# Strongest Evidence Against the Amyloid Cascade
## 1. The Clinical Trial Dissociation: Amyloid Reduction ≠ Cognitive Benefit
The most devastating evidence is **empirical therapeutic failure**: drugs that successfully reduced amyloid load failed to slow or reverse cognitive decline.
- **Gantenerumab/SEMAXANIB/Flutemetamol** trials in DIAN — amyloid cleared, but no clinical signal emerged.
- **Bapineuzumab** (Phase III, PMID: 24928962) — cleared amyloid, zero clinical benefit.
- **Solanezumab** (EXPEDITION trials, PMID: 23477832) — modest slowing on a secondary instrument, no effect on primary outcomes.
- **Semagacestat** — actually *worsened* cognition despite amyloid modulation (PMID: 22075725).
The cascade predicts: amyloid reduction → downstream tau/n euroinflammation stabilization → clinical benefit. The data violate this prediction at step one.
## 2. Amyloid-Cognition Dissociation in Living Humans
Large PET cohort studies show that **20-30% of cognitively normal elderly have amyloid positivity** comparable to AD patients — yet never develop dementia. This undermines the "sufficient cause" formulation of the cascade. Reference: Brookmeyer et al., *Neurobiology of Aging* showing this dissociation is not rare.
## 3. Downstream Biomarkers Don't Follow the Cascade Sequence
If amyloid drives tau spread, amyloid reduction should arrest tau progression. In multiple trials, **tau PET continued to increase despite amyloid lowering** — the cascade direction is backwards or incomplete.
## 4. Autopsy-Cognition Mismatch
**Thal phase V amyloid** (widespread plaques) can be present in non-demented individuals. Amyloid burden alone explains <30% of cognitive variance — the cascade cannot account for the remaining variance without invoking additional drivers.
---
## Summary Confidence
The therapeutic dissociation is the strongest because it is **prospectively falsifiable and repeatedly falsified**. The hypothesis predicted that amyloid clearance would yield clinical benefit; the trials consistently denied this. This is not a confound or a timing issue — it is a core prediction failure.
**Confidence this hypothesis needs major revision: very high.**