Antidiabetic Drugs and Their Links to Mild Memory Problems and Alzheimer's Disease: A Review of Metformin, Pioglitazone, and GLP-1 Treatments

Type 2 diabetes mellitus (T2DM) is consistently associated with increased risk of developing AD and MCI. Epidemiological studies have reported higher rates of cognitive impairment and dementia in individuals with T2DM [11,12,13]. The risk of developing AD increases by approximately 1.5-fold (RR ≈ 1.56) in diabetic patients compared to non-diabetic populations [14]. This association likely reflects shared metabolic and vascular processes linking T2DM with AD and MCI. Insulin resistance and chronic metabolic stress can impair neuronal function. Diabetes-related microvascular disease also heightens vulnerability to vascular and mixed forms of dementia [9]. In addition, recurrent hypoglycemic episodes in individuals with diabetes may further contribute to neurodegeneration and promote the development of AD [15,16]. Neuroimaging and neuropathological studies demonstrate cerebrovascular and neurodegenerative changes in diabetic populations [17].

Although a definitive causal pathway has not been established, the consistent relationship between metabolic dysfunction and cognitive decline supports targeting metabolic pathways to reduce risk or slow progression in AD and MCI.

Metformin is a first-line therapy for T2DM that lowers blood glucose by suppressing hepatic gluconeogenesis and improving peripheral insulin sensitivity. This is mainly achieved by activating adenosine monophosphate-activated protein kinase (AMPK), a central regulator of cellular energy balance [18]. Activation of AMPK reduces lipogenesis and enhances insulin sensitivity in hepatic and skeletal muscle tissues. Metformin also inhibits mitochondrial complex I, increasing the AMP:ATP ratio and activating AMPK-dependent signalling while exerting AMPK-independent effects on gluconeogenic and redox pathways [18,19]. These mechanisms are relevant to AD and MCI, as impaired insulin signalling, mitochondrial dysfunction, oxidative stress, and chronic low-grade inflammation contribute to neuronal injury and synaptic failure [20].

In addition to glycemic control, metformin modulates mitochondrial function by partially inhibiting mitochondrial complex I. It also influences mitochondrial respiration and biogenesis through both AMPK-dependent and AMPK-independent pathways [19,21]. Metformin exerts anti-inflammatory effects by activating AMPK and inhibiting the mammalian target of rapamycin (mTOR), which affects mitochondrial activity and autophagy in immune cells [21]. These mechanisms help explain why the impact of metformin may vary across the course of Alzheimer’s disease. Metformin may be more effective in individuals with insulin resistance or T2DM during the early stages of AD, when metabolic dysfunction is a driving factor in neurodegeneration. In contrast, in non-diabetic patients or those with later-stage disease of AD, even if relevant pathways are engaged, clinical benefit may be limited [20]. These actions provide a mechanistic rationale for metformin’s potential relevance in neurodegenerative disease (Figure 1A).

Pioglitazone, a thiazolidinedione insulin sensitizer, acts as an agonist of peroxisome proliferator-activated receptor gamma (PPARγ), a nuclear receptor regulating metabolic and inflammatory gene expression relevant to Alzheimer’s disease pathology. Activation of PPARγ attenuates neuroinflammation by suppressing pro-inflammatory enzymes such as cyclooxygenase-2 and inducible nitric oxide synthase, reducing microglial and astrocytic activation, and lowering cytokines, including interleukin-1β, in experimental models [22,23].

PPARγ activation also regulates genes involved in glucose metabolism, lipid handling and mitochondrial function, linking peripheral insulin sensitization to pathways relevant to neuronal energy balance and inflammatory signalling in AD [23]. This provides a mechanistic basis for the metabolic-status dependency observed in clinical studies, as these effects are more biologically relevant in patients with baseline insulin resistance or T2DM, in whom metabolic dysfunction directly contributes to neuronal stress and cognitive decline [22,23].

In addition to its immunomodulatory effects, pioglitazone influences amyloid processing. Experimental data show that PPARγ activation downregulates β-secretase 1, therefore reducing production and accumulation of amyloid-β, particularly the Aβ1–42 species, in transgenic Alzheimer’s models [24,25]. However, much of this evidence is derived from preclinical systems, and it is still unclear whether therapeutic concentrations in humans are sufficient to reproduce these central effects to the same extent. This may explain why clinical benefits appear more evident in metabolically impaired populations, whereas in non-diabetic patients or later-stage disease, pathway engagement may occur without sufficient impact to modify established neurodegeneration or amyloid burden [22,23,24,25].

The combined anti-inflammatory and amyloid-related effects of pioglitazone provide a biological rationale for its investigation in Alzheimer’s disease and mild cognitive impairment (Figure 1B).

Glucagon-like peptide-1 receptor agonists (GLP-1 RAs) exert neuroprotective effects by engaging G-protein-coupled receptors (GLP-1R) expressed in key brain regions involved in cognition, including the hippocampus and cortex [26,27]. Upon binding, GLP-1 RAs activate intracellular signalling pathways, especially the cyclic adenosine monophosphate (cAMP)/protein kinase A (PKA) and phosphoinositide 3-kinase (PI3K)/Akt axes, which promote neuronal survival, enhance synaptic plasticity, and support synaptic integrity [26,27].

The metabolic-status dependency observed in clinical studies may reflect the interaction between GLP-1 receptor signalling and impaired insulin pathways in the AD brain. In insulin-resistant states, GLP-1 receptor activation engages downstream signalling pathways linked to neuronal survival and synaptic function, which may remain more responsive in metabolically dysregulated conditions [26,27]. Preclinical evidence supports this, with liraglutide shown to reverse memory impairment, reduce synaptic loss, and decrease amyloid plaque burden in AD models [28].

Regarding central target engagement, GLP-1 RAs such as liraglutide have been shown to cross the blood–brain barrier, although cerebrospinal fluid concentrations in humans are low, reported at approximately 0.02% of plasma levels [27]. Despite this, functional engagement is supported by neuroimaging evidence, with clinical studies demonstrating preservation of cerebral glucose metabolism and improved blood–brain glucose transport in patients with AD [27].

In addition, GLP-1 RAs have been associated with reduced neuroinflammation, oxidative stress, and apoptosis in experimental models of neurodegeneration and cerebral ischaemia [26,27,29]. These effects, alongside evidence of enhanced neuronal survival and synaptic preservation, suggest that GLP-1 RAs act on multiple pathways relevant to neurodegeneration. Their effects are therefore better understood as modulation of neuro-metabolic and neurovascular dysfunction rather than as direct alteration of core AD pathology [26,27,28,29,30].

In addition to the drug classes discussed above, other antidiabetic therapies have also been explored in the context of Alzheimer’s disease. Amylin, a pancreatic peptide co-secreted with insulin, has been implicated in Alzheimer’s disease due to its structural similarity to amyloid-β and its potential role in amyloid aggregation and clearance [31]. Amylin analogs, such as pramlintide, have demonstrated neuroprotective and anti-amyloid effects in preclinical studies, although clinical evidence remains limited. Other approaches, including insulin-based interventions and newer metabolic agents, have also been investigated, but findings remain heterogeneous. Against this broader background, the present review focuses on metformin, pioglitazone, and GLP-1 receptor agonists, which currently have the most developed clinical evidence in MCI and AD.

Eligibility was defined using the Population, Intervention, Comparator, Outcomes, and Study design (PICOS) framework (Table 1). Searches used controlled vocabulary and keywords for AD/MCI, antidiabetic medications, and cognitive/neurodegenerative outcomes. Searches were limited to human studies published in English from 2000s to the present. Database-specific strategies were applied.

Search string can be found in. Supplementary Table S1

All records were imported into Covidence for duplicate removal. Titles and abstracts were independently screened by two reviewers, followed by independent full-text assessment. Disagreements were resolved through discussion and consensus (Figure 2).

Two reviewers independently extracted data using a standardized form, including study design, setting, sample size, participant demographics, diagnostic criteria, intervention details (agent, dose, duration), comparator, outcome measures and follow-up duration. Discrepancies were resolved by discussion.

Two reviewers independently assessed risk of bias. Randomized trials were evaluated using the Cochrane Risk of Bias 2.0 tool, and non-randomized studies were assessed using ROBINS-I (). Disagreements were resolved through discussion. Supplementary Table S2

Given substantial heterogeneity in study designs, participant populations, and particularly in reported outcome measures, quantitative pooling was not appropriate. After consultation with a data analyst, it was determined that a meta-analysis was not feasible due to the lack of comparable cognitive and biomarker endpoints across studies. Findings were therefore synthesized narratively and organized by intervention class (metformin, pioglitazone, GLP-1 receptor agonists).

Following the influential 2013 review by the Senior group [43], which summarized the early reasons for targeting metabolism in neurodegenerative disease, the treatment field has advanced. It has moved from possible connections to clinical trials confirmed by biological markers. In 2013, the focus was mostly on testing widely used drugs like metformin and thiazolidinediones in large patient groups. Since then, studies from 2013 to 2026 have shown a major shift toward tracking how these drugs affect the brain [32,33,37].

A notable advancement in the field is the emergence of GLP-1 receptor agonists. While this drug class was in its early stages in 2013, recent randomized trials have yielded new findings; specifically, these studies demonstrate its capacity to restore maximal blood-brain glucose transfer and prevent regional declines in cerebral glucose metabolism [32,33,40].

Recent research has advanced the use of high-resolution neuroimaging, proteomics, and neuronal-derived extracellular vesicles. These tools help monitor target engagement within the central nervous system [35,37]. Moreover, this era is characterized by a shift toward metabolic stratification, as researchers now acknowledge that drug efficacy depends greatly on a participant’s baseline insulin resistance and specific disease stage [36,42]. Over the past thirteen years, it has become clear that metabolic pathways in the human brain are now modifiable. However, the gap between restored glucose handling and immediate cognitive improvement remains a significant challenge for the field [41].

Overall, metformin showed the most consistent cognitive effects among the therapies reviewed, although the improvements were modest and largely confined to specific domains. Improvements were seen primarily in episodic memory and executive function [34,35], particularly in early-stage disease and in individuals with insulin resistance or T2DM [22,34]. Rather than producing broad cognitive change, the effects appeared selective, emerging most clearly in recall-based tasks that rely on hippocampal integrity.

Episodic memory decline reflects early hippocampal vulnerability in prodromal Alzheimer’s disease. The finding that total recall improved in amnestic MCI [34] suggests that metformin may exert benefit at a stage when neural networks remain structurally intact but are metabolically strained. In that setting, improving metabolic efficiency may enhance synaptic performance without shifting overall clinical classification. However, the lack of consistent change at global scales, such as the ADAS-Cog, suggests that these effects are subtle. Moreover, repeated cognitive testing can produce practice-related improvements even in placebo groups, potentially masking small treatment effects [34].

Findings from observational cohorts further support this pattern. In the ADNI cohort, metformin use among individuals with MCI and T2DM was associated with better composite memory performance, relative preservation of hippocampal volume, and more favorable CSF profiles [36,37]. While these associations cannot establish causality and may reflect residual confounding, the convergence of structural, biomarker, and cognitive findings strengthens the case that metformin may mitigate metabolically driven vulnerability in early disease.

Imaging and biomarker studies offer additional support. In randomized crossover trials, metformin was linked to increased orbitofrontal cerebral blood flow [35], providing a plausible neural correlate for improved executive performance. Proteomic analyses showed modulation of inflammatory and apoptosis-related pathways in plasma and CSF [37], indicating central biological engagement even when amyloid-β and tau markers remained unchanged over short follow-up [35,37]. Together, these findings suggest that metformin acts primarily through metabolic and inflammatory pathways that support neuronal resilience, rather than directly reducing amyloid deposition.

Importantly, the impact of metformin appears to vary with disease severity. Benefits were more apparent in MCI and early AD [34,35,36]. In more advanced disease, effects were attenuated and largely limited to isolated executive measures [35]. Once neuronal loss becomes substantial, metabolic support alone is unlikely to reverse impairment. This stage-dependent pattern suggests that metformin is unlikely to function as a universal cognitive enhancer but may have relevance as an early adjunctive strategy in metabolically vulnerable populations.

Lastly, it is important to note that the existing studies have clear limitations. The RCTs included are small and brief, limiting power to detect sustained effects [34,35]. Furthermore, heterogeneity in cognitive endpoints complicates comparison across studies, and observational findings are vulnerable to confounding by baseline metabolic control and comorbidity burden [36]. Most trials have also enrolled relatively selected participants, restricting generalizability.

Even with these caveats, a coherent picture emerges. Metformin’s potential appears to be strongest in early-stage disease with demonstrable metabolic dysfunction. The effects are domain-specific and biologically supported by converging imaging and inflammatory data, even in the absence of short-term changes in core Alzheimer’s biomarkers. Larger and longer trials, with prespecified metabolic stratification and sensitive domain-level outcomes, are needed to determine whether these early improvements translate into durable slowing of disease progression.

Pioglitazone’s cognitive effects were less consistent than those observed with metformin and seemed to depend heavily on the presence and severity of metabolic dysfunction. Across the included trials, the most convincing evidence of benefit was limited to individuals with mild Alzheimer’s disease and established T2DM [38]. In this subgroup, improvements in global cognition and episodic memory were accompanied by reductions in fasting insulin and in Homeostatic Model Assessment of Insulin Resistance (HOMA-R), along with increased parietal cerebral perfusion. The fact that cognitive, metabolic, and imaging changes moved in the same direction strengthens the impression that, in the context of clear metabolic dysregulation, insulin sensitization may meaningfully influence brain function. However, this finding should be interpreted cautiously, as Sato et al. was also the only pioglitazone study to report cognitive benefit and used an open-label design, making it more vulnerable to performance and observer bias.

However, this signal did not generalize beyond diabetic cohorts. In non-diabetic patients with mild-to-moderate AD, prolonged treatment did not result in meaningful improvements across cognitive scales, daily functioning, or global status [39]. Similarly, in individuals with MCI and insulin resistance but without diabetes, pioglitazone improved clamp-derived insulin sensitivity, yet these physiological improvements were not accompanied by measurable cognitive progress [39]. This suggests that peripheral insulin sensitization, independently, may not be sufficient to influence cognitive changes unless metabolic dysfunction reaches a threshold that directly contributes to disease expression. Importantly, the two blinded placebo-controlled trials, Geldmacher et al. [39] and Hildreth et al. [42], did not replicate the benefit reported by Sato et al. [38], which further weakens the case for a consistent cognitive effect.

Neuroimaging and biomarker findings reinforce this interpretation. In the mild AD diabetic trial, improved parietal cerebral blood flow and stabilization of plasma Aβ40/Aβ42 ratios were observed with treatment [38], suggesting biological engagement beyond peripheral glycemic correction. In contrast, comparable effects were not observed in non-diabetic cohorts [39,42], in which neither structural imaging nor Alzheimer-specific biomarkers showed consistent changes. Taken together, these findings imply that pioglitazone may influence cerebral perfusion and amyloid-related markers primarily in metabolically compromised individuals.

Disease stage further contextualizes these results. While some improvement was observed in mild AD with T2DM [38], no meaningful benefit was detected in more advanced disease [39]. At later stages, when synaptic loss and neuronal degeneration are more extensive, restoring insulin sensitivity may not reverse established structural damage. Conversely, in early MCI without overt diabetes, metabolic modulation alone may be insufficient to alter cognition over relatively short follow-up periods [42]. This pattern suggests that both the presence of significant metabolic dysfunction and the degree of neurodegeneration shape therapeutic responsiveness.

Several limitations constrain interpretation. All included studies were relatively small, and the most favorable findings arose from an open-label design [38], raising concerns regarding bias. Randomized, placebo-controlled trials in non-diabetic populations did not replicate these benefits [39,42]. Moreover, metabolic heterogeneity across cohorts complicates comparisons. Insulin resistance was more pronounced in diabetic AD cohorts than in MCI populations with milder metabolic disturbance [42], which may explain differential responsiveness.

Based on the current evidence, pioglitazone cannot be viewed as a broadly effective disease-modifying therapy for Alzheimer’s disease or MCI. Rather, its potential appears more limited and context-dependent. The available data point toward a possible role in early-stage Alzheimer’s disease when substantial metabolic dysfunction, particularly insulin resistance or T2DM, is present.

Future trials should reflect this narrower therapeutic window. Rather than enrolling broadly defined AD populations, studies should focus on patients with early-stage disease and clear metabolic impairment. Moreover, longer follow-up periods, combined with serial imaging and biomarker assessments, would help determine whether changes in perfusion or metabolic parameters translate into durable clinical stability. Without this level of clarity and precision in trial design, any real benefit in metabolically defined subgroups is likely to be diluted within heterogeneous samples and ultimately overlooked.

GLP-1 receptor agonists demonstrated the most robust central metabolic engagement, but the least convincing cognitive translation. Across trials, liraglutide and exenatide preserved cerebral glucose metabolism and improved measures of blood–brain glucose transport [32,33]. Baseline glucose transport correlated with disease duration and cognition [33], suggesting that impaired glucose handling may contribute to progression rather than simply reflect downstream neurodegeneration.

Despite this physiological engagement, cognitive outcomes remained largely unchanged [32,40,41]. In the 26-week liraglutide trial in mild to moderate AD, placebo-treated participants exhibited the expected regional decline in cerebral glucose metabolism, whereas this decline was not observed with treatment; however, global cognitive measures did not differ between groups [32]. Similarly, in the 18-month exenatide trial in early AD, no significant between-group differences were observed on ADAS-Cog or other neuropsychological outcomes, and both groups progressed in line with the natural course of the disease [40]. Although reductions in neuronal-derived extracellular vesicle Aβ42 were observed [40], these biomarker changes did not translate into measurable cognitive or structural benefit.

When evaluated earlier in the disease continuum, results remained neutral. In MCI, long-acting exenatide failed to improve cognition over 32 weeks [41]. ADAS-Cog and other global measures did not shift meaningfully. A treatment-by-sex interaction suggested possible worsening ADAS-Cog trajectories among women receiving exenatide; although the sample size was small and the study was open-label, these findings highlight potential heterogeneity in response and underscore the need for adequately powered, blinded trials [41].

The consistent dissociation between preserved glucose metabolism and unchanged cognition is notable. Restoration of cerebral glucose transport and metabolic activity suggests that neuronal energy utilization remains pharmacologically modifiable even in symptomatic disease. However, these changes may stabilize function rather than reverse established synaptic loss or structural damage. The absence of change in amyloid PET signal and CSF Aβ42, total tau, and phosphorylated tau over six to eighteen months [32,33,40] further suggests that metabolic engagement does not directly modify core pathological aggregates within the timeframes studied.

Disease stage likely influences this pattern. Intervention during symptomatic AD may occur after substantial synaptic and neuronal loss has already taken place, limiting the capacity for functional recovery. Even in MCI, short- to intermediate-duration trials failed to demonstrate cognitive improvement [40,41], raising the possibility that metabolic stabilization requires longer exposure to influence clinical outcomes or that intervention must occur at an even earlier, preclinical stage.

Future trials of GLP-1 receptor agonists should shift earlier in the disease course. Enrolling biomarker-defined preclinical or very early MCI populations, with follow-up extending beyond one to two years, may be necessary to detect meaningful effects. Stratifying participants by the presence and severity of metabolic impairment could also clarify whether those with insulin resistance are more likely to benefit.

Safety and tolerability are important when considering the use of antidiabetic therapies in older adults with cognitive impairment. Across the included studies, metformin was generally well tolerated and not associated with serious adverse events. The main issue was gastrointestinal side effects, such as nausea and poor tolerance, which led some participants to reduce the dose or stop treatment altogether [34,35,37]. Pioglitazone also appeared relatively safe, but peripheral edema was reported more frequently, consistent with its known fluid-retention effect [38,39,42]. While this was usually mild, it does raise concerns in older patients, especially those having underlying cardiovascular disease. GLP-1 receptor agonists were similarly well tolerated overall, but gastrointestinal symptoms, including nausea and reduced appetite, were common and sometimes led to discontinuation [32,33,40,41].

From a clinical perspective, these drugs appear reasonably safe, but tolerability may become a limiting factor in this population. Many patients with cognitive impairment are frail, on multiple medications, and more sensitive to side effects. As a result, even mild adverse effects could affect adherence. Consequently, although these therapies are promising, their use in Alzheimer’s disease will demand rigorous patient selection, gradual dose titration, and vigilant long-term monitoring to optimize outcomes and minimize risk.

The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author.