Alzheimer's Disease — What Actually Happens, and What the New Drugs Actually Do

October 1, 2026

Alzheimer's disease is the second post in our series explaining the dementia diagnoses, after mild cognitive impairment. It is also the one where the most has changed in the last three years, in two directions at once: the diagnosis got dramatically easier, and the first drugs that genuinely alter the underlying disease arrived — delivering a benefit that most of the people who measured it describe as being at or below the threshold a patient would notice.

Both halves of that are true, and the gap between them is where the honest version of this article lives. We are not doctors, and nothing here is treatment advice.

What actually happens in the brain

Two proteins define the disease pathologically.

Amyloid-beta is a fragment cut from a larger membrane protein. In Alzheimer's it misfolds and aggregates outside neurons into plaques. Amyloid accumulation begins decades before symptoms — this is the single most important fact about the timeline, and the reason prevention research targets people in midlife.

Tau normally stabilises the internal scaffolding that neurons use to move material along their length. In Alzheimer's it becomes abnormally phosphorylated, detaches, and tangles up inside the cell. Unlike amyloid, tau spread tracks closely with where symptoms appear and how fast they progress.

The conventional account — the amyloid cascade hypothesis — says amyloid comes first and drives tau pathology, inflammation, synapse loss and neuronal death downstream. It has been contested for years, largely because amyloid burden correlates poorly with symptom severity, and because drug after drug cleared amyloid without helping anyone.

That last objection has partly resolved, and it is worth being precise about how. Removing amyloid now does measurably slow decline. The effect is small. So the hypothesis survives in a weakened form: amyloid appears to be causally involved rather than incidental, and also not the whole story — which is roughly what you would expect of a disease that has been accumulating for twenty years before anyone intervenes.

Diagnosis has genuinely changed

Until recently, confirming Alzheimer's pathology in a living person meant an amyloid PET scan or a lumbar puncture — expensive, limited in availability, and in the second case unpleasant. Blood tests have changed that, and the accuracy is now well characterised.

A 2026 meta-analysis of 25 studies and 18,073 participants assessed plasma p-tau217 against amyloid PET as the reference standard.¹ On the Simoa platform, pooled sensitivity was 0.894 (95% CI 0.861–0.920) with specificity 0.875 (0.845–0.900); on CLIA, sensitivity 0.886 and specificity 0.865. Which measurement platform a lab uses measurably changes the answer, which is why the review stratified by it.

A separate network meta-analysis of 18 studies, 24 independent datasets and 4,736 participants ranked the options: p-tau217 measured by mass spectrometry performed best, followed by the p-tau217 ratio and then automated immunoassays — and all of them significantly outperformed the older p-tau181 immunoassays.²

So a blood test can now identify amyloid pathology with roughly 89% sensitivity and 88% specificity. Useful, and not perfect: at those figures, roughly one in eight people without amyloid pathology will still test positive. A blood result is a strong input to a diagnosis, not a diagnosis.

The anti-amyloid antibodies: what three 2026 meta-analyses found

Lecanemab and donanemab are monoclonal antibodies that bind amyloid and recruit the immune system to clear it. They are the first treatments that change the underlying pathology rather than the symptoms. Three separate meta-analyses published in 2026 pooled the trial evidence, and they agree closely.

The main outcome is CDR-SB — Clinical Dementia Rating, Sum of Boxes — an 18-point scale where a higher score is worse. Lower numbers below mean less decline over the trial.

Six randomised comparisons from four phase III trials, 7,695 participants: pooled CDR-SB difference −0.42 points (95% CI −0.59 to −0.25), I² = 78%. The effect was driven by lecanemab and donanemab; aducanumab gave discordant results across its trials and gantenerumab showed no clinically meaningful benefit.³
Seven trials across all three main agents: CDR-SB −0.41 (−0.63 to −0.18), ADAS-Cog standardised mean difference −0.15 (−0.21 to −0.10), MMSE +0.44 (0.03 to 0.86). The authors describe the effects as small.⁴
Donanemab specifically, two trials: CDR-SB −0.66 (−0.90 to −0.42) and iADRS +2.93 (1.52–4.33), rising to +3.80 (2.10–5.50) in people with low or medium tau burden. Amyloid clearance was dramatic — risk ratio 234.46 versus placebo, with 76.4% reaching amyloid-negative status.⁵

Now the part that matters, and all three say it independently. A minimal clinically important difference is the smallest change a patient or clinician would actually notice. On CDR-SB in early Alzheimer's, these results sit at or below it:

The four-trial analysis: the pooled effect "approached the lower boundary of the minimal clinically important difference."³
The seven-trial analysis: a "statistically significant but small slowing of decline that may not reach patient-perceptible clinical meaningfulness."⁴
The donanemab analysis: the improvements "did not approach" their minimal clinically important difference thresholds.⁵

Three independent teams, three slightly different datasets, the same conclusion. These drugs do something real to the disease. Whether a person living with it would feel the difference over 18 months is not established, and the researchers who pooled the data are the ones saying so.

One genuine signal of a subgroup effect: donanemab's benefit was larger in people with lower tau burden — consistent with the broader principle that intervening earlier in the pathology does more.

The harms, and why the published rates disagree

ARIA — amyloid-related imaging abnormalities — is the characteristic risk of this drug class. ARIA-E is brain swelling; ARIA-H is small bleeds. Both are detected on MRI, and trials mandate regular scanning to find them.

Here the published figures diverge sharply, and the reason is instructive:

The seven-trial analysis of the three main agents at therapeutic doses: ARIA-E 23.9% versus 1.9% on placebo (RR 11.65, 9.06–14.99) and ARIA-H 16.8% versus 6.8% (RR 2.45, 1.94–3.09).⁴
A network meta-analysis spanning 22 trials and up to 23,120 participants, including phase II studies and lower doses: pooled ARIA-E 6.8% and ARIA-H 15.8%, with substantial heterogeneity — odds ratios of 7.93 (4.50–13.98) for ARIA-E and 1.87 (1.28–2.72) for ARIA-H.⁶

A 23.9% and a 6.8% ARIA-E rate are both correctly calculated. They differ because the second pools a much wider set of trials, agents and doses. If you see one quoted without the other, you are seeing a slice.

Proportion, honestly stated: the four-trial review found that most ARIA-E events were asymptomatic and picked up only because the protocol required MRI surveillance.³ That matters a great deal — an abnormality nobody would have noticed is not the same as an injury. It also cuts the other way: the monitoring that detects it is part of the cost and burden of the treatment.

APOE ε4 is the big risk modifier. Carriers had ARIA-E in 29.3% of cases versus 13.8% of non-carriers (RR 2.10, 1.67–2.64).⁴ The network analysis found the same for both ARIA types — ARIA-E odds ratio 2.28, ARIA-H 2.07 — with a clear gene-dose effect in homozygotes, meaning two copies carries more risk than one.⁶ Risk also differs by drug: highest ARIA-E with donanemab and aducanumab, then gantenerumab and lecanemab.

Two further things reported rather than buried. Treatment discontinuation ran at RR 3.26 (2.38–4.47) for donanemab, and all-cause mortality was not significantly different (RR 1.44, 0.69–3.00) — a confidence interval wide enough that it neither establishes nor rules out a mortality effect.⁵ And the seven-trial review noted brain-volume loss, ventricular enlargement and treatment-related deaths as signals identified narratively but not pooled.⁴ Not pooled means not quantified, which means unresolved rather than absent.

The older drugs

Donepezil, rivastigmine and galantamine are cholinesterase inhibitors: they slow the breakdown of acetylcholine, a neurotransmitter depleted in Alzheimer's. Memantine works on glutamate signalling instead. All four are symptomatic treatments — they may modestly improve or stabilise symptoms for a period without altering the underlying disease process. That is the settled position, and it is why the arrival of the antibodies mattered despite their small effect sizes: they are the first drugs in this illness that do anything to the pathology itself.

We looked at one meta-analysis reporting a notably large pooled effect for cholinesterase inhibitors and decided not to print the figure, because it is far out of line with the rest of the literature and its confidence interval was very wide. Where a number looks too good, the honest move is to leave it out rather than borrow its authority.

Where we come in, which is nowhere

The same thing we said about MCI applies with more force here.

We sell nothing that treats, slows or prevents Alzheimer's disease, and no supplement does. No ingredient in our range has been tested against this pathology. Anyone selling you a capsule for Alzheimer's — on amyloid, on tau, on "brain detox" — is selling you something that has not been shown to do it, in a category where the actual disease-modifying drugs required phase III trials of thousands of people to demonstrate an effect small enough to argue about.

What does have evidence is the risk side, before anyone is ill: blood pressure, hearing, vision, activity, alcohol, smoking, sleep, social contact, education. Those are covered properly in modifiable dementia risk and the fourteen factors. The honest comparison between that and what we sell is in exercise versus supplements.

And if what brought you here is worry about your own memory rather than a diagnosis, start with memory worries: normal ageing or not, because the overlap between being worried and being impaired is weaker than most people assume.

The short version

Amyloid plaques build up for decades before symptoms. Tau tangles track much more closely with symptoms and progression.
The amyloid cascade hypothesis survives in weakened form: clearing amyloid does slow decline, by a little.
Blood p-tau217 detects amyloid pathology at roughly 89% sensitivity and 88% specificity against PET. Strong input, not a diagnosis.
Three 2026 meta-analyses put the anti-amyloid antibody effect at CDR-SB −0.41, −0.42 and −0.66 points. All three say it sits at or below the minimal clinically important difference.
Donanemab cleared amyloid in 76.4% of treated patients. Dramatic biology, modest clinical effect — those are different things.
ARIA-E rates are quoted as 23.9% or 6.8% depending on which trials are pooled. Most events were asymptomatic and found only by mandated MRI.
APOE ε4 roughly doubles ARIA risk, with a gene-dose effect in homozygotes.
Brain-volume loss, ventricular enlargement and treatment-related deaths were flagged but not quantified. Unresolved, not absent.
No supplement treats, slows or prevents this, ours included.

The rest of this series

Six diagnoses, each written the same way: what it is, what the evidence actually shows, and what we do and do not sell for it.

This article is for informational purposes only and has not been evaluated by the FDA. It is not intended to diagnose, treat, cure, or prevent any disease. We are not doctors. Alzheimer's disease requires specialist diagnosis, and decisions about anti-amyloid therapy involve genetic testing, MRI surveillance and individual risk assessment that only a treating clinician can weigh. Nothing here is a reason to start, stop or change any prescribed medication. If you are worried about memory or thinking changes in yourself or someone else, that is a conversation for a doctor.

References

Yuan C, Zhang H, Chen X, Zhu L, Wang Z, Wang Y, Ge S. "Diagnostic accuracy of plasma p-tau217 against amyloid PET: A meta-analysis of technical platform variability and ratio versus single marker comparisons." Neurological Sciences, 2026. doi:10.1007/s10072-026-09322-8
Chen X, Huang T, Shi C, Xu S, Fan S. "Diagnostic performance of plasma p-Tau217, p-Tau181, and p-Tau231 across the Alzheimer's disease continuum: a network meta-analysis." Frontiers in Aging Neuroscience, 2026. doi:10.3389/fnagi.2026.1834591
Vamanu A, Mastaleru A, Schreiner TG, Popescu G, Roceanu AM, Cucu AI, et al. "Pharmacological and Clinical Heterogeneity of Anti-Amyloid Monoclonal Antibodies in Early Alzheimer's Disease: A Systematic Review and Meta-Analysis of Randomized Trials." Medical Sciences, 2026, 14(3):337. doi:10.3390/medsci14030337
Ali M, Shabbir H, Shaharyar M, Mukesh S, Rodriguez A, Afshani M, Kalra D, Koriesh A. "Efficacy and safety of aducanumab, lecanemab, and donanemab in Alzheimer's disease: A meta-analysis of randomized controlled trials." Journal of Alzheimer's Disease, 2026. doi:10.1177/13872877261488812
Hu G, Zhang M. "Efficacy and Safety of Donanemab in the Treatment of Alzheimer's Disease: A Systematic Review and Meta-Analysis." Current Alzheimer Research, 2026. doi:10.2174/0115672050425914260119063736
Sharma A, Suresh V, Kola PR, Raj R, Javairia F, Nithyanandam A, Kaur A, Arora H, et al. "Comparative risk of amyloid-related imaging abnormalities with anti-amyloid-β monoclonal antibodies: A systematic review and penalized likelihood network meta-analysis of randomized trials." Journal of Alzheimer's Disease, 2026. doi:10.1177/13872877261466097