Published on 19 September 2026
Donepezil, galantamine, rivastigmine: what does a comparison of effect sizes actually show?
The essentials
This systematic review pooled, molecule by molecule, randomized placebo-controlled trials of the three cholinesterase inhibitors used in Alzheimer’s disease. On cognition, all three outperform placebo, with standardized mean differences of −0.33 for donepezil (95% CI −0.52 to −0.13), −0.48 for galantamine (95% CI −0.58 to −0.38) and −0.65 for rivastigmine (95% CI −1.06 to −0.23). Set side by side, these figures suggest a ranking. They do not establish one: these are separate meta-analyses with no direct comparison between molecules, their confidence intervals overlap, and much of the size of the rivastigmine estimate rests on a single 42-patient phase II trial. On functioning, only donepezil reaches significance (0.24, 95% CI 0.12 to 0.36). On adverse events, the excess is significant with galantamine (odds ratio 2.34, 95% CI 1.35 to 4.08) and non-significant with donepezil (1.22, 95% CI 0.98 to 1.52), which does not rule out an added risk. Rivastigmine could not be analyzed on either outcome. The authors conclude that donepezil should be favored. That preference rests on no formal comparison, the review reports neither a tool nor a result for risk-of-bias assessment, and its own selection criteria are contradicted by its own trial characteristics table. The choice between the three molecules cannot be based on this work.
Context
Cholinesterase inhibitors are among the most studied drugs in Alzheimer’s disease, and the Cochrane reviews on donepezil and its reimbursement in France and on galantamine in Alzheimer’s disease and mild cognitive impairment have already been analyzed here. Yet one question keeps coming up in consultations, often from families: which of the three works best?
This meta-analysis appears to answer it by reporting an effect size for each molecule, and its authors go so far as to recommend donepezil. What it mainly illustrates is why figures from separate analyses do not amount to a comparison.
In France, donepezil, galantamine and rivastigmine retain marketing authorization, but their formulations were removed from the list of reimbursable medicines from August 1, 2018, under a ministerial order dated May 29, 2018. In its 2016 reassessment, published with the October 19, 2016 opinion on donepezil, the Transparency Committee of the French National Authority for Health (Haute Autorité de Santé) judged the medical benefit rendered to be insufficient, found that the three cholinesterase inhibitors and memantine no longer had a place in the therapeutic strategy, and recommended that they be delisted from reimbursement (Légifrance, the public medicines database, and the Haute Autorité de Santé website, accessed September 10, 2026). The example is French, but the structure of the problem is not: marketing authorization, commercial availability and collective reimbursement are three distinct decisions, and a national health authority’s finding of insufficient added value does not necessarily settle how a clinician should weigh a modest but real effect at the bedside. Readers outside France should check the reimbursement and coverage status of these molecules in their own jurisdiction.
The study at a glance
| Question (PICO) | |
|---|---|
| Population | |
| Alzheimer’s disease, from mild to severe stage depending on the trial: in the trial characteristics table, ten trials at mild to moderate stage, four at severe stage, one at moderate to severe stage and one with no stage specified. All ages eligible, mean ages from 70 to 86 years. This table does not describe a single rivastigmine trial. | |
| Interventions | |
| Donepezil, galantamine, rivastigmine; doses not detailed, dosage forms not analyzed for lack of data | |
| Comparator | |
| Placebo | |
| Outcomes | |
| Cognition and functioning as standardized mean difference, adverse events as odds ratio; neither the scales used nor the definition of adverse events is specified in the text; functioning and adverse events judged insufficiently documented for rivastigmine alone | |
| Design | |
| Systematic review reported as PRISMA-compliant, pairwise random-effects meta-analyses in Review Manager 5.3. Double-blind randomized trials published in English, PubMed search from inception to October 2022. Sixteen trials for efficacy (six donepezil, six galantamine, four rivastigmine), fifteen for safety (nine donepezil, six galantamine). Heterogeneity assessed with Cochran’s Q and I². |
Quality control
| Criterion | Status |
|---|---|
| Stated method | PRISMA |
| FindingA systematic approach is reported, with a flow diagram | |
| Literature search | Restricted |
| FindingPubMed only per the stated methods and limitations; the abstract also cites clinical trial websites and the flow diagram an Indian registry, with no strategy described; English-language articles only | |
| Risk of bias of the trials | Said to be assessed, not disclosed |
| FindingA quality assessment of the trials is mentioned, with no tool named and no result given | |
| Certainty of evidence | Not reported |
| FindingNo GRADE assessment reported: it is impossible to weigh the confidence in each estimate | |
| Funding and conflicts of interest | None declared |
| FindingTwo Chinese institutional research projects (Hebi, Henan); article under a CC BY 3.0 licence | |
Results
| Outcome, molecule versus placebo | Pooled estimate (95% CI) |
|---|---|
| Cognition, donepezil | Standardized mean difference −0.33 (−0.52 to −0.13) |
| PEB readingSmall, significant effect six trials, 750 patients, I² 38% | |
| Cognition, galantamine | Standardized mean difference −0.48 (−0.58 to −0.38) |
| PEB readingSmall-to-moderate, precise effect six trials, 2,466 patients, I² 29% | |
| Cognition, rivastigmine | Standardized mean difference −0.65 (−1.06 to −0.23) |
| PEB readingImprecise and highly heterogeneous four trials, 1,503 patients, I² 92%, one 42-patient trial at −3.22 | |
| Functioning, donepezil | Standardized mean difference 0.24 (0.12 to 0.36) |
| PEB readingSmall, significant effect six trials, 1,128 patients, I² 0%, three at severe stage | |
| Functioning, galantamine | Standardized mean difference 0.17 (−0.08 to 0.43) |
| PEB readingNot significant, highly heterogeneous three trials, I² 83%: an inconclusive result, which does not establish an absence of effect | |
| Adverse events, donepezil | Odds ratio 1.22 (0.98 to 1.52) |
| PEB readingNot significant, an excess not excluded nine trials, 498 patients with an event out of 869 versus 459 out of 862, I² 0% | |
| Adverse events, galantamine | Odds ratio 2.34 (1.35 to 4.08) |
| PEB readingSignificant, heterogeneous excess six trials, 208 patients with an event out of 1,386 versus 83 out of 1,246, I² 73% | |
| Functioning and tolerability, rivastigmine | Data judged insufficient for a meta-analysis |
| PEB readingNot analyzed which does not mean without effect or without risk | |
A standardized mean difference expresses the gap in units of standard deviation, which allows different scales to be pooled. It does not say whether the patient feels better day to day. Comparing −0.33 with −0.65 would assume that the trials for each molecule involved comparable patients, durations and doses, something the review does not allow one to verify: its trial characteristics table does not describe a single rivastigmine trial. The rivastigmine figure also calls for a specific caveat. It pools three trials with effects ranging from −0.41 to −0.23 and a 42-patient phase II trial (Forette 1999) credited with an effect of −3.22, with a standard deviation of 1.4 in each arm against 6.2 to 7.3 in the other three trials, a discrepancy the review does not comment on and that cannot be verified without the original publication of that trial. With a weight of 12.0% in the analysis, it markedly pulls the pooled estimate. The apparent order of the three molecules is therefore not established.
Two further points limit the reading. On the functioning forest plot, the axis places the label “favors control” on the side of positive values, which the text interprets as a treatment benefit, and the direction of the scales is never described anywhere. For adverse events, the review does not state what was actually counted: the proportions of events, 57% under donepezil and 15% under galantamine, suggest different definitions from one molecule to the next.
Critical appraisal
| AMSTAR 2 domain | Judgment |
|---|---|
| Pre-registered protocol | Not reported |
| FindingNo protocol or registration mentioned anywhere in the full text | |
| Completeness of the search | Insufficient |
| FindingA single database, PubMed, per the limitations stated by the authors themselves, with restriction to English | |
| Consistency of selection criteria | Contradicted by the data |
| FindingA minimum treatment duration of 52 weeks is stated, while the trial table shows durations of 12 to 24 weeks or 3 to 6 months | |
| Risk of bias of included trials | Not reported |
| FindingQuality assessment mentioned with no tool and no result, on a domain critical to AMSTAR 2 | |
| Heterogeneity | High, unexplored |
| FindingI² of 92% (rivastigmine, cognition), 83% and 73% (galantamine, functioning and tolerability), with no sensitivity or subgroup analysis | |
| Publication bias | Not assessed |
| FindingRisk acknowledged by the authors, not tested for lack of a sufficient number of trials | |
| Reliability of the report | Multiple inconsistencies |
| FindingReference citations that do not match the figures, rivastigmine replaced by galantamine in two summary sentences | |
| Comparison between molecules | Not tested |
| FindingNeither a network meta-analysis nor a formal indirect comparison | |
| Adequacy of the conclusion | More assertive than the data |
| FindingThe authors recommend favoring donepezil with no comparison between molecules, while calling for further trials on galantamine and rivastigmine | |
Level of evidence
PEB assessment: reasonable confidence in a cognitive effect for donepezil and galantamine, weaker for rivastigmine, low on everything else. That each of the three cholinesterase inhibitors performs a little better than placebo on cognitive scales is consistent with the Cochrane reviews. Whether one molecule is more effective or better tolerated than another is not something this work can establish. On functioning and tolerability, which are what decide practice, it provides estimates for donepezil and galantamine, but with no reported risk-of-bias assessment, undefined criteria, and nothing at all for rivastigmine.
The colleague test
What an experienced colleague would say if shown this study in two minutes, between two consultations.
What this mainly confirms is a small cognitive effect for all three drugs. But the size of the rivastigmine figure, resting on a wide interval and one small trial pulling the average, is not a reason to reach for it first, any more than the authors’ recommendation is a reason to switch everyone to donepezil without a real comparison between molecules. The choice still comes down to tolerability, formulation and the individual patient, and for that the Cochrane reviews remain the better reference.
Translation for practice: nothing changes in how the molecule is chosen. The study mainly helps explain to a resident, or to a family, why a ranking by effect size is not enough.
What you can do with this
- Continue choosing between the three molecules based on the adverse-effect profile, the route of administration and the patient’s comorbidities.
- When a family asks which one is “the strongest”: this review does not compare the molecules with each other and does not allow any one of them to be called superior.
- Keep in mind that these treatments are no longer reimbursed in France; check the coverage and reimbursement status that applies in your own jurisdiction, since it can weigh on the discussion with the patient and their family.
- Faced with a meta-analysis that lines up effect sizes molecule by molecule, check whether a direct or network comparison was actually done before drawing a ranking from it, and look for whether a single trial is carrying the estimate.
Frequently asked questions
Is rivastigmine more effective than donepezil?
This work cannot say. Both figures come from separate analyses, the confidence interval for rivastigmine is wide, and its estimate depends heavily on one small trial with atypical data.
Should donepezil be favored, as the authors suggest?
Not on the basis of this review. Donepezil is the only molecule with a significant effect on functioning and a non-significant excess of adverse events, but these results come from separate analyses with no comparison between molecules, and the odds-ratio interval (0.98 to 1.52) does not rule out an added number of events.
Is an effect of −0.33 clinically useful?
It is a small effect in standard-deviation units. This review does not report how it translates into a benefit the patient can perceive.
Why does the absence of GRADE matter?
Because an estimate can be precise and still rest on low-quality trials. Without a certainty assessment, there is no way to know how much weight to give each figure.
What does this review say about tolerability?
It reports a significant excess of adverse events with galantamine (odds ratio 2.34, 95% CI 1.35 to 4.08) and a non-significant excess with donepezil (1.22, 95% CI 0.98 to 1.52), without specifying which events were counted. For rivastigmine, the authors judged the data insufficient to pool, which means neither an absence of adverse effects nor good tolerability.
Annotated bibliography
Source study. Gao Y, Liu Y, Li Y. Safety and efficacy of acetylcholinesterase inhibitors for Alzheimer’s disease: a systematic review and meta-analysis. Advances in Clinical and Experimental Medicine, 2024, volume 33, issue 11, pages 1179 to 1187. DOI 10.17219/acem/176051 · PMID 38439609, record established in the PubMed registry on September 10, 2026. Published online on March 1, 2024. Funding: Project in High-Tech Research Centre for home care products for the elderly patients with dementia and disability in Hebi (project Heke (2021)45-14) and Young Backbone Teacher’s Project of Henan Higher Vocational School (project 2020GZGG121). Conflicts of interest: none declared. No registration or protocol mentioned in the publication. No supplementary material accompanied the document consulted: the supplementary figures and tables of this publication were not reviewed, and no data appearing only there is used here.
Editorial collections
Tags
Verified on September 10, 2026 against the full text of the publication and its supplementary material where available. This analysis underwent an independent double reading. The English version was checked for conformity on September 19, 2026, against the figures of the French version and against the source. How we verify what we publish
This analysis is intended for healthcare professionals. It does not constitute a prescribing recommendation and does not replace individual clinical judgment.
Analysis from Psychiatry Evidence Base, evidence-based psychiatry, explained with rigor.