Published on 15 September 2026
Galantamine in Alzheimer’s disease: a certain effect, and none in mild cognitive impairment
In brief
This Cochrane systematic review brings together 21 randomised placebo-controlled trials and 10,990 participants, of which 19 trials and 10,497 participants enter the meta-analysis. The headline results concern the recommended dose, 16 to 24 mg per day. In Alzheimer’s disease, galantamine improves cognition measured with the ADAS-cog at six months (mean difference −2.86, 95% CI −3.29 to −2.43; 6 trials, 3,049 participants), at a certainty the authors grade as high. That difference crosses the lower bound of the range of clinical relevance stated for this scale, 2.6 to 4 points, but it does not reach the upper bound. The authors also report a lower death rate at six months under galantamine (1.3% against 2.3%, OR 0.56, 95% CI 0.33 to 0.96), a result resting on a very small number of deaths and whose upper bound borders on no effect. The price is gastrointestinal: the odds of nausea are multiplied by nearly three (20.9% against 8.4%, OR 2.89) and the odds of premature discontinuation by 1.41 (22.7% against 17.2%). In mild cognitive impairment, at twenty-four months, no cognitive gain is shown (expanded ADAS-cog, mean difference −0.21, 95% CI −0.78 to 0.37, low certainty), nor any gain on activities of daily living, and adverse events are more frequent; the authors do report a probable reduction in progression to dementia (OR 0.74, 95% CI 0.58 to 0.94, moderate certainty). Two limitations frame the whole. The authors themselves show that the imbalance in treatment discontinuation, combined with imputation by last observation carried forward, may bias the cognitive results in favour of galantamine, and 16 of the 21 included trials had an industry sponsor.
The context
The question comes up twice in the consulting room, and not in the same way. Faced with an isolated memory complaint or a documented mild cognitive impairment, the patient and the family often ask whether a treatment exists that would slow the decline. Faced with established Alzheimer’s disease, the question becomes the place of a cholinesterase inhibitor, in a health system where these drugs may no longer be publicly funded: in France, they have not been since 1 August 2018.
This Cochrane review answers each of these two situations separately, and the two answers do not point the same way. It says nothing, on the other hand, about the funding decision: marketing authorisation, actual marketing and collective coverage are three distinct questions, judged on different criteria. We come back to this further down, as an open question.
The mechanism
What the review measures, and what it does not
The introduction to the review recalls that galantamine is a selective, competitive and rapidly reversible inhibitor of acetylcholinesterase, and that it also acts as an allosteric modulator at nicotinic receptor sites. This pharmacological reminder lies outside the results: that dual property is often invoked to explain a superior cognitive effect, but no datum in the review allows it to be established.
| Element | What the review says about it |
|---|---|
| Starting hypothesis | The cholinergic hypothesis of cognitive deficit |
| FindingThe therapeutic reasoning rests on partial restoration of cholinergic transmission, an old and symptomatic hypothesis. The authors themselves place galantamine among the drugs that do not modify the disease. | |
| What is measured | Clinical scales, not a mechanism |
| FindingCognition (ADAS-cog), the clinician’s global impression (CIBIC-plus), activities of daily living, functional disability (DAD), behaviour (NPI), adverse events and death. These are clinical outcomes, which is a strength and not a weakness. | |
| What is not measured | No marker of the disease process |
| FindingNo central cholinergic activity, no amyloid or tau biomarker, no imaging of progression retained as a synthesis outcome. A gain on the ADAS-cog says nothing about the trajectory of the lesions, and nowhere does this review claim that it does. | |
The study at a glance
| Population, intervention, comparator, outcomes | |
|---|---|
| Population | |
| PPatients with probable or possible Alzheimer’s disease, or with mild cognitive impairment. Three trials enrolled participants with mild cognitive impairment and a Clinical Dementia Rating of 0.5, eighteen concerned Alzheimer’s disease, one of them severe forms in a nursing home setting. Mean age 74 years, 37% men. The two populations are analysed separately | |
| Intervention | |
| IOral galantamine. The summary results cover the recommended dose of 16 to 24 mg per day, that is 8 to 12 mg twice daily. Other dosing regimens, from 8 to 36 mg per day, are analysed separately in the body of the review | |
| Comparator | |
| CPlacebo. No active comparator, therefore no direct comparison with another cholinesterase inhibitor. The authors explain that choice by the fact that the included trials were conducted at a time when no other drug was approved in this indication | |
| Outcomes | |
| OCognition (ADAS-cog, the primary outcome of the review), global impression of change (CIBIC-plus), activities of daily living, functional disability (DAD), behaviour (NPI), adverse events, death. In mild cognitive impairment, dementia severity rated on the CDR-SB is added | |
| Design | |
| DSystematic review with meta-analysis of double-blind, parallel-group randomised trials lasting more than four weeks. Fixed-effect model, Peto odds ratios, risk of bias assessed with the first-generation Cochrane tool, certainty graded with GRADE. Search closed on 14 December 2022 | |
| Numbers | |
| N21 trials included, 10,990 participants. Meta-analysis covering 19 trials and 10,497 participants. The numbers randomised per trial range from 18 to 2,051 | |
| Horizon | |
| TTreatment durations range from eight weeks to two years, twenty-four weeks being the most frequent. The summary of findings tables are set at six months for Alzheimer’s disease and at twenty-four months for mild cognitive impairment |
Quality control
| Checkpoint | Judgement |
|---|---|
| Stated level of certainty | High on several outcomes |
| FindingIn Alzheimer’s disease, cognition, functional disability, behaviour, death and tolerability are graded at high certainty. That is rare in this field, where most syntheses remain at moderate or low certainty. | |
| Evidence base | 21 trials, 10,990 participants |
| FindingVolume sufficient for precise estimates on the cognitive outcomes, distinctly less so for rare events. The summary analyses draw on far narrower subsets, from 1,043 to 3,616 participants depending on the outcome. | |
| Funding of the synthesis | Public and academic support |
| FindingNo internal source of support is declared. The external sources cited are public or academic: National Institute of Mental Health, National Institute on Aging, the Australian National Health and Medical Research Council, a dementia research fellowship. | |
| Authors’ declared interests | One author linked to industry |
| FindingTwo authors declare no link. The third declares research funding from Biogen, Eli Lilly, Merck, Novartis, Roche and Genentech, TauRx, and consultancy fees from Alpha-cognition and Corium. None of those links concerns galantamine, which is worth stating, but the information deserves to reach the reader. | |
| Funding of the included trials | 16 of 21 trials sponsored by Janssen |
| FindingThe review writes it explicitly: pharmaceutical companies, Janssen and Johnson & Johnson, sponsored 16 of the 21 included trials. This is not an accusation, it is the context in which the evidence base was produced, and it justifies reading the chapter on bias attentively. | |
| Size of the cognitive effect | Lower bound of the stated range |
| Finding2.86 ADAS-cog points, against a range of clinical relevance stated as 2.6 to 4 points. The authors judge that change clinically meaningful, while noting in their discussion that the observed differences, 2.4 to 3.3 points depending on the regimen, fall short of the 4-point threshold initially set by the Food and Drug Administration. | |
| Signal on death | Few events, bound close to 1 |
| Finding1.3% against 2.3% at six months, on 3,493 participants and 6 trials. The absolute difference is about one percentage point and the upper bound of the confidence interval reaches 0.96. The result is to be handled as a favourable signal, never as an argument for prescribing. | |
| Transposition to a funding decision | Question distinct from efficacy |
| FindingThe review is international, conducted in the United Kingdom, the United States, Europe, Canada and Japan, and it does not address reimbursement, which is a national decision distinct from the assessment of efficacy. | |
The findings
| Outcome | Result |
|---|---|
| Cognition, ADAS-cog, Alzheimer’s disease, 6 months | Mean difference −2.86 (95% CI −3.29 to −2.43); 6 trials, 3,049 participants; high certainty |
| Functional disability, DAD, Alzheimer’s disease, 6 months | Mean difference 2.12 (95% CI 0.75 to 3.49); 3 trials, 1,275 participants; high certainty |
| Behaviour, NPI, Alzheimer’s disease, 6 months | Mean difference −1.63 (95% CI −3.07 to −0.20); 2 trials, 1,043 participants; high certainty |
| Global impression, CIBIC-plus, Alzheimer’s disease, 6 months | OR 1.58 (95% CI 1.36 to 1.84); 6 trials, 3,002 participants. The review grades this outcome as moderate certainty in its summary of findings table and as low certainty in its abstract, a discrepancy we flag without being able to settle it |
| Cognition, MMSE, Alzheimer’s disease, 24 months | Mean difference 0.73 (95% CI 0.34 to 1.12); 1 trial, 1,812 participants; high certainty |
| Death at six months, Alzheimer’s disease | 1.3% against 2.3%, OR 0.56 (95% CI 0.33 to 0.96); 6 trials, 3,493 participants; high certainty |
| Premature discontinuation, Alzheimer’s disease, 6 months | 22.7% against 17.2%, OR 1.41 (95% CI 1.19 to 1.68); 6 trials, 3,336 participants; high certainty |
| Nausea, Alzheimer’s disease, 6 months | 20.9% against 8.4%, OR 2.89 (95% CI 2.40 to 3.49); 7 trials, 3,616 participants; high certainty |
| Cognition, expanded ADAS-cog, mild cognitive impairment, 24 months | Mean difference −0.21 (95% CI −0.78 to 0.37); 2 trials, 1,901 participants; low certainty |
| Activities of daily living, ADCS-ADL, mild cognitive impairment, 24 months | Mean difference 0.30 (95% CI −0.26 to 0.86); 2 trials, 1,901 participants; low certainty |
| Progression to dementia, CDR-SB from 0.5 to 1.0 or above, mild cognitive impairment, 24 months | OR 0.74 (95% CI 0.58 to 0.94), that is a rate about 26% lower; 2 trials, 1,903 participants; moderate certainty |
| Tolerability, mild cognitive impairment, 24 months | Nausea 29.4% against 10.7% (OR 3.49, 95% CI 2.75 to 4.44); premature discontinuation 40.7% against 28.6% (OR 1.71, 95% CI 1.42 to 2.05); 2 trials, 2,057 participants; moderate certainty |
| Death, mild cognitive impairment, 24 months | 0.5% against 0.1%, OR 5.03 (95% CI 0.87 to 29.10); 2 trials, 2,057 participants; low certainty |
What is demonstrated. In Alzheimer’s disease, galantamine does better than placebo on cognition at six months, by an amount the authors judge clinically meaningful. The gain is established with the same certainty on functional disability and on behaviour. The gastrointestinal cost is demonstrated as well, and it is not trivial. These conclusions remain subject to a reservation the authors set out themselves, and which we return to below: the method of imputing missing data used by the included trials mechanically favours the arm that loses the most participants.
What is suggested. The lower death rate at six months. It comes from pooling the safety data of trials none of which was designed or powered to test survival. A difference of one percentage point on rare deaths, with an interval that almost touches 1, calls for a cautious reading: the result is coherent and points towards an absence of excess mortality, which is already useful information, but it cannot become an argument for prescribing, nor a promise made to a family. Also suggested, in mild cognitive impairment, is a reduction of about a quarter in progression to dementia at twenty-four months, graded at moderate certainty after downgrading for risk of attrition, and resting on two trials.
What is not demonstrated. In mild cognitive impairment, the words have to be precise: on cognition and on activities of daily living, the review shows an absence of demonstrated benefit, with point estimates very close to zero, a low certainty and a base of two trials, and not formal proof that no effect exists. The symmetry holds the other way round: an absence of significant difference on an outcome is no more a proof of safety, and the small signal of excess mortality observed at twenty-four months, statistically non-significant and graded at low certainty, illustrates that. The practical conclusion therefore rests on a whole and not on a single figure: with no cognitive or functional gain, and a clear excess of nausea and of discontinuation, the balance does not justify prescribing, even taking the signal on progression to dementia into account.
Critical appraisal
| Domain | Judgement |
|---|---|
| Method of the synthesis | Cochrane review, GRADE grading |
| FindingDouble-blind randomised trials against placebo, meta-analysis on 19 of the 21 trials, certainty graded outcome by outcome. All the trials are at low or unclear risk for randomisation, allocation concealment and blinding; four are judged at high risk for attrition and two for selective outcome reporting. | |
| Precision of the estimates | Narrow interval on cognition |
| FindingThe interval around the ADAS-cog difference is tight. That of the odds ratio for death, from 0.33 to 0.96, is wide and its upper bound borders on no effect. No funnel plot asymmetry was detected on the two outcomes with enough trials to assess it. | |
| Withdrawals and imputation | Unbalanced discontinuation, LOCF imputation |
| FindingThis is the major limitation, and it comes from the authors. All the intention-to-treat trials imputed missing data by last observation carried forward, a method that favours the arm losing the most participants in a disease that declines. Discontinuation was more frequent under galantamine. The meta-regression shows that the ratio of discontinuation rates between arms is associated with the cognitive difference (coefficient −3.18, 95% CI −5.53 to −0.82, P = 0.01), whereas the overall level of discontinuation is not (P = 0.08). In other words, it is the imbalance, and not the size of the losses, that goes with the most favourable effects. | |
| Time horizon | Two years at most, a single trial |
| FindingContrary to what is often read, the review does not stop at six months in Alzheimer’s disease: a two-year trial shows the cognitive gain on the MMSE maintained and less functional disability at twelve and twenty-four months. But that long-term datum rests on a single trial, terminated early once it had reached the number of deaths pre-specified for the interim mortality analysis. The authors themselves report the absence of data beyond twenty-four months. Nothing is established on time to institutional care. | |
| Evidence in mild cognitive impairment | Three trials, analyses on two |
| FindingThree trials enrolled participants with mild cognitive impairment, but the analyses at twenty-four months bring together only two of them. A narrow base, limited power, and every grading downgraded for high risk of attrition. The overall picture remains unfavourable for practice, without amounting to a demonstration of inefficacy. | |
| Tolerability | Excess nausea and discontinuation |
| FindingThe odds of nausea are multiplied by nearly three in Alzheimer’s disease and by 3.49 in mild cognitive impairment, the odds of premature discontinuation by 1.41 and 1.71 respectively. In a frail or undernourished older patient, this profile weighs more heavily than odds ratios suggest. | |
| Comparison with the other cholinesterase inhibitors | No direct comparison |
| FindingThe only comparator is placebo. Setting this effect size beside that of another synthesis covering another drug would be an indirect comparison, across different trials, periods and populations. A superiority of galantamine over donepezil cannot be presented as demonstrated on the basis of these data. | |
| External validity | Selected trial populations |
| FindingThe authors note that most of the trials enrolled patients with no associated comorbidity, a situation uncommon in practice. They also report an apparently different response in Japanese participants, without being able to explain why. Transposition to a patient of 84 with several comorbidities remains an extrapolation, and the search stops at December 2022. | |
Level of evidence
Confidence is high on one precise and limited point: in Alzheimer’s disease, galantamine does better than placebo on cognition, function and behaviour at six months, by an amount that is clinically perceptible. It is high as well on the gastrointestinal side, which points the unfavourable way. It is tempered by the demonstration, made by the authors themselves, that the imbalance in treatment discontinuation combined with imputation by last observation carried forward pulls the cognitive results towards galantamine. It is lower on death, where the stated certainty sits alongside a small number of events and an interval that almost touches no effect. It is low in mild cognitive impairment, where two trials are enough neither to establish a cognitive benefit nor to rule out an effect on progression to dementia. Finally, no confidence can be given to what the review did not study: the effect beyond two years, the time to institutional care, quality of life, cost-effectiveness, and the relative place of the different drugs against one another.
Solid evidence, a drug no longer reimbursed: three separate questions
This is the point that makes the review interesting beyond its subject. In France, the drugs for Alzheimer’s disease, galantamine among them, have not been covered by the statutory health insurance scheme since 1 August 2018, following a reassessment by the transparency committee of the Haute Autorité de santé, made public on 21 October 2016, which concluded that the medical value was insufficient to justify coverage by national solidarity. A Cochrane synthesis of high certainty concludes, in the same decade, that there is a cognitive benefit crossing the lower bound of the relevance threshold. The juxtaposition is striking, but it is not necessarily contradictory. The example is French; the structure of the problem is not, and it recurs in every system that funds medicines collectively.
Three questions are in fact answered separately. Marketing authorisation judges whether the benefit-to-risk ratio is acceptable for a given indication. Marketing then depends on a decision by the company holding the licence, and it can stop without anything scientific having changed: in France, the public medicines database carries today, for the galantamine products we were able to consult, either a withdrawn authorisation or a declared discontinuation of marketing, the most recent in 2025. Reimbursement, finally, arbitrates the value of collective coverage, and brings in other criteria: the size of the effect set against cost and against the organisation of care, the period over which the benefit is documented, transposability to the population actually treated, safety in real-world conditions, and place in the treatment strategy. The same datum can therefore be held scientifically reliable and judged insufficient to ground reimbursement.
This does not close the discussion, it moves it. The open question, to our mind, is whether a gain documented at six months on a cognitive scale, extended by a single trial to two years, with no datum on entry into institutional care nor on quality of life, is a sufficient basis for a collective decision. It deserves to be put in those terms, without turning it into a quarrel between a national agency and an international collaboration: they are not answering the same question.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ The message that interests me is mild cognitive impairment: I do not prescribe it there, there is nothing to gain on memory or on independence, and there is a great deal more nausea and discontinuation. The small signal on progression to dementia, I note it, but two trials and a moderate certainty do not tip a balance that is already negative. In Alzheimer’s disease, fine, the cognitive evidence is solid, almost three ADAS-cog points, that sits inside the range of the threshold we use. The signal on death I handle with tongs: there are very few deaths and the interval almost touches 1, I am not going to sell that to a family. What cools me a little is that the trials where people stop most in the treated arm are the ones that give the finest effects, and the authors say so themselves. And in any case, where I practise, it is no longer reimbursed. What I mainly take away is to watch the nausea and the weight. ”
What this means in practice: management changes in mild cognitive impairment, where holding off is now supported by direct evidence of an absence of cognitive and functional gain. In Alzheimer’s disease, what changes is not the prescription, it is the quality of what you can tell the patient and the family about the order of magnitude of the benefit, about the period over which it is documented, and about its gastrointestinal cost.
What you can do with this
- Faced with mild cognitive impairment, do not start galantamine. The argument is no longer only the absence of a recommendation: there is direct evidence of an absence of cognitive and functional gain at two years, together with a marked excess of nausea and of discontinuation. The favourable signal on progression to dementia, at moderate certainty and resting on two trials, is not enough to reverse that balance, but it is worth knowing if you are asked the question.
- Know how to give the order of magnitude. In Alzheimer’s disease, about 2.86 ADAS-cog points at six months against placebo, at a dose of 16 to 24 mg per day, inside the range of the relevance threshold used. A single trial extends the observation to two years. This is a symptomatic gain, not a slowing of the lesion process, which is the claim assessed for the anti-amyloid antibodies on a separate body of evidence.
- Know how to answer the reimbursement question by separating the three planes: authorisation, marketing, collective coverage. A reimbursement decision does not say that a treatment is ineffective, it says that the documented benefit was not judged sufficient for coverage by the community.
- Never use the signal on death as an argument for prescribing. It rests on rare deaths in trials that were not built for that purpose. It is reassuring, it is not demonstrative.
- If treatment is under way, monitor the gastrointestinal side: nausea, vomiting, appetite, weight, particularly in a frail or undernourished older patient. Before any initiation, check the contraindications specific to the class, cardiac and rhythm disorders in particular, and the combinations with other bradycardic or anticholinergic drugs.
Frequently asked questions
Is galantamine superior to donepezil?
These data do not allow that to be asserted. The review compares galantamine only with placebo, and its authors state that direct comparison with the other drugs fell outside its scope. Setting its effect size against one drawn from another synthesis covering another drug is an indirect comparison, across different trials and populations. The question would call for head-to-head trials or a network meta-analysis.
Should a cholinesterase inhibitor be offered for a memory complaint or mild cognitive impairment?
No. This is the most directly usable conclusion of the review: no cognitive or functional gain demonstrated at two years, and a great deal more gastrointestinal adverse events and treatment discontinuation. The authors do mention less progression to dementia under galantamine, at moderate certainty, but that isolated result does not reverse an overall unfavourable balance and grounds no recommendation. Clinical energy is better placed elsewhere, on characterising the disorder, looking for reversible causes, following the course over time, and taking vascular risk factors into account.
Why is this drug no longer reimbursed in some countries when the effect is demonstrated?
Because the two judgements do not bear on the same question. The review establishes that an effect exists and that it crosses the lower bound of a relevance threshold at six months. A reimbursement decision weighs that size of effect against the period over which it is documented, the functional impact, tolerability in real life and the cost to the community. In France, the transparency committee concluded in October 2016 that the medical value was insufficient, and coverage ceased on 1 August 2018.
Does the result on death change the benefit-to-risk balance?
It lightens it, it does not reverse it. A death rate of 1.3% against 2.3% at six months, with a confidence interval whose upper bound reaches 0.96, should be read as an absence of any signal of excess mortality rather than as a survival benefit. No included trial was designed to test that outcome. In mild cognitive impairment, the ratio reverses numerically, 0.5% against 0.1%, but the interval runs from 0.87 to 29.10: there is nothing to conclude from it.
How long should treatment be continued?
This review does not answer directly. Its summary of findings tables stop at six months in Alzheimer’s disease, a single trial runs to two years, and the authors explicitly report the absence of data beyond that. An absence of answer is not a negative answer: it is an uncovered area, which leaves the decision to continue or to stop to periodic clinical assessment, to tolerability and to the guidelines in force. A separate Cochrane review, devoted to withdrawing or continuing cholinesterase inhibitors, addresses that question specifically.
Annotated bibliography
Source study. Lim AWY, Schneider L, Loy C. Galantamine for dementia due to Alzheimer’s disease and mild cognitive impairment. Cochrane Database of Systematic Reviews, 2024; 2024(11): CD001747. DOI 10.1002/14651858.CD001747.pub4 · PMID 39498781. Contribution: 21 randomised trials, 10,990 participants, GRADE grading outcome by outcome, with high certainty on cognition, function, behaviour, death and tolerability in Alzheimer’s disease, and a clean separation between Alzheimer’s disease and mild cognitive impairment. Limitations: summary of findings tables at six months in Alzheimer’s disease, two-year data from a single trial, only two trials in the analyses of mild cognitive impairment, no active comparator, imputation by last observation carried forward in all the intention-to-treat trials.
Context, another synthesis in the same class. Birks JS, Harvey RJ. Donepezil for dementia due to Alzheimer’s disease. Cochrane Database of Systematic Reviews, 2018; 6(6): CD001190. DOI 10.1002/14651858.CD001190.pub3. Contribution: it sets the order of magnitude of the effects of the class on the same scales. Limitations: any comparison with the present review remains indirect, with no adjustment for differences in population, in scale and in period.
Context, withdrawing or continuing treatment. Parsons C, Lim WY, Loy C, McGuinness B, Passmore P, Ward SA, Hughes C. Withdrawal or continuation of cholinesterase inhibitors or memantine or both, in people with dementia. Cochrane Database of Systematic Reviews, 2021; 2(2): CD009081. DOI 10.1002/14651858.CD009081.pub2. Contribution: this is the synthesis that addresses the question of duration, absent from the review analysed here. Limitations: it covers the whole class and not galantamine alone.
Regulatory context, the French example. Haute Autorité de santé, reassessment of the drugs for Alzheimer’s disease by the transparency committee, communication of 21 October 2016 concluding that the medical value was insufficient to justify coverage by national solidarity. Order of 29 May 2018 delisting pharmaceutical products, published in the Journal officiel of 1 June 2018, taking effect on 1 August 2018. Contribution: it documents the criteria of a collective coverage decision, which are not the criteria of an efficacy assessment. Limitations: the decision covers the whole class, donepezil, rivastigmine, galantamine and memantine, and not galantamine alone, and it applies in France only.