Skip to main content
NCBI home page
Alzheimer's & Dementia logoLink to Alzheimer's & Dementia
. 2025 May 30;21(6):e70313. doi: 10.1002/alz.70313

Heterogeneous treatment effects of GLP‐1RAs and SGLT2is on risk of Alzheimer's disease and related dementia in patients with type 2 diabetes: Insights from a real‐world target trial emulation

Huilin Tang 1, William T Donahoo 2, Steven T DeKosky 3,4, Yao An Lee 1, Pareeta Kotecha 1, Mikael Svensson 1, Jiang Bian 5, Jingchuan Guo 1,6,✉
PMCID: PMC12125485  PMID: 40448382

Abstract

INTRODUCTION

This study assessed the heterogeneous treatment effects (HTEs) of glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) and sodium‐glucose cotransporter‐2 inhibitors (SGLT2is) on the risk of Alzheimer's disease and related dementias (ADRD).

METHODS

This target trial emulation study included adults (≥ 50 years) with type 2 diabetes (T2D) and newly prescribed a GLP‐1RA, SGLT2i, or other second‐line glucose‐lowering drugs (GLDs). A doubly robust learning approach was deployed to estimate the risk difference (RD) of ADRD and identify key subgroups.

RESULTS

Both GLP‐1RAs (RD, −1.5%) and SGLT2is (−1.7%) were associated with a reduced ADRD risk compared to other GLDs. Key subgroups were determined based on cardiovascular disease (CVD), cerebrovascular disease (CeVD), chronic kidney disease, and Hispanic ethnicity. Patients with CVD and CeVD had the greatest benefits from GLP‐1RAs (−4.8%) and SGLT2is (−4.6%). No overall difference was observed between GLP‐1RAs and SGLT2i.

DISCUSSION

These findings suggest the importance of personalized treatment in diabetes management regarding ADRD risk.

Highlights

  • Glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) were associated with a decreased risk of Alzheimer's disease and related dementias (ADRD), while the protective association varied across subgroups defined by cardiovascular disease (CVD), cerebrovascular disease (CeVD), and chronic kidney disease (CKD).

  • Similarly, sodium‐glucose cotransporter‐2 inhibitors (SGLT2is) were associated with a decreased risk of ADRD, with the protective association varying among subgroups defined by CVD, CeVD, and Hispanic ethnicity.

  • There was no difference between GLP‐1RAs and SGLT2is in the risk of ADRD.

Keywords: Alzheimer's disease and related dementias, GLP‐1RAs, SGLT2is, type 2 diabetes

1. BACKGROUND

Type 2 diabetes (T2D) is a prevalent metabolic disorder with significant global public health implications. 1 Increasingly, T2D is recognized as a significant risk factor for cognitive decline and Alzheimer's disease (AD) and related dementia (ADRD). 2 The interplay between diabetes and cognitive decline underscores the need for management strategies that extend beyond glycemic control to mitigate the long‐term risk of ADRD in this vulnerable population. 3

Recent advances in glucose‐lowering drugs (GLDs), particularly glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) and sodium‐glucose transport protein 2 inhibitors (SGLT2is), have demonstrated benefits beyond lowering glucose, including cardiovascular and renal protective effects. 4 , 5 , 6 Intriguingly, preclinical and clinical studies suggested that GLP‐1RAs and SGLT2is may also influence the pathophysiology of AD, such as mitigating neuroinflammation and oxidative stress. 7 , 8 Moreover, population‐based studies have reported potential associations between the use of GLP‐1RAs and SGLT2is and a reduced risk of dementia. 9 , 10 , 11 , 12 Our recent meta‐analysis of observational studies found that GLP‐1RA users (risk ratio [RR], 0.72; 95% confidence interval [CI], 0.54–0.97) and SGLT2i users (RR, 0.62; 95% CI, 0.39–0.97) had a lower risk of dementia compared to nonusers. 13 Despite these promising findings, the observed high level of heterogeneity (I 2: 82.5‐91.3%) suggests that the effectiveness of these GLDs in reducing dementia risk varies across different studies, indicating their effects on dementia risk may differ among various patient subgroups. 13

Understanding which subpopulations benefit most from GLP‐1RAs and SGLT2is regarding ADRD risk reduction is crucial for advancing precision medicine. While some studies have attempted subgroup analyses, there is limited evidence on which specific subpopulations might gain the greatest benefits from these GLDs. Our previous study has shown potential heterogeneous treatment effects (HTEs) of SGLT2is on the risk of dementia. 14 This study aimed to address this gap by determining HTEs of GLP‐1RAs and SGLT2is compared to other second‐line GLDs in people with T2D. Additionally, we investigated potential HTEs between GLP‐1RAs and SGLT2is. By enhancing our understanding of the cognitive benefits associated with these newer GLDs across different patient subgroups, this research seeks to uncover novel strategies for the prevention and treatment of dementia. Such insights could lead to more personalized and effective interventions, ultimately improving outcomes for individuals at risk of or living with ADRD.

2. METHODS

2.1. Study design and data source

We conducted three target trial emulation studies comparing the effectiveness of GLP‐1RAs with other second‐line GLDs, including sulfonylurea, thiazolidinedione, dipeptidyl peptidase‐4 inhibitor (DPP4i), α‐glucosidase inhibitor, and meglitinide; SGLT2is with other GLDs, and GLP‐1RAs with SGLT2is, on risk of ADRD among people with T2D. The exposures of interest are detailed in Table S1. Target trial emulation involves simulating a randomized controlled trial (RCT) using real‐world data, with prespecification of protocol elements including eligibility criteria, treatment strategies, assignment, outcomes, follow‐up, causal contrasts, and statistical analysis. 15 , 16 This study adhered to the target trial emulation framework, with key components summarized in Table S2. The overview of the study design is shown in Figure S1. This study was approved by the University of Florida Institutional Review Board (IRB202201196).

The data for this study were extracted from OneFlorida+ Data Trust, a centralized research patient data repository created and managed by the OneFlorida + Clinical Research Consortium (“OneFlorida+”). 17 OneFlorda+ includes longitudinal electronic health records linked with various other data sources, such as Medicaid and Medicare claims, death data, and tumor registry data. 17 As of 2023, OneFlorida+ encompassed over 21 million individuals, including approximately 18 million residents from Florida, 2 million from Georgia, and 1 million from Alabama. It covers a broad range of patient characteristics, including demographics, diagnoses, medications, procedures, vital signs, and lab test results. 18

2.2. Eligibility criteria

We included people who initiated treatment with a GLP‐1RA, SGLT2i, or other GLDs in OneFlorida+ between January 1, 2014 and June 30, 2023. Other GLDs were selected as the comparator group to help mitigate confounding by indication. 19 The cohort entry (index date) was defined as the first prescription for a GLP‐1RA, SGLT2i, or other GLDs, defined as no prior prescription for these drugs in the previous 1 year. Eligible people were required to have a diagnosis of T2D before or on the index date. People were excluded if they met the following criteria: (1) age < 50 years; (2) diagnosis code of ADRD before or on index date; (3) use of FDA‐approved anti‐ADRD medications (e.g., donepezil, memantine, rivastigmine, galantamine, and aducanumab) before or on index date; (4) diagnosis of gestational diabetes, type 1 diabetes, or end‐stage renal disease (ESRD)/dialysis before or on index date; (5) no healthcare encounters in 2 years before the index date; (6) initiation of treatment with SGLT2i or GLP‐1RA and comparator on the index date; (7) No healthcare encounter during the follow‐up period. Details regarding the diagnosis codes are present in Table S3.

2.3. Outcome and follow‐up

The primary outcome was the development of ADRD, including AD as well as other forms of dementia such as vascular dementia, frontotemporal dementia, and Lewy body dementia. ADRD was identified using the Chronic Conditions Warehouse (CCW) chronic condition algorithms developed by the Centers for Medicare & Medicaid Services (CMS). 20 The International Classification of Diseases (ICD) diagnosis codes used for ADRD identification are detailed in Table S3.

An intention‐to‐treat approach was employed (Figure S1), whereby individuals remained in their initially assigned treatment group from baseline until the end of the follow‐up period, regardless of adherence to the treatment regimen. Follow‐up continued until the occurrence of the study outcome, death, or the end of the study period (June 30, 2023), whichever came first.

2.4. Baseline covariates

Baseline covariates were selected based on previous research 21 , 22 and clinical experience, as detailed in Table S4. The demographic characteristics included age (continuous value), sex (male vs. female), race/ethnicity (Hispanic, non‐Hispanic Black, non‐Hispanic White, and other individuals), and diabetes complications (within 2 years before or on index date and classified as yes vs. no) included diabetes retinopathy, diabetic neuropathy, peripheral vascular disease (PVD), hypoglycemia, or hyperglycemic emergency. Comorbidities (within 2 years before or on index date) were classified as presence vs. absence of disease, including cardiovascular disease (CVD), cerebrovascular disease (CeVD), obesity, and chronic kidney disease (CKD). Medication use was also collected within 1 year before or on the index date and classified as yes versus no, including metformin, insulin, and antihypertensive drugs. The most recent values of hemoglobin A1C (HbA1c) and body mass index (BMI) within 1 previous year were also collected.

RESEARCH IN CONTEXT

  1. Systematic review: The authors reviewed the literature using PubMed. Previous studies indicated that both glucagon‐like peptide‐1 receptor agonists (GLP‐1RAs) and sodium‐glucose cotransporter‐2 inhibitors (SGLT2is) were associated with a reduced risk of Alzheimer's disease and related dementias (ADRD). However, the current research lacks clarity on which specific patient populations might derive the most benefit from these treatments.

  2. Interpretation: Our study found a potential decreased risk of ADRD associated with GLP‐1RAs and SGLT2is, while the protective association varied across different subgroups defined by cardiovascular disease (CVD), cerebrovascular disease (CeVD), chronic kidney disease (CKD), or Hispanic ethnicity.

  3. Future directions: These findings highlight the importance of personalized treatment in diabetes management regarding the ADRD risk.

2.5. Statistical analysis

We conducted data analyses for GLP‐1RA vs. other GLD cohorts, SGLT2i vs. other GLD cohorts, and GLP‐1RA vs. SGLT2i cohorts, separately. In each cohort, patient characteristics were summarized using frequency and percentage for categorical variables and mean with standard deviation for continuous covariates. To address missing data for HbA1c and BMI, we employed multiple imputation by chained equations. 23 The balance of baseline covariates between groups was assessed using standard mean difference (SMD), with an SMD < 0.1 indicating negligible differences. 24

In each cohort, we employed a doubly robust learning framework to estimate the conditional average treatment effects (CATEs), 25 which represent the average treatment effect within specific subgroups defined by key baseline characteristics (e.g., CVD, CeVD, or CKD). Understanding CATEs is crucial to identifying populations who may derive the most benefit from treatment, thus informing more personalized clinical decision‐making. This approach mitigates the risk of model misspecification by combining the propensity score (PS) model with the outcome regression model, ensuring accurate estimates even if one model is misspecified. 21 All baseline covariates were included as confounders in the PS, outcome regression, and final models to account for any potential imbalances and to identify subgroups with different CATEs. The dataset was randomly split into training (70%) and test (30%) sets. The training set was used for model development and hyperparameter tuning, while the test set evaluated model performance. Models were developed in two stages. In the first stage, we estimated the predicted probability of receiving the medication of interest for each subject using a PS model with an inverse probability of treatment weighting (IPTW) approach. We also estimated the risk of ADRD for both the exposure group and the non‐exposure group using an outcome regression model. Logistic regression and least absolute shrinkage and selection operator‐type regularized regression (LASSO) were used for the PS model and outcome regression model, respectively. Optimal hyperparameters were selected using cross‐validated models with “GridSearchCV” based on mean squared error (MSE). In the second stage, we integrated the two models from the first stage into the final model to estimate treatment effects using “LinearDRLearner,” a supervised machine learning model developed in the EconML package. 25

Treatment effects, including CATEs and predicted individualized treatment effects (ITEs, treatment effect on person level), were reported as absolute risk differences (RDs) with 95% CIs for the development of ADRD between the two groups within each cohort. Negative values (e.g., ITEs) indicated a reduced absolute risk of ADRD, while positive values (e.g., ITEs) indicated an increased absolute risk. A decision tree was employed to identify the most important covariates and to explore the drivers of heterogeneity, with at least 5% of the total study sample required in each leaf (subgroup). The decision tree model was trained to maximize the difference in treatment effects between subgroups, identifying patient subgroups with varying responses to the exposure regarding the risk of ADRD. Model performance was validated using five‐fold cross‐validation for internal validation and the test set for external validation.

To assess the robustness of the findings, we conducted several sensitivity analyses. First, we excluded people with mild cognitive impairment (MCI) at baseline. Second, we excluded people with Parkinson's disease at baseline. Third, we excluded people diagnosed with ADRD within 0.5 years after the index date, given the potential lag period for a diagnosis of ADRD. Additionally, to address the competing risk of death, we applied a doubly robust estimation of the hazard difference for competing risk data using the R Package “HazardDiff.” 26 Statistical analyses were performed using SAS version 9.4 (SAS Institute Inc.), Python version 3.9.12 (Python Software Foundation), and R version 4.3.3 (R Foundation).

3. RESULTS

3.1. Study cohort

According to the eligibility criteria, we included a total of 33,858 individuals initiating treatment with either GLP1‐RA or other GLD, 34,185 individuals initiating treatment with either SGLT2i or other GLD, and 24,117 individuals initiating treatment with either GLP‐1RA or SGLT2i from OneFlorida+ (Figure S2). The baseline demographic and clinical characteristics of all patients are presented in Table 1 and Table S5.

TABLE 1.

Selected baseline characteristics of patients with T2D within GLP‐1RA vs. other GLD cohort, SGLT2i vs. other GLD cohort, and GLP‐1RA vs. SGLT2i cohort.

GLP‐1RA vs. other GLD cohort SGLT2i vs. other GLD cohort GLP‐1RA vs. SGLT2i cohort
Characteristics GLP‐1RA (n = 10,212) Other GLDs (n = 23,646) SMD SGLT2i (n = 10,524) Other GLDs (n = 23,661) SMD GLP‐1RA (n = 11,395) SGLT2i (n = 12,722) SMD
Age at index date, years, mean(sd) 62.3 (8.0) 66.2 (9.5) −0.451 64.8 (9.0) 66.2 (9.6) −0.152 62.5 (8.1) 64.9 (9.0) −0.282
Race/ethnicity
Non‐Hispanic White 4096 (40.1%) 10,108 (42.7%) 0.094 4526 (43.0%) 10,137 (42.8%) 0.060 4678 (41.1%) 5420 (42.6%) 0.069
Non‐Hispanic Black 2964 (29.0%) 6583 (27.8%) 2834 (26.9%) 6566 (27.8%) 3280 (28.8%) 3347 (26.3%)
Hispanics 1346 (13.2%) 3258 (13.8%) 1373 (13.0%) 3283 (13.9%) 1503 (13.2%) 1696 (13.3%)
Other 1806 (17.7%) 3697 (15.6%) 1791 (17.0%) 3675 (15.5%) 1934 (17.0%) 2259 (17.8%)
Female 6003 (58.8%) 11,981 (50.7%) 0.164 4862 (46.2%) 12,002 (50.7%) −0.044 6692 (58.7%) 5778 (45.4%) 0.269
Diabetes complications
Diabetic retinopathy 572 (5.6%) 894 (3.8%) 0.086 486 (4.6%) 904 (3.8%) 0.040 634 (5.6%) 591 (4.6%) 0.042
Diabetic neuropathy 1345 (13.2%) 2559 (10.8%) 0.072 1456 (13.8%) 2568 (10.9%) 0.091 1487 (13.0%) 1724 (13.6%) −0.015
Peripheral vascular disease 713 (7.0%) 2005 (8.5%) −0.056 1565 (14.9%) 1993 (8.4%) 0.202 784 (6.9%) 1779 (14.0%) −0.234
Hypoglycemia 110 (1.1%) 163 (0.7%) 0.042 101 (1.0%) 164 (0.7%) 0.030 136 (1.2%) 120 (0.9%) 0.024
Hyperglycemic emergency 162 (1.6%) 447 (1.9%) −0.023 151 (1.4%) 447 (1.9%) −0.036 180 (1.6%) 181 (1.4%) 0.013
Comorbidities
Ever smoking 65 (0.6%) 562 (2.4%) −0.143 184 (1.7%) 562 (2.4%) −0.091 69 (0.6%) 219 (1.7%) −0.104
Mild cognitive impairment 48 (0.5%) 137 (0.6%) −0.015 65 (0.6%) 138 (0.6%) 0.005 53 (0.5%) 78 (0.6%) −0.020
Parkinson's disease 39 (0.4%) 137 (0.6%) −0.029 50 (0.5%) 137 (0.6%) −0.014 39 (0.3%) 56 (0.4%) −0.016
Cardiovascular disease 1976 (19.3%) 6073 (25.7%) −0.152 3801 (36.1%) 6061 (25.6%) 0.229 2147 (18.8%) 4395 (34.5%) −0.361
Atrial fibrillation 653 (6.4%) 2094 (8.9%) −0.093 1559 (14.8%) 2090 (8.8%) 0.186 699 (6.1%) 1751 (13.8%) −0.257
Heart failure 799 (7.8%) 2163 (9.1%) −0.048 2778 (26.4%) 2140 (9.0%) 0.467 816 (7.2%) 3036 (23.9%) −0.474
Cerebrovascular disease 578 (5.7%) 1930 (8.2%) −0.099 946 (9.0%) 1925 (8.1%) 0.031 643 (5.6%) 1108 (8.7%) −0.119
Hyperlipidemia 6423 (62.9%) 13,720 (58.0%) 0.100 7198 (68.4%) 13,701 (57.9%) 0.219 7199 (63.2%) 8710 (68.5%) −0.112
Traumatic brain injury 45 (0.4%) 166 (0.7%) −0.035 53 (0.5%) 166 (0.7%) −0.026 53 (0.5%) 54 (0.4%) 0.006
Epilepsy/seizures 92 (0.9%) 288 (1.2%) −0.031 128 (1.2%) 285 (1.2%) 0.001 93 (0.8%) 137 (1.1%) −0.027
Post‐traumatic stress disorder 78 (0.8%) 138 (0.6%) 0.022 99 (0.9%) 138 (0.6%) 0.041 86 (0.8%) 108 (0.8%) −0.011
Bipolar 125 (1.2%) 310 (1.3%) −0.008 150 (1.4%) 312 (1.3%) 0.009 144 (1.3%) 173 (1.4%) −0.008
Schizophrenia 58 (0.6%) 162 (0.7%) −0.015 63 (0.6%) 161 (0.7%) −0.010 65 (0.6%) 78 (0.6%) −0.006
Depression 1358 (13.3%) 2457 (10.4%) 0.090 1232 (11.7%) 2464 (10.4%) 0.041 1495 (13.1%) 1436 (11.3%) 0.056
Anxiety 1415 (13.9%) 2422 (10.2%) 0.111 1471 (14.0%) 2425 (10.2%) 0.115 1525 (13.4%) 1688 (13.3%) 0.003
Hypertension 7440 (72.9%) 17,227 (72.9%) 0.000 8330 (79.2%) 17,205 (72.7%) 0.151 8330 (73.1%) 10,001 (78.6%) −0.129
Chronic kidney disease 1225 (12.0%) 3469 (14.7%) −0.079 2105 (20.0%) 3461 (14.6%) 0.142 1425 (12.5%) 2439 (19.2%) −0.183
Vitamin B12 deficiency 54 (0.5%) 92 (0.4%) 0.021 55 (0.5%) 91 (0.4%) 0.021 61 (0.5%) 70 (0.6%) −0.002
Sleep disorder 2713 (26.6%) 4359 (18.4%) 0.196 2877 (27.3%) 4355 (18.4%) 0.214 2922 (25.6%) 3287 (25.8%) −0.004
Hearing impairment 404 (4.0%) 841 (3.6%) 0.021 400 (3.8%) 837 (3.5%) 0.014 457 (4.0%) 470 (3.7%) 0.016
Vision impairment 42 (0.4%) 130 (0.5%) −0.020 53 (0.5%) 130 (0.5%) −0.006 46 (0.4%) 61 (0.5%) −0.011
Alcohol use disorder 138 (1.4%) 531 (2.2%) −0.067 273 (2.6%) 527 (2.2%) 0.024 151 (1.3%) 303 (2.4%) −0.078
Obesity 3536 (34.6%) 4411 (18.7%) 0.367 5939 (56.4%) 11,794 (49.8%) 0.132 3821 (33.5%) 3589 (28.2%) 0.115
Pancreatitis 64 (0.6%) 183 (0.8%) −0.018 117 (1.1%) 182 (0.8%) 0.036 69 (0.6%) 142 (1.1%) −0.055
NAFLD 672 (6.6%) 1036 (4.4%) 0.097 654 (6.2%) 1030 (4.4%) 0.083 731 (6.4%) 751 (5.9%) 0.021
Thyroid disease 1820 (17.8%) 3560 (15.1%) 0.075 1745 (16.6%) 3575 (15.1%) 0.040 1987 (17.4%) 2068 (16.3%) 0.032
Cancer 985 (9.6%) 3005 (12.7%) −0.097 1171 (11.1%) 2994 (12.7%) −0.047 1130 (9.9%) 1412 (11.1%) −0.039
Medications
ACEIs 2351 (23.0%) 6996 (29.6%) −0.150 2476 (23.5%) 6987 (29.5%) −0.136 2884 (25.3%) 3320 (26.1%) −0.018
Beta‐blockers 2238 (21.9%) 6841 (28.9%) −0.162 3713 (35.3%) 6823 (28.8%) 0.138 2634 (23.1%) 4438 (34.9%) −0.262
Calcium channel blockers 2309 (22.6%) 6454 (27.3%) −0.108 2420 (23.0%) 6456 (27.3%) −0.099 2723 (23.9%) 3085 (24.2%) −0.008
Diuretics 2954 (28.9%) 7198 (30.4%) −0.033 4071 (38.7%) 7184 (30.4%) 0.176 3485 (30.6%) 4845 (38.1%) −0.158
Angiotensin receptor blockers 2336 (22.9%) 5101 (21.6%) 0.031 3180 (30.2%) 5090 (21.5%) 0.200 2725 (23.9%) 3857 (30.3%) −0.144
Statins 4897 (48.0%) 12,070 (51.0%) −0.062 5793 (55.0%) 12,053 (50.9%) 0.082 5805 (50.9%) 7315 (57.5%) −0.132
NSAIDS 2107 (20.6%) 4544 (19.2%) 0.036 1823 (17.3%) 4541 (19.2%) −0.048 2483 (21.8%) 2307 (18.1%) 0.092
Proton pump inhibitors 1985 (19.4%) 6019 (25.5%) −0.145 2568 (24.4%) 6022 (25.5%) −0.024 2364 (20.7%) 3113 (24.5%) −0.089
Antidepressant 1409 (13.8%) 2811 (11.9%) 0.057 1254 (11.9%) 2821 (11.9%) 0.000 1619 (14.2%) 1508 (11.9%) 0.070
Antipsychotics 386 (3.8%) 1387 (5.9%) −0.098 494 (4.7%) 1382 (5.8%) −0.051 431 (3.8%) 571 (4.5%) −0.036
Anti‐Parkinson agents 643 (6.3%) 1971 (8.3%) −0.078 797 (7.6%) 1965 (8.3%) −0.027 724 (6.4%) 901 (7.1%) −0.029
Benzodiazepines 1400 (13.7%) 4463 (18.9%) −0.140 1862 (17.7%) 4454 (18.8%) −0.029 1562 (13.7%) 2160 (17.0%) −0.091
Hormone replacement therapy 197 (1.9%) 274 (1.2%) 0.063 136 (1.3%) 274 (1.2%) 0.012 231 (2.0%) 160 (1.3%) 0.061
Oral steroids 2719 (26.6%) 6559 (27.7%) −0.025 2949 (28.0%) 6561 (27.7%) 0.007 3173 (27.8%) 3564 (28.0%) −0.004
Opioid 2166 (21.2%) 7598 (32.1%) −0.249 2821 (26.8%) 7595 (32.1%) −0.116 2486 (21.8%) 3314 (26.0%) −0.099
Aspirin 1179 (11.5%) 5002 (21.2%) −0.262 2451 (23.3%) 4994 (21.1%) 0.053 1403 (12.3%) 2843 (22.3%) −0.268
Metformin 4110 (40.2%) 10,855 (45.9%) −0.115 3932 (37.4%) 10845 (45.8%) −0.173 4956 (43.5%) 5363 (42.2%) 0.027
Insulin 3538 (34.6%) 7930 (33.5%) 0.023 3802 (23.3%) 7989 (33.8%) 0.050 3932 (34.5%) 4276 (33.6%) 0.019
SGLT2is 368 (3.6%) 319 (1.3%) 0.146 − − − − − −
GLP‐1RAs − − − 619 (5.9%) 334 (1.4%) 0.240 − − −
Other GLDs − − − − − − 1586 (13.9%) 1855 (14.6%) −0.019
HbA1c and BMI
Baseline HbA1c% 7.8 (1.6) 7.7 (1.5) 0.040 7.7 (1.5) 7.7 (1.4) −0.002 7.9 (1.6) 7.8 (1.6) 0.050
Baseline BMI 33.8 (6.8) 30.7 (6.4) 0.458 31.6 (6.5) 30.7 (6.3) 0.144 33.9 (6.9) 31.5 (6.7) 0.359
Open in a new tab

Abbreviations: ACEIs, angiotensin‐converting‐enzyme inhibitors; BMI, body mass index; GLDs, glucose‐lowering drugs; GLP‐1RAs, glucagon‐like peptide‐1 receptor agonists; HbA1c, hemoglobin A1C; HIV/AIDS, human immunodeficiency virus/acquired immunodeficiency syndrome; NAFLD, nonalcoholic fatty liver disease; NSAIDS, nonsteroidal anti‐inflammatory drugs; SGLT2is, sodium‐glucose cotransporter 2 inhibitors; SMD, standard mean difference; T2D, type 2 diabetes.

3.2. GLP‐1RA vs. other GLD cohort

GLP‐1RA initiators were generally younger (62.3 vs. 66.2 years) and more likely to be female with a higher BMI at baseline. They had higher rates of hyperlipidemia, anxiety, sleep disorders, and obesity, while having lower proportions of smoking, CVD, and anemia, compared to other GLD initiators. GLP‐1RA initiators more frequently used SGLT2i but were less likely to use metformin, antihypertensives (e.g., angiotensin‐converting enzyme inhibitors (ACEis), beta‐blockers, and calcium channel blockers), proton pump inhibitors, benzodiazepines, opioids, and antiplatelet agents.

Seventy‐five of 10,212 patients developed ADRD in the GLP‐1RA group over a mean follow‐up of 2.22 years, and 639 of 23,646 patients developed ADRD in the other GLD group over a mean follow‐up of 3.74 years. The mean time to ADRD onset was 2.59 years. In the overall cohort, GLP‐1RAs were significantly associated with a decreased risk of ADRD compared to other GLDs (RD, −1.5%; 95% CI, −1.8% to −1.2%).

We estimated the HTEs of GLP‐1RAs on the risk of ADRD; deciles of predicted ITEs of GLP‐1RAs vs. other GLDs on ADRD risk are presented in Figure 1. Among all analytic records, 26,615 (78.6%) had a decreased risk of ADRD. HTE analysis using a decision tree model identified three key covariates: CVD, CeVD, and CKD, which could be used to divide the overall cohort into four HTE subgroups (Figure 2). GLP‐1RAs showed the most substantial effect in reducing risk of ADRD in people with CVD and CeVD (RD, −4.8%; 95% CI, −6.7% to −3.0%), followed by in people with CVD but without CeVD (RD, −2.6%; 95% CI, −3.8% to −1.5%), in people without CVD but with CKD (RD, −2.6%; 95% CI, −3.6% to −1.5%), and in people without CVD or CKD (RD, −0.8%; 95% CI, −1.1% to −0.5%). The MSEs of the final models for the training set and testing sets are presented in Table S6.

FIGURE 1.

FIGURE 1

Open in a new tab

Predicted individualized treatment effect (ITE, treatment effect on person level) of GLP‐1RA vs. other GLDs (A), SGLT2is vs. other GLDs (B), and GLP‐1RA vs. SGLT2i on the risk of developing ADRD. The predicted ITE is presented as RD in the risk of ADRD between groups (Y‐axis). The predicted ITE is stratified into deciles (X‐axis). ADRD, Alzheimer's disease and related dementias; CeVD, cerebrovascular disease; CKD, chronic kidney disease; CVD, cardiovascular disease; GLDs, glucose‐lowering drugs; GLP‐1RA, glucagon‐like peptide‐1 receptor agonists; RD, risk difference; SGLT2i, sodium‐glucose cotransporter 2 inhibitors.

FIGURE 2.

FIGURE 2

Open in a new tab

A single decision tree was developed to identify the absolute risk change in risk of ADRD between GLP‐1RA and other GLDs (A), between SGLT2i and other GLDs (B), and between GLP‐1RAs and SGLT2is (C) in people with T2D. Negative values indicate reduced absolute risk of ADRD, whereas positive values indicate increased absolute risk of ADRD. ADRD, Alzheimer's disease and related dementias; GLDs, glucose‐lowering drugs; GLP‐1RA, glucagon‐like peptide‐1 receptor agonists; SGLT2i, sodium‐glucose cotransporter 2 inhibitors; T2D, type 2 diabetes.

The study findings were consistent across a range of sensitivity analyses (Figure 3). GLP‐1RA use was significantly associated with a lower risk of ADRD than other GLD use in the following sensitivity analysis: (1) excluding those with MCI at baseline (RD, −1.4%; 95% CI, −1.8% to −1.1%); (2) excluding those with PD at baseline (RD, −1.4%; 95% CI, −1.8% to −1.1%); (3) excluding those having a diagnosis of ADRD within 0.5 years after index date (RD, −1.2%; 95% CI, −1.5% to −0.9%). The result was consistent even after accounting for the competing risk of death (hazard difference, −0.14%, 95% CI, −0.27% to −0.01%).

FIGURE 3.

FIGURE 3

Open in a new tab

Sensitivity analyses of absolute RD for ADRD for GLP‐1RA vs. other GLDs(A), SGLT2i vs. other GLDs(B), and GLP‐1RA vs. SGLT2is(C) in the overall population and subgroups identified using a single decision tree model. *no HTE subgroups were identified. ADRD, Alzheimer's disease and related dementias; CI, confidence interval; CKD, chronic kidney disease; CeVD, cerebrovascular disease; GLDs, glucose‐lowering drugs; RD, risk difference; SGLT2is, sodium‐glucose cotransporter 2 inhibitors.

3.3. SGLT2i vs. other GLD cohorts

In the SGLT2i vs. other GLDs cohort (Table 1), SGLT2i initiators were generally younger (64.8 vs. 66.2 years) with a higher BMI (31.6 vs. 30.7) at baseline. They had higher rates of PVD, CVD, atrial fibrillation, heart failure, hyperlipidemia, anxiety, hypertension, CKD, sleep disorders, and obesity. SGLT2i initiators more frequently used beta‐blockers, diuretics, angiotensin receptor blockers, non‐aspirin antiplatelet agents, and GLP‐1RAs, but were less likely to use ACEis, opioids, and direct oral anticoagulants.

One hundred and one of 10,524 patients developed ADRD in the SGLT2i group over a mean follow‐up of 1.95 years, and 642 of 23,661 patients developed ADRD in the other GLD group over a mean follow‐up of 3.76 years. The mean time to ADRD onset was 2.44 years. In the overall cohort, SGLT2is were significantly associated with a decreased risk of ADRD compared to other GLDs (RD, −1.7%; 95% CI, −2.0% to −1.4%).

We estimated the HTEs of SGLT2i on the risk of ADRD; deciles of predicted ITEs of GLP‐1RAs vs. other GLDs on ADRD risk are presented in Figure 1. Among all analytic records, 28,510 (83.4%) had a decreased risk of ADRD. HTE analysis using a single decision tree model identified three key covariates: CVD, CeVD, and Hispanic ethnicity, which could be used to divide the overall cohort into four HTE subgroups (Figure 2). SGLT2i showed the most substantial effect in reducing risk ADRD in the subgroup with CVD and CeVD(RD, −4.6%; 95% CI, −6.0% to −3.3%), followed by in the subgroup with CVD but without CeVD (RD, −2.6%; 95% CI, −3.5% to −1.8%), in Hispanic patients without CVD (RD, −2.1%; 95% CI, −2.9% to −1.3%), and in non‐Hispanic patients without CVD (RD, −1.0%; 95% CI, −1.3% to −0.6%). The MSEs of the final models for the training set and testing sets are presented in Table S6.

GLP‐1RA use was significantly associated with a lower risk of ADRD than other GLD in the following sensitivity analysis: (1) excluding those with MCI at baseline (RD, −1.7%; 95% CI, −2.0% to −1.4%); (2) excluding those with PD at baseline (RD, −1.7%; 95% CI, −2.0% to −1.4%); (3) excluding those having a diagnosis of ADRD within 0.5 years after the index date (RD, −1.6%; 95% CI, −1.9% to −1.3%)(Figure 3). However, osteoporosis rather than Hispanic ethnicity was identified as an important covariate to subgroup division when excluding those with PD at baseline and excluding those having a diagnosis of ADRD within 0.5 years after the index date. The result was consistent even after accounting for the competing risk of death (hazard difference, −0.06%, 95% CI, −0.17% to 0.05%).

3.4. GLP‐1RA vs. SGLT2i cohort

In the GLP‐1RA vs. SGLT2i cohort (Table 1), GLP‐1RA initiators were generally younger (62.5 vs. 64.9 years) and more likely to be female (58.7% vs. 45.4%) with a higher BMI (33.9 vs. 31.5) at baseline. They had higher rates of obesity while having lower proportions of smoking, PVD, CVD, atrial fibrillation, heart failure, CeVD, hypertension, chronic obstructive pulmonary disease, CKD, anemia, and benign prostatic hyperplasia, compared to SGLT2I initiators. GLP‐1RA initiators were less likely to use beta‐blockers, diuretics, angiotensin receptor blockers, statins, anticoagulants, and antiplatelet agents.

Ninety of 11,395 patients developed ADRD in the GLP‐1RA group over a mean follow‐up of 2.39 years, and 130 of 12,722 patients developed ADRD in the SGLT2i group over a mean follow‐up of 2.07 years. The mean time to ADRD onset was 2.02 years. In the overall cohort, there was no significant difference between GLP‐1RAs and SGLT2is regarding the risk of ADRD (RD, 0.06%; 95 %CI, −0.20% to 0.32%).

We estimated the HTEs of GLP‐1RAs on the risk of ADRD compared to SGLT2is; deciles of predicted ITEs of GLP‐1RAs vs. SGLT2is on ADRD risk are presented in Figure 1. Among all analytic records, 12,028 (49.9%) had a decreased risk of ADRD. HTE analysis using a decision tree model identified three key covariates: obesity, Black people, and CKD, which could be used to divide the overall cohort into four HTE subgroups (Figure 2). GLP‐1RAs were significantly associated with a decreased risk of ADRD in the patients without obesity but with CKD (RD, −0.9%; 95% CI, −1.6% to −0.1%), but increased risk in the non‐Black patients with obesity (RD, 0.4%; 95% CI, 0.0% to 0.8%), as compared to SGLT2is. There was no significant difference in the other two subgroups: Black patients with obesity (RD, 0.0%; 95% CI, −0.3% to 0.4%) and patients without obesity or CKD (RD, −0.2%; 95% CI, −0.6% to 0.3%). The MSEs of the final models for the training set and testing sets are presented in Table S6.

GLP‐1RA use was significantly associated with a lower risk of ADRD than SGLT2i use in the following sensitivity analysis: (1) excluding those with MCI at baseline (RD, 0.08%; 95% CI, −0.19% to 0.35%); (2) excluding those with PD at baseline (RD, 0.06%; 95% CI, −0.21% to 0.34%); (3) excluding those having a diagnosis of ADRD within 0.5 years after the index date (RD, 0.25%; 95% CI, −0.01% to 0.49%) (Figure 3). The result was consistent even after accounting for the competing risk of death (hazard difference, −0.48%, 95% CI, −0.65% to −0.31%).

4. DISCUSSION

This target trial emulation using real‐world data from the OneFlorida+ provides compelling evidence that initiating GLP‐1RAs or SGLT2is is associated with a reduced risk of ADRD compared to other GLDs in adults with T2D. Our primary analysis revealed that GLP‐1RA initiation was associated with a 1.5% absolute risk reduction in ADRD compared to other GLDs, while SGLT2i initiation was associated with a 1.7% risk reduction, indicating approximately 15 and 17 fewer cases of ADRD per 1000 patients treated with GLP‐1RAs and SGLT2i than other GLDs over a follow‐up period of about 3 years, respectively. Given the devastating burden of ADRD and the lack of effective preventive therapies, even a modest absolute risk reduction could be clinically meaningful. We identified CVD, CeVD, CKD (for GLP‐1RA), and Hispanic people (for SGLT2i) as key covariates for HTE subgroup divisions. While no significant difference in risk of ADRD was observed between GLP‐1RA and SGLT2i initiators, three key covariates—obesity, CKD, and Black race—were identified for HTE subgroup divisions. GLP‐1RAs had a decreased risk in nonobese patients with CKD but increased risk in the non‐Black obese patients compared to SGLT2is.

These findings align with and extend previous observational studies suggesting a potential protective effect of GLP‐1RAs and SGLT2is on reducing the risk of ADRD. 12 , 27 Our recent meta‐analysis of observational studies reported RRs of 0.62 (95% CI, 0.39–0.97) for SGLT2i users and 0.72 (95% CI, 0.54‐0.97) for GLP‐1RA users compared to nonusers regarding ADRD risk. 13 Our study provides more robust evidence by using a target trial emulation and advanced causal inference methods to minimize confounding. Several potential mechanisms could explain the protective effects of GLP‐1RAs and SGLT2is on cognitive function. Both drug classes have demonstrated neuroprotective and anti‐inflammatory properties in preclinical studies. 28 , 29 GLP‐1RAs may reduce amyloid‐β accumulation and tau phosphorylation, key pathological hallmarks of AD. 30 , 31 They may also enhance neurogenesis and synaptic plasticity. 32 , 33 , 34 SGLT2is may improve cerebral blood flow, reduce oxidative stress, 35 and enhance ketone body production, which could serve as an alternative energy source for the brain. 36 , 37 Additionally, the cardiovascular and metabolic benefits of GLP‐1RAs and SGLT2is, including improved glycemic control, weight loss, and blood pressure reduction, 6 , 38 , 39 may contribute to their cognitive effects by reducing vascular risk factors associated with cognitive decline. 2 , 40 , 41

Importantly, we identified significant heterogeneity in treatment effects, with certain subgroups deriving greater benefits from GLP‐1RAs. This HTE provided crucial insights into which patients might benefit most from GLP‐1RAs in terms of ADRD risk reduction. We identified three key covariates—CVD, CeVD, and CKD—that stratified the population into four subgroups with varying levels of benefits. The most substantial risk reduction was observed in people with both CVD and CeVD, with a 4.8% absolute risk reduction for GLP‐1RA initiators, while the least benefit was observed in those without CVD or CKD, with an RD of −0.8%. This finding is particularly important as it suggests that patients with the highest cardiovascular risk may derive the greatest cognitive benefit from GLP‐1RAs. The strong association of CVD and CeVD with ADRD suggests that the cognitive benefits of GLP‐1RAs and SGLT2is may be largely mediated through vascular mechanisms. This aligns with the established role of CVD and CeVD as risk factors for ADRD. 2 , 40 , 41 The patients with CVD but without CeVD (RD, −2.6%) and those without CVD but with CKD (RD, −2.6%) have similar benefits from GLP‐1RAs. Poor and declining kidney function is an important risk factor for ADRD. 42 , 43 GLP‐1RAs are associated with decreased risk of kidney outcomes, 44 potentially leading to reduced risk of ADRD.

Similarly, when comparing SGLT2is with other GLDs, we identified three key covariates—CVD, CeVD, and Hispanic ethnicity—that were applied to divide the population into four subgroups. The most substantial risk reduction was observed in people with both CVD and CeVD (RD, −4.6%), while the least benefit was observed in non‐Hispanic people without CVD, which may be explained by the improved cardiovascular and cerebrovascular outcomes with SGLT2is. 6 , 45 Hispanic ethnicity was identified as an important covariate for subgroup division, consistent with our previous study. 14 Hispanic individuals are more likely to have risk factors for dementia, such as CVD, hypertension, and diabetes. 46 Our previous meta‐analysis of randomized outcome trials showed that compared with non‐Hispanic people, Hispanic people with T2D appeared to obtain a greater benefit of lowered major adverse cardiovascular events with SGLT2is, which may partly explain greater benefits from SGLT2is on reducing the risk of ADRD in Hispanic individuals.

We did not find a significant overall difference in the risk of ADRD between users of GLP‐1RAs and SGLT2is. However, our HTE analysis revealed important subgroup differences. GLP‐1RAs were associated with a decreased risk of ADRD compared to SGLT2is in the subgroup of patients without obesity but with CKD. In contrast, GLP‐1RAs were associated with a slightly increased risk of ADRD compared to SGLT2is in non‐Black people with obesity. The mechanisms underlying these HTEs are not fully understood. Recent evidence suggests that GLP‐1RAs, such as semaglutide, could induce widespread changes in the circulating proteome, affecting pathways related to inflammation, lipid metabolism, adipose tissue function, and cardiovascular health, independent of weight loss and glycemic control. 47 These pleiotropic effects may interact differently with underlying comorbidities, such as CKD and obesity, potentially leading to variation in cognitive outcomes. Additionally, evidence suggests that late‐life obesity may be associated with a decreased risk of ADRD, whereas weight loss may increase the risk. 48 , 49 Since GLP‐1RAs generally induce greater weight loss than SGLT2is, 50 particularly among White individuals, this may partly explain the observed subgroup differences. 51 Furthermore, although both drug classes improve kidney outcomes, SGLT2is have shown superior effects in preserving renal function, 52 , 53 which may also influence cognitive trajectories. These findings highlight the complex interplay among obesity, kidney function, drug‐specific effects, and potential racial differences, underscoring the need for further research to optimize personalized strategies for ADRD prevention.

Our study has several strengths, including the large sample size, the use of real‐world data from a diverse population with high external validity and representative of actual clinical practice, and the application of sophisticated statistical techniques, including target trial emulation and doubly robust learning, strengthens the robustness of our causal inferences. The HTE analysis provides crucial insights into subgroup‐specific benefits, contributing significantly to the field of personalized medicine in diabetes management. Multiple sensitivity analyses were conducted to test the robustness of our findings. However, several limitations should be noted. First, despite our rigorous methodology, as an observational study, we cannot completely rule out the possibility of residual confounding from unmeasured variables, such as social economic and genetic factors. Second, our reliance on diagnostic codes for identifying ADRD cases may have led to misclassification. This method may not capture all cases, especially those in early stages, and cannot differentiate between subtypes of ADRD. Third, the relatively short follow‐up duration (ranging from 1.95 to 3.76 years) in our study may limit our ability to fully capture the long‐term effects of GLP‐1RAs and SGLT2is on the risk of ADRD. Future studies with extended follow‐up periods are necessary to better understand the sustained impact of these treatments over time and to assess any potential delayed effects on ADRD development. Fourth, we did not account for variations in drug dosing, adherence, or cumulative treatment duration, which could impact the risk of ADRD. Detailed information on GLP‐1RA and SGLT2i dose intensity and longitudinal exposure patterns was not consistently available in our dataset. Future studies incorporating detailed dose–response analyses are warranted to better understand these relationships. Fifth, while selecting users of other GLDs as the comparator group helped mitigate confounding by indication, the heterogeneity within this group remains a limitation. Different GLDs may have varying baseline risks for ADRD and other clinical outcomes, potentially introducing residual confounding. More importantly, as other GLDs have been recommended as third‐forth line choice following the introduction of GLP‐1RAs and SGLT2is, 19 differences in patient characteristics and treatment decisions may further impact the comparability of groups. Sixth, while our sample was diverse and identified from OneFlorida+, findings may not be fully generalizable to populations with significantly different characteristics (e.g., those without T2D) or healthcare systems.

The HTEs of GLP‐1RAs and SGLT2is on risk of ADRD have significant clinical implications for T2D management. These findings can inform more tailored treatment strategies. Clinicians could prioritize GLP‐1RAs or SGLT2is for patients who fall into the high‐benefit subgroups, potentially offering dual benefits of glycemic control and ADRD risk reduction. The identified key covariates—CVD, CeVD, CKD, Hispanic ethnicity—could be incorporated into risk stratification tools to identify patients most likely to benefit cognitively from these medications. Patients with CVD and CeVD, in particular, should be considered for GLP‐1RA or SGLT2i prescriptions with regard to ADRD risk. Future research should include RCTs specifically designed to assess the cognitive effects of GLP‐1RAs and SGLT2is, with longer follow‐up periods, and investigations into the underlying mechanisms of the observed effects. Furthermore, exploring the potential cognitive benefits of these medications in nondiabetic populations at high risk for ADRD would be valuable.

In conclusion, this study provides strong evidence that initiation of GLP‐1RAs and SGLT2is is associated with a reduced risk of ADRD in adults with T2D, with certain subgroups deriving greater benefits. While no significant difference was observed between GLP‐1RA and SGLT2i initiators, the risk of ADRD varies among different subgroups. These findings support the potential repurposing of these medications as a strategy for ADRD prevention and highlight the importance of personalized treatment approaches in diabetes management. However, further research, including RCTs, is needed to confirm these findings and establish GLP‐1RAs and SGLT2is as effective interventions for maintaining cognitive health in individuals with T2D.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

This study was approved by the University of Florida Institutional Review Board (IRB202201196). This is a secondary analysis and obtaining informed consent for this study was not necessary.

Supporting information

Supporting Information

ALZ-21-e70313-s002.docx (661.3KB, docx)

Supporting Information

ALZ-21-e70313-s001.pdf (1,001.2KB, pdf)

ACKNOWLEDGMENTS

This work was supported by The Pharmaceutical Research and Manufacturers of America (PhRMA) Foundation Predoctoral Fellowship (2023PDVH1064032), National Institutes of Health (NIH)/National Institute on Aging (NIA) (R01AG089445 and R01AG076234), and NIH/National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (R01DK133465).

Tang H, Donahoo WT, DeKosky ST, et al. Heterogeneous treatment effects of GLP‐1RAs and SGLT2is on risk of Alzheimer's disease and related dementia in patients with type 2 diabetes: Insights from a real‐world target trial emulation. Alzheimer's Dement. 2025;21:e70313. 10.1002/alz.70313

REFERENCES

  • 1. Khan MAB, Hashim MJ, King JK, Govender RD, Mustafa H, Kaabi JA. Epidemiology of type 2 diabetes – Global Burden of Disease And Forecasted Trends. J Epidemiol Glob Health. 2020;10(1):107. doi: 10.2991/jegh.k.191028.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Livingston G, Huntley J, Liu KY, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet Standing Commission. Lancet. 2024;404(10452):572‐628. doi: 10.1016/S0140-6736(24)01296-0 [DOI] [PubMed] [Google Scholar]
  • 3. Kale MB, Bhondge HM, Wankhede NL, et al. Navigating the intersection: diabetes and Alzheimer's intertwined relationship. Ageing Res Rev. 2024;100:102415. doi: 10.1016/j.arr.2024.102415 [DOI] [PubMed] [Google Scholar]
  • 4. Braunwald E. Gliflozins in the management of cardiovascular disease. N Engl J Med. 2022;386(21):2024‐2034. doi: 10.1056/NEJMra2115011 [DOI] [PubMed] [Google Scholar]
  • 5. Heerspink HJL, Perkins BA, Fitchett DH, Husain M, Cherney DZI. Sodium glucose cotransporter 2 inhibitors in the treatment of diabetes mellitus: cardiovascular and kidney effects, potential mechanisms, and clinical applications. Circulation. 2016;134(10):752‐772. doi: 10.1161/CIRCULATIONAHA.116.021887 [DOI] [PubMed] [Google Scholar]
  • 6. McGuire DK, Shih WJ, Cosentino F, et al. Association of SGLT2 inhibitors with cardiovascular and kidney outcomes in patients with type 2 diabetes: a meta‐analysis. JAMA Cardiol. 2021;6(2):148‐158. doi: 10.1001/jamacardio.2020.4511 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. He D, Aleksic S. Is it time to repurpose geroprotective diabetes medications for prevention of dementia? J Am Geriatr Soc. 2023;71(7):2041‐2045. doi: 10.1111/jgs.18405 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Bello‐Chavolla OY, Antonio‐Villa NE, Vargas‐Vázquez A, Ávila‐Funes JA, Aguilar‐Salinas CA. Pathophysiological mechanisms linking type 2 diabetes and dementia: review of evidence from clinical, translational and epidemiological research. Curr Diabetes Rev. 2019;15(6):456‐470. doi: 10.2174/1573399815666190129155654 [DOI] [PubMed] [Google Scholar]
  • 9. Mui JV, Zhou J, Lee S, et al. Sodium‐glucose cotransporter 2 (sglt2) inhibitors vs. dipeptidyl peptidase‐4 (dpp4) inhibitors for new‐onset dementia: a propensity score‐matched population‐based study with competing risk analysis. Front Cardiovasc Med. 2021;8:747620. doi: 10.3389/fcvm.2021.747620 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Wium‐Andersen IK, Osler M, Jørgensen MB, Rungby J, Wium‐Andersen MK. Antidiabetic medication and risk of dementia in patients with type 2 diabetes: a nested case‐control study. Eur J Endocrinol. 2019;181(5):499‐507. doi: 10.1530/EJE-19-0259 [DOI] [PubMed] [Google Scholar]
  • 11. Wu CY, Iskander C, Wang C, et al. Association of sodium‐glucose cotransporter 2 inhibitors with time to dementia: a population‐based cohort study. Diabetes Care. 2023;46(2):297‐304. doi: 10.2337/dc22-1705 [DOI] [PubMed] [Google Scholar]
  • 12. Tang H, Donahoo WT, DeKosky ST, et al. GLP‐1RA and SGLT2i medications for type 2 diabetes and Alzheimer disease and related dementias. JAMA Neurol. 2025;82(5):439‐449. doi: 10.1001/jamaneurol.2025.0353 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Tang H, Shao H, Shaaban CE, et al. Newer glucose‐lowering drugs and risk of dementia: a systematic review and meta‐analysis of observational studies. J Am Geriatr Soc;71(7):2096‐2106. doi: 10.1111/jgs.18306 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Tang H, Donahoo WT, Svensson M, et al. Heterogeneous treatment effects of sodium‐glucose cotransporter 2 inhibitors on risk of dementia in people with type 2 diabetes: a population‐based cohort study. Alzheimers Dement;20(8):5528‐5539. doi: 10.1002/alz.14048 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183(8):758‐764. doi: 10.1093/aje/kwv254 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Hernán MA, Wang W, Leaf DE. Target trial emulation: a framework for causal inference from observational data. JAMA. 2022;328(24):2446‐2447. doi: 10.1001/jama.2022.21383 [DOI] [PubMed] [Google Scholar]
  • 17. Hogan WR, Shenkman EA, Robinson T, et al. The OneFlorida data trust: a centralized, translational research data infrastructure of statewide scope. J Am Med Inform Assoc. 2022;29(4):686‐693. doi: 10.1093/jamia/ocab221 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. OneFlorida+Clinical Research Network . Data summary–OneFlorida+. Accessed August 15, 2024. https://onefloridaconsortium.org/data/
  • 19. ElSayed NA, Aleppo G, Aroda VR, et al. 9. Pharmacologic approaches to glycemic treatment: standards of care in diabetes‐2023. Diabetes Care. 2023;46(Suppl 1):S140‐S157. doi: 10.2337/dc23-S009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Chronic Conditions . Chronic Conditions Data Warehouse. Accessed August 16, 2024. https://www2.ccwdata.org/condition‐categories‐chronic
  • 21. Li Q, Yang X, Xu J, et al. Early prediction of Alzheimer's disease and related dementias using real‐world electronic health records. Alzheimers Dement J Alzheimers Assoc. 2023;19(8):3506‐3518. doi: 10.1002/alz.12967 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Aschwanden D, Aichele S, Ghisletta P, et al. Predicting cognitive impairment and dementia: a machine learning approach. J Alzheimers Dis JAD. 2020;75(3):717‐728. doi: 10.3233/JAD-190967 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Jakobsen JC, Gluud C, Wetterslev J, Winkel P. When and how should multiple imputation be used for handling missing data in randomised clinical trials—a practical guide with flowcharts. BMC Med Res Methodol. 2017;17(1):162. doi: 10.1186/s12874-017-0442-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity‐score matched samples. Stat Med. 2009;28(25):3083‐3107. doi: 10.1002/sim.3697 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Doubly Robust Learning—econml 0.14.0 documentation. Accessed June 16, 2023. https://econml.azurewebsites.net/spec/estimation/dr.html
  • 26. Rava D, Xu R. Doubly robust estimation of the hazard difference for competing risks data. Stat Med. 2023. doi: 10.1002/sim.9644. Published online January 3, 2023. [DOI] [PubMed] [Google Scholar]
  • 27. Tang H, Donahoo WT, DeKosky ST, et al. Glycated hemoglobin and body mass index as mediators of GLP‐1RAs and Alzheimer's disease and related dementias in patients with type 2 diabetes. Alzheimers Dement. 2025;21(4):e70161. doi: 10.1002/alz.70161 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Diz‐Chaves Y, Mastoor Z, Spuch C, González‐Matías LC, Mallo F. Anti‐Inflammatory effects of glp‐1 receptor activation in the brain in neurodegenerative diseases. Int J Mol Sci. 2022;23(17):9583. doi: 10.3390/ijms23179583 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Yoon G, Kim YK, Song J. Glucagon‐like peptide‐1 suppresses neuroinflammation and improves neural structure. Pharmacol Res. 2020;152:104615. doi: 10.1016/j.phrs.2019.104615 [DOI] [PubMed] [Google Scholar]
  • 30. Li Y, Duffy KB, Ottinger MA, et al. GLP‐1 receptor stimulation reduces amyloid‐β peptide accumulation and cytotoxicity in cellular and animal models of Alzheimer's disease. J Alzheimers Dis JAD. 2010;19(4):1205‐1219. doi: 10.3233/JAD-2010-1314 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Hansen HH, Barkholt P, Fabricius K, et al. The GLP‐1 receptor agonist liraglutide reduces pathology‐specific tau phosphorylation and improves motor function in a transgenic hTauP301L mouse model of tauopathy. Brain Res. 2016;1634:158‐170. doi: 10.1016/j.brainres.2015.12.052 [DOI] [PubMed] [Google Scholar]
  • 32. McClean PL, Gault VA, Harriott P, Hölscher C. Glucagon‐like peptide‐1 analogues enhance synaptic plasticity in the brain: a link between diabetes and Alzheimer's disease. Eur J Pharmacol. 2010;630(1‐3):158‐162. doi: 10.1016/j.ejphar.2009.12.023 [DOI] [PubMed] [Google Scholar]
  • 33. McClean PL, Hölscher C. Liraglutide can reverse memory impairment, synaptic loss and reduce plaque load in aged APP/PS1 mice, a model of Alzheimer's disease. Neuropharmacology. 2014;76(Pt A):57‐67. doi: 10.1016/j.neuropharm.2013.08.005 [DOI] [PubMed] [Google Scholar]
  • 34. Abbas T, Faivre E, Hölscher C. Impairment of synaptic plasticity and memory formation in GLP‐1 receptor KO mice: interaction between type 2 diabetes and Alzheimer's disease. Behav Brain Res. 2009;205(1):265‐271. doi: 10.1016/j.bbr.2009.06.035 [DOI] [PubMed] [Google Scholar]
  • 35. Hasan R, Lasker S, Hasan A, et al. Canagliflozin ameliorates renal oxidative stress and inflammation by stimulating AMPK‐Akt‐ENOS pathway in the isoprenaline‐induced oxidative stress model. Sci Rep. 2020;10(1):14659. doi: 10.1038/s41598-020-71599-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36. Pawlos A, Broncel M, Woźniak E, Gorzelak‐Pabiś P. Neuroprotective effect of SGLT2 inhibitors. Molecules. 2021;26(23):7213. doi: 10.3390/molecules26237213 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Tharmaraja T, Ho JSY, Sia CH, et al. Sodium‐glucose cotransporter 2 inhibitors and neurological disorders: a scoping review. Ther Adv Chronic Dis. 2022;13:20406223221086996. doi: 10.1177/20406223221086996 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38. Nelson AJ, Pagidipati NJ, Aroda VR, et al. Incorporating SGLT2i and GLP‐1RA for cardiovascular and kidney disease risk reduction: call for action to the cardiology community. Circulation. 2021;144(1):74‐84. doi: 10.1161/CIRCULATIONAHA.121.053766 [DOI] [PubMed] [Google Scholar]
  • 39. Kristensen SL, Rørth R, Jhund PS, et al. Cardiovascular, mortality, and kidney outcomes with GLP‐1 receptor agonists in patients with type 2 diabetes: a systematic review and meta‐analysis of cardiovascular outcome trials. Lancet Diabetes Endocrinol. 2019;7(10):776‐785. doi: 10.1016/S2213-8587(19)30249-9 [DOI] [PubMed] [Google Scholar]
  • 40. Knopman D, Boland LL, Mosley T, et al. Cardiovascular risk factors and cognitive decline in middle‐aged adults. Neurology. 2001;56(1):42‐48. doi: 10.1212/WNL.56.1.42 [DOI] [PubMed] [Google Scholar]
  • 41. Gorelick PB, Scuteri A, Black SE, et al. Vascular contributions to cognitive impairment and dementia. Stroke J Cereb Circ. 2011;42(9):2672‐2713. doi: 10.1161/STR.0b013e3182299496 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Singh‐Manoux A, Oumarou‐Ibrahim A, Machado‐Fragua MD, et al. Association between kidney function and incidence of dementia: 10‐year follow‐up of the Whitehall II cohort study. Age Ageing. 2022;51(1):afab259. doi: 10.1093/ageing/afab259 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Xu H, Garcia‐Ptacek S, Trevisan M, et al. Kidney function, kidney function decline, and the risk of dementia in older adults. Neurology. 2021;96(24):e2956‐e2965. doi: 10.1212/WNL.0000000000012113 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Granata A, Maccarrone R, Anzaldi M, et al. GLP‐1 receptor agonists and renal outcomes in patients with diabetes mellitus type 2 and diabetic kidney disease: state of the art. Clin Kidney J. 2022;15(9):1657‐1665. doi: 10.1093/ckj/sfac069 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45. Proietti R, Rivera‐Caravaca JM, López‐Gálvez R, et al. Cerebrovascular, cognitive and cardiac benefits of SGLT2 inhibitors therapy in patients with atrial fibrillation and type 2 diabetes mellitus: results from a global federated health network analysis. J Clin Med. 2023;12(8):2814. doi: 10.3390/jcm12082814 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Gomez S, Blumer V, Rodriguez F. Unique cardiovascular disease risk factors in Hispanic individuals. Curr Cardiovasc Risk Rep. 2022;16(7):53‐61. doi: 10.1007/s12170-022-00692-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47. Maretty L, Gill D, Simonsen L, et al. Proteomic changes upon treatment with semaglutide in individuals with obesity. Nat Med. 2025;31(1):267‐277. doi: 10.1038/s41591-024-03355-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Emmerzaal TL, Kiliaan AJ, Gustafson DR. 2003‐2013: a decade of body mass index, Alzheimer's disease, and dementia. J Alzheimers Dis. 2015;43(3):739‐755. doi: 10.3233/JAD-141086 [DOI] [PubMed] [Google Scholar]
  • 49. Li J, Joshi P, Ang TFA, et al. Mid‐ to late‐life body mass index and dementia risk: 38 years of follow‐up of the Framingham study. Am J Epidemiol. 2021;190(12):2503‐2510. doi: 10.1093/aje/kwab096 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50. Ma H, Lin YH, Dai LZ, Lin CS, Huang Y, Liu SY. Efficacy and safety of GLP‐1 receptor agonists versus SGLT‐2 inhibitors in overweight/obese patients with or without diabetes mellitus: a systematic review and network meta‐analysis. BMJ Open. 2023;13(3):e061807. doi: 10.1136/bmjopen-2022-061807 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51. Rubino D, Angelene H, Fabricatore A, Ard J. Efficacy and safety of semaglutide 2.4 mg by race and ethnicity: a post hoc analysis of three randomized controlled trials. Obes Silver Spring Md. 2024;32(7):1268‐1280. doi: 10.1002/oby.24042 [DOI] [PubMed] [Google Scholar]
  • 52. Lui DTW, Au ICH, Tang EHM, et al. Kidney outcomes associated with sodium‐glucose cotransporter 2 inhibitors versus glucagon‐like peptide 1 receptor agonists: a real‐world population‐based analysis. eClinicalMedicine. 2022;50:101510. doi: 10.1016/j.eclinm.2022.101510 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53. Sohn M, Nam S, Nauck MA, Lim S. Long‐term comparison of renal and metabolic outcomes after sodium–glucose co‐transporter 2 inhibitor or glucagon‐like peptide‐1 receptor agonist therapy in type 2 diabetes. BMC Med. 2024;22:273. doi: 10.1186/s12916-024-03483-z [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supporting Information

ALZ-21-e70313-s002.docx (661.3KB, docx)

Supporting Information

ALZ-21-e70313-s001.pdf (1,001.2KB, pdf)

Articles from Alzheimer's & Dementia are provided here courtesy of Wiley

RESOURCES