling Phase 2 signal in Alzheimer’s outside of Leqembi. It is also a Phase 2 signal in a disease where Phase 2 signals have a historical failure rate north of 95%. Biogen plans registrational development. Every structural risk applies: the dose-response failure complicates Phase 3 design, and 18-month trial duration with intrathecal administration means dropout risk is severe. The full diranersen analysis covers this in depth.

Novo Nordisk semaglutide (NVO, $51.58, $223.7B market cap): Already failed. EVOKE and EVOKE+ showed a 0.06-point CDR-SB difference (p=0.77). The GLP-1 neuroinflammation hypothesis is dead as a treatment approach. Semaglutide produced real anti-inflammatory activity (24 to 29% hsCRP reduction) and measurable biomarker changes. None of it translated to clinical benefit. This is the textbook surrogate-endpoint trap. At $223.7B market cap, the failure is immaterial to NVO shareholders. The full failure analysis is here.

Acadia remlifanserin (ACAD, $26.34, $4.7B market cap): FDA Fast Track for Alzheimer’s disease psychosis. Phase 2 RADIANT data expected September to October 2026. This is a different category from the cognitive decline drugs. Remlifanserin targets hallucinations and delusions, not memory. It is a 5-HT2A inverse agonist, a mechanism with precedent in psychiatric indications. The risk is lower because the endpoint is behavioral symptom control, not disease modification. But ACAD’s history with pimavanserin, which the FDA rejected for Alzheimer’s psychosis in 2021 after approving it for Parkinson’s psychosis, means the regulatory pathway is unproven. A $4.7B market cap company with one approved product (Nuplazid) needs this to work.

The Investor Framework

When you see an Alzheimer’s biotech catalyst, ask five questions:

1. What mechanism? Amyloid, tau, neuroinflammation, or metabolic. Each has a different failure rate. Anti-tau antibodies have a 0% Phase 3 success rate. Diranersen is an ASO, not an antibody. The mechanism matters.
2. What endpoint? Cognitive (CDR-SB, ADAS-Cog) or biomarker (amyloid PET, CSF tau). Biomarker-only results predict Phase 3 failure. Ask whether the trial measures what patients experience, not what scans show. Our clinical trial endpoints guide breaks this down.
3. What phase? The gap between Phase 2 and Phase 3 in Alzheimer’s is the widest in drug development. A Phase 2 win means roughly nothing. A Phase 3 win is a binary stock event.
4. What patient population? Early-stage patients show larger effects because they have less brain damage. Late-stage patients are harder to treat. Always check whether the trial enrolled early or late disease.
5. Who is the competitor? Leqembi set the bar at 0.45 points on CDR-SB. Any new cognitive drug needs to approach or exceed that. Diranersen’s 0.54 points at 18 months is the first tau drug to clear the Leqembi bar in Phase 2. Semaglutide’s 0.06 points did not come close.

The Contrarian Takeaway

The 99% attrition rate is not a reason to avoid Alzheimer’s biotech entirely. It is a reason to size positions correctly and demand evidence before buying. If you follow the 2 to 5% position sizing rule for biotech investing for any single pre-approval biotech, Alzheimer’s is the sector where that rule matters most. A Phase 2 Alzheimer’s data readout is a coin flip with a 95% chance of tails. A Phase 3 readout with good Phase 2 data and a validated biomarker is closer to 40% odds. Diranersen is the most interesting Alzheimer’s catalyst since Leqembi because it is the first drug to show that targeting tau produces both biomarker and clinical effects. But it missed its primary endpoint, the dose-response is inverted, and Phase 3 has not started. Biogen at $30.8B is not pricing in diranersen success. The market is pricing in the 99% failure rate. That is the correct base rate.

The companies that beat Alzheimer’s will own a moat nobody else can cross. The odds say most will not get there. Invest accordingly.