Published on 16 September 2026
Stopping cholinesterase inhibitors in Alzheimer’s disease: what do withdrawal trials show?
In brief
Seven randomised trials, 955 participants randomised, 759 of them allocated to arms relevant to the question, brought together in a Cochrane systematic review with meta-analysis and GRADE assessment for each outcome. The question is the one asked in the consulting room: in a patient taking a cholinesterase inhibitor or memantine, stop or continue. At twelve months, stopping is associated with a cognitive score 2.09 points lower on the standardised version of the MMSE (95% CI −3.43 to −0.75) and a functional score 3.38 points lower on the BADLS (95% CI −6.67 to −0.10), with certainty judged moderate in both cases. The major limitation follows at once: these twelve-month data come from a single trial, with 108 participants analysed for cognition and 109 for function, and therefore without replication. In the short term, the cognitive difference is also present (SMD −0.42; 95% CI −0.64 to −0.21) but certainty is only low; in the medium term, between three and eleven months, it becomes very uncertain (SMD −0.40; 95% CI −0.87 to 0.07). Five of the seven trials stopped treatment abruptly, which is not what practice does. Finally, every participant had Alzheimer’s disease, and the authors write that their findings are not transferable to other types of dementia. For adverse events and mortality, the pooled analyses show no difference, but their certainty is low and they could not be split by time point, which does not allow a conclusion of no difference; dropouts, on the other hand, are more frequent after stopping (OR 1.48; 95% CI 1.01 to 2.17).
The context
The question of stopping a cholinesterase inhibitor comes up several times a week in an old age psychiatry clinic, and it rarely comes up for good reasons. It arises when the disease progresses despite treatment, when the family no longer sees any benefit, when the medication list grows longer, when an intercurrent event calls for a shorter prescription, or when another prescriber is surprised that the drug is still there. In all these cases the implicit reasoning is the same: if the effect cannot be seen, there is none, so it can be stopped without risk.
Where public funding of these drugs has been withdrawn, that reasoning is reinforced by a factor that is not scientific in nature. France is the example we can document. The three cholinesterase inhibitors marketed in France, donepezil, rivastigmine and galantamine, together with memantine, were removed from the list of medicines reimbursable to insured persons by the order of 29 May 2018, published in the Journal officiel of 1 June 2018, taking effect on 1 August 2018 (Légifrance, consulted on 12 August 2026). The delisting followed the reassessment conducted by the transparency committee of the Haute Autorité de santé, whose opinions of 19 October 2016 concluded that the medical benefit was insufficient (has-sante.fr, consulted on 12 August 2026). Three notions are regularly confused here and deserve to be separated, and the distinction holds in any system that funds medicines collectively. Marketing authorisation is a decision assessing the benefit-to-risk ratio: in France it remains in force, the products concerned still being listed with an active authorisation in the public medicines database (consulted on 12 August 2026). Marketing is a commercial decision: the same database declares these products marketed, and therefore obtainable from pharmacies (same source, consulted on 12 August 2026). Reimbursement is a decision on collective coverage, distinct from the other two: in France the products are now freely priced and not reimbursable. A drug that is not reimbursed is neither a withdrawn drug nor a drug whose ineffectiveness has been demonstrated. The French delisting rested on a medical benefit judged insufficient, that is on the size and clinical relevance of the benefit and on the place of the product in the treatment strategy, and not on a demonstration that there is no effect.
The review analysed here does not deal with reimbursement. It deals with a strictly clinical question that comes first: what happens, measurably, when treatment is stopped. It deserves to be read on its own terms, independently of what a funding decision, such as the French one, imposes on the patient and the family in terms of cost.
The mechanism
What a withdrawal trial actually tests
A pharmacological reminder, outside the scope of the review analysed. Cholinesterase inhibitors are symptomatic treatments. The pharmacological hypothesis is that of a cholinergic deficit: by inhibiting acetylcholinesterase, synaptic availability of acetylcholine is increased and a failing transmission is partly compensated, without acting on the neurodegenerative process itself. Memantine acts through a different mechanism, non-competitive antagonism of NMDA receptors, with the same symptomatic aim.
A testable prediction follows from this hypothesis: if treatment produces a persistent symptomatic effect, stopping it should reveal a loss of performance compared with continuation, and that loss should be all the more visible the longer the follow-up. That is exactly what a withdrawal trial measures, and it is also its limitation: it says nothing about what happens in the brain, only about what is observed at the level of a group score.
| What the design allows us to conclude | Status |
|---|---|
| Loss of performance after stopping | Measured, at group level |
| Finding The difference between arms is measured on clinical, cognitive and functional scales, not on a biological marker. | |
| Residual cholinergic activity | Not measured |
| Finding No enzyme assay, no cholinergic imaging, no biomarker objectifies the presumed mechanism. The mechanistic inference remains an expert hypothesis. | |
| Reversibility after restarting | Not measured |
| Finding The trials compare stopping with continuing. None answers the question of the patient who stopped six months ago and in whom restarting is being considered. | |
| Identification of responders | Not measured |
| Finding A mean group difference allows no individual prediction. Nothing makes it possible to identify in advance the patient who will deteriorate after stopping. | |
The study at a glance
| Design, population, comparison, outcomes | |
|---|---|
| Design | |
| Detail Systematic review with meta-analysis, Cochrane methodology: protocol published in 2011, independent double data extraction, risk of bias assessed with the Cochrane tool, GRADE assessment of certainty for each outcome. Search of the Specialised Register of the Cochrane Dementia and Cognitive Improvement Group up to 17 October 2020. Seven randomised trials included, 955 participants randomised, 759 of them in arms relevant to the question. Trial durations range from 6 weeks to 24 months according to the body of the review, whereas the abstract gives 6 weeks to 12 months: this discrepancy within the publication is due to the GAL-ITA-2 trial, followed for 24 months. Results are grouped into short term (up to 2 months), medium term (3 to 11 months) and long term (12 months or more). | |
| Population (P) | |
| Detail Patients with Alzheimer’s disease, from mild to very severe, taking a cholinesterase inhibitor and with no known adverse effect at inclusion. The published title refers to dementia in general, whereas every participant in the seven trials had dementia due to Alzheimer’s disease: the authors say so themselves and conclude that their findings are not transferable to other types of dementia. This mismatch between the title and the content should be pointed out whenever the review is cited. | |
| Intervention (I) | |
| Detail Withdrawal of the cholinesterase inhibitor or of memantine, with or without placebo substitution. Abrupt withdrawal in 5 of 7 trials, stepwise withdrawal in 2 of 7 trials. Six trials concerned withdrawal of a cholinesterase inhibitor alone; the seventh mixed donepezil and memantine without distinguishing the two, and was left out of the meta-analyses. No trial assessed withdrawal of memantine alone. | |
| Comparator (C) | |
| Detail Continuation of the same treatment. | |
| Outcomes (O) | |
| Detail Cognition (MMSE and its standardised version, the SMMSE), function in activities of daily living (BADLS, DAD), behavioural and psychological symptoms (NPI), institutionalisation, adverse events, mortality, quality of life, impact on carers. | |
| Numbers randomised and analysed | |
| Detail 955 participants randomised across the seven trials, 759 of them in the arms relevant to this question. By trial: DOMINO-AD 295 randomised, 146 of them in the two relevant arms; Johannsen 2006, 202 randomised in the double-blind phase; GAL-ITA-2 Scarpini 139; GAL-USA-5 Gaudig 118; Holmes 2004, 96 randomised out of 134 entered; Hong 2018, 65 randomised; Herrmann 2016, 40. The numbers actually analysed then vary by outcome and time point: 344 participants and 4 trials for cognition in the short term, 411 and 3 trials in the medium term, 108 and 1 trial at twelve months. |
Quality control
| Item | Judgement |
|---|---|
| Level of evidence | Systematic review and meta-analysis |
| Finding A Cochrane synthesis of randomised trials, the highest level in the hierarchy of evidence for a therapeutic question. The methodology of the review is not in question. | |
| Explicit GRADE assessment | Present for every outcome |
| Finding The authors report a certainty ranging from very low to moderate depending on the outcome and the time point, instead of presenting a single, undifferentiated result. No outcome reaches high certainty. This is what allows readers to calibrate their confidence result by result. | |
| Twelve-month data | Moderate certainty |
| Finding The outcome most relevant to practice, the cognitive difference at one year, is also the one with the highest certainty: 2.09 SMMSE points (95% CI −3.43 to −0.75), in 108 participants from a single trial. Three outcomes are graded moderate at this time point, and the third, the neuropsychiatric score, shows no difference (mean difference −0.87 NPI points; 95% CI −8.42 to 6.68). | |
| Total volume of evidence | Small, 759 participants |
| Finding Seven trials and 759 participants in the relevant arms, for a question that applies to millions of patients worldwide. The subgroup analyses planned in the protocol, by dementia severity, by duration of treatment before inclusion and by mode of withdrawal, could not be carried out: there were too few trials available. These are therefore not negative analyses, but missing ones. | |
| Replication of the main result | A single trial at twelve months |
| Finding The best-graded result comes from a single trial, DOMINO-AD, which randomised 295 patients into four arms, of which only 146 belong to the relevant comparison, and in which 108 to 109 were analysed at twelve months. A moderate GRADE certainty does not make up for the lack of independent replication. | |
| Mode of withdrawal | Abrupt in 5 of 7 trials |
| Finding Abrupt withdrawal is not recommended practice. What the review measures is therefore the effect of a more aggressive mode of withdrawal than the one used in the field, which limits how directly the figures can be applied. | |
| Breadth of the population | Alzheimer’s disease only |
| Finding The review had nonetheless made vascular dementia, mixed dementias, dementia with Lewy bodies and Parkinson’s disease dementia eligible. No corresponding trial was found: there are therefore no usable data for these forms. Yet it is precisely in dementia with Lewy bodies that the question of stopping carries the highest stakes. | |
The findings
| Outcome | GRADE certainty |
|---|---|
| Cognition, short term | Low |
| Result Standardised mean difference of −0.42 (95% CI −0.64 to −0.21) to the disadvantage of stopping, across 4 trials and 344 participants. The authors define the short term as the period up to 2 months after randomisation. The interval excludes no effect, but low certainty means that the estimate may be substantially changed by new data. | |
| Cognition, medium term | Very low |
| Result Between 3 and 11 months, standardised mean difference of −0.40 (95% CI −0.87 to 0.07) across 3 trials and 411 participants, with substantial heterogeneity (I² = 81%). The interval includes no effect and certainty is very low: at this time point, the review reaches no conclusion. When the trial that had recruited only patients with a poor response to donepezil is set aside, heterogeneity falls (I² = 24%) and the result turns against stopping (SMD −0.62; 95% CI −0.94 to −0.31): this is a sensitivity analysis, not the main result. | |
| Cognition, twelve months | Moderate |
| Result Mean difference of −2.09 SMMSE points (95% CI −3.43 to −0.75) in 108 participants from a single trial. On a scale scored from 0 to 30, two points represent a modest gap. The review sets no threshold of clinical relevance for this outcome and does not discuss one: interpreting the size of the effect is therefore left to the reader. The lower bound of the interval is compatible with a gap of more than three points. | |
| Function, before one year | Low, then very low |
| Result In the short term, the review concludes that stopping may make little or no difference to independence (SMD −0.25; 95% CI −0.54 to 0.04; 2 trials, 183 participants; low certainty). In the medium term, SMD −0.38 (95% CI −0.74 to −0.01; 2 trials, 314 participants), but with very low certainty: the upper bound touches no effect. | |
| Function, twelve months | Moderate |
| Result Mean difference of −3.38 BADLS points (95% CI −6.67 to −0.10) in 109 participants from a single trial, to the disadvantage of the withdrawal arm. The direction of reading is set by the authors: in their tables, a lower difference corresponds to a more marked functional deterioration relative to baseline. The upper bound comes close to no difference, which makes the result fragile. | |
| Neuropsychiatric symptoms | Low, no effect at one year |
| Result Contrary to what one might think, this outcome was indeed analysed. In the short term, SMD −0.48 (95% CI −0.82 to −0.13; 2 trials, 136 participants; low certainty); in the medium term, SMD −0.27 (95% CI −0.47 to −0.08; 3 trials, 410 participants; low certainty). At twelve months, by contrast, mean difference of −0.87 NPI points (95% CI −8.42 to 6.68; 108 participants; moderate certainty), that is little or no difference. The apparent behavioural benefit at the shorter time points is therefore not found at one year. | |
| Adverse events and mortality | Low certainty |
| Result The analyses exist and show no difference: all adverse events OR 0.85 (95% CI 0.57 to 1.27; 4 trials, 446 participants), serious adverse events OR 0.80 (95% CI 0.46 to 1.39; 4 trials, 390 participants), death OR 0.75 (95% CI 0.36 to 1.55; 5 trials, 598 participants). All are graded low certainty, for risk of bias and imprecision, and could not be split by time point for lack of data. This finding therefore does not demonstrate equivalence. There is, however, a signal on dropouts from any cause, which are more frequent after stopping (OR 1.48; 95% CI 1.01 to 2.17; 6 trials, 694 participants). | |
| Institutionalisation and quality of life | Very limited data |
| Result A single trial measured entry into institutional care. Across the two relevant arms, the cumulative probability of entering institutional care does not differ significantly (HR 1.46; 95% CI 0.94 to 2.29). Pooling the four arms of that trial, which goes beyond the comparison under review, the authors report an excess of entries into institutional care in the first year after stopping donepezil (HR 2.09; 95% CI 1.29 to 3.39), with no difference over the following three years; the trial was not powered for this outcome. The patient’s quality of life and the carer’s health status, measured in two trials, show little or no difference. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Conduct of the review | Low risk of bias |
| Finding Protocol published as early as 2011, double extraction, risk of bias and certainty assessed with validated tools, deviations between protocol and review declared. The synthesis is sound; it is the available trials that are few, not the method that pools them. The risk of bias of the primary trials, for its part, is far from low: two trials are judged at high risk of attrition bias, two at high risk of reporting or outcome selection bias, one at high risk of performance bias for lack of blinding, and all domains include many unclear ratings. | |
| Imprecision | Insufficient sample size |
| Finding 759 participants in the relevant arms, spread across seven trials, and numbers analysed that fall to 108 for the twelve-month outcome. Subgroup analyses by severity, by duration of treatment before inclusion and by mode of withdrawal were planned in the protocol and could not be done: any subgroup reading therefore lies outside what these data can support. | |
| Replication | Main result not replicated |
| Finding The three results graded moderate come from the same twelve-month trial, DOMINO-AD: two favour continuation, the third, the neuropsychiatric score, shows no difference. A single trial, even a well-conducted one, remains exposed to the particularities of its recruitment, its site and its period. Moderate certainty concerns the quality of the estimate, not its reproducibility. | |
| External validity, mode of withdrawal | Not representative |
| Finding Five of the seven trials stop treatment all at once. In practice, withdrawal is done in steps. The measured gap is therefore probably an upper estimate of what a gradual taper would produce, but this comparison has not been tested directly. | |
| External validity, population | Alzheimer’s disease only |
| Finding Extrapolating these figures to dementia with Lewy bodies or to vascular dementia would be a matter of opinion, not of evidence. The published title, which refers to dementia, is broader than the content. | |
| Safety and mortality | Low-certainty analyses |
| Finding The analyses exist and show no difference, but their certainty is low and the authors could not split them by time point, for lack of usable data: they therefore pool trial durations ranging from 6 weeks to 24 months. This part cannot be treated as reassuring. The point matters particularly for cholinesterase inhibitors, whose cardiovascular and gastrointestinal profile is one of the usual reasons for stopping: a pharmacological reminder, outside the scope of the review. | |
| Independence | Structurally strong |
| Finding The review is supported by public and academic funding: Queen’s University Belfast, HSC R&D of the Public Health Agency for Northern Ireland, NIHR infrastructure funding. Five of the seven authors declare no conflict of interest; one author declares having been an investigator in a donepezil registration trial and having received honoraria and grants from manufacturers of cholinesterase inhibitors and memantine, another a Wellcome Trust travel award. The Cochrane format limits allegiance bias at the level of the synthesis but does not neutralise it in the primary trials: four of the seven trials were funded by industry, two by public or charitable funds, one by mixed funding, and the review authors themselves note that all but two of the trials were sponsored or funded by industry. Note that the direction of any industry bias would here favour continuing treatment. | |
Level of evidence
What is established with the highest certainty in the review, judged moderate and not high: in patients with moderate to severe Alzheimer’s disease treated with donepezil, stopping is followed at twelve months, at group level, by cognitive and functional performance below that of the group that continues. The result comes from randomised trials, so the comparison is causal within this experimental setting, but here it rests on a single trial and about a hundred participants analysed.
What is suggested without being established: that stopping worsens cognition and independence at shorter time points, where certainty falls from low to very low and where the authors write that stopping “may” lead to worsening, not that it does; that the size of the loss is clinically noticeable to the family; that the gap is smaller with a gradual taper; and that the phenomenon applies to other forms of dementia.
What is contradicted by the one-year data: the idea that the behavioural benefit observed at the shorter time points is maintained. At twelve months, the difference on the neuropsychiatric score is nil, with moderate certainty, like the two other results at the same time point.
What is expert opinion: attributing the loss to the disappearance of a symptomatic cholinergic effect, what to do about restarting, and the choice of the threshold beyond which a very severely affected patient no longer derives benefit. The review brings nothing decisive on these three points.
Three inferential risks deserve to be named. First, low power on the safety outcomes, which easily turns an absence of demonstration into false reassurance. Second, a single-source effect on the most cited result: a meta-analysis whose long-term arm rests on a single trial does not have the robustness that the word meta-analysis suggests to the hurried reader. Third, a selection effect at entry into the primary trials: two of them had sorted their participants on their previous response to treatment, in one direction for the first, which kept only responders, and in the other for the second, which kept only non-responders. This heterogeneity of recruitment weighs directly on the medium-term estimate.
The colleague test
What a geriatrician or old age psychiatrist would say if you put this review to them in two minutes, between two consultations.
“ All right, stopping is not neutral, I knew that anyway. What I mostly take away is that the one-year figure comes from a single trial, and that treatment was stopped all at once in almost all the others. So I am not going to tell the family he will lose two MMSE points, I am going to say that stopping probably costs something, and that if we stop it will be gradually and with monitoring. And I have nothing for Lewy body disease, which is exactly my most difficult case. ”
What this means in practice: the usable message is not a figure to announce to the patient, it is a reversal of the burden of proof. Stopping is no longer the default option without consequences, it becomes a decision that has to be justified, planned and monitored, just like starting treatment.
What you can do with this
- Do not treat stopping as a neutral option. In a patient with Alzheimer’s disease who is stable and free of adverse effects, continuation remains the option supported by the data. Absence of visible benefit is not absence of benefit, since the relevant comparator is the patient’s course without treatment, which cannot be observed.
- Justify stopping with an explicit reason. Intolerance, bradycardia or conduction disorder, syncope, repeated falls, undernutrition or vomiting, interaction, end of life, the patient’s refusal, no treatment goal left at all. That reason is written in the notes, it is not inferred from how long the patient has been on treatment.
- If stopping is decided, do it in steps and with monitoring. The data come mostly from abrupt withdrawals, which are not recommended. Plan a cognitive and functional reassessment after an interval, and tell the family what to report.
- Keep the reimbursement question separate from the clinical question. Where these drugs are no longer reimbursed, as in France, where delisting took effect on 1 August 2018, the change alters the cost borne by the family, not the results of the trials. This distinction should be stated explicitly in the consultation, otherwise it is understood as a judgement of ineffectiveness.
- Do not extrapolate beyond Alzheimer’s disease. For dementia with Lewy bodies, Parkinson’s disease dementia, vascular dementia or frontotemporal degeneration, the decision rests on other arguments, which this review does not provide.
Frequently asked questions
Is a two-point MMSE difference clinically noticeable?
Not necessarily in a given patient. It is a mean difference between two groups on a 30-point scale, with a confidence interval running from about 0.75 to 3.43 points. Some patients will not change, others will lose more, and nothing in these data makes it possible to know in advance which. It is an argument for monitoring after stopping, not a figure to announce to a family.
If cholinesterase inhibitors are no longer reimbursed, does that mean they do not work?
No, and the confusion is common. Take the French example: the delisting ordered by the order of 29 May 2018 rests on a medical benefit judged insufficient by the Haute Autorité de santé in October 2016, that is on the size and clinical relevance of the benefit and on the place of the product in the treatment strategy; it does not say that an effect is non-existent, and it is distinct from the marketing authorisation, still in force in France, and from marketing, still ongoing there (sources consulted on 12 August 2026). The review analysed here measures something else: what happens when treatment is stopped in a patient already on it.
Should cholinesterase inhibitors be continued indefinitely in very severe Alzheimer’s disease?
The review includes patients of mild to very severe severity, but the comparison between stages, planned in the protocol, could not be carried out: there were too few trials, and the authors state explicitly that they were unable to determine whether the effect of stopping varies with baseline severity. There is therefore no evidence-based severity threshold beyond which stopping has been shown to be without consequences. The decision is made case by case, on the goals of care and tolerability, accepting that it is a clinical judgement and not a trial conclusion.
Is the loss smaller if cholinesterase inhibitors are tapered gradually?
It is plausible but not demonstrated. Only two of the seven trials used gradual withdrawal, which allows no reliable comparison between the two modes. Stepwise withdrawal remains recommended for reasons of pharmacological caution, not because these data have established that it is superior.
Does stopping cholinesterase inhibitors increase adverse events or mortality?
The analyses were indeed carried out and show no difference, whether for adverse events, serious adverse events or mortality. But their certainty is low, and the authors could not separate them by time point: they mix trials of 6 weeks and trials of 24 months. This result should therefore be read as an absence of demonstration, not as evidence of safety. The same caveat applies in the other direction to continuing treatment. One point is worth flagging: dropouts from any cause are more frequent in the withdrawal arm.
Annotated bibliography
Source study. 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; 2021(2): CD009081. DOI 10.1002/14651858.CD009081.pub2 · PMID 35608903. The only Cochrane systematic synthesis devoted to this question. Contribution: pooling of seven randomised trials with GRADE assessment by outcome and by time point. Limitations: small volume of evidence, a long-term arm resting on a single trial, a population restricted to Alzheimer’s disease.
The trial behind the twelve-month data. Howard R, McShane R, Lindesay J, Ritchie C, Baldwin A, Barber R, et al. Donepezil and memantine for moderate-to-severe Alzheimer’s disease. The New England Journal of Medicine 2012; 366(10): 893-903. DOI 10.1056/NEJMoa1106668. The DOMINO-AD trial, a two-by-two factorial design, 295 patients randomised into four arms, of which only two belong to the comparison of continuing or stopping donepezil on memantine placebo. Contribution: it is the source of the three results the review grades moderate. Limitations: a single, unreplicated trial, whose inclusion criteria, moderate to severe Alzheimer’s disease with an SMMSE between 5 and 13 in patients living at home, shape the whole long-term message.
Analysis of entry into institutional care in the same trial. Howard R, McShane R, Lindesay J, Ritchie C, Baldwin A, Barber R, et al. Nursing home placement in the donepezil and memantine in moderate to severe Alzheimer’s disease (DOMINO-AD) trial: secondary and post-hoc analyses. The Lancet Neurology 2015; 14(12): 1171-81. Contribution: it is the source of the institutionalisation data used in the Cochrane review. Limitations: secondary and post hoc analyses, in a trial that was not powered for this outcome.
Efficacy of donepezil in Alzheimer’s disease. Birks JS, Harvey RJ. Donepezil for dementia due to Alzheimer’s disease. Cochrane Database of Systematic Reviews 2018, issue 6, CD001190. DOI 10.1002/14651858.CD001190.pub3. Contribution: it sets out the size of the initial benefit, without which the question of stopping has no frame. Limitations: a modest symptomatic benefit, measured on group scales, with substantial heterogeneity between trials.
Earlier meta-analysis on the same question. O’Regan J, Lanctôt KL, Mazereeuw G, Herrmann N. Cholinesterase inhibitor discontinuation in patients with Alzheimer’s disease: a meta-analysis of randomized controlled trials. Journal of Clinical Psychiatry 2015; 76(11): e1424-31. PMID 26646039. Contribution: it had already concluded that there was cognitive and behavioural worsening after stopping, and placed most of the difference within the six weeks following withdrawal, which bears directly on monitoring. Limitations: its authors judged the included trials to be of good overall quality, whereas the Cochrane review grades certainty as low; this is a disagreement of appraisal, not of data.
Regulatory context, the French example. Haute Autorité de santé, transparency committee, opinions of 19 October 2016 on ARICEPT, EXELON, REMINYL and EBIXA: insufficient medical benefit. Delisting ordered by the order of 29 May 2018 removing pharmaceutical products from the list referred to in the first paragraph of article L. 162-17 of the French social security code, Journal officiel of 1 June 2018, taking effect on 1 August 2018; appeal dismissed by the Conseil d’État on 16 December 2019 (decision no. 422672). Contribution: these texts explain the prescribing context in France and the frequency of requests to stop. Limitations: this is a decision on collective coverage, not a demonstration of ineffectiveness, it applies in France only, and it does not address the question of withdrawal in a patient already on treatment.
Practice guideline on dementia. National Institute for Health and Care Excellence. Dementia: assessment, management and support for people living with dementia and their carers, guideline NG97, published on 20 June 2018. Contribution: it formalises the principle of not stopping a cholinesterase inhibitor solely on the grounds of disease severity. Limitations: a guideline written for a particular health system, whose arrangements for access and follow-up do not transfer as they stand to other systems, the French one among them.
Rational deprescribing. Reeve E, Farrell B, Thompson W, Herrmann N, Sketris I, Magin PJ, et al. Deprescribing cholinesterase inhibitors and memantine in dementia: guideline summary. Medical Journal of Australia 2019; 210(4): 174-179. DOI 10.5694/mja2.50015 · PMID 30771226. A summary of the full guideline published in February 2018 by the Cognitive Decline Partnership Centre and the Bruyère Research Institute, itself cited by the Cochrane review. Contribution: it proposes a structured approach to stopping, with tapering and monitoring, where the Cochrane review provides only comparative figures. Limitations: recommendations resting partly on expert consensus, with the same limited base of trials.