Impact of GLP-1 Receptor Agonists on Cognitive Performance in Adults at Risk of Alzheimer's Disease: A Systematic Review and Meta-Analysis of Randomized Controlled Trials.

Impact of GLP-1 Receptor Agonists on Cognitive Performance in Adults at Risk of Alzheimer’s Disease: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Background/Objectives: Type 2 diabetes, a widely recognized risk factor of Alzheimer’s disease (AD), is mainly due to the role played by insulin resistance on neuronal degeneration. Mild Cognitive Impairment (MCI) also significantly increases the risk of developing AD by approximately 40%. Epidemiological studies have demonstrated correlations between glucagon-like peptide-1 agonists (GLP-1RAs) on reducing the risk of AD. We synthesize current data from randomized controlled trials (RCTs) to summarize the effects of GLP-1RA treatment in adults with Type 2 diabetes or with MCI on cognitive performance. Methods: We searched databases from inception to 1 January 2026 investigating effects of GLP-1RA(s) on cognition. PubMed, Web of Science and Scopus databases were explored following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). We selected only RCTs in adults investigating GLP-1RA(s) on cognitive performance. Results: Overall mean age was 64.5 yrs with an average follow-up period of 28.4 months. A total of 13,711 participants from five RCTs were included. Following a source-data audit, four randomized controlled trials contributed analytically compatible data to at least one cognitive domain. Cognitive domains were analyzed separately using Hedges’ g. Random-effects models used restricted maximum likelihood (REML) estimation with Knapp–Hartung inference when at least two compatible studies contributed. For global cognition, three studies yielded g = 0.014 (95% CI: −0.061 to 0.088; p = 0.515; I2 = 0%). For attention, two studies yielded g = 0.072 (95% CI: −2.164 to 2.308; p = 0.753; I2 = 31.1%). For delayed memory, executive function, and verbal function, only one compatible study was available per domain; these were therefore reported as single-study estimates rather than pooled effects. A sensitivity analysis excluding the dominant Cukierman trial yielded an imprecise and non-significant effect estimate for global cognition (g = −0.205, 95% CI: −2.678 to 2.269) indicating that the robustness of the pooled result could not be established. Conclusions: Current randomized evidence is insufficient to establish a consistent beneficial effect of GLP-1RA(s) on cognitive function in adults at risk of AD. The available findings should be considered exploratory and hypothesis-generating because few analytically compatible trials were available within each cognitive domain.

1. Introduction

Alzheimer’s disease (AD) currently burdens over 55 million people on a worldwide scale and is predicted to triple by 2050 [1]. Type 2 diabetes mellitus has been shown to negatively affect cognitive performance and is associated with an approximately twofold increased risk of AD [2], with chronic hyperglycemia and insulin resistance potentially contributing to neuronal dysfunction and accelerated cognitive decline [3,4]. Insulin plays a pivotal role in normal cognitive functioning and directly regulates brain glucose metabolism. Insulin resistance and altered insulin/insulin growth factor (IGF) signaling have been implicated in reduced cognitive performance [5,6], as well as in AD’s main etiopathogenic cascades, namely beta-amyloid (AB) deposition and tau protein hyperphosphorylation [7]. In addition to insulin resistance, other mechanisms implicated in the development of AD include neuro-inflammation, oxidative stress, mitochondrial dysfunction, and vascular dysfunction.

Given these shared metabolic and molecular pathways, the potential of glucose-lowering therapies that specifically improve insulin sensitivity as disease-modifying strategies for cognitive decline has emerged as a growing area of research [8]. Preclinical experimental models and AD clinical trials suggest that selected antidiabetic agents may exert neuroprotective effects [9]. Among these, glucagon-like peptide-1 agonists (GLP-1RA(s)) have gained increasing attention due to numerous epidemiological studies reporting associations between GLP-1RA and a reduced risk of dementia, thus raising interest in potential cognitive benefits beyond glycemic control [10]. Considering their neurotrophic, neuroprotective actions, GLP-1RA(s) are evolving as potential disease-modifying treatments even in those without Type 2 diabetes, who are afflicted with early AD and Mild Cognitive Impairment (MCI) [11,12,13]. GLP-1RA(s) not only improve insulin sensitivity but also exhibit anti-inflammatory and antioxidant properties that may attenuate neuro-inflammation [14,15]. In addition, GLP-1RA(s) can cross the blood–brain barrier and thus, may directly improve central insulin signaling [16].

MCI also significantly increases the risk of developing AD by approximately 40% [17]. Cognitive impairment in individuals at risk of AD commonly involves specific areas related to deficits in memory, attention, verbal abilities, and executive functioning. Neuropsychological testing of these domains may reflect metabolic and neurofunctional status of specific brain areas and may be sensitive to therapeutic interventions targeting insulin resistance. At the moment, there is a need to recognize adults at a high risk of developing AD and test the potential impact of GLP-1RA(s) on cognitive performance.

Given this background, the present systematic review and meta-analysis aimed at pooling evidence from available randomized controlled trials (RCTs) to evaluate the effects of GLP-1RA(s) on specific cognitive outcomes in adults at high risk of developing AD (Type 2 diabetes or MCI).

2. Materials and Methods

2.1. Search Strategy

We performed a systematic review with meta-analysis, in accordance with the Cochrane Collaboration guidelines, and reported the findings following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [18]. A protocol was preregistered on PROSPERO (number: CRD420261301187) and accessed at https://www.crd.york.ac.uk/PROSPERO/view/CRD420261301187 (accessed 5 March 2026) [19].

Three electronic bibliographic databases (PubMed, Scopus and Web of Science) were searched from inception to 1 January 2026 with the search strings adapted to each database’s syntax.

Key terms included combinations of diverse condition specific terms (e.g., “glucagon-like peptide-1 receptor agonists”, “glp-1 agonist”, “glp-1ra”, “glp-1 analog”, “incretin mimetic” “lixisenatide”, “liraglutide”, “semaglutide”, “albiglutide”, “dulaglutide”, “exenatide” and “Dementia”, “Cognitive impairment”, “MCI”, “Neurocognitive disorder”, “Cognitive decline”, “Memory disorder”, “Memory impairment”, “Cognitive dysfunction”, “Brain degeneration”, “Mental deterioration”, “Alzheimer”, “MMSE”, “Mini Mental State Examination”, “neuropsychological test battery”, “NTB”, “neurobehavioral test battery”, “neuropsychological evaluation”, “neuropsychological assessment”, “GP-COG”, “assessment of cognition”, “evaluation of cognition”, “cognitive assessment”, “cognitive evaluation”, “cognitive test”, “short test of mental status”, “STMS”, “short portable mental status questionnaire”, SPMSQ”, “Pfeiffer test”, “clock drawing test”, “CDT”, “MDB”, “mental deterioration battery”, “FAB”, “frontal assessment battery”, “ADAS-Cog”, “AD assessment scale-cognition”, “Digit span”, “NCSD”, “BPSD”, “global deterioration scale”, “neuropsychological tests”). The search strategies detailed by database are reported in Supplementary File S1.

2.2. Hand Search Was Conducted by Reviewing the Bibliographies of Included Studies

We built the search string following this PICO question: P—adults over the age of 18 years with Type 2 diabetes or with MCI (no Type 2 diabetes); I—RCT testing treatment with GLP-1RA; C—compare treatment with GLP-1RA to placebo or other anti-diabetic agents; O—what is the effect of GLP-1RA treatment according to diverse areas of neuropsychological testing.

2.3. Study Selection

Duplicate articles were removed, and the title and abstract of all collected records were screened by 2 review authors (A.M.A. and G.M.) to identify articles that were potentially eligible for inclusion. Studies that clearly did not meet the inclusion criteria were excluded. Full texts of potentially relevant articles were retrieved and independently assessed for eligibility by 2 review team members (A.M.A. and G.M.). Disagreements were resolved through discussion and the reasons for exclusion were recorded.

Inclusion criteria were: (i) adults over 18 years of age; (ii) treatment with GLP-1RA; (iii) reporting information on cognitive performance tests evaluating different cognitive domains related to global cognitive functioning, attention, verbal, executive functioning, and memory; (iv) studies in humans; (v) written in English. Eligible GLP-1RA agents included: lixisenatide, liraglutide, semaglutide, albiglutide, dulaglutide, exenatide in the search. We included studies evaluating the use of GLP-1RA(s) independently of nutritional status, on cognitive performance evaluation tests following treatment. When differences in assessment arose between authors regarding study eligibility, a third author was consulted to review the differing opinions and make the final decision on study inclusion. Studies providing raw mean and standard deviation of cognitive tests for the outcome of interest were also included in the meta-analysis.

Exclusion criteria included: (i) non-original research (including reviews, comments, editorials, perspectives), (ii) non-English language because the review team did not have sufficient linguistic resources to reliably assess eligibility, methods and data extraction, (iii) retracted publications, (iv) animal or in vitro studies, (v) conference abstracts, (vi) lack of specific information for performing a meta-analysis.

2.4. Data Extraction and Outcomes

Data were abstracted independently by two researchers (F.D. and G.M.) using a standardized data collection form with a third independent investigator (A.M.A.) controlling the data collected. For every scientific paper, we abstracted the title, year of publication, authors, type of study design (open, placebo-controlled), intervention, numbers, cognitive performance tests and domains, and changes in cognitive test scores over time. We extracted data from the follow-up assessments at the time points reported in the included studies. When more than one assessment time point was reported, results at 18–24 months, or at the nearest time point available, were extracted. Included RCT protocols registered at ClincalTrial.gov website were searched for unreported data. The extracted data were cross-checked for inconsistencies and discrepancies, merged into a preliminary dataset, and independently reviewed by a third investigator (A.M.A.).

Given that specific areas of cognitive performance correspond to discrete areas of cerebral function, we investigated changes in neuropsychological measures of global cognitive functioning, attention, verbal, executive functioning, and memory. We evaluated performance scores in these five cognitive domains according to treatment or control groups. Cognitive function was assessed using the cognitive domains and instruments reported in the individual studies. No single cognitive domain, cognitive instrument, or assessment time point was prespecified as the primary cognitive outcome in the registered protocol. Therefore, the analyses of individual cognitive domains were considered exploratory. For global cognitive functioning, we evaluated the performance scores of the Mini-Mental State Examination (MMSE) or the Montreal Cognitive Assessment (MoCA). When more than one instrument assessing the same cognitive domain was reported within a study, a single instrument was selected to avoid double-counting correlated outcomes. For global cognition, the MMSE was preferentially selected when available; the MoCA was selected only when MMSE data were not reported. This selection rule was established post hoc after examination of the eligible studies and was therefore not prespecified in the original protocol. We included the Digit Symbol Substitution Test (DSST), the attention subdomain in the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) as measures of attention. Verbal domain tests included the Verbal Fluency Test. Trail making tests A and B were used to test for measures of executive functioning. Cognitive performance tests of memory included Digit Span Forward (DSF), delayed memory subdomain in the RBANS, and Spanish English Verbal Learning Test (SEVLT) (see Supplementary Table S1). Primary outcomes were comparative measures of cognitive performance between intervention and control groups.

2.5. Statistical Analysis

All statistical analyses were performed using Stata version 18 (StataCorp LLC, College Station, TX, USA). Cognitive domains were analyzed separately. To avoid unit-of-analysis errors, one cognitive outcome per trial within each domain was included in the quantitative synthesis. Continuous outcomes were summarized using Hedges’ g when compatible group-level means, participant-level standard deviations, and sample sizes were available. Model-based adjusted means and their confidence intervals or standard errors were not treated as raw means and participant-level standard deviations and not included in further analysis. For instruments in which lower scores indicate better performance, the direction of the effect was reversed so that positive Hedges’ g values consistently favored GLP-1RA treatment.

Random-effects models used restricted maximum likelihood (REML) estimation, with Knapp–Hartung inference when at least two compatible studies contributed to a pooled estimate. Between-study heterogeneity was described using tau2 and I2 and interpreted cautiously because of the small number of studies. No overall pooled estimate combining distinct cognitive domains was calculated. Formal funnel-plot asymmetry testing using Egger’s test was not performed because fewer than 10 independent studies contributed to each domain. Because one trial contributed 99.3% of the weight to the global cognition analysis, a sensitivity analysis excluding this dominant trial was also performed.

Statistical significance was set at a two-sided p-value < 0.05. Pooled results were expressed as Hedges’ g with 95% confidence intervals (95% CI) and displayed using forest plots. Domains represented by a single compatible study were reported as single-study effect estimates and were not interpreted as pooled meta-analytic effects. The dataset and statistical code used for the analyses are available from the corresponding author upon reasonable request.

2.6. Quality Assessment

The quality of each outcome of interest included was assessed independently by two authors (A.M.A. and G.M.) using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2) [20]. The RoB 2 tool consists of five domains evaluating: (i) bias arising from the randomization process, (ii) bias due to deviations from intended interventions, (iii) bias due to missing outcome data, (iv) bias in outcome measurement, and (v) bias in selection of the reported results. Based on the individual domain risk-of-bias judgments, the overall risk-of-bias judgment is defined as “low risk of bias,” “some concerns,” or “high risk of bias”. As RoB 2 is outcome-specific, outcomes were initially considered at the outcome level; however, when multiple cognitive outcomes within the same trial shared the same methodological features and yielded identical domain-level judgments, they were presented jointly to avoid unnecessary duplication.

2.7. Certainty of Evidence Assessment

The certainty of evidence for each outcome of interest was evaluated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [21]. GRADE assessments were performed both for outcomes included in quantitative meta-analyses and for outcomes represented by a single study. For the latter, the certainty assessment was based on the individual study estimate. As no specific cognitive domain, instrument, or follow-up time point was prespecified as the primary outcome of the review, domain-specific analyses were considered exploratory. As all included studies were randomized controlled trials, the certainty of evidence was initially rated as high and was subsequently downgraded by one or two levels when concerns were considered serious or very serious, respectively, across five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Risk of bias was informed by the RoB 2 assessments of the studies contributing to each outcome; for pooled estimates, their relative contribution to the meta-analysis was also considered. Inconsistency was assessed considering the magnitude and direction of individual study effects, overlap of confidence intervals, and statistical heterogeneity, including the I2 statistic. For outcomes represented by a single study, inconsistency was considered not assessable because no between-study comparison was possible and therefore did not result in downgrading. Indirectness was evaluated in relation to the prespecified review question, considering whether differences in study populations, GLP-1RA agents, comparators, cognitive instruments, and follow-up duration remained within the intended scope of the review. Imprecision was judged on the width of the 95% confidence interval around the pooled or individual study effect estimates and whether it included both no effect and clinically relevant benefit or harm. Publication bias was assessed qualitatively, considering the number and size of the available studies and the possibility of selective non-publication. For single-study outcomes, publication bias was evaluated qualitatively on the basis of the available evidence, acknowledging that formal assessment was not possible. The overall certainty of evidence was classified as high, moderate, low, or very low.

3. Results

A literature search of three databases—PubMed, Web of Science and Scopus produced 3319 results, of which 1353 duplicates were removed, leaving 1966 records to be screened. Title and abstract screening excluded 1950 records. Full-text review excluded 11 articles. A total of five articles were included in the systematic review, and four were included in the quantitative synthesis. PRISMA flow chart is illustrated in Figure 1. Please see Supplementary Table S2, which provides detailed information on study selection and exclusion in the Supplementary Files.

Overall, five studies included data from different cognitive domains in 13,711 participants (Table 1). The mean duration follow-up was 28.4 months (range from 4 to 64). All eligible studies were RCTs comparing the use of GLP-1RA drugs (exenatide, dulaglutide, liraglutide) to a control group.

Of note, we reached out to the authors of the pilot study using exenatide [12] and to the authors of the GRADE trial [22], to ask for data regarding memory, verbal and executive functioning between groups; however, we did not receive any response regarding this information.

Participant baseline data, patient characteristics and neuropsychological tests analyzed in reports testing the use of GLP-1RA treatment are reported in Table 1. We found an overall mean age of 64.5 years. There were not any substantial differences according to age regarding intervention and comparison groups. Studies that included patients with MCI without T2DM used the GLP-1RA agent exenatide. The other two RCTS in those with Type 2 diabetes investigated the use of GLP-1RA agents, liraglutide and dulaglutide.

The study and instrument underlying each single-study estimate for delayed memory, executive function, and verbal function are reported in Table 2.

According to the outcome-specific RoB 2 assessment for the five trials included in the systematic review, outcomes from three trials [22,23,24] were judged at low risk of bias and two [12,13] raised some concerns (Figure 2). Specifically, insufficient reporting of the randomization procedures in Dei Cas et al. [13] raised some concerns regarding bias arising from the randomization process, whereas the proportion of randomized participants without baseline cognitive assessments in Mullins et al. [12] raised some concerns regarding bias due to missing outcome data. No included trial-outcome assessment was classified as high risk of bias in the revised analytical set.

Following source-data audit, only analytically compatible continuous outcomes were used to calculate Hedges’ g, with one outcome per trial within each cognitive domain. Three studies contributed to the global cognition analysis [12,13,24]. The pooled random-effects estimate showed no evidence of a treatment effect (Hedges’ g = 0.014, 95% CI: −0.061 to 0.088; p = 0.515), with tau2 = 0 and I2 = 0%. However, the estimate was largely driven by the substantially larger Cukierman trial [24], which accounted for 99.3% of the meta-analytic weight (Figure 3).

In the sensitivity analysis excluding the dominant Cukierman trial [24] from global cognition, the remaining two studies yielded g = −0.205 (95% CI: −2.678 to 2.269; p = 0.484), with tau2 = 0 and I2 = 0%. This estimate was highly imprecise and did not provide evidence of either benefit or harm (Figure 4).

Two studies contributed to the attention domain [23,24]. The pooled estimate was g = 0.072 (95% CI: −2.164 to 2.308; p = 0.753), with tau2 = 0.036 and I2 = 31.1%. The very wide Knapp–Hartung confidence interval reflects the substantial uncertainty associated with pooling only two studies (Figure 5).

For memory, executive function, and verbal function, only one compatible study was available per domain; these findings were therefore reported as single-study effect estimates rather than pooled meta-analytic effects [12,13,23]. The corresponding estimates were g = 1.10 (0.27 to 1.93) for delayed memory, g = 0.611 (95% CI: −0.181 to 1.403) for executive function, and g = −0.554 (95% CI: −1.257 to 0.148) for verbal function. No pooled forest plot was generated for domains represented by a single study.

GRADE assessment, reported in Table 3, indicated high-certainty evidence for global cognition, suggesting little to no effect of treatment on this outcome. No downgrade for indirectness was applied because differences in study population, GLP-1RA agent, comparator, cognitive instrument, and follow-up remained within the prespecified scope of the review. Also, although Cukierman-Yaffe et al. contributed approximately 99% of the meta-analytic weight, this was not considered a source of indirectness because the trial population, intervention, comparator, and outcome were directly relevant to the prespecified review question. No downgrade for risk of bias was applied because Cukierman-Yaffe et al., which contributed almost all the statistical weight, was judged at low risk of bias, while the concerns identified in the two smaller studies were considered unlikely to materially affect the pooled estimate. No downgrade for inconsistency was applied because the individual study effects showed no important discordance and statistical heterogeneity was negligible (I2 = 0%). No downgrade for imprecision was applied because the pooled estimate was close to the null and the 95% confidence interval was narrow (Hedges’ g 0.01, 95% CI: −0.06 to 0.09), excluding effects of a magnitude considered clinically important. Finally, although formal assessment of publication bias was limited by the small number of studies, we found no specific indication warranting downgrading. In contrast, the certainty of evidence for attention was rated as low because of very serious imprecision, indicating substantial uncertainty regarding the direction and magnitude of the effect. Among single-study findings, certainty of evidence was rated as low for executive function due to very serious imprecision, moderate for memory due to serious imprecision, and low for verbal fluency due to serious risk-of-bias concerns and serious imprecision. As these outcomes were represented by a single trial, inconsistency was not assessable and did not result in downgrading.

4. Discussion

This systematic review and meta-analysis evaluated the effects of GLP-1RA treatment on cognitive performance in adults at risk of AD, specifically with Type 2 diabetes or without Type 2 diabetes (MCI). We found no evidence supporting GLP-1RA(s) on cognitive performance for adults at risk of AD. Global cognition showed no evidence of a treatment effect, whereas the attention estimate remained not significant even following sensitivity analysis. Cognitive performance domains evaluating memory, executive function, and verbal function were each represented by a single compatible study and therefore, could not be interpreted as pooled domain-specific treatment effects.

General cognitive assessments with MMSE or MoCA lack sensitivity to modest domain-specific changes, particularly in the early stages of cognitive decline [25]. Because in-depth neuropsychological testing of specific domains, including attention, verbal abilities, memory, and executive functioning, can reflect the integrity of areas of cerebral functioning, we specifically examined the available literature of RCTs on these individual outcomes. Clinically, subtle impairment of one or more of these domains often occurs before they can be detected using the MMSE or MoCA [25]. It is important to note that domain-specific analyses of cognitive outcomes are gaining increasing attention in the literature, as illustrated by the recent meta-analysis by Xu et al. [26], which similarly evaluated distinct cognitive domains separately rather than combining heterogeneous neuropsychological outcomes into a single overall estimate. However, this meta-analysis [26] was designed to specifically evaluate cognitive performance in patients with diagnosed neurodegenerative diseases, namely AD or Parkinson’s disease, as well as in patients with Type 2 Diabetes, following GLP-1RA intervention. These authors also included a paper by Li et al. [27] in their analyses, whereas we did not include it in the present meta-analysis because it was not considered an RCT, as stated by the authors in the original article.

Chronic hyperglycemia has been consistently associated with an increased risk of cognitive impairment and suboptimal glycemic control may be associated with modest decrements of cognitive performance [28]. Although both Type 2 diabetes trials included in our paper showed comparable improvements in glycemic control over time in both the GLP-1RA and control groups, participants receiving GLP-1RA in the paper by Cheng et al. were the only group with available data highlighting improvements in delayed memory, attention and executive function, while no outcomes were found for global cognition even after adjustment for glycemic control [23]. Another report by Luchsinger et al. [22] did not find any significant associations between time-weighted HbA1c and cognitive outcomes over a 4.3-year follow-up period. Unfortunately, we were unable to compare these studies due to the lack of data regarding cognitive test scores.

It is widely recognized that determinants of low-grade inflammation, including overnutrition, obesity, and aging, contribute to increased cardiovascular risk and cognitive decline [5,6,29]. Given that GLP-1RA interventions reduce body weight and improve metabolic pathways involved in glucose homeostasis, oxidative stress, and inflammation, GLP-1RA(s) may improve cardiovascular risk profiles, especially in those overweight and/or with T2DM [30]. In fact, the trial by Cukierman-Yaffe et al. [24] specifically included participants with a history of cardiovascular disease. Even though the results from the sensitivity analysis for global cognition were performed after excluding the study by Cukierman-Yaffe, we did not find any significant treatment effect change (Figure 4).

GLP-1 receptors are highly concentrated in the hypothalamus, hippocampus, and cerebral cortex [31,32]. GLP-1RA(s) can cross the blood–brain barrier and preclinical studies have suggested neuroprotective mechanisms, including effects on amyloid-related pathways, tau phosphorylation, oxidative stress, synaptic function, and neuronal plasticity [7,33,34]. These biological observations provide mechanistic plausibility but should not be interpreted as evidence of a clinically meaningful cognitive benefit.

In the revised quantitative analysis, the favorable delayed-memory estimate derives from a single small trial [23] and therefore requires confirmation in larger, specifically designed randomized trials. The revised domain-specific analysis does not support a consistent treatment effect across cognitive domains. Attention was pooled from only two compatible studies and had a very wide confidence interval, while executive and verbal outcomes were represented by single studies. Individual trials have reported signals in selected cognitive measures at follow-up times, but these findings remain heterogeneous and should be interpreted descriptively rather than as evidence of a reproducible domain-specific benefit.

As previously mentioned, Type 2 diabetes is associated with an increased risk of AD and is frequently observed in individuals who progress from MCI to AD. However, we identified and included only two small RCTs in MCI in our analysis. Mullins et al. [12] evaluated exenatide administered at 5 mcg twice daily, titrated to 10 mcg twice daily, over 18 months, while Dei Cas et al. [13] investigated a long-acting formulation of exenatide at 2 mg once weekly for 8 months. Given heterogeneity in study populations, pharmacological formulations, dosing regimens, and intervention periods, current evidence remains insufficient to draw substantiated conclusions about the efficacy of GLP-1RA(s) in patients with MCI. Future trials are needed to test a potential role of GLP-1RA(s) on specific cognitive domains in MCI with and without Type 2 Diabetes.

In this paper, we specifically aimed to investigate cognitive domain outcomes among individuals at risk of AD. Given the potential of GLP-1RA(s) to improve cognitive performance, as suggested by previous preclinical and epidemiological evidence [9,10], future RCTs specifically designed to assess their effects on different cognitive domains in well-defined at-risk populations may be warranted.

Unfortunately, different RCTs testing the impact of GLP-1RA use on improving cognitive abilities or stabilizing cognitive performance in persons with early AD have not shown positive findings [35]. A recent report regarding findings from two large multicenter and randomized trials, the Evoke and Evoke+ trials, included adults aged between 55 and 85 years with mild dementia due to an AD diagnosis (semaglutide, n = 928; placebo, n = 927; and semaglutide, n = 976; placebo, n = 977, respectively) [35]. Both trials tested once daily oral semaglutide 14 mg or placebo for 156 weeks [35]. The authors did not find that semaglutide was efficacious in slowing clinical progression, as measured by cognitive and functional changes on the Clinical Dementia Rating—sum of boxes (CDR-SB) in early AD. The authors found no differences in global cognition based on MoCA or MMSE results, which also confirms our findings from previous trials in those at risk of AD. This may be explained by the fact that neurodegeneration process has already begun, and the potential neuroprotective effect of GLP-1RA(s) is most likely limited due to the presence of neuronal loss and synaptic degeneration [36,37]. Indeed, GLP-1RA intervention might have potential for those individuals at risk of AD who are asymptomatic. Considering the overall safety profile, specifically designed RCTs may be warranted.

Another recent multicenter RCT, the ELAD study, evaluated the use of liraglutide in adults with mild to moderate AD with no diabetes and did not find any significant differences in cerebral glucose metabolism after 12 months; however, they found a slight improvement in executive function tests [38].

The principal limitations are the small number of analytically compatible trials within each cognitive domain, differences in study populations, GLP-1RA agents, comparators, follow-up duration, and neuropsychological instruments, and the disproportionate influence of one large trial on global cognition. With only two or three studies in pooled analyses, estimates of between-study heterogeneity are imprecise; therefore, tau2 and I2 should be interpreted cautiously. Formal funnel-plot asymmetry testing was not performed because fewer than 10 independent studies contributed to each domain. The sensitivity analysis excluding the dominant Cukierman trial [24] produced a highly imprecise estimate and should not be interpreted as confirming robustness. An additional limitation is that no single cognitive domain, instrument, or assessment time point was prespecified as the primary cognitive outcome in the registered protocol. Accordingly, the domain-specific analyses should be regarded as exploratory and may be subject to multiplicity and selective outcome reporting. These findings should therefore be interpreted cautiously and considered hypothesis-generating.

5. Conclusions

Current randomized evidence is insufficient to establish a consistent beneficial effect of GLP-1RA(s) on cognitive function. Global cognition showed no evidence of a treatment effect, while the pooled attention estimate was highly imprecise. Favorable findings for delayed memory and executive function arose from a single small, randomized trial [23] and should therefore be considered exploratory and hypothesis-generating. The evidence is limited by the small number of analytically compatible trials, heterogeneous populations, interventions and cognitive measures, and risk-of-bias concerns. Larger randomized controlled trials specifically designed to assess cognition, with standardized and prespecified cognitive outcomes and adequate follow-up, are required to determine whether GLP-1RA(s) hold potential in adults at risk of developing AD.

References

1.Dementia Available online: https://www.who.int/news-room/fact-sheets/detail/dementia(accessed on 15 January 2026)

2.Wang Y. Li C. Liang J. Gao D. Pan Y. Zhang W. Zhang Y. Zheng F. Xie W. Onset age of diabetes and incident dementia: A prospective cohort study J. Affect. Disord. 2023 329 493 499 10.1016/j.jad.2023.02.138

3.Abbatecola A.M. Rizzo M.R. Barbieri M. Grella R. Arciello A. Laieta M.T. Acampora R. Passariello N. Cacciapuoti F. Paolisso G. Postprandial plasma glucose excursions and cognitive functioning in aged type 2 diabetics Neurology 2006 67 235 240 10.1212/01.wnl.0000224760.22802.e8

4.Arnold S.E. Arvanitakis Z. Macauley-Rambach S.L. Koenig A.M. Wang H.Y. Ahima R.S. Craft S. Gandy S. Buettner C. Stoeckel L.E. Brain insulin resistance in type 2 diabetes and Alzheimer disease: Concepts and conundrums Nat. Rev. Neurol. 2018 14 168 181 10.1038/nrneurol.2017.185

5.Abbatecola A.M. Paolisso G. Lamponi M. Bandinelli S. Lauretani F. Launer L. Ferrucci L. Insulin resistance and executive dysfunction in older persons J. Am. Geriatr. Soc. 2004 52 1713 1718 10.1111/j.1532-5415.2004.52466.x

6.Geroldi C. Frisoni G.B. Paolisso G. Bandinelli S. Lamponi M. Abbatecola A.M. Zanetti O. Guralnik J.M. Ferrucci L. Insulin resistance in cognitive impairment: The InCHIANTI study Arch. Neurol. 2005 62 1067 1072 10.1001/archneur.62.7.1067

7.Mullins R.J. Diehl T.C. Chia C.W. Kapogiannis D. Insulin Resistance as a Link between Amyloid-Beta and Tau Pathologies in Alzheimer’s Disease Front. Aging Neurosci. 2017 3 11810.3389/fnagi.2017.00118

8.Abbatecola A.M. Lattanzio F. Molinari A.M. Cioffi M. Mansi L. Rambaldi P. DiCioccio L. Cacciapuoti F. Canonico R. Paolisso G. Rosiglitazone and cognitive stability in older individuals with type 2 diabetes and mild cognitive impairment Diabetes Care 2010 33 1706 1711 10.2337/dc09-2030

9.Katsenos A.P. Davri A.S. Simos Y.V. Nikas I.P. Bekiari C. Paschou S.A. Peschos D. Konitsiotis S. Vezyraki P. Tsamis K.I. New treatment approaches for Alzheimer’s disease: Preclinical studies and clinical trials centered on antidiabetic drugs Expert Opin. Investig. Drugs 2022 31 105 123 10.1080/13543784.2022.2022122

10.Kuate Defo A. Bakula V. Pisaturo A. Labos C. Wing S.S. Daskalopoulou S.S. Diabetes, antidiabetic medications and risk of dementia: A systematic umbrella review and meta-analysis Diabetes Obes. Metab. 2024 26 441 462 37869901 10.1111/dom.15331

11.Gejl M. Gjedde A. Egefjord L. Møller A. Hansen S.B. Vang K. Rodell A. Brændgaard H. Gottrup H. Schacht A. In Alzheimer’s Disease, 6-Month Treatment with GLP-1 Analog Prevents Decline of Brain Glucose Metabolism: Randomized, Placebo-Controlled, Double-Blind Clinical Trial Front. Aging Neurosci. 2016 8 10827252647 10.3389/fnagi.2016.00108 PMC4877513

12.Mullins R.J. Mustapic M. Chia C.W. Carlson O. Gulyani S. Tran J. Li Y. Mattson M.P. Resnick S. Egan J.M. A Pilot Study of Exenatide Actions in Alzheimer’s Disease Curr. Alzheimer Res. 2019 16 741 752 31518224 10.2174/1567205016666190913155950 PMC7476877

13.Dei Cas A. Micheli M.M. Aldigeri R. Gardini S. Ferrari-Pellegrini F. Perini M. Messa G. Antonini M. Spigoni V. Cinquegrani G. Long-acting exenatide does not prevent cognitive decline in mild cognitive impairment: A proof-of-concept clinical trial J. Endocrinol. Investig. 2024 47 2339 2349 38565814 10.1007/s40618-024-02320-7 PMC11368991

14.Kopp K.O. Glotfelty E.J. Li Y. Greig N.H. Glucagon-like peptide-1 (GLP-1) receptor agonists and neuroinflammation: Implications for neurodegenerative disease treatment Pharmacol. Res. 2022 186 106550 10.1016/j.phrs.2022.106550

15.Bray J.J. Foster-Davies H. Salem A. Hoole A.L. Obaid D.R. Halcox J.P. Stephens J.W. Glucagon-like peptide-1 receptor agonists improve biomarkers of inflammation and oxidative stress: A systematic review and meta-analysis of randomised controlled trials Diabetes Obes. Metab. 2021 23 1806 1822 10.1111/dom.14399

16.Hölscher C. Novel dual GLP-1/GIP receptor agonists show neuroprotective effects in Alzheimer’s and Parkinson’s disease models Neuropharmacology 2018 136 251 259 10.1016/j.neuropharm.2018.01.040

17.Salemme S. Lombardo F.L. Lacorte E. Sciancalepore F. Remoli G. Bacigalupo I. Piscopo P. Zamboni G. Rossini P.M. Cappa S.F. The prognosis of mild cognitive impairment: A systematic review and meta-analysis Alzheimer’s Dement. Diagn. Assess. Dis. Monit. 2025 17 e70074 Erratum in Alzheimer’s Dement. Diagn. Assess. Dis. Monit. 2025, 17, e70150. https://doi.org/10.1002/dad2.7015010.1002/dad2.70074

18.Page M.J. McKenzie J.E. Bossuyt P.M. Boutron I. Hoffmann T.C. Mulrow C.D. Shamseer L. Tetzlaff J.M. Akl E.A. Brennan S.E. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews BMJ 2021 372 n71 10.1136/bmj.n71

19.Abbatecola A.M. Migliara G. D’Agostino F. Paolisso G. Impact of GLP-1 Receptor Agonists on Cognitive Performance and Alzheimer’s Disease-Related Pathways in Type 2 Diabetes: A Meta-Analysis of Randomized Clinical Trials 2026 Available online: https://www.crd.york.ac.uk/PROSPERO/view/CRD420261301187(accessed on 5 March 2026)

20.Sterne J.A. Savović J. Page M.J. Elbers R.G. Blencowe N.S. Boutron I. Cates C.J. Cheng H.Y. Corbett M.S. Eldridge S.M. RoB 2: A revised tool for assessing risk of bias in randomised trials BMJ 2019 366 l4898 10.1136/bmj.l4898

21.Neumann I. Schünemann H. The GRADE Book (Version 1.0). GRADE Work Group 2024 Available online: https://book.gradepro.org(accessed on 15 September 2026)

22.Luchsinger J.A. Rosin S.P. Kazemi E.J. Younes N. Suratt C.E. Fattaleh B.N. Florez H.J. Gonzalez J.S. Hollander P. Hox S.H. Glucose-Lowering Medications, Glycemia, and Cognitive Outcomes: The GRADE Randomized Clinical Trial JAMA Intern. Med. 2025 185 778 787 10.1001/jamainternmed.2025.1189

23.Cheng H. Zhang Z. Zhang B. Zhang W. Wang J. Ni W. Miao Y. Liu J. Bi Y. Enhancement of Impaired Olfactory Neural Activation and Cognitive Capacity by Liraglutide, but Not Dapagliflozin or Acarbose, in Patients with Type 2 Diabetes: A 16-Week Randomized Parallel Comparative Study Diabetes Care 2022 45 1201 1210 10.2337/dc21-2064

24.Cukierman-Yaffe T. Gerstein H.C. Colhoun H.M. Diaz R. García-Pérez L.E. Lakshmanan M. Bethel A. Xavier D. Probstfield J. Riddle M.C. Effect of dulaglutide on cognitive impairment in type 2 diabetes: An exploratory analysis of the REWIND trial Lancet Neurol. 2020 19 582 590 Erratum in Lancet Neurol. 2020, 19, e9. https://doi.org/10.1016/S1474-4422(20)30303-310.1016/s1474-4422(20)30173-3

25.Biessels G.J. Despa F. Cognitive decline and dementia in diabetes mellitus: Mechanisms and clinical implications Nat. Rev. Endocrinol. 2018 14 591 604 10.1038/s41574-018-0048-7

26.Xu H. Zhang J. Wang Y. Ma G. Zhao Y.J. The Efficacy of Glucagon-like Peptide-1 Receptor Agonists on Different Cognitive Domains: A Meta-analysis of Randomized Controlled Trials Curr. Neuropharmacol. 2026 ahead of print10.2174/011570159X439787260312090015

27.Li Q. Jia M. Yan Z. Li Q. Sun F. He C. Li Y. Zhou X. Zhang H. Liu X. Activation of Glucagon-Like Peptide-1 Receptor Ameliorates Cognitive Decline in Type 2 Diabetes Mellitus Through a Metabolism-Independent Pathway J. Am. Heart Assoc. 2021 10 e020734 10.1161/JAHA.120.020734

28.Crane P.K. Walker R. Hubbard R.A. Li G. Nathan D.M. Zheng H. Haneuse S. Craft S. Montine T.J. Kahn S.E. Glucose levels and risk of dementia N. Engl. J. Med. 2013 369 540 548 Erratum in N. Engl. J. Med. 2013, 369, 147610.1056/nejmoa1215740

29.Song R. Hu M. Qin X. Qiu L. Wang P. Zhang X. Liu R. Wang X. The Roles of Lipid Metabolism in the Pathogenesis of Chronic Diseases in the Elderly Nutrients 2023 15 343337571370 10.3390/nu15153433 PMC10420821

30.Gallo G. Desideri G. Savoia C. Update on Obesity and Cardiovascular Risk: From Pathophysiology to Clinical Management Nutrients 2024 16 278110.3390/nu16162781

31.Diz-Chaves Y. Herrera-Pérez S. González-Matías L.C. Mallo F. Effects of Glucagon-like peptide 1 (GLP-1) analogs in the hippocampus Vitamins and Hormones ElsevierAmsterdam, The Netherlands2022 Volume 118 457 478 10.1016/bs.vh.2021.12.005

32.Cork S.C. Richards J.E. Holt M.K. Gribble F.M. Reimann F. Trapp S. Distribution and characterisation of Glucagon-like peptide-1 receptor expressing cells in the mouse brain Mol. Metab. 2015 4 718 731 10.1016/j.molmet.2015.07.008

33.Chen S. Sun J. Zhao G. Guo A. Chen Y. Fu R. Deng Y. Liraglutide Improves Water Maze Learning and Memory Performance While Reduces Hyperphosphorylation of Tau and Neurofilaments in APP/PS1/Tau Triple Transgenic Mice Neurochem. Res. 2017 42 2326 2335 10.1007/s11064-017-2250-8

34.Gejl M. Brock B. Egefjord L. Vang K. Rungby J. Gjedde A. Blood-Brain Glucose Transfer in Alzheimer’s disease: Effect of GLP-1 Analog Treatment Sci. Rep. 2017 7 17490 10.1038/s41598-017-17718-y

35.Cummings J.L. Atri A. Sano M. Zetterberg H. Scheltens P. Knop F.K. Johannsen P. Wichmann C.A. Abschneider R.M. Leon T. Efficacy and safety of oral semaglutide 14 mg (flexible dose) in early-stage symptomatic Alzheimer’s disease (evoke and evoke+): Two phase 3, randomised, placebo-controlled trials Lancet 2026 407 2167 2179 10.1016/S0140-6736(26)00459-9

36.Biswas P. Rahman M.H. Tabassum A. Shoyshob T.Z. Jalouli M. Islam Tareq M.M. Imtiaz M. Dona H.A. Harrath A.H. Rahman M.A. Novel Drug Targets for Neurodegenerative Diseases of Elderly People to Develop Effective Therapeutics: A Comprehensive Analysis Adv. Pharmacol. Pharm. Sci. 2025 2025 8847508 10.1155/adpp/8847508

37.Tesco G. Lomoio S. Pathophysiology of neurodegenerative diseases: An interplay among axonal transport failure, oxidative stress, and inflammation? Seminars in Immunology Academic PressCambridge, MA, USA2022 Volume 59 101628 10.1016/j.smim.2022.101628

38.Edison P. Femminella G.D. Ritchie C. Nowell J. Holmes C. Walker Z. Ridha B. Raza S. Livingston N.R. Frangou E. Liraglutide in mild to moderate Alzheimer’s disease: A phase 2b clinical trial Nat. Med. 2026 32 353 361 10.1038/s41591-025-04106-7