Citation Chain Audit — Case Study #5
The Amyloid Hypothesis — How the Costliest Wrong Hypothesis in Drug History Survived Twenty Years of Failures
04-23-2026
The drug worked. Plaques were removed. PET imaging proved it. Cognition didn't improve in any trial, at any stage, in any patient population. The error was not in the science -- it was in the step from correlation to causation that nobody required evidence for.
The Citation Chain
1984 — Foundational Observation
Glenner & Wong — Biochemical and Biophysical Research Communications
Amyloid beta protein first isolated and sequenced from the cerebrovascular deposits of Alzheimer's patients. Establishes the molecular identity of amyloid plaques. Correlation between amyloid deposition and Alzheimer's disease is real and reproducible.
✓ Validated — amyloid is present in Alzheimer's brains
↓
correlation interpreted as causation
1992 — The Hypothesis Is Named
Hardy & Higgins — Science
The amyloid cascade hypothesis formally published. Central claim: amyloid beta accumulation is the initiating event in Alzheimer's disease. All other pathology -- tau tangles, neuronal death, cognitive decline -- follows downstream from amyloid. The unstated premise: correlation implies causation.
⚠ Hypothesis — causal mechanism asserted, not demonstrated
↓
hypothesis drives therapeutic strategy
Late 1990s — Biomarker Becomes the Surrogate Outcome
PET Imaging for Amyloid — Adopted as Clinical Endpoint
Amyloid PET imaging developed and validated. For the first time, plaques can be visualized and quantified in living patients. The field adopts amyloid clearance as the primary endpoint for clinical trials. The implicit assumption: if the drug removes amyloid, the drug is working.
⚠ Biomarker trap — amyloid clearance accepted as proxy for cognitive benefit without validation
↓
multi-billion-dollar trials designed on biomarker endpoint
2000s — Therapeutic Programs Launched
Solanezumab (Eli Lilly), Bapineuzumab (Pfizer / J&J), Gantenerumab (Roche), and Others
Multiple anti-amyloid antibody programs initiated across the industry. Each targets amyloid clearance. Trial designs center on PET-confirmed amyloid reduction as primary or surrogate endpoint. The foundational assumption -- that amyloid causes cognitive decline -- is not a condition of investment. It is the premise of investment.
⚠ $10B+ committed on an assumption that was never gated as a condition of funding
Eli Lilly solanezumab program public disclosure · Multiple Phase III IND filings
The Break — Three Unvalidated Leaps
Leap 1: Correlation to causation. Amyloid plaques are present in Alzheimer's brains. This is a correlation. The cascade hypothesis asserted causation. No controlled experiment established that amyloid deposition directly causes neuronal death in humans (as opposed to co-occurring with the actual cause). This gap was visible in 1992 and never closed.
Leap 2: Biomarker to outcome. "Amyloid is reduced on PET" was treated as evidence the drug was working. A surrogate endpoint is only valid if reducing the surrogate has been shown to produce the desired clinical outcome. For amyloid, that validation was never performed -- it was assumed.
Leap 3: Early intervention to disease modification. When late-stage trials failed, the explanation was "we treated too late." The strategy shifted to earlier and earlier intervention. The assumption: the drug works, the timing is wrong. The alternative explanation -- the hypothesis is wrong -- was systematically not considered.
Failures — The Sequential Record
2012
Solanezumab Phase III fails in mild cognitive impairment
Eli Lilly's solanezumab Phase III EXPEDITION 1 and EXPEDITION 2 trials fail to show cognitive benefit in mild-to-moderate Alzheimer's. Amyloid was cleared. Cognition was unchanged. Field explanation: "We need to intervene earlier, before symptoms appear."
The hypothesis survives via explanation. The explanation is untested.
ClinicalTrials.gov NCT00905372 · NCT00904683
2016
Solanezumab fails in asymptomatic patients
EXPEDITION 3 trial tests solanezumab in patients with mild Alzheimer's and confirmed amyloid on PET. Despite amyloid clearance, no statistically significant cognitive benefit. The earlier-intervention thesis is tested -- and fails. Field explanation: "We need to treat before any cognitive changes, in truly preclinical disease."
The hypothesis survives via a second, untested explanation.
2020
A4 Study fails definitively
The Anti-Amyloid Treatment in Asymptomatic Alzheimer's (A4) study -- the definitive test of the early-intervention hypothesis. Participants: cognitively normal adults with confirmed amyloid on PET. Treatment: solanezumab for 4.5 years. Result: amyloid was reduced. Cognitive decline was unchanged. The plaques were not causing the decline.
Every version of the earlier-intervention explanation is now exhausted.
Sperling et al. ·
PMID 37458272 · NEJM 389:1096, 2023 (results published 2023, trial ran 2014-2020)
June 2021
Biogen's aducanumab approved despite 10-of-11 advisory panel voting against
FDA grants accelerated approval to aducanumab (Aduhelm) based on amyloid reduction as a surrogate endpoint -- the same surrogate that twenty years of trials showed does not correlate with cognitive benefit. 10 of 11 members of the FDA's advisory panel voted against approval. Three panel members resigned in protest. The approval price: $56,000 per year. The institutional logic: the surrogate is all we have, so we approve on the surrogate.
Peak Biogen market cap: $56 billion. The biomarker that failed as a clinical endpoint is approved as the basis for a drug.
January 2024
Aducanumab withdrawn from market
After clinical failure in real-world use, widespread reimbursement denials (Medicare initially refused coverage), and serious safety concerns (ARIA -- amyloid-related imaging abnormalities, including brain swelling and bleeding), Biogen voluntarily withdrew aducanumab from the market. 20 years of trials. $10B+ spent across the industry. The drug that proved amyloid removal was achievable also proved it was irrelevant to the disease.
Peak market cap: $56 billion. End state: withdrawn. One unvalidated causal leap, sustained for twenty years.
Why It Took Twenty Years
Pattern 1
Sunk cost at institutional scale
When a hypothesis has consumed $10B+ across multiple companies, no individual has the authority or incentive to say "we were wrong." Careers have been built on amyloid. Departments organized around it. Tenure decisions made on it. Grant portfolios built on it. Killing the hypothesis means admitting that decades of work and billions of dollars were misdirected. The institutional logic of the hypothesis persists long after the scientific evidence turns against it.
Pattern 2
Each failure was explained, not acknowledged
"We treated too late." "We treated the wrong patients." "The dose was wrong." "The antibody wasn't specific enough." "We need biomarker-confirmed patients." Each explanation was scientifically plausible. Each one was also untested before the next trial was designed around it. The pattern: negative evidence is always refiled as a reason to run the next trial, never as evidence against the hypothesis. The hypothesis is immune to falsification because every failure produces a new version of the hypothesis rather than a reconsideration of it.
Pattern 3
The biomarker trap
PET imaging showed plaques disappearing. Beautiful, quantifiable, reproducible evidence that the drug was "working." This created a perverse dynamic: the more clearly you could measure amyloid reduction, the harder it was to accept that the measurement didn't matter. The measurement became a substitute for the outcome. When the FDA approved aducanumab on amyloid reduction as a surrogate, they institutionalized the biomarker trap -- approving on the measurement that twenty years of trials proved does not predict the actual outcome patients care about.
What Was in the Literature — Available Evidence Against the Hypothesis
2006
Findings from DIAN and other autosomal dominant AD studies
Studies of patients with familial Alzheimer's caused by APP and presenilin mutations show that amyloid accumulation precedes cognitive symptoms by 15-20 years. If amyloid causes the disease, the timeframe between amyloid and symptoms raised questions about mechanism. Critics noted that tau pathology -- not amyloid -- better predicted the pattern of neuronal loss and cognitive decline in multiple study cohorts.
Available before major trial failures
2009
Jack et al. — Lancet Neurology — Biomarker trajectory model
Proposed model of Alzheimer's biomarker trajectories: amyloid accumulates first, but tau and neurodegeneration markers (not amyloid) correlate more closely with clinical decline. The implication: amyloid may be an early event that does not directly drive the downstream pathology. The model was widely cited -- and then largely ignored by trial design, which continued to focus on amyloid clearance.
Available before A4 study design was finalized
2014
Amyloid-positive cognitively normal individuals — multiple cohort studies
Multiple cohort studies identified individuals with high amyloid burden on PET who remained cognitively normal for years. If amyloid causes cognitive decline, the existence of amyloid-positive cognitively normal people is unexplained by the hypothesis. The alternative: amyloid is a risk factor or co-occurring process, not the driver. The hypothesis predicts this population should not exist, or should rapidly decline. They didn't.
Available before aducanumab approval
$2B+
Eli Lilly spend on solanezumab alone, 2000-2020
$56B
Biogen peak market cap on aducanumab -- drug later withdrawn
$10B+
Estimated industry total across all amyloid programs
20 yrs
Duration from hypothesis to definitive disproof (A4 study, 2020)
What forensics would have flagged at investment time
The foundational assumption -- amyloid is causal, not correlational -- was the load-bearing claim that was never validated before billions were spent. A citation chain forensic audit in the late 1990s would have asked one question: "What is the evidence that amyloid causes cognitive decline, versus that it correlates with cognitive decline?" The answer was epidemiological association. No controlled mechanism proving amyloid deposition directly causes neuronal death in humans (as opposed to co-occurring with the actual cause). That gap was visible in 1992 in the Hardy & Higgins paper itself -- the authors knew they were proposing a cascade hypothesis, not a demonstrated mechanism.
The forensic finding would not kill investment. It would reframe the condition: fund the first amyloid-clearing trial with a specific protocol to test whether amyloid clearance produces cognitive benefit. Gate subsequent investment on that result. If the first trial clears amyloid and cognition improves, proceed. If the first trial clears amyloid and cognition is unchanged, the hypothesis fails -- after one trial, not twenty.
That gate was never set. The biomarker became the endpoint. The correlation became the causation. Twenty years and $10B+ later, the A4 study confirmed what the 1992 hypothesis should have required evidence for before the first dollar was spent.