Metformin use is associated with a reduced risk of cognitive impairment in adults with diabetes mellitus: A systematic review and meta-analysis

Objective Controversy exists regarding the impact of metformin and whether it prevents or promotes the incidence of cognitive dysfunction. This systematic review and meta-analysis were conducted to identify the effect of metformin therapy on cognitive function in patients with diabetes. Methods Electronic databases (PubMed, EMBASE, PsycINFO, the Cochrane Library, and Web of Science) were systematically searched by two investigators from the date of inception until March 1, 2022. The study followed PRISMA guidelines. Inclusion criteria were defined according to the PECOS model. Eligible studies investigated cognitive dysfunction in metformin users compared with non-users in adults with diabetes. Only observational study designs (such as cohort, cross-section, and case-control) were included. Results A systematic search identified 1,839 articles, of which 28 (17 cohort, 8 case-control, and 3 cross-sectional studies) were included in the meta-analysis. Metformin reduced the occurrence of cognitive impairment in patients with diabetes [unadjusted hazard ratio (HR) = 0.67, 95% CI: 0.62–0.73; adjusted hazard ratio (aHR) = 0.92, 95% CI: 0.85–0.99]. In addition, the use of metformin was associated with a decreased risk of dementia (HR = 0.64, 95% CI: 0.59–0.69; aHR = 0.90, 95% CI: 0.84–0.96), while a random-effects meta-analysis indicated no significant effect of metformin on the risk of Alzheimer's disease (AD) (HR = 0.85, 95% CI: 0.60–1.22; aHR = 1.10, 95% CI: 0.95–1.28). Conclusion Metformin therapy decreased the occurrence risk of cognitive decline in patients with diabetes mellitus. Moreover, the use of metformin by adults with diabetes for the prevention of dementia, but not AD, is supported by the available evidence.


Introduction
Cognitive dysfunction, which includes delirium, mild cognitive deficits, and dementia, is characterized by a significant decline from a previously attained cognitive functional level (Sachdev et al., 2014;Zhang et al., 2020). Numerous epidemiological studies have increasingly recognized cognitive impairment as important comorbidity and complication of diabetes and it has become a major public health concern (Gispen and Biessels, 2000;Biessels and Despa, 2018;Biessels and Whitmer, 2020). A systematic review reported that patients with diabetes have a 73% increase in the risk of dementia and a 56% increase in the risk of Alzheimer's disease (AD) (Diniz Pereira et al., 2021). Moreover, the etiology of cognitive impairment in patients with diabetes is potentially multifactorial (Campbell et al., 2018;Jash et al., 2020;Yuan et al., 2021). For example, poor glycemic control and the presence of microvascular complications, such as neuropathy and retinopathy, have also been associated with cognitive dysfunction (Moheet et al., 2015); insulin resistance also may increase the occurrence risk of AD (Baker et al., 2011;Lyu et al., 2020).
Metformin is a primary oral hypoglycemic agent widely used for treating diabetes since 1950s (Flory and Lipska, 2019). Metformin functions predominantly by improving the sensitivity of insulin receptors to insulin, which enhances glucose uptake and decreases hepatic glycogen synthesis at low glucose (Hundal et al., 2000;Satoh, 2014). However, the function of metformin is not confined to glucose reduction (Liu et al., 2014). Increasing evidence has emerged indicating that metformin can penetrate the blood-brain barrier to improve cerebral energy metabolism in some regions of the brain associated with semantic memory and some white matter in adults with diabetes Sritawan et al., 2020). Moreover, an animal experiment supported the fact that metformin treatment prevents amyloid plaque deposition and reduces memory impairment (Ou et al., 2018).
Several studies have reported that metformin could negatively impact cognitive function (Hsiao et al., 2014;Ha et al., 2021). For example, Chen et al. (2009) found that the activation of AMP-activated protein kinase (AMPK) by metformin raised the production of β-secretase to promote the deposition of β-amyloid peptides (Aβ), which leads to cognitive dysfunction (Chen et al., 2009). In addition, a case-control study of patients aged 65 years or older indicated that the long-term metformin use increased the risk of AD [(OR): 1.71, 95% CI: 1.12-2.60] (Imfeld et al., 2012). Therefore, the effect of metformin and whether it is associated with the prevention or promotion of the incidence of cognitive impairment is controversial.
This meta-analysis aimed to analyze the available evidence on the use of metformin and cognitive function in adults with diabetes and ascertain the relationship between the two.

Search strategy
The databases (PubMed, EMBASE, PsycINFO, Cochrane Library, and Web of Science) were screened independently by two investigators (JHZ and YQS) from their inception date until March 1, 2022. The search strategy for the PubMed database is presented in Table 1. In addition, reference lists from identified and relevant reviews were manually searched.

Selection criteria
This systematic review and meta-analysis were conducted based on the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines (Moher et al., 2009). The protocol for this systematic review and metaanalysis was registered at INPLASY (registration number: INPLASY202250065). Studies were only selected for inclusion in accordance with the following PECOS criteria. Participants: all patients are individuals with diabetes aged 18 years or older and have no history of cognitive disorder. Exposure: taking metformin monotherapy at any dosage for any duration. Comparator: participants received other antidiabetic drugs rather than metformin or no therapy as the control group. Outcomes: studies that investigated the risk (or incidence) of cognitive dysfunction were eligible for inclusion. Study: only published observational study designs-such as cohort, case-control, or cross-sectional studies-were eligible for inclusion. In addition, detailed meeting summary information was included. Studies of randomized controlled trials (RCTs), case reports/series, basic science, and reviews were excluded.

Data extraction and quality assessment of each study
Data from all eligible studies were extracted onto a standardized Excel spreadsheet independently by two investigators (JHZ and YQS). The following data were abstracted from each included study: publication details (such as first author and year of publication), study design, number of participants, participant characteristics (mean age and age range), gender, comparator, exposure, number of events, years enrolled, and outcomes (diagnosis and diagnostic criteria). For any discrepancies, a consensus was reached via discussion between the two investigators; if any uncertainty remained Zhang et al. . /fnins. . regarding inclusion, a senior author (XYZ) was consulted. To acquire relevant missing data from included studies, the first and/or corresponding authors of the studies were contacted. Subsequently, JHZ and YQS independently evaluated the quality of each included study. The Newcastle-Ottawa Quality Assessment Scale (NOS) was used for cohort studies and casecontrol studies (Stang, 2010), while cross-sectional studies were appraised by the Agency for Healthcare Research and Quality (AHRQ) .

Data synthesis
The Review Manager software (version 5.4) was used to conduct the meta-analysis and sensitivity analysis, and publication bias was performed by STATA software (version 15.0). In accordance with the study of Jatho et al. (2021), the odds ratio (OR), relative risk (RR), or hazard ratio (HR) with a 95% confidence interval (CI) were selected as the effect size for included studies. Adjusted OR/HR/RR (accounting for confounding variables) and unadjusted OR/HR/RR were conducted. Heterogeneity was assessed using Higgins I-squared (I 2 ) (I 2 > 50% was regarded as significant heterogeneity) (Higgins et al., 2003). Publication bias was examined by Begg's funnel plot and Egger's test (Macaskill et al., 2001). Publication bias is present if Begg's funnel plot shows asymmetry or the p-value of Egger's test is less than 0.05. A sensitivity analysis was conducted by moving each study individually. All statistical significance was set at p < 0.05 (two-tailed).

Sample characteristics
The characteristics of the 28 included studies on the occurrence risk of cognitive dysfunction in diabetes with oral metformin are shown in Table 2. The 28 studies comprised seventeen cohort studies (Hsu et al., 2011;Moore et al., 2013;Whitmer et al., 2013;Cheng et al., 2014;Naharci et al., 2016;Orkaby et al., 2017;Kim et al., 2019;Koo et al., 2019;Porter . /fnins. .  (Moore et al., 2013;Naharci et al., 2016) for which the enrollment periods were not reported. Sample sizes varied from 278,290  to 193 (Naharci et al., 2016). Additional details on the covariates that were adjusted for in the statistical analyses are included in Supplementary Table 1. . /fnins.
.      (a) Results of quality assessment using the Newcastle-Ottawa Scale for cohort studies.

Study
Representativeness of the exposed cohort Selection of non-exposed cohort

Quality assessment
As shown in Table 3, the NOS score for all the cohort studies and case-control studies ranged from 7 to 9 points. The AHRQ score for each included cross-sectional study ranged from 8 to 10 points.

Sensitivity analysis and publication bias
Sensitivity analyses were performed to investigate the influence of each individual study on the overall meta-analysis summary estimate and the validity of the effect size. Excluding the included studies one by one demonstrated that no single study had a significant impact on the outcome of the combined analysis, suggesting that the results of this meta-analysis were stable ( Figure 5). Egger's test and Begg's funnel plot did not find evidence of publication bias (p < 0.05) (Figure 6).

Discussion
The aim of this systematic review and meta-analysis was to evaluate the impact of metformin on cognitive impairment in adults diagnosed with diabetes. A total of 28 observational studies met the inclusion criteria, and 24 studies were deemed appropriate in the pooled analysis for this systematic review and meta-analysis, the main findings of which included: (1) a meta-analysis reported that the protective effect of metformin therapy decreases the risk of cognitive dysfunction in patients with diabetes; (2) subgroup analyses found that oral metformin was associated with a decreased risk of dementia in patients with diabetes; (3) a subgroup analysis of meta-analysis on metformin and AD found that metformin could be associated with no significant effect on the decreased risk of AD.
Congruent with the findings that oral metformin was associated with a lower prevalence of dementia in the current study, there is some research suggesting that metformin initiation is associated with a substantially lower risk of dementia among younger African American patients (Scherrer et al., 2019a). However, the results from Salas et al. (2020) did not support initiating metformin earlier to prevent cognitive decline. The discrepancies in the results of these studies might .
/fnins. . be attributed to patient populations that differed in clinical and demographic characteristics and treatment timing. Therefore, RCTs with an optimal sample size need to be performed on the use of metformin for diabetes to confirm and extend these findings. A subgroup analysis in the current study, based on observational studies, indicated that metformin was not significantly associated with a decreased risk of AD. In contrast, Ha et al. (2021) reported that metformin use was related to an increased risk of AD after adjusting for comorbidities and cardiometabolic risk profile by multivariable regression analyses. In addition, the treatment of diabetes with metformin cumulatively for more than 4 years significantly increased the risk of developing AD (Hsiao et al., 2014). However, Sluggett et al. (2020) showed that long-term (≥10 years) and high-dose metformin therapy had a lower risk of incidence of AD in older people with diabetes. Taken together, these results show that future RCTs with a larger sample size focusing on AD and metformin use in adults diagnosed with diabetes are warranted to explain these mixed findings.
The potential mechanisms of the relationship between metformin and cognitive performance have yet to be elucidated. Previous studies in animals have indicated that metformin could reduce cognitive impairment by reversing the harmful effects of impaired insulin signaling that causes a cascade of deleterious events, such as oxidative stress, inflammation, and tau hyper-phosphorylation (Farr et al., 2019;Gorgich et al., 2021). Moreover, the current study provides primary evidence suggesting that adults with diabetes facing a high risk of cognitive dysfunction should consider metformin as a firstline therapy.
Heterogeneity was detected in the meta-analysis of the present study. This variation could potentially be related to differences in the cumulative dose and duration of metformin, the race of the participants, and the duration and severity of diabetes. In addition, the uncertain accuracy of AD diagnoses in administrative data should be considered.
. /fnins. . Studies were included that used reliable neuropsychological cognitive assessment tools (i.e., Repeatable Battery for the Assessment of Neuropsychological Status and Frontal Assessment Battery) to report cognitive impairment instead of Minimum Mental State Examination (MMSE) in metformin users. However, the current study showed that the effect of metformin on the incidence of cognitive impairment remained effective when data were adjusted for potential confounding factors.

Limitations
There were three main limitations in this meta-analysis. First, the findings of this review provided only very weak support for the hypothesis that metformin could prevent cognitive impairment in people without diabetes. However, the subgroup analysis of these factors could not be conducted due to the limited amount of data. Second, a subgroup analysis of studies adequately controlled for diabetes severity at baseline could not .
be conducted due to the limited amount of data. Finally, studies with vs. without an active comparator were not performed in the subgroup analysis in this meta-analysis.

Conclusion
Metformin reduces the incidence of cognitive impairment but not AD in adults with diabetes. Future trials should examine the role of metformin in patients with diabetes in an RCT with a larger sample size, well-controlled confounding factors, sufficient follow-up time, and more accurate assessment of metformin exposure levels.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author/s.