Association Between Use of Sodium-Glucose Co-Transporter-2 (SGLT2) Inhibitors and Cognitive Function in a Longitudinal Study of Patients with Type 2 Diabetes

Abstract

Background:

The association between sodium-glucose cotransporter-2 inhibitors (SGLT2i) use and cognitive function in type 2 diabetes remains unclear.

Objective:

Explore the association between SGLT2i and longitudinal changes in cognitive function in adults with type 2 diabetes (T2DM) and assessed the cognitive domains which were impacted by SGLT2i.

Methods:

We conducted a prospective cohort study of 476 patients aged 60.6±7.4 years with follow-up period up to 6.4 years. Data on SGLT2i use was derived from questionnaire and verified with clinical database. We used Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) to assess cognition. The association between SGLT2i use and rate of RBANS score change was examined using multiple linear regression.

Results:

There were 138 patients (29.0%) on SGLT2i, including 84 (17.7%) for < 3 years and 54 (11.3%) for ≥3 years. SGLT2i use was positively associated with RBANS total score increase in language (coefficient 0.60; 95% CI 0.10–1.11; p = 0.019) in unadjusted analysis. This positive association persisted in fully adjusted model (coefficient 0.74; 95% CI 0.12 to 1.36; p = 0.019). SGLT2i use for ≥3 years was positively associated with RBANS score increase globally and in language domain in fully adjusted analysis with coefficients 0.54 (95% CI 0.13 to 0.95; p = 0.010) and 1.12 (95% CI 0.27 to 1.97; p = 0.010) respectively.

Conclusion:

Our findings revealed a previously unobserved association between ≥3 years SGLT2i use and improved cognitive scores globally and in language domain and executive function. Future studies should investigate the role of SGLT2i in ameliorating cognitive decline.

Keywords

Cognitive function sodium-glucose co-transporter-2 inhibitors type 2 diabetes

INTRODUCTION

Diabetes mellitus (DM) is a significant public health concern worldwide. The International Diabetes Federation reported that 10.5% of the world’s population or 537 million adults have DM today. It was projected that the number of adults with DM will increase to 12.2% or 783 million by 2045 [1]. The most common type of DM is type 2 DM (T2DM) which is characterized by hyperglycemia resulting from insulin resistance or relative insulin insufficiency [1, 2]. T2DM brings about a myriad of microvascular complications including retinopathy, nephropathy, neuropathy, and cognitive impairment [2].

According to a pooled analysis involving 2.3 mil-lion subjects from 14 studies, DM is associated with 60% higher risk of dementia [3, 4]. The pathophy-siological mechanisms underlying the association between DM and cognitive impairment include dysfunction of blood-brain barrier, increased amyloid-β (Aβ)_1 - 40/42 and p-tau level due to cerebral insulin resistance, and cerebral pathological angiogenesis [5].

In view of multiple effects of oral antidiabetic medications in lowering blood glucose levels and controlling insulin resistance, there is a strong impetus to understand the effect of these medications on risk of cognitive impairment in patients with T2DM [6]. Studies have investigated the effects of anti-diabetic medications to manage cognitive dysfunction. For example, liraglutide (glucagon-like peptide-1 (GLP-1) mimetics) and lixisenatide (GLP-1 analog) are reported to reverse loss of memory, prevent reduction of synaptic plasticity in hippocampus and decrease Aβ plaque in the cortex of mice with Alzheimer’s disease [4 , 8]. More than six years’ use of metformin was associated with lower odds of cognitive impairment in a longitudinal study of community-living older adults (odds ratio (OR) 0.27; 95% CI 0.12–0.60) [9]. Patients with DM and Alzheimer’s disease-related cognitive impairment (ADCI) treated with dipeptidyl peptidase-4 inhibitor (DPP4i) showed lower levels of Aβ and experienced slower cognitive decline than patients with ADCI not treated with DPP4i [10]. A Cochrane review in 2017, however, concluded that there was no evidence of effect of antidiabetic agent on neurocognitive function in patients with T2DM [11].

Sodium-glucose transporter-2 inhibitors (SGLT2i) are one of the newer anti-diabetic agents for T2DM treatment. SGLT2, a glucose transporter, is found in proximal convoluted renal tubules and is responsible for reabsorption of urinary glucose [12 –14]. SGLT2i inhibits glucose reabsorption in the kidney through SGLT2, thereby resulting in increased urinary excretion of glucose [12, 13]. Research has demonstrated the beneficial effect of dapagliflozin, a SGLT2i, in reducing body weight and improving glycemic control [12 , 15–17]. Interestingly, studies recently demonstrated neuroprotective effects of SGLT2i [12, 18]. The study by Lin and colleagues in 2014 showed that empagliflozin, a SGLT2i, ameliorated cognitive decline in db/db mice by reducing cerebral oxidative stress and elevating cerebral brain-derived neurotrophic factor [19]. Another study by Hayden and colleagues in 2019 showed that empagliflozin conferred a protective effect of the neurovascular unit in female db/db mice [2]. A randomized controlled trial conducted on 39 elderly participants with T2DM who were treated with SGLT2i and incretin showed no evidence of deterioration of cognitive function with either agent at 12 months’ follow-up [20]. To the best of our knowledge, there is no published longitudinal study which has observed a definitive longer term neuroprotective effect of SGLT2i in humans. We therefore explored the association between SGLT2i and longitudinal changes in cognitive function in adults with T2DM and assessed the cognitive domains which were impacted by SGLT2i.

MATERIALS AND METHODS

Participants

This was a prospective study of patients who participated in the Singapore Study of Macro-angiopathy and Micro-vascular Reactivity in Type 2 Diabetes (SMART2D) between September 2014 and January 2019. The SMART2D study aimed to study the impact of T2DM on vascular function and diabetic complications. The present study is a secondary analysis of the SMART2D cohort. The participants were recruited during their visits for T2DM at a Diabetes Centre in a public hospital and two primary care polyclinics in the Northern Region of Singapore. The participants were followed up during the period July 2019 and May 2021.

Patients were excluded from SMART2D study for the following: overt untreated malignancy, active inflammation (e.g., Systemic Lupus Erythematosus, intake of non-steroidal anti-inflammatory medications on the same day of study at baseline, intake of oral steroids equivalent to prednisolone more than 7.5 mg per day, history of dementia from case notes, and inability to give written informed consent. For the purpose of this analysis, we also excluded participants who were aged < 45 years or with history of stroke, duration of SGLT2i < 3 months, non-availability of RBANS score at baseline or follow-up. The flowchart of participants was shown in Supplementary Figure 1. The study was done in accordance with the principles of Helsinki. Ethics approval for this study was granted by the National Healthcare Group National Healthcare Group Domain Specific Review Board in Singapore (DSRB Ref: 2014/00667). Written informed consent was given by all the participants.

Data collection

Trained research nurses collected data on demographics, education, medications, and clinical history from a standardized questionnaire. The information on medications was verified with clinical records. Of note, data on the use SGLT2i was verified with medications data extracted from the pharmacy records. The specific SGLT2i agents were canagliflozin, dapagliflozin, and empagliflozin. Participants were deemed to be users of SGLT2i if they had been on the agent for 3 months or longer. Depressive symptoms were assessed using the Geriatric Depression Scale (GDS). GDS score ≥5 indicated presence of depressive symptoms [9].

Blood pressure was measured using a standard automated sphygmomanometer (HEM-C7011-C1, OMRON Corp., Kyoto, Japan) with the participant in a seated position after a resting period of at least 10 min.

The participants gave blood and spot urine samples. These were sent to the hospital laboratory which used the following assays for measurement: immunoturbidmimetric assay for urinary albumin; enzymatic colorimetric test for serum creatinine; and Tinaquant Haemoglobin A1c Gen.3 for Haemoglobin A1c (HbA1c) (COBAS-Roche, Mannheim, Germany). Rate of change in HbA1c was calculated as follow-up HbA1c minus baseline HbA1c and divided by the length of follow-up period in years. Apolipoprotein E (APOE) ɛ4 allele was assessed with Illumina OmniExpress 24 array. We calculated the estimated glomerular filtration rate (eGFR) with the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation [21].

Cognitive function was measured at baseline and at one follow-up over periods from 1.6 years to 6.4 years using the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) which has undergone translation with validation for use in Singapore [22, 23]. There are five domains of cognition, namely immediate and delayed memory, visuospatial/ constructional, attention and language. They were assessed from 12 subsets of cognition tests which include list learning, story memory, figure copy, line orientation, picture naming, semantic fluency, digit span, coding, list recall, list recognition, story recall, and figure recall [22, 23]. Although a specific Executive Functioning Index is not derived by subtests in the RBANS, Picture Naming and Semantic Fluency (Language domain) and Digit Span Coding (Attention domain) are executive function tasks.

Statistical analysis

Continuous variables were presented as means±standard deviation (SD) or median with interquartile range (IQR). Categorical variables were presented as number with percentages. The baseline characteristics were compared between non-use and use of SGLT2i using student t-test or Wilcoxon rank-sum test for continuous variables, depending on distribution of variables, and Chi-squared test for categorical variable.

The change in RBANS score was calculated as follow-up RBANS score minus baseline RBANS score. Least square mean change in RBANS score was compared between SGLT2i users and non-users.

We used linear regression to examine the association between use of SGLT2i and rate of change in RBANS score, adjusting for age, gender, ethnicity, education < 7 years and ≥7 years, GDS score ≥5, DM duration, rate of change of HbA1c, systolic blood pressure (SBP), eGFR, log-transformed urinary albumin-to-creatinine ratio (uACR), use of metformin, use of insulin, presence of APOE ɛ4 allele and follow-up duration. The rate of change in RBANS score was chosen as the outcome of interest as it was necessary to take into account the follow-up period which was different for each participant. Such an outcome measure was also used in earlier research. [24] The choice of covariates for adjustment was made in view of their biological plausibility or by observed univariate association with p < 0.10. The analysis was repeated with stratification by use of SGLT2i for < 3 years and ≥3 years. The threshold of 3 years was chosen as the mean duration of SGLT2i use was 2.5 years (SD 1.6).

The association between use of SGLT2i and rate of change in RBANS score was also examined using propensity score matching for the same covariates. Propensity score matching was done with a one-to-one matching protocol using a nearest neighbor matching approach without replacement. All matching procedures were conducted using the PSMATCH2 package. [25]

RESULTS

As shown in Table 1 on baseline characteristics, the mean age of the participants was 60.6±7.4 years. There was a slight preponderance of males (56.3%). The mean DM duration was 15.4±9.0 years. In Supplementary Table 1, there was no statistically significant difference in the baseline RBANS score between SGLT2i users and non-users (p > 0.05). The mean duration of follow-up was not statistically different between SGLT2i users (4.5±0.9 years) and non-SGLT2i users (4.3±1.1 years) (p = 0.124). There was no statistically significant difference in mean RBANS score between SGLT2i users and non-users before and after propensity score matching (results not shown).

Table 1

Baseline characteristics stratified by use of SGLT2 inhibitors

	Use of SGLT2 inhibitor
Variable	All	No	Yes	p
Number	476	338 (71.0)	138 (29.0)
Age (y)	60.6±7.4	61.5±7.5	58.6±6.8	< 0.001
Male (%)	268 (56.3)	185 (54.7)	83 (60.1)	0.280
Ethnicity (%)				0.747
Chinese	252 (52.9)	184 (54.4)	68 (49.3)
Malay	86 (18.1)	59 (17.5)	27 (19.6)
Indian	115 (24.2)	80 (23.7)	35 (25.4)
Other	23 (4.8)	15 (4.4)	8 (5.8)
Education (%)				0.123
≥7 y	364 (76.5)	252 (74.6)	112 (81.2)
< 7 y	112 (23.5)	86 (25.4)	26 (18.8)
GDS score≥5 (%)	22 (4.6)	16 (4.7)	6 (4.4)	0.868
DM duration (y)	15.4±9.0	14.8±9.2	16.8±8.4	0.028
SBP (mmHg)	138.9±16.6	139.7±17.0	137.1±15.4	0.127
HbA1c (%)	7.9±1.6	7.7±1.5	8.5±1.7	< 0.001
HbA1c (mmol/mol)	62.8±17.5	60.7±16.4	69.4±18.6	< 0.001
eGFR (ml/min/1.73 m²)	82.0±29.7	82.1±30.9	81.6±26.7	0.860
uACR (mg/g)	23.2 (6.5–114.5)	18.7 (6.0–87.0)	43.0 (8.0–235.7)	< 0.001
Use of metformin (%)	409 (86.1)	278 (82.3)	131 (95.6)	< 0.001
Use of insulin (%)	163 (34.4)	88 (26.1)	75 (54.7)	< 0.001
Presence of APOE ɛ4 allele	71 (18.2)	48 (17.0)	23 (21.1)	0.348

SGLT2i, Sodium-glucose cotransporter-2 inhibitors; GDS, Geriatric Depression Scale; DM, diabetes mellitus; SBP, systolic blood pressure; HbA1c, haemoglobin A1c; eGFR, estimated glomerular filtration rate; uACR, urinary albumin-to-creatinine ratio; APOE, Apolipoprotein E.

There were 138 patients (29.0%) on SGLT2i, of which 17.7% were on SGLT2i for < 3 years and 11.3% were on SGLT2i for ≥3 years. In Supplementary Table 2, the participants on SGLT2i for ≥3 years were younger, had higher baseline HbA1c and uACR levels, and higher proportion of metformin use than those on SGLT2i for < 3 years and non-SGLT2i users (p≤0.001). In Supplementary Table 3, there was no statistically significant difference in baseline RBANS score among the three groups (none, SGLT2i use for < 3 years and SGLT2i use for ≥3 years) (p > 0.05).

The median follow-up period was 4.6 years (IQR 4.1–5.0; max 6.4 years). In Table 2, RBANS total score and scores for all the domains decreased in the non-SGLT2i users (p = 0.006 for visuo-spatial/ construction and p < 0.001 for total and other domain-specific scores). RBANS total score and scores for delayed memory, visuo-spatial/ constructional ability and attention decreased in the SGLT2i users (p < 0.05) but there was no statistically significant drop in the scores for immediate memory (p = 0.512) and language for this group (p = 0.884). The change in RBANS total score was significantly different between the two groups in the domain of language where the non-SGLT2i users experienced significantly greater reduction in RBANS score compared to SGLT2i users (p = 0.005). The difference in change of total score and immediate memory score were of borderline statistical significance between the two groups (p = 0.052 and p = 0.056 respectively).

Table 2

Baseline and follow up RBANS score in users and non-users of SGLT2 inhibitor

		No use of SGLT2 inhibitor				Use of SGLT2 inhibitor			p ^a
RBANS	Baseline	Follow-up	Difference (95% CI)	p	Baseline	Follow-up	Difference (95% CI)	p
Total	102.3±7.9	100.0±9.0	–2.3 (–2.9 to –1.8)	< 0.001	103.2±7.6	101.8±7.5	–1.4 (–2.2 to –0.5)	0.002	0.052
Immediate memory	104.3±10.3	102.2±10.7	–2.1 (–3.1 to –1.2)	< 0.001	104.5±10.1	104.0±9.6	–0.5 (–1.8 to 0.9)	0.512	0.056
Delayed memory	102.7±10.1	99.6±12.0	–3.1 (–4.0 to –2.1)	< 0.001	103.9±9.8	101.4±8.9	–2.5 (–3.8 to –1.1)	< 0.001	0.497
Visuo-spatial/construction	101.5±10.7	100.0±12.1	–1.5 (–2.6 to –0.4)	0.006	102.4±9.9	100.2±12.5	–2.1 (–4.1 to –0.2)	0.031	0.578
Language	101.1±11.1	98.5±12.6	–2.7 (–3.7 to –1.6)	< 0.001	102.6±9.8	102.7±10.1	0.1 (–1.5 to 1.7)	0.884	0.005
Attention	102.1±11.8	99.7±11.9	–2.4 (–3.3 to –1.5)	< 0.001	102.7±11.2	100.8±11.7	–1.8 (–3.2 to –0.5)	0.008	0.535

SGLT2i, Sodium-glucose cotransporter-2 inhibitors; RBANS, Repeatable Battery for the Assessment of Neuropsychological Status. ^ap-value for the difference in change of RBANS score between the non-SGLT2i users and SGLT2i users.

In Supplementary Figure 2, the SGLT2i users had higher mean RBANS total score than the non-SGLT2i users over the years. The decline also appeared less steep among the SGLT2i users compared to the non-SGLT2i users from year 4 of follow-up onwards. In Supplementary Figure 3, the SGLT2i users for ≥3 years had higher RBANS total score compared to the non-SGLT2i users and SGLT2i users for < 3 years.

Table 3 showed that compared to non-use of SGLT2i, use of SGLT2i was positively associated with increase in rate of RBANS score in language in unadjusted analysis. This association persisted in Model 1 which was adjusted for age, gender, ethnicity, and education, and in Model 2 which was additionally adjusted for GDS score, DM duration, rate of HbA1c change, eGFR, uACR, use of metformin, use of insulin, presence of APOE ɛ4 allele and follow-up duration with corresponding coefficients 0.65 (95% CI 0.15 to 1.15; p = 0.011) and 0.74 (95% CI 0.12 to 1.36; p = 0.019). There was no evidence of association between use of SGLT2i and rate of change in the other cognitive domains.

Table 3

Association between SGLT2 inhibitors and rate of RBANS score change

	Coefficient (95% CI) p
	Unadjusted	Model 1^a	Model 2^b
SGLT2 inhibitors (versus none)
Total	0.20 (–0.03 to 0.44) p = 0.091	0.17 (–0.06 to 0.41) p = 0.153	0.19 (–0.12 to 0.49) p = 0.223
Immediate memory	0.38 (–0.04 to 0.79) p = 0.074	0.38 (–0.05 to 0.80) p = 0.081	0.11 (–0.41 to 0.64) p = 0.675
Delayed memory	0.09 (–0.34 to 0.52) p = 0.677	–0.06 (–0.49 to 0.37) p = 0.776	–0.09 (–0.64 to 0.47) p = 0.759
Visuo-spatial/construction	–0.11 (–0.67 to 0.46) p = 0.703	–0.10 (–0.67 to 0.47) p = 0.730	0.12 (–0.61 to 0.85) p = 0.748
Language	0.60 (0.10 to 1.11) p = 0.019	0.65 (0.15 to 1.15) 0.011	0.74 (0.12 to 1.36) p = 0.019
Attention	0.05 (–0.35 to 0.44) p = 0.806	0.00 (–0.40 to 0.40) p = 0.999	0.05 (–0.49 to 0.59) p = 0.851
SGLT2 inhibitors (versus none)
< 3 y
Total	–0.01 (–0.29 to 0.27) p = 0.959	–0.05 (–0.33 to 0.24) p = 0.752	–0.03 (–0.37 to 0.32) p = 0.869
Immediate memory	0.42 (–0.08 to 0.91) p = 0.102	0.43 (–0.08 to 0.94) p = 0.098	0.09 (–0.52 to 0.70) p = 0.766
Delayed memory	–0.24 (–0.75 to 0.27) p = 0.354	–0.39 (–0.90 to 0.12) p = 0.134	–0.46 (–1.09 to 0.18) p = 0.159
Visuo-spatial/construction	–0.48 (–1.16 to 0.20) p = 0.168	–0.45 (–1.14 to 0.24) p = 0.198	–0.15 (–0.99 to 0.69) p = 0.720
Language	0.56 (–0.05 to 1.16) p = 0.072	0.56 (–0.04 to 1.16) p = 0.066	0.51 (–0.20 to 1.22) p = 0.160
Attention	–0.29 (–0.76 to 0.18) p = 0.231	–0.38 (–0.85 to 0.10) p = 0.119	–0.14 (–0.76 to 0.48) p = 0.660
≥3 y
Total	0.53 (0.19 to 0.87) p = 0.002	0.51 (0.17 to 0.85) p = 0.003	0.54 (0.13 to 0.95) p = 0.010
Immediate memory	0.32 (–0.28 to 0.91) p = 0.302	0.29 (–0.32 to 0.90) p = 0.352	0.14 (–0.58 to 0.87) p = 0.694
Delayed memory	0.61 (–0.01 to 1.22) p = 0.053	0.46 (–0.16 to 1.07) p = 0.147	0.52 (–0.24 to 1.27) p = 0.180
Visuo-spatial/construction	0.46 (–0.35 to 1.28) p = 0.266	0.45 (–0.38 to 1.28) p = 0.287	0.56 (–0.44 to 1.56) p = 0.268
Language	0.68 (–0.05 to 1.41) p = 0.068	0.78 (0.06 to 1.51) p = 0.034	1.12 (0.27 to 1.97) p = 0.010
Attention	0.57 (0.01 to 1.14) p = 0.048	0.59 (0.02 to 1.16) p = 0.042	0.36 (–0.38 to 1.10) p = 0.336

^aModel 1 adjusted for age, gender, ethnicity, and years of education < 7. ^bModel 2 adjusted for age, gender, ethnicity, years of education < 7, GDS score≥5, diabetes duration, rate of HbA1c change, SBP, eGFR, log-transformed uACR, use of metformin, use of insulin, presence of APOE ɛ4 allele and follow-up duration.

Compared with non-use of SGLT2i, use of SGLT2i for ≥3 years was associated with increase in RBANS total score in unadjusted analysis (coefficient 0.53; 95% CI 0.19 to 0.87; p = 0.002), Model 1 (coefficient 0.51; 95% CI 0.17 to 0.85; p = 0.003), and Model 2 (coefficient 0.54; 95% CI 0.13 to 0.95; p = 0.010) in Table 3. The positive association between use of SGLT2i for ≥3 years and increase in RBANS score was also evident in the domain of language where the coefficients were 0.78 (95% CI 0.06 to 1.51; p = 0.034) in Model 1 and 1.12 (95% CI 0.27 to 1.97; p = 0.010) in Model 2. There was no evidence of association between use of SGLT2i for < 3 years and rate of change in RBANS score.

In Supplementary Table 4, SGLT2i use was positively associated with increase in rate of RBANS score in language in a propensity score matched analysis with matching for demographics and education (coefficient 0.74; 95% CI 0.06 to 1.42; p = 0.034) and additionally for clinical covariates, medications, and duration of follow-up (coefficient 0.71; 95% CI 0.22 to 1.20; p = 0.005). The positive association persisted in a repeated analysis stratified by duration of SGLT2i use (< 3 years and ≥3 years) with corresponding coefficients 1.42 (95% CI 0.73 to 2.11; p < 0.001) and 1.44 (95% CI 0.95 to 1.94; p < 0.001) upon full propensity score matching. SGLT2i use for ≥3 years was also positively associated with increase in rate of RBANS score in attention with coefficient 0.76 (95% CI 0.56 to 0.97; p < 0.001) upon full propensity score matching.

DISCUSSION

To the best of our knowledge, no prior publications have reported a possible positive effect of SGLT2i use on cognitive function in patients with T2DM. We demonstrate for the first time that prolonged use of SGLT2i over 3 or more years was associated with increased RBANS performance globally and in the domain of language. The association was independent of demographics, education, clinical covariates, other medications, and presence of APOE ɛ4 allele. The finding on the positive association between SGLT2i use and increased RBANS score in language was also observed upon propensity score matching. The positive association between SGLT2i use for ≥3 years and increased RBANS score in attention in the analysis upon propensity score matching may need confirmation in future research with larger sample size.

The observed association is consistent with an euglycemic state that is known to be associated with reduced inflammation and oxidative stress which are substrates for cognitive impairment [20]. Beyond this established mechanism of action, however, recent studies on db/db or high-fat diet-fed mice have shown tantalizingly results on the direct positive effect of SGLT2i on cognition [2 , 18]. Glucotoxicity is postulated to play a pivotal role in aberrant remodeling of microvascular neurovascular unit and neuroglia. Studies show that empagliflozin prevents abnormalities of the neurovascular unit such as aberrant mitochondria (absence of crista and mitochondrial matrix electron density), invasion of activated microglia, astrocyte foot processes detachment and thickening of basement membrane female db/db mice [2]. In a mixed murine model of Alzheimer’s disease and T2DM (APP/PS1 x db/db mice), empagliflozin causes a decreased rate of brain atrophy, amyloid pathology, cortical microglia burden, tau phosphorylation, and spontaneous hemorrhage, and improved learning and memory [18]. Similarly, empagliflozin reduces oxidative stress and increases brain-derived neurotrophic factor in the brain in db/db mice [19]. Compared to vildagliptin (dipeptidy peptidase-4 inhibitor), dapagliflozin shows higher efficacy in preserving synaptic plasticity, and a combination of dapaglilozin and vildagliptin was shown to improve insulin sensitivity and decrease oxidative stress in the brain [12].

We observed that the prolonged use of SGLT2i over 3 or more years was associated specifically with increased RBANS performance domain of language. This cognitive domain is derived from the Picture Naming and Semantic Fluency test subtests, measuring the ability to respond verbally to either naming or retrieving learned material, which are also indicative of executive function performance. Patients with T2DM show decline in executive function, memory, learning, attention, and psychomotor speed [26]. Executive function impairment is particularly associated with cerebral microvascular dysfunction due to small vessel disease evidenced by magnetic resonance imaging features of white matter hyperintensities in the subcortical periventricular areas adjacent to the hypothalamus and pre-frontal cortex, causing disconnection of fronto-parietal-subcortical circuits [27 –29]. SGLT2i reduces hyperglycemia by inhibiting glucose reabsorption in the kidney and increasing urinary excretion of glucose [12, 13]. The reduced glucotoxicity in the brain and its protective effect on the neurovascular unit mentioned above may explain the apparently favorable effect of SGLT2 on executive function. Further research is needed to understand the exact pathophysiological mechanism underlying the neuroprotective effect of SGLT2i. Future randomized controlled trials should be conducted to provide further evidence for SGLT2i as a disease modifying treatment to slow cognitive decline in T2DM.

Our findings, derived from real-world setting of diabetes care in the outpatient setting, have particular clinical relevance in and development of future DM treatment guidelines and diabetes management. Increasing awareness among healthcare providers on the potential benefits of anti-diabetic treatment such as SGLT2i in ameliorating cognitive decline is crucial for global strategies to prevent dementia. This especially important in this group of middle-aged adults, who are economically and socially active with significant life-years ahead for aging without dementia.

A strength of this study is its longitudinal design determining changes in cognitive function from patients’ follow up. We adjusted for a comprehensive range of covariates including demographics, education, depressive symptoms, clinical parameters, and other anti-diabetic medications and APOE ɛ4 allele. The use of RBANS is a well-established validated brief clinical tool for assessing neuro-cognitive domains in Alzheimer’s disease and other neurodegenerative and cerebrovascular disease. There are some limitations. First, the study lacked brain imaging data which corroborate cognitive findings with the cerebral structural and functional impact of SGLT2i. Also, there are possibly residual confounding from known risk factors for cognitive decline such as hearing impairment, physical, social, and mentally stimulating activity, vascular disease, and psychotropic drugs, which were not available for analysis. The follow up period was relatively short and might have accounted for small magnitude in changes between baseline and follow-up RBANS score. Hence it is uncertain if these changes translate to clinically meaningful differences in cognitive function in the real world. A longer follow up of these patients may help to determine clinical progression to dementia.

In conclusion, our findings revealed a novel, previously unobserved positive association between use of SGLT2i over 3 or more years and increased cognitive score, both globally and in the domains of language and executive function. More studies in future should investigate the potential clinical benefits of SGLT2i in ameliorating cognitive decline.

Footnotes

ACKNOWLEDGMENTS

This SMART2D cohort was supported by the Singapore Ministry of Health’s National Medical Research Council under its CS-IRG (MOH-000066). The corresponding author is supported by the Singapore Ministry of Health’s National Medical Research Council under its Clinician Scientist Award (MOH-000714-01). The first author is supported by the Singapore Ministry of Health’s National Medical Research Council under its Research Training Fellowship (NMRC/MOH000226). The authors declare that they have competing conflicts of interest.

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

International Diabetes Federation (2021) International Diabetes Federation Diabetess Atlas 10th edition. https://diabetesatlas.org/atlas/tenth-edition, Last updated 8 November 2021, Accessed 15 December 2021.

Hayden

, Grant

, Aroor

, DeMarco

(2019) Empagliflozin ameliorates type 2 diabetes-induced ultrastructural remodeling of the neurovascular unit and neu-roglia in the female db/db mouse. Brain Sci 9, 57.

Chatterjee

, Peters

, Woodward

, Mejia Arango

, Batty

, Beckett

, Beiser

, Borenstein

, Crane

, Haan

, Hassing

, Hayden

, Kiyohara

, Larson

, Li

, Ninomiya

, Ohara

, Peters

, Russ

, Seshadri

, Strand

, Walker

, Xu

, Huxley

(2016) Type 2 diabetes as a risk factor for dementia in women compared with men: A pooled analysis of 2.3 million people comprising more than 100,000 cases of dementia. Diabetes Care 39, 300–307.

Zilliox

, Chadrasekaran

, Kwan

, Russell

(2016) Diabetes and cognitive impairment. Curr Diab Rep 16, 87.

Lyu

, Wu

, Wei

, Wu

(2020) Vascular cognitive impairment and dementia in type 2 diabetes mellitus: An overview. Life Sci 254, 117771.

Zhou

, Tang

, Han

, Yang

, Simó

(2020) Impact of antidiabetic agents on dementia risk: A Bayesian network meta-analysis. Metabolism 109, 154265.

McClean

, Hölscher

(2014) Lixisenatide, a drug developed to treat type 2 diabetes, shows neuroprotective effects in a mouse model of Alzheimer’s disease. Neuropharmacology 86, 241–258.

McClean

, Parthsarathy

, Faivre

, Hölscher

(2011) The diabetes drug liraglutide prevents degenerative processes in a mouse model of Alzheimer’s disease. J Neurosci 31, 6587–6594.

, Feng

, Yap

, Lee

, Tan

, Winblad

(2014) Long-term metformin usage and cognitive function among older adults with diabetes. J Alzheimers Dis 41, 61–68.

10.

Jeong

, Kim

, Hong

, Jung

, Baik

, Cho

, Lyoo

, Ye

, Sohn

, Seong

, Lee

(2021) Association of dipeptidyl peptidase-4 inhibitor use and amyloid burden in diabetic patients with AD-related cognitive impairment. Neurology 97, e1110–e1122.

11.

Areosa Sastre

, Vernooij

, González-Colaço Harmand

, Martínez

(2017) Effect of the treatment of Type 2 diabetes mellitus on the development of cognitive impairment and dementia. Cochrane Database Syst Rev 6, CD003804.

12.

Sa-Nguanmoo

, Tanajak

, Kerdphoo

, Jaiwongkam

, Pratchayasakul

, Chattipakorn

(2017) SGLT2-inhibitor and DPP-4 inhibitor improve brain function via attenuating mitochondrial dysfunction, insulin resistance, inflammation, and apoptosis in HFD-induced obese rats. Toxicol Appl Pharmacol 333, 43–50.

13.

Jabbour

, Goldstein

(2008) Sodium glucose co-transporter 2 inhibitors: Blocking renal tubular reabsorption of glucose to improve glycaemic control in patients with diabetes. Int J Clin Pract 62, 1279–1284.

14.

Mather

, Pollock

(2011) Glucose handling by the kidney. Kidney Int Suppl S1–6.

15.

Bailey

, Gross

, Hennicken

, Iqbal

, Mansfield

, List

(2013) Dapagliflozin add-on to metformin in type 2 diabetes inadequately controlled with metformin: A randomized, double-blind, placebo-controlled 102-week trial. BMC Med 11, 43.

16.

Han

, Hagan

, Taylor

, Xin

, Meng

, Biller

, Wetterau

, Washburn

, Whaley

(2008) Dapagliflozin, a selective SGLT2 inhibitor, improves glucose homeostasis in normal and diabetic rats. Diabetes 57, 1723–1729.

17.

Wilding

, Woo

, Soler

, Pahor

, Sugg

, Rohwedder

, Parikh

(2012) Long-term efficacy of dapagliflozin in patients with type 2 diabetes mellitus receiving high doses of insulin: A randomized trial. Ann Intern Med 156, 405–415.

18.

Hierro-Bujalance

, Infante-Garcia

, Del Marco

, Herrera

, Carranza-Naval

, Suarez

, Alves-Martinez

, Lubian-Lopez

, Garcia-Alloza

(2020) Empagliflozin reduces vascular damage and cognitive impairment in a mixed murine model of Alzheimer’s disease and type 2 diabetes. Alzheimers Res Ther 12, 40.

19.

Lin

, Koibuchi

, Hasegawa

, Sueta

, Toyama

, Uekawa

, Ma

, Nakagawa

, Kusaka

, Kim-Mitsuyama

(2014) Glycemic control with empagliflozin, a novel selective SGLT2 inhibitor, ameliorates cardiovascular injury and cognitive dysfunction in obese and type 2 diabetic mice. Cardiovasc Diabetol 13, 148.

20.

Perna

, Mainardi

, Astrone

, Gozzer

, Biava

, Bacchio

, Spadaccini

, Solerte

, Rondanelli

(2018) 12-month effects of incretins versus SGLT2-Inhibitors on cognitive performance and metabolic profile. A randomized clinical trial in the elderly with Type-2 diabetes mellitus. Clin Pharmacol 10, 141–151.

21.

Levey

, Stevens

, Schmid

, Zhang

, Castro

3rd, Feldman

, Kusek

, Eggers

, Van Lente

, Greene

, Coresh

(2009) A new equation to estimate glomerular filtration rate. Ann Intern Med 150, 604–612.

22.

, Nyunt

MSZ

, Shuvo

, Eng

, Yap

, Hee

, Chan

, Scherer

(2018) The neighborhood built environment and cognitive function of older persons: Results from the Singapore Longitudinal Ageing Study. Gerontology 64, 149–156.

23.

Randolph

, Tierney

, Mohr

, Chase

(1998) The Repeatable Battery for the Assessment of Neuropsychological Status (RBANS): Preliminary clinical validity. J Clin Exp Neuropsychol 20, 310–319.

24.

Dickerson

, Schroeder

, Stallings

, Origoni

, Katsafanas

, Schwienfurth

, Savage

, Khushalani

, Yolken

(2014) A longitudinal study of cognitive functioning in schizophrenia: Clinical and biological predictors. Schizophr Res 156, 248–253.

25.

Leuven

, Sianesi

PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching. Available from: http://ideas.repec.org/boc/bocode/s432001.html. Accessed February 2, 2022.

26.

Okereke

, Kang

, Cook

, Gaziano

, Manson

, Buring

, Grodstein

(2008) Type 2 diabetes mellitus and cognitive decline in two large cohorts of community-dwelling older adults. J Am Geriatr Soc 56, 1028–1036.

27.

Sudo

, Alves

CEO

, Alves

, Ericeira-Valente

, Tiel

, Moreira

, Laks

, Engelhardt

(2012) Dysexecutive syndrome and cerebrovascular disease in non-amnestic mild cognitive impairment: A systematic review of the literature. Dement Neuropsychol 6, 145–151.

28.

van Sloten

, Sedaghat

, Carnethon

, Launer

, Stehouwer

CDA

(2020) Cerebral microvascular complications of type 2 diabetes: Stroke, cognitive dysfunction, and depression. Lancet Diabetes Endocrinol 8, 325–336.

29.

Wang

, Wang

, Wei

, Xia

, Tian

, Cui

, Li

(2020) Relationship between type 2 diabetes and white matter hyperintensity: A systematic review. Front Endocrinol (Lausanne) 11, 595962.