Published on 14 September 2026, updated on 15 September 2026
Anti-amyloid antibodies: does clearing the plaques change anything for the patient?
In brief
This Cochrane review brings together 17 randomised trials and 20,342 participants with mild cognitive impairment or mild dementia due to Alzheimer’s disease, exposed to seven monoclonal antibodies directed against amyloid beta. On cognition measured with the ADAS-Cog, the standardised mean difference is −0.11 (95% CI −0.16 to −0.06; 13 studies, 9,895 participants; moderate certainty): the gap with placebo is statistically present and trivial in size. On severity measured with the CDR-SB, the upper bound of the interval touches zero and certainty is low. On function, the authors conclude overall that the effect is “small at best”, and two of the three functional scales carry low certainty. Against that, amyloid-related imaging abnormalities increase: 107 more cases of oedema per 1,000 patients treated across 11 trials, and, in the only two trials that document the symptomatic form, 29 more symptomatic oedemas per 1,000. Serious adverse events and mortality are not increased, at high certainty. The authors conclude that “successful removal of amyloid from the brain does not seem to be associated with clinically meaningful effects”, and that research should explore other mechanisms of action.
The context
The question is no longer theoretical. Lecanemab has held a European marketing authorisation since 15 April 2025, donanemab since 25 September 2025, both restricted to patients who are non-carriers or heterozygous for the ApoE ε4 allele. A marketing authorisation is a regulatory decision on quality, safety and efficacy as filed. It is not a judgement on how much a patient stands to gain, and funding decisions are taken separately, country by country.
A patient or a family who has read the press therefore arrives with a precise question: a treatment exists, why am I not being offered it? The Cochrane review supplies the answer in figures, and it supplies it for the class as a whole rather than for one isolated molecule.
The mechanism
What the review tests, and what it does not test
The amyloid cascade hypothesis holds that accumulation of the amyloid beta peptide triggers tau pathology, synaptic loss and then clinical decline. Monoclonal antibodies are its direct therapeutic application: they target the cerebral amyloid load, and a reduction in that load is reported in the individual trials.
Amyloid load is not among the outcomes of critical importance in this review. Those outcomes are exclusively clinical: cognition, dementia severity, functional ability, amyloid-related imaging abnormalities, serious adverse events, mortality. The review therefore does not answer the question of whether the amyloid has been removed, which the individual trials have already answered in the affirmative. It answers the next one: once the amyloid is gone, what happens to the patient.
| Link in the cascade | Status in this review | Reading |
|---|---|---|
| Reduction in amyloid load | Not retained as an outcome of critical importance | A surrogate outcome, documented elsewhere, not assessed here |
| Cognition, severity, function | Primary outcomes, meta-analysed | This is where the demonstration is decided, and the effect there runs from trivial to small |
| The link between the two | Not demonstrated by these data | What the review calls into question is not the target, it is the value of that target in predicting clinical benefit |
The distinction matters for interpretation. These results do not refute the causal role of amyloid in the disease. They indicate that acting on this link, with these molecules, at this stage of the disease and over this time horizon, produces no benefit the patient can perceive.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| People with mild cognitive impairment or mild dementia due to Alzheimer’s disease. 17 studies, 20,342 participants. Mean age across the studies ranged from 70 to 74 years | |
| Intervention | |
| Monoclonal antibodies targeting amyloid beta. Seven molecules appear in the included trials: aducanumab (3 studies), bapineuzumab (4), crenezumab (2), donanemab (1), gantenerumab (4), lecanemab (1), solanezumab (2). Ponezumab and remternetug were named in the objectives, with no eligible trial | |
| Comparator | |
| Placebo or no treatment under the eligibility criteria, all 17 included trials having used a placebo | |
| Outcomes of critical importance | |
| Cognitive function, dementia severity, functional ability, any amyloid-related imaging abnormality (oedema and haemorrhage), symptomatic abnormalities, symptomatic brain haemorrhage, serious adverse events, any-cause mortality | |
| Design | |
| Cochrane systematic review with meta-analysis of randomised trials lasting at least 12 months. Eleven trials lasted 18 months, four lasted 24 months, two lasted more than 24 months. The review analyses data at 12, 18, 24 and over 24 months. The published abstract reports only the 18-month point: the analyses at 24 months and beyond were not consulted. Random-effects model, inverse variance method. Most recent search date 7 August 2025. Protocol registered on PROSPERO, CRD420251114325 · CEBM 1a · RoB 2 and GRADE |
Quality control
| Criterion | Status |
|---|---|
| Protocol registration | Sound |
| FindingPROSPERO CRD420251114325, protocol published before the review | |
| Risk of bias assessment | Reservation |
| FindingRoB 2 applied outcome by outcome rather than globally by study. The picture is mixed: low overall risk of bias for serious adverse events and mortality, but residual concerns for every efficacy outcome, mainly because of functional unblinding | |
| Grading of confidence | Sound |
| FindingGRADE for each outcome, with levels running from low to high depending on the result | |
| Independence of the review | Sound |
| FindingPublic funding: the review is funded in part by the Regione Emilia-Romagna, and publication of the article is supported by Ricerca Corrente funding from the Italian Ministry of Health. The authors’ declarations of interest do not appear on the record consulted and have not been verified | |
| Funding of the included trials | Reservation |
| FindingAll 17 trials were funded by the pharmaceutical industry. The review is independent, its raw material is not | |
| Completeness of outcome reporting | Reservation |
| FindingThe authors state explicitly that benefit outcomes and adverse events alike were reported inconsistently from one trial to another | |
| Time horizon | Reservation |
| FindingMost trials stop at 18 months. A disease that unfolds over ten years is not judged on eighteen months | |
The findings
| Outcome | Published result |
|---|---|
| Cognition (ADAS-Cog) | SMD −0.11 (95% CI −0.16 to −0.06); 13 studies, 9,895 participants; moderate certainty |
| PEB readingStatistically present, clinically trivial the interval excludes zero, which says nothing about usefulness to the patient | |
| Severity (CDR-SB) | SMD −0.12 (95% CI −0.24 to −0.00); 9 studies, 8,053 participants; low certainty |
| PEB readingUpper bound touching zero GRADE certainty falls to low on this outcome, as it does on two of the three functional scales | |
| Function (ADCS-ADL) | SMD 0.09 (95% CI 0.03 to 0.16); 3 studies, 3,478 participants; moderate certainty |
| PEB readingLittle to no difference that is the wording the authors use for this scale, on three studies only | |
| Function (ADCS-iADL) | SMD 0.21 (95% CI 0.10 to 0.32); 1 study, 1,252 participants; low certainty |
| PEB readingLow certainty, a single study the most favourable functional estimate is also the least replicated | |
| Function (ADCS-ADL-MCI) | SMD 0.23 (95% CI 0.12 to 0.33); 4 studies, 2,802 participants; low certainty |
| PEB readingLow certainty the same order of magnitude, but confidence in the estimate stays low. The phrase “small at best” applies to the authors’ overall conclusion on function, not to any single scale | |
| Amyloid-related oedema, all forms (ARIA-E) | Absolute risk difference 107 more per 1,000 (95% CI 77 more to 148 more); 11 studies, 13,595 participants; moderate certainty |
| PEB readingAbout one patient in ten develops an abnormality that would not exist without the treatment | |
| Symptomatic oedema (ARIA-E) | 29 more per 1,000 (95% CI 22 more to 38 more); 2 studies, 3,522 participants; moderate certainty |
| PEB readingAbout 1 patient in 34 develops a clinically evident form. A discrepancy worth flagging: in words the authors call this result “probably little to no difference in symptomatic ARIA E”, which the interval, lying entirely above zero, does not support | |
| Symptomatic amyloid-related haemorrhage (ARIA-H) | 4 more per 1,000 (95% CI 1 fewer to 31 more); 1 study, 1,795 participants; moderate certainty |
| PEB readingA single study, a very wide interval the imprecision forbids any conclusion in either direction | |
| Amyloid-related haemorrhage, all forms (ARIA-H) | Heterogeneous results (I² = 81%) across 3 studies, pooling not possible |
| PEB readingMissing data, not reassuring data | |
| Serious adverse events | 6 more per 1,000 (95% CI 10 fewer to 26 more); 9 studies, 11,904 participants; high certainty |
| PEB readingNo increase detected the data rule out a large excess, not a modest one | |
| Any-cause mortality | 2 more per 1,000 (95% CI 3 fewer to 11 more); 7 studies, 9,733 participants; high certainty |
| PEB readingNo increase detected with the same caveat on interpretation | |
One point of method, because everything else depends on it. The published abstract reports standardised mean differences and absolute risk differences, not scale points and not risk ratios. The figures circulating elsewhere on this file, a fall of 0.85 points on the ADAS-Cog, a risk ratio of 10 for oedema, a risk ratio of 52 for symptomatic oedema, do not appear in the published abstract. PEB does not carry them. The absolute risk difference of 107 per 1,000 can be used directly at the bedside, which a risk ratio cannot, since its value depends on the baseline risk.
Caution is called for on thresholds of clinical relevance too. A frequently cited piece of work, by Andrews and colleagues in 2019, places the decline an observer judges meaningful at around 1 to 2 points of increase on the CDR-SB. Two reservations: that threshold describes decline within one individual over time, not a gap between two trial arms, and the work proposes no value for the ADAS-Cog. The threshold of 2 to 4 points on the ADAS-Cog mentioned in our own working notes could not be traced back to a verifiable primary source, so it is not advanced here.
Critical appraisal
| Domain | Judgement |
|---|---|
| Level of evidence of the synthesis | Sound |
| FindingTop of the hierarchy: 17 randomised trials, pre-registered protocol, risk of bias assessed outcome by outcome, independent funding | |
| Selective reporting of results | Reservation |
| FindingInconsistent reporting of benefit outcomes and of adverse events alike. For symptomatic oedema, 2 studies out of 17 report the outcome; for symptomatic haemorrhage, one. When an adverse event is documented by only a minority of manufacturer-sponsored trials, underestimation is more likely than overestimation | |
| Heterogeneity | Reservation |
| FindingI² = 81% for haemorrhage in all forms. The absence of a pooled figure on this outcome is a gap, not a negative result | |
| Functional unblinding | Reservation |
| FindingOedema visible on monitoring MRI, infusion reactions, an intensified imaging protocol: blinding can break in practice. On scales rated by an examiner, that pushes the result towards the treatment, which makes the observed benefit more likely inflated than understated | |
| External validity | Reservation |
| FindingMean age in the trials between 70 and 74 years, whereas Alzheimer’s disease largely presents beyond that. Real patients, older and more comorbid, are poorly represented, including for haemorrhagic risk | |
| Class effect or molecule effect | Reservation |
| FindingThe pooled set mixes molecules with very different profiles, several of them abandoned after failure. Lecanemab and donanemab, the only two marketed, contribute one trial each. The conclusion at class level is robust, the conclusion molecule by molecule is not robust to the same degree | |
| Observation horizon | Reservation |
| FindingEighteen months for most trials. The review also analyses data at 12, 24 and over 24 months, but the published abstract reports only the 18-month point, and the analyses at 24 months and beyond were not consulted. The hypothesis of a benefit that widens at five years is neither demonstrated nor refuted by these data. It remains a hypothesis | |
Two reading errors lie in wait here, in opposite directions. On one side, presenting the absence of any increase in serious events and mortality as proof of safety: certainty is high, the interval remains compatible with 26 more serious events per 1,000, and cerebral oedema is genuinely increased. On the other, concluding that amyloid has nothing to do with Alzheimer’s disease: these trials test one intervention, at one stage, over one duration. What is solidly established, at moderate certainty on cognition and low certainty on severity, is the absence of clinically relevant benefit at 18 months. What is suggested is that amyloid load is a poor surrogate outcome. What belongs to opinion is whether the target should be abandoned or the window of intervention moved.
Level of evidence
PEB appraisal: moderate confidence in the absence of clinically perceptible cognitive and functional benefit at 18 months, low confidence on severity measured with the CDR-SB, moderate confidence in the increase in oedema-type imaging abnormalities, high confidence only in the absence of an excess of serious adverse events and deaths. The methodological apparatus is what one expects of a Cochrane review, with a conclusion calibrated on what the data allow. What lowers confidence lies not in the synthesis but in its raw material: trials all funded by industry, inconsistent reporting of adverse events, a short horizon, and a population younger than the one seen in clinic.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ The result is significant, fine, but I am looking for what it changes for my patient and I cannot find it. Against that, one more case of cerebral oedema for every ten patients treated, monitoring MRIs, infusions. I am willing to be told this is a beginning and that we should treat earlier, but then show me, because over eighteen months I do not see it. ”
What this means in practice: the question to put to any positive result is not whether it is statistically significant, but how large it is and against which threshold. This file is the textbook case, and it will serve you elsewhere.
What you can do with this on Monday morning. Whether or not you can prescribe these antibodies where you practise, the conversation reaches the consulting room.
- Answer a family with figures rather than with a position. Across the 11 trials and 13,595 participants that document oedema in all its forms, there are around 107 more cases of oedema per 1,000 patients treated compared with placebo. Across the only two trials that document the symptomatic form, 3,522 participants, there are around 29 more symptomatic cases per 1,000, again compared with placebo. These are excesses attributable to the treatment, not prevalences, and the two figures do not nest inside each other: they come from different sets of trials. On cognition, the authors call the effect trivial, and no validated threshold of clinical relevance exists for the ADAS-Cog.
- Set out the regulatory position without approximation: European marketing authorisation for lecanemab since April 2025 and for donanemab since September 2025, restricted to non-carriers or heterozygotes for ApoE ε4. An authorisation states that a dossier met the regulatory standard. It does not state that the effect is worth having, and funding decisions are taken separately.
- If a patient raises treatment abroad, recall the contraindications and the monitoring burden. The higher-risk situations are known: current anticoagulation, microbleeds or amyloid angiopathy seen on MRI, ApoE ε4 homozygosity, which falls outside the European indication. The monitoring burden is real: confirmation of amyloid status, repeated infusions, scheduled follow-up MRI.
- Steer the consultation back towards what remains available and useful: treatment of cardiovascular comorbidity and of sleep disorders, correction of hearing and vision, physical activity, support for carers, advance care planning. These measures do not come from the Cochrane review analysed here: they are external good practice recommendations, notably those of the 2024 report of the Lancet standing Commission on dementia for modifiable risk factors and carer support. The place of cholinesterase inhibitors and memantine is settled country by country, which does not prevent discussing their value case by case.
- Take away the reading grid itself: standardised mean difference on one side, threshold of clinical relevance on the other, absolute risk difference for adverse events. Three tools that serve well beyond Alzheimer’s disease.
Frequently asked questions
Are these treatments available to patients?
Both hold a European marketing authorisation, restricted to patients who are non-carriers or heterozygous for ApoE ε4. An authorisation is not the same thing as access: whether a treatment is funded is decided country by country, on an appraisal of the size of the benefit against the risks and the practical burden, and the authorisation itself says nothing about clinical value.
How can a result be statistically significant and of no clinical interest?
Statistical significance says that the observed gap sits poorly with chance. It says nothing about its size. With nearly 10,000 participants, a minute gap becomes detectable. The next question, the only useful one, is whether that gap crosses the threshold at which a patient or a relative notices a change.
Does this review bury the amyloid hypothesis?
No, and saying so would go beyond the data. It shows that these antibodies, at this stage of the disease and over this horizon, produce no clinically perceptible benefit, despite the reduction in amyloid load reported in the individual trials. That is a strong argument against the value of amyloid load as a surrogate outcome. It is not a refutation of the role of amyloid in the disease.
Are amyloid-related imaging abnormalities serious?
They range from an asymptomatic finding on a monitoring MRI to symptomatic oedema and cerebral haemorrhage. The review documents a clear increase in oedema, part of it symptomatic. On haemorrhage, the data are too scattered to be pooled and a single trial reports the symptomatic form. The absence of a reliable figure does not mean the absence of risk.
Can these molecules be called safe, since mortality does not rise?
No. Certainty is high and that is reassuring on the heaviest events, but a confidence interval compatible with 26 more serious events per 1,000 does not exclude a modest excess risk, and cerebral oedema is genuinely increased. An absence of significant difference is never proof of an absence of effect.
Could a benefit appear later, or earlier in the disease?
This is the main argument put against the review, and it is not unreasonable. It remains a hypothesis: eleven of the seventeen trials stop at 18 months, and none provides data on intervention in the preclinical phase. A plausible hypothesis is no substitute for a demonstration.
Annotated bibliography
Nonino F, Minozzi S, Sambati L, Del Giovane C, Baldin E, Bassi MC, De Santis C, Gonzalez-Lorenzo M, Vignatelli L, Filippini G, Richard E (2026). Amyloid-beta-targeting monoclonal antibodies for people with mild cognitive impairment or mild dementia due to Alzheimer’s disease. Cochrane Database of Systematic Reviews, 4(4), CD016297. DOI 10.1002/14651858.CD016297 · PMID 41985900. Source study analysed here. Review funded in part by the Regione Emilia-Romagna, publication of the article supported by the Ricerca Corrente programme of the Italian Ministry of Health; the 17 included trials are all industry-funded. The authors’ declarations of interest do not appear on the record consulted and have not been verified. PROSPERO protocol CRD420251114325, most recent search 7 August 2025.
Andrews JS, Desai U, Kirson NY, et al. (2019). Disease severity and minimal clinically important differences in clinical outcome assessments for Alzheimer’s disease clinical trials. Alzheimer’s & Dementia: Translational Research & Clinical Interventions, 5, 354-363. DOI 10.1016/j.trci.2019.06.005 · PMID 31417957. Reference work on thresholds of clinical relevance, which places a decline judged meaningful at around 1 to 2 points of increase on the CDR-SB. The limitation to keep in mind: these thresholds describe the course of one individual over time rather than a gap between two trial arms, and the ADAS-Cog does not feature in them. The method used to derive them has also been the subject of a published exchange in the literature, which reinforces the caution already stated here.
Livingston G, Huntley J, Liu KY, et al. (2024). Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. The Lancet, 404(10452), 572-628. DOI 10.1016/S0140-6736(24)01296-0 · PMID 39096926. Source of the non-drug measures cited in the practice box: modifiable risk factors and support for carers. A commission report, therefore an expert synthesis rather than a graded systematic review; it lies outside the Cochrane review analysed here.
European Medicines Agency. Lecanemab, European public assessment report. ema.europa.eu. Marketing authorisation of 15 April 2025, indication restricted to patients who are non-carriers or heterozygous for ApoE ε4. A regulatory source, without independent appraisal of the benefit-risk balance in the sense of evidence-based medicine.