Abstract
OBJECTIVE
Glucagon-like peptide 1 agonists (GLP-1s) compared with dipeptidyl peptidase 4 inhibitors (DPP-4s) are associated with reduced risk of dementia in the general population with diabetes, but whether this association is true for patients requiring hemodialysis is unknown.
RESEARCH DESIGN AND METHODS
Using the U.S. Renal Data System and Medicare Parts A, B, and D claims data from 2011 to 2021, we used the active comparator, new-user design to evaluate incident dementia comparing GLP-1s versus DPP-4s among individuals with both diabetes and hemodialysis dependence. We used inverse probability of treatment weights (IPTW) to balance baseline characteristics and Fine-Gray models to estimate subdistribution hazard ratios (sHRs) accounting for competing risks of death and kidney transplantation. We estimated intention-to-treat and as-treated effects.
RESULTS
We identified 3,619 GLP-1 users and 11,502 DPP-4 users. After IPTW, the average individual was 63 years old, 63% were White, and mean BMI was 31 kg/m2. The median (interquartile interval) follow-up was 1.5 (0.6–2.9) years, and 2,014 patients received a dementia diagnosis. In the intention-to-treat analysis, the IPTW-sHR for dementia was 0.82 (95% CI 0.67–0.98), and after 2 years of follow-up, the cumulative incidence of dementia was 10.2% on GLP-1s vs 11.2% on DPP-4s. As-treated and subgroup analyses were consistent. GLP-1s were also associated with an increased risk of ketoacidosis (sHR 1.52, 95% CI 1.14–2.02; 2-year cumulative incidence: 3.1% vs. 2.2%).
CONCLUSIONS
In patients with diabetes requiring hemodialysis, GLP-1s (vs. DPP-4s) may be a promising therapy to reduce the risk of dementia.
Graphical Abstract
Introduction
Alzheimer disease and related dementias (ADRDs) are life-limiting illnesses with significant comorbidity and mortality (1), and for patients with end-stage kidney disease (ESKD), ADRD diagnoses are associated with increased risk of hospitalization, dialysis withdrawal, and mortality (2). Compared with the general population, the incidence of ADRDs is higher in patients with ESKD (3), and the increased disease burden may be related to shared kidney and brain susceptibility to vascular damage, inflammation, atherogenesis, and oxidative stress (4). Glucagon-like peptide 1 agonists (GLP-1s) are a class of medications used to treat kidney disease, diabetes, obesity, and cardiovascular disease (5). Given encouraging evidence of a protective effect against ADRDs among patients with diabetes in observational studies (6,7), multiple randomized clinical trials evaluating GLP-1 use and ADRD reduction are ongoing to confirm these findings (8–10). Whether the results of these clinical trials will generalize to patients requiring dialysis, however, warrants evaluation.
For patients requiring dialysis, GLP-1s are recommended as a first-line glucose-lowering medication by the combined American Diabetes Association (ADA) and the Kidney Disease Improving Global Outcomes (KDIGO) guidelines (5), but GLP-1 use remains limited (11,12). Dipeptidyl peptidase 4 inhibitors (DPP-4s), which are another class of glucose-lowering medication, have higher use in this population (12). Studies evaluating the comparative effectiveness of different diabetes management strategies may help shift clinical practice and provide additional support for guideline recommendations. Furthermore, the comparative safety of these medications needs to be evaluated. GLP-1s have been associated with risk of ketoacidosis in postmarketing surveillance in the general population (13,14), but the absolute risk is unclear. Two observational studies have evaluated GLP-1s versus DPP-4s in patients with advanced kidney disease (15,16), and GLP-1s were shown to be associated with decreased mortality. ADRD outcomes, however, were not assessed.
Given the recommendations to use GLP-1s (5), limited safety data, and potential benefit of GLP-1s against ADRDs onset, we used the United State Renal Data System (USRDS) database (11) to emulate a target trial (17) evaluating the effectiveness and safety of GLP-1s versus DPP-4s on incident ADRDs among patients with diabetes requiring hemodialysis. We also quantified the risk of ketoacidosis.
Research Design and Methods
Data Source
We used Medicare Part A, B, and D claims files from the USRDS database (11), which is a national surveillance system that follows U.S. patients requiring dialysis or kidney transplantation. Medicare Part A contains hospitalizations; Part B contains physician inpatient and outpatient claims; and Part D contains medication dispense data. Medicare claims can then be combined with USRDS data related to dialysis initiation, treatment, and clinical outcomes.
The Jefferson Health Institutional Review Board (Philadelphia, PA) approved the study and waived the need for informed consent given use of a limited data set. We followed the International Society for Pharmacoeconomics and Outcomes Research (IPSOR) reporting guidelines for comparative effectiveness research.
Emulating a Target Trial
Eligibility Criteria and Study Population
We emulated a hypothetical target trial comparing new users of GLP-1s versus new users of DPP-4s among patients with diabetes requiring in-center hemodialysis (Supplementary Table 1). Enrollment started 1 January 1 2011, and participants were eligible at treatment strategy initiation (i.e., baseline) if they were aged ≥18 years, survived ≥90 days on in-center hemodialysis (according to USRDS recommendations to ensure stable dialysis treatment) (11), and without a GLP-1 or DPP-4 dispense in the previous 180 days. Exclusion criteria at enrollment included off in-center hemodialysis (e.g., renal recovery, renal transplant, transition to other dialysis modality, or unconfirmed dialysis modality), <180 days of continuous Medicare Part A, B, and D primary payer, and missing BMI or KT/V (dialysis clearance ∗ time/volume) from either the Centers for Medicare & Medicaid Services (CMS) Medical Evidence Form 2728 or monthly dialysis claims files. We additionally excluded individuals with a history of dementia or those without type 2 diabetes (both were defined via ICD claims in the previous 180 days) (Supplementary Table 2). Figure 1 shows how the analytic cohort was constructed.
Figure 1.
Study flow diagram.
Treatment Strategies
We compared new users of GLP-1s with new users of DPP-4s between 1 January 2011 and 31 December 2021. Individuals who had a dispense for both a GLP-1 and DPP-4 at baseline were excluded.
Covariate Ascertainment
We used the CMS Form 2728, which is filled out at dialysis initiation by the patient’s nephrologist, to abstract information regarding demographics, race, initial dialysis access, comorbidities at dialysis initiation (i.e., atherosclerotic heart disease, cancer, chronic pulmonary disease, cerebrovascular disease, diabetes, dysrhythmia, heart failure, hypertension, myocardial infarction, other cardiac disease, and peripheral vascular disease [PVD]), smoking status, end-stage renal disease network region (i.e., geographic region), and etiology of kidney failure (diabetes or not). We then ascertained additional comorbidities (i.e., atherosclerotic heart disease, cerebrovascular disease, heart failure, PVD, other cardiac disease, respiratory disease, liver disease, dysrhythmia, cancer, and diabetes) (18) by the presence of one inpatient ICD claim or two outpatient claims in the 180 days prior to baseline (Supplementary Table 2). We additionally abstracted the number of hospitalizations in the previous 180 days, and we used the most updated BMI, KT/V (measure of dialysis adequacy), and history of erythropoietin stimulating agent use from monthly dialysis claims. We also noted baseline medication use (90 days prior to study initiation) (Supplementary Table 3), Medicare Part D subsidy status, dialysis vintage (baseline minus dialysis start date), and baseline calendar year of medication initiation. Finally, in the previous 180 days, we also evaluated for history of hypoglycemia hospitalization, hyperglycemia hospitalization, diabetic ketoacidosis (outpatient or hospitalization), outpatient endocrinology care, and outpatient primary care (defined as at least one visit to “Internal Medicine”, “Family Practice,” “General Medicine” or “Geriatric Medicine”) from Medicare claims files.
Treatment Assignment (Emulation of Randomization)
We used inverse probability of treatment weighting (IPTW) methods (19) to account for the observational, nonrandomized study design. First, we estimated the probability of treatment (i.e., GLP-1 vs DPP-4) for each patient using a logistic regression model that included all covariates listed in Table 1 (i.e., propensity score). We then used the inverse of this propensity score with the marginal probability of GLP-1 use (i.e., stabilized weight) to create a pseudopopulation which ideally has equal covariate balance between treatment groups replicating a randomized clinical trial.
Table 1.
Baseline characteristics of new GLP-1 versus new DPP-4 users pre- and post-IPTW
| Pre-IPTW | Post-IPTW | |||||
|---|---|---|---|---|---|---|
| GLP-1 (n = 3,619) | DPP-4 (n = 11,502) | SMD | GLP-1 (n = 3,492) | DPP-4 (n = 11,609) | SMD | |
| Age, mean (SD), years | 60 (12) | 64 (12) | 0.37 | 63 (12) | 63 (13) | 0.04 |
| Male sex | 1,827 (50) | 6,123 (53) | 0.06 | 1,798 (52) | 6,069 (52) | 0.02 |
| Race | 0.14 | 0.07 | ||||
| White | 2,340 (65) | 6,989 (61) | 2,254 (65) | 7,201 (62) | ||
| Black/African American | 1,028 (28) | 3,441 (30) | 987 (28) | 3,411 (29) | ||
| Asian | 120 (3) | 675 (6) | 140 (4) | 599 (5) | ||
| Native Hawaiian or Pacific Islander | 63 (2) | 200 (2) | 55 (2) | 198 (2) | ||
| American Indian or Alaska Native | 50s* (2) | 148 (1) | 40 s (1) | 156 (1) | ||
| Other or multiracial | <11* (0) | 49 (0) | <11* (0) | 43 (0) | ||
| Year of dialysis initiation | 0.69 | 0.08 | ||||
| 2011 | <11* (0) | 82 (1) | 21 (1) | 67 (1) | ||
| 2012 | 20s* (1) | 333 (3) | 93 (3) | 274 (2) | ||
| 2013 | 53 (1) | 563 (5) | 110 (3) | 471 (4) | ||
| 2014 | 85 (2) | 900 (8) | 196 (6) | 749 (6) | ||
| 2015 | 151 (4) | 1,252 (11) | 308 (9) | 1,068 (9) | ||
| 2016 | 199 (6) | 1,417 (12) | 401 (11) | 1,234 (11) | ||
| 2017 | 357 (10) | 1,583 (14) | 403 (12) | 1,470 (13) | ||
| 2018 | 533 (15) | 1,598 (14) | 499 (14) | 1,636 (14) | ||
| 2019 | 856 (24) | 1,629 (14) | 620 (18) | 1,914 (16) | ||
| 2020 | 845 (23) | 1,473 (13) | 558 (16) | 1,760 (15) | ||
| 2021 | 509 (14) | 672 (6) | 282 (8) | 965 (8) | ||
| Dialysis vintage, years | 2.3 (1.17–4.0) | 1.9 (0.9–3.5) | 0.19 | 2.0 (1.0–3.7) | 2.0 (0.9–3.6) | 0.02 |
| KT/V | 1.6 (1.4–1.7) | 1.6 (1.4–1.8) | 0.01 | 1.6 (1.4–1.8) | 1.6 (1.4–1.8) | 0.01 |
| Smoking status at dialysis initiation | 182 (5) | 561 (5) | 0.01 | 190 (5) | 577 (5) | 0.02 |
| Medicare Part D low-income subsidy | 2,912 (80) | 9,086 (79) | 0.04 | 2,745 (79) | 9,232 (80) | 0.02 |
| ESKD from diabetes | 2,898 (80) | 8,708 (76) | 0.11 | 2,716 (78) | 8,926 (77) | 0.02 |
| USRDS network region | 0.09 | 0.02 | ||||
| West | 974 (27) | 2,961 (26) | 919 (26) | 3,091 (27) | ||
| Northeast | 673 (19) | 2,574 (22) | 722 (21) | 2,455 (21) | ||
| South | 1,291 (36) | 3,935 (34) | 1,214 (35) | 3,994 (34) | ||
| Midwest | 681 (19) | 2,032 (18) | 638 (18) | 2,069 (18) | ||
| Nephrology care prior to dialysis initiation | 0.13 | 0.02 | ||||
| None | 1,187 (33) | 4,460 (39) | 1,284 (37) | 4,286 (37) | ||
| <6 months | 548 (15) | 1,740 (15) | 526 (15) | 1,753 (15) | ||
| 6–12 months | 809 (22) | 2,292 (20) | 711 (20) | 2,439 (21) | ||
| >12 months | 1,075 (30) | 3,010 (26) | 972 (28) | 3,132 (27) | ||
| Initial dialysis access via catheter | 2,704 (75) | 8,807 (77) | 0.04 | 2,672 (77) | 8,875 (76) | <0.01 |
| BMI, kg/m2 | 34 (29–40) | 29 (25–34) | 0.64 | 31 (26–36) | 30 (25–36) | 0.01 |
| Atherosclerosis | 880 (24) | 3,827 (33) | 0.20 | 1,054 (30) | 3,572 (31) | 0.01 |
| Cancer | 213 (6) | 887 (8) | 0.07 | 267 (8) | 830 (7) | 0.02 |
| Arrhythmia | 685 (19) | 2,705 (24) | 0.11 | 773 (22) | 2,645 (23) | 0.02 |
| Cerebrovascular disease | 618 (17) | 2,470 (21) | 0.11 | 717 (21) | 2,356 (20) | 0.01 |
| Congestive heart failure | 1,122 (31) | 4,448 (39) | 0.16 | 1,308 (37) | 4,249 (37) | 0.02 |
| History of GI bleeding | 154 (4) | 832 (7) | 0.13 | 211 (6) | 747 (6) | 0.02 |
| Hypertension | 3,268 (90) | 10,431 (91) | 0.01 | 3,183 (91) | 10,534 (91) | 0.01 |
| Nonambulatory | 204 (6) | 698 (6) | 0.02 | 232 (7) | 698 (6) | 0.03 |
| Liver disease | 215 (6) | 1,370 (12) | 0.21 | 337 (10) | 1,206 (10) | 0.02 |
| Other cardiac disease | 463 (13) | 1,971 (17) | 0.12 | 552 (16) | 1,848 (16) | <0.01 |
| PVD | 1,348 (37) | 5,581 (49) | 0.23 | 1,529 (44) | 5,270 (45) | 0.03 |
| Respiratory disease | 438 (12) | 1,531 (13) | 0.04 | 480 (14) | 1,505 (13) | 0.02 |
| No. of hospitalizations, <6 months | 0.37 | 0.04 | ||||
| None | 2,255 (62) | 5,277 (46) | 1,745 (50) | 5,830 (50) | ||
| 1 | 756 (21) | 2,771 (24) | 788 (23) | 2,684 (23) | ||
| 2 | 324 (9) | 1,626 (14) | 427 (12) | 1,486 (13) | ||
| 3 | 144 (4) | 862 (7) | 244 (7) | 765 (7) | ||
| ≥4 | 140 (4) | 966 (8) | 288 (8) | 844 (7) | ||
| Outpatient primary care visit, <6 months | 2,152 (59) | 6,586 (57) | 0.04 | 2,024 (58) | 6,684 (58) | 0.01 |
| Outpatient endocrinology visit, <6 months | 919 (25) | 2,759 (24) | 0.03 | 804 (23) | 2,797 (24) | 0.03 |
| Hyperglycemia hospitalization, <6 months | 293 (8) | 1,149 (10) | 0.07 | 346 (10) | 1,102 (9) | 0.01 |
| Hypoglycemia hospitalization, <6 months | 83 (2) | 594 (5) | 0.15 | 176 (5) | 518 (4) | 0.03 |
| Diabetic ketoacidosis, <6 months | 17 (0) | 64 (1) | 0.01 | 19 (1) | 63 (1) | <0.01 |
| No. of unique medications, <90 days | 11 (8–15) | 10 (7–14) | 0.15 | 11 (8–14) | 11 (8–14) | 0.04 |
| ACE/ARB | 1,102 (30) | 4,138 (36) | 0.12 | 1,276 (37) | 4,006 (35) | 0.04 |
| Aldosterone antagonist | 46 (1) | 140 (1) | <0.01 | 40 (1) | 136 (1) | <0.01 |
| α-Glucosidase inhibitors | <11* | 18 (0) | 0.01 | <11* | 21 (0) | 0.01 |
| Antiplatelet | 682 (19) | 2,314 (20) | 0.03 | 722 (21) | 2,347 (20) | 0.01 |
| β-Blocker | 1,997 (55) | 6,827 (59) | 0.08 | 2,107 (60) | 6,761 (58) | 0.04 |
| Calcium channel blocker | 1,498 (41) | 5,509 (48) | 0.13 | 1,573 (45) | 5,357 (46) | 0.02 |
| Diuretic | 1,159 (32) | 3,012 (26) | 0.13 | 1,009 (29) | 3,203 (28) | 0.03 |
| Direct oral anticoagulants | 255 (7) | 635 (6) | 0.06 | 215 (6) | 747 (6) | 0.01 |
| History of any Epo use | 2,910 (80) | 8,628 (75) | 0.13 | 2,664 (76) | 8,865 (76) | <0.01 |
| Insulin | 2,551 (70) | 4,698 (41) | 0.63 | 1,737 (50) | 5,607 (48) | 0.03 |
| Phosphate binder | 2,267 (63) | 7,005 (61) | 0.04 | 2,117 (61) | 7,061 (61) | <0.01 |
| Sodium–glucose cotransporter 2 | 17 (0) | 20 (0) | 0.05 | <11* (0) | 29 (0) | <0.01 |
| Sulfonylurea | 246 (7) | 1,517 (13) | 0.21 | 383 (11) | 1,347 (12) | 0.02 |
| Thiazolidinediones | 84 (2) | 343 (3) | 0.04 | 124 (4) | 335 (3) | 0.04 |
| Warfarin | 214 (6) | 806 (7) | 0.04 | 217 (6) | 779 (7) | 0.02 |
Data reported as n (%), median (IQI), or as indicated otherwise as mean (SD).
Epo, erythropoietin stimulating agent.
*Cells with <10 were censored according to USRDS reporting policy or rounded down (if multiple categories).
Follow-up, Outcomes, and Causal Contrasts
Follow-up started the day after medication dispense, and individuals were monitored until study outcome, kidney transplant, renal recovery off dialysis, loss of Medicare Primary Payer status, or death or 31 December 2021, whichever came first. The primary study outcome was the first ADRD diagnosis, and for secondary outcomes, we evaluated specific ADRD ICD diagnoses: dementia not otherwise specified (NOS), vascular dementia, Alzheimer dementia, Lewy body dementia, and frontotemporal dementia. We additionally evaluated the risk of ketoacidosis. Outcomes were defined by the date of first ICD code in either hospitalization (any position) or outpatient Medicare claims (Supplementary Table 4).
We estimated the intention-to-treat effect, and given the possibility of medication discontinuation, we additionally estimated as-treated effects, which accounts for medication discontinuation. In as-treated analysis, individuals were additionally censored at “end of medication dispense,” defined as no refill within 30 days after end date of last dispense or at “medication switch” (i.e., a GLP-1 user initiating a DPP-4 or a DPP-4 user initiating a GLP-1).
Statistical Analysis
Baseline characteristics between GLP-1 and DPP-4 users were compared pre- and post-IPTW, with absolute standardized mean differences (SMDs) <0.10 considered good balance (20).
We used Fine-Gray models (21) to account for the competing risk of kidney transplant and death on the IPTW-weighted cohort to estimate both the IPTW subdistribution hazard ratios (sHRs) and cumulative incidences of study outcomes. We also used cause-specific Cox proportional hazards regression, where competing risk was treated as censoring, and Poisson regression to assess IPTW-hazard ratios (HRs) and IPTW-incidence rates (IR) of GLP-1s (vs. DPP-4s) and study outcomes. We then examined for a subgroup interaction between baseline insulin use and ketoacidosis (13,14).
Analyses were performed using R 4.2.2 software (R Foundation for Statistical Computing) (22) and Stata 17 software (StataCorp) (23). Significant results were determined by a two-tailed P value < 0.05.
Sensitivity Analyses
We additionally evaluated ADRD risk by subgroups (age [≥65 or <65], sex, race, BMI [by quartile], history of cerebrovascular disease, insulin use, and Medicare Part D low-income subsidy) using the intention-to-treat and as-treated approach. For subgroup analysis, we included an interaction term between the treatment variable and subgroup variable in the IPTW Fine-Gray competing-risk model. Then, instead of defining the study outcome as any claim (inpatient or outpatient), we repeated the main analyses defining ADRD diagnosis as hospitalization only, hospitalization or two outpatient claims, and two claims (inpatient or outpatient). We also repeated analyses after removing individuals who had any ADRD diagnosis within the Medicare claim files (i.e., any time point before GLP-1 or DPP-4 initiation). We then repeated our primary and secondary as-treated analyses using a 60-day (instead of 30-day) refill window to define medication discontinuation. We estimated the E-value (24), which is the minimum relative risk (RR) that an unmeasured confounder would need to have with both the outcome and the exposure to explain the observed association between GLP-1 use and ADRDs. To provide some context for interpretation of the E-value, we report the top three HRs of study covariates for ADRDs (i.e., the outcome) using multivariate Cox proportional hazards models adjusted for all covariates in Table 1.
Positive and Negative Controls
We tested our target trial emulation approach by examining positive and negative control outcomes. For the positive control, we used gastrointestinal (GI) symptoms, which is a commonly reported adverse event. For the negative controls, we used lower respiratory diseases, hearing loss, and lens disorders (Supplementary Table 4) (7) as these outcomes should not differ by GLP-1 and DPP-4 use.
Data and Resource Availability
We are not able to provide the data due to privacy reasons. The data are available with an approved proposal through the USRDS.
Results
Patient Characteristics
We identified 3,619 GLP-1 and 11,502 DPP-4 new users who met study criteria (Fig. 1). Prior to IPTW, GLP-1 users (vs. DPP-4) were more likely to be younger, White, have higher dialysis vintage, have seen a nephrologist prior to dialysis initiation, have a higher BMI, use certain medications (i.e., insulin, erythropoietin, and diuretics). They were less likely to have ESKD from diabetes, medical comorbidities (i.e., arrhythmia, atherosclerosis, cerebrovascular disease, congestive heart failure, history of GI bleeding, liver disease, PVD), history of hypoglycemia hospitalization, and use certain medications (i.e., ACE/ARBs, calcium channel blockers, direct-acting oral anticoagulants, and sulfonylureas). After IPTW, all variables were balanced with SMDs <0.10 (Table 1 and Supplementary Fig. 1).
In the weighted cohort, there were 3,492 GLP-1 and 11,609 DPP-4 users. The mean age of the weighted study population was 63 years, 63% were White, and 52% were men. Median (interquartile interval [IQI]) dialysis vintage was 2.0 (1.0–3.6) years, and 78% had kidney failure attributed to diabetes (Table 1).
Follow-up and Outcomes in the Weighted Cohort
The median (IQI) follow-up was 1.5 (0.6–2.9) years. In the unweighted cohort, the primary outcome of ADRD occurred in 253 individuals on GLP-1s and in 1,761 on DPP-4s, with an IPTW-sHR of 0.82 (95% CI 0.67–0.98). The IPTW-IR was 56 events per 1,000 person-years versus 69 events per 1,000 person-years. The Fine-Gray cumulative incidences of ADRD were 6.6%, 10.2%, and 14.5% at 1 year, 2 years, and 5 years for individuals on GLP-1s, respectively, and were 7.2%, 11.4%, and 17.7% for individuals on DPP-4s, respectively (Fig. 2). See Supplementary Fig. 2 for IPTW–Kaplan-Meier plots. When we looked at ADRD subtypes (i.e., NOS, vascular, or Alzheimer), the point-estimates were similar, but only “dementia NOS” was statistically significant (Table 2).
Figure 2.
IPTW-cumulative incidence of ADRD by GLP-1 (vs. DPP-4) use. Cumulative incidence curves were estimated from Fine-Gray models accounting for competing risk of mortality and kidney transplantation. IPTW-risk tables are shown.
Table 2.
HRs of ADRDs and ketoacidosis by GLP-1 (versus DPP-4) use in patients requiring hemodialysis
| IPTW-No. of events | IPTW-IR (95% CI) per 1,000 patient-years | ||||||
|---|---|---|---|---|---|---|---|
| GLP-1 (n = 3,492) | DPP-4 (n = 11,609) | GLP-1 | DPP-4 | IPTW-IRD | IPTW-HR | IPTW-sHR | |
| Intention-to-treat | |||||||
| Any dementia | 382 | 1,558 | 56 (44, 72) | 69 (66, 72) | −12 (−23, 0) | 0.82 (0.67, 0.99) | 0.81 (0.67, 0.98) |
| Dementia NOS | 339 | 1,405 | 50 (38, 65) | 62 (58, 65) | −12 (−22, 0) | 0.81 (0.66, 0.99) | 0.81 (0.66, 0.99) |
| Vascular | 72 | 325 | 10 (6, 17) | 13 (12, 15) | −3 (−6, 1) | 0.75 (0.51, 1.11) | 0.75 (0.51, 1.11) |
| Alzheimer | 58 | 254 | 8 (5, 14) | 11 (9, 12) | −3 (−6, 2) | 0.77 (0.49, 1.21) | 0.76 (0.49, 1.19) |
| Lewy body | <11 | 21 | 1 (0, 3) | 1 (1, 1) | 0 (−1, 2) | 0.75 (0.21, 2.67) | 0.73 (0.21, 2.60) |
| Frontotemporal | <11 | <11 | — | — | — | — | — |
| Ketoacidosis | 112 | 252 | 16 (10, 25) | 10 (9, 12) | 5 (1, 11) | 1.51 (1.11, 2.07) | 1.52 (1.14, 2.02) |
| As-treated | |||||||
| Any dementia | 139 | 906 | 45 (32, 62) | 70 (65, 74) | −25 (−35, −11) | 0.61 (0.47, 0.79) | 0.61 (0.47, 0.78) |
| Dementia NOS | 117 | 789 | 37 (26, 53) | 60 (56, 64) | −23 (−32, −11) | 0.59 (0.45, 0.78) | 0.60 (0.45, 0.78) |
| Vascular | 20 | 176 | 6 (3, 14) | 13 (11, 15) | −7 (−10, −1) | 0.47 (0.25, 0.88) | 0.46 (0.25, 0.87) |
| Alzheimer | 26 | 135 | 8 (4, 19) | 10 (8, 12) | −2 (−6, 6) | 0.79 (0.41, 1.55) | 0.78 (0.40, 1.53) |
| Lewy body | <11 | 11 | — | — | — | — | — |
| Frontotemporal | <11 | <11 | — | — | — | — | — |
| Ketoacidosis | 71 | 123 | 22 (12, 41) | 9 (7, 11) | 13 (6, 25) | 2.40 (1.56, 3.69) | 2.30 (1.56, 3.40) |
IPTW-HR estimated using cause-specific Cox proportional hazards regression. IPTW-sHR estimated using Fine-Gray models accounting for competing risk of mortality and kidney transplantation.
IRD, incidence rate difference.
In as-treated analysis, censoring due to medication discontinuation was high, with 1,892 (52%) stopping GLP-1s after a median (IQI) 120 (60–265) days, and 6,019 (52%) stopping DPP-4s after 152 (60–422) days. At 1 year, 1,123 (31%) remained on GLP-1s and 4,366 (38%) on DPP-4s. GLP-1s were associated with a lower risk of ADRDs (sHR 0.61, 95% CI 0.47–0.78) in as-treated analysis, and findings by ADRD subtype in as-treated analysis were also consistent with the intention-to-treat analysis. Cause-specific HRs were also similar in both intention-to-treat and as-treated analyses (Table 2).
In the unweighted cohort, the primary outcome of ketoacidosis occurred in 112 patients on GLP-1s versus 252 patients on DPP-4s. The weighted sHR was 1.52 (95% CI 1.14–2.02), and the cumulative incidence of ketoacidosis was 2.1%, 3.1%, and 4.6% at 1 year, 2 years, and 5 years for the GLP-1 group versus 1.4%, 2.2%, and 3.2% for the DPP-4 group, respectively (Supplementary Fig. 3). As-treated estimates were similar. There was no interaction between GLP-1s and ketoacidosis by baseline insulin use (insulin use: sHR 1.45 [95% CI 1.07–1.97] vs. no-insulin use: sHR 1.58 [95% CI 0.87–2.87]; P value for interaction = 0.77). (See Supplementary Table 5 for the number of events.)
Sensitivity Analysis
Subgroup analyses by age, sex, race, BMI, history of cerebrovascular disease, insulin use, or Medicare Part D low-income subsidy status did not demonstrate significant interaction across both intention-to-treat and as-treated analysis (Supplementary Figs. 4 and 5). Baseline covariates within subgroups were generally well balanced, with SMDs <0.10. Results were qualitatively unchanged when we defined study outcomes by hospitalization only, hospitalization or outpatient with two claims, or at least two claims regardless of type. When we examined all historic Medicare claim files prior to baseline, we excluded an additional 332 individuals with a prior ADRD diagnosis and both intention-to-treat and as-treated analyses remained consistent. Results were also unchanged when redefining the medication discontinuation window to 60 days. The E-value was 1.77 in the intention-to-treat analysis and was 2.66 in the as-treated analysis. The strongest associations of study variables and ADRDs were observed with a history of four or more hospitalizations (HR 1.59), a history of cerebrovascular disease (HR 1.58), and smoking at dialysis initiation (HR 1.44). Each of these associations was weaker than the E-values.
Positive/Negative Control Outcomes
GLP-1 use was associated with increased GI symptoms (sHR 1.17, 95% CI 1.07–1.28), and there was no association for pneumonia (sHR 1.08, 95% CI 0.98–1.19), hearing loss (sHR 1.05, 95 CI 0.88–1.27), or lens disease (sHR 0.98, 95% CI 0.87–1.10). As-treated estimates were similar.
Conclusions
In this target trial emulation study of patients with diabetes on hemodialysis, new use of GLP-1s (vs. DPP-4s) was associated with a lower risk of ADRDs. GLP-1s were also associated with an increased risk of ketoacidosis. GLP-1 adherence at 1 year was low at 31%, but the as-treated effect estimates were consistent with the intention-to-treat analyses. GLP-1s may be a promising therapy to lower the risk of ADRDs for patients with diabetes and dialysis dependence.
Our study supports the growing literature of GLP-1s and ADRD reduction (6), and we were able to quantify this protective association among patients with diabetes requiring in-center hemodialysis, a population excluded from many of the GLP-1 clinical trials (25–27) and retrospective studies of GLP-1s and ADRDs (7). Our estimate of GLP-1 ADRD risk reduction is lower than secondary analyses of GLP-1 clinical trials (i.e., HR 0.81 in our study vs. 0.55 seen in clinical trial secondary data) (6), but our 1-year medication adherence was significantly less (i.e., 31% vs. 80%) (25), which may affect effect size. The mechanism of GLP-1 protection against ADRDs is not fully understood but may be mediated by the ability of GLP-1s to cross the blood-brain barrier, exerting direct effects on the central nervous system. Preclinical studies suggest GLP-1s may lead to improvement in brain insulin signaling, inflammatory cytokines, amyloid plaque clearance, oxidate phosphorylation, antioxidant enzymes, and neural cell function (28). GLP-1s also confer significant benefits on weight loss and glycemic control (25,26), but whether ADRD protection is related to changes in hemoglobin A1c or BMI is less clear (29,30). In our subgroup analyses, we did not observe a significant interaction by BMI or insulin use at baseline, but we were unable to account for subsequent changes in weight or glucose control. GLP-1s also improve lipid profiles (31), but for patients requiring dialysis, improvements in lipid profiles have not translated to improved clinical outcomes (32).
In terms of safety, GLP-1s (vs. DPP-4s) were associated with increased risk of ketoacidosis. This is seemingly paradoxical given GLP-1 benefits on glycemic control and ADRD risk (33), but the absolute risk of ketoacidosis was low, with a 2-year cumulative incidence of 3.1% on GLP-1s versus 2.2% on DPP-4s (vs. ADRD 2-year cumulative incidence of 10.2% on GLP-1s and 11.2% on DPP-4s). The risk of ketoacidosis with GLP-1s has been observed in postmarketing surveillance systems in both the U.S. (13) and U.K. (14). This increased risk of ketoacidosis has been attributed to insulin titration after GLP-1s initiation (14), but we did not observe an interaction effect by insulin use at baseline, potentially due to low event size. Although GLP-1 use does not appear to be causally linked to ketoacidosis, our results indicate that dialysis-specific strategies after GLP-1 (vs. DPP-4) initiation may be needed to mitigate risk of ketoacidosis. Ongoing studies in this population evaluating glucose time in range and GLP-1 use will provide further clarification regarding glycemic complications (34,35), which appear to be common (36).
Further, our results support the joint ADA and KDIGO guidelines that recommend GLP-1 use prior to other diabetes medications (5). Data from the USRDS (11,12), however, shows that insulin remains the most prescribed therapy for patients requiring dialysis, which may be related to treatment inertia, lack of dialysis-specific diabetes studies (15,25,26), and lack of GLP-1–specific dialysis data (15). Our study addresses some of these knowledge gaps by comparing two classes of contemporary glucose-lowering medications and their association on the incidence of ADRDs, which has remained fourfold higher for patients requiring in-center hemodialysis compared with the general population (i.e., 60–70 events per 1,000 person-years vs. 10–15 events per 1,000 person-years) (2,37,38). Dementia prevention strategies for patients requiring in-center hemodialysis are needed, and GLP-1s may be an immediate, already recommended therapy to prevent ADRD onset and associated morbidity and mortality. Randomized trials evaluating GLP-1 use and ADRD reduction in the general population will provide crucial efficacy data (8–10), but dialysis-specific studies may be needed given the high mortality and morbidity seen in this population (11).
Our study also has limitations. There may still be confounding (39) (e.g., confounding by indication, where healthier patients preferentially received GLP-1s vs. DPP-4s) despite both the IPTW method and use of positive/negative control outcomes. There is also risk of residual and/or unmeasured confounding when using observational data. For example, hemoglobin A1c, diabetes duration, frailty status, and baseline-measured cognitive function, which may affect ADRD incidence, were not measured in claims data. We used the E-value method, which estimated that an unmeasured confounder would need a risk ratio >1.77 (vs. 2.7 for as-treated analysis) with both ADRD diagnosis and GLP-1 use to explain our observed benefit. The strongest association with ADRD within our study covariates was a history of four or more hospitalizations (HR 1.59), which was less than the E-value, but it is possible that the combination of unmeasured and/or residual confounders could exceed this number. There is also a risk of misclassification as we defined all study outcomes and covariates using ICD codes. Patients were also not systematically examined for ADRD diagnosis, which may also lead to underestimation of disease incidence. Although this should affect treatment groups equally, GLP-1s are recommended over DPP-4s among patients requiring hemodialysis, so patients on GLP-1s may be receiving more guideline concordant care and have more opportunities to receive an ADRD diagnosis. This ascertainment bias, however, would bias our results toward the null or show increased risk with GLP-1s. We were also underpowered to examine specific ADRD diagnoses, but ICD-specific diagnoses have low validity (40). We were also unable to assess dosing patterns. Our generalizability is limited as all patients had Medicare Part A, B, and D primary payer for at least 6 months prior to medication dispense.
Despite these limitations, strengths of this study include use of a national dialysis registry, emulated trial design, competing-risk analysis (to account for the high rate of mortality) (11), prescription dispense data, robust covariate ascertainment, likely complete outcome ascertainment, and consistent positive and negative control outcomes.
In this target trial emulation study with >15,000 patients with diabetes and in-center hemodialysis dependence, we found that new use of GLP-1s (vs. DPP-4s) was associated with a reduced risk of ADRDs and increased risk of diabetic ketoacidosis. Our results suggest the potential effectiveness for GLP-1s in reducing ADRD risk for patients requiring in-center hemodialysis.
This article contains supplementary material online at https://doi.org/10.2337/figshare.30438470.
Article Information
Acknowledgments. The data reported here have been supplied by the USRDS. The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as an official policy or interpretation of the U.S. government.
Study funders did not have any role in study design, collection, analysis, interpretation of data, writing of the report, or decision for submission for publication.
Duality of Interest. No potential conflicts of interest relevant to this article were reported.
Author Contributions. D.L., M.K., W.K.K., M.E.G., B.G.J., and J.-I.S. contributed to the research idea and study design. D.L., M.K., W.K.K., M.E.G., B.G.J., and J.-I.S. analyzed and interpreted the data. D.L. and J.S. performed statistical analysis. W.K.K., M.E.G., B.G.J., J.-I.S. contributed to supervision or mentorship. D.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Prior Presentation. Parts of this study were presented in abstract form at the American Society of Nephrology Annual Meeting, Houston, TX, 6–9 November 2025.
Handling Editors. The journal editors responsible for overseeing the review of the manuscript were John B. Buse and Csaba P. Kovesdy.
Funding Statement
D.L. was supported by a Young Investigator Grant of the National Kidney Foundation. J.-I.S. was supported by R01DK139324 and R01DK115534 from the National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases.
Supporting information
References
- 1. Nandi A, Counts N, Chen S, et al. Global and regional projections of the economic burden of Alzheimer’s disease and related dementias from 2019 to 2050: a value of statistical life approach. EClinicalMedicine 2022;51:101580. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. McAdams-DeMarco MA, Daubresse M, Bae S, Gross AL, Carlson MC, Segev DL. Dementia, Alzheimer’s disease, and mortality after hemodialysis initiation. Clin J Am Soc Nephrol 2018;13:1339–1347 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Kuo Y-T, Li C-Y, Sung J-M, et al. Risk of dementia in patients with end-stage renal disease under maintenance dialysis—a nationwide population-based study with consideration of competing risk of mortality. Alzheimers Res Ther 2019;11:31. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Luca M, Luca A, Calandra C. The role of oxidative damage in the pathogenesis and progression of Alzheimer’s disease and vascular dementia. Oxid Med Cell Longev 2015;2015:504678. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. de Boer IH, Khunti K, Sadusky T, et al. Diabetes management in chronic kidney disease: a consensus report by the American Diabetes Association (ADA) and Kidney Disease: improving Global Outcomes (KDIGO). Diabetes Care 2022;45:3075–3090 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Seminer A, Mulihano A, O’Brien C, et al. Cardioprotective glucose-lowering agents and dementia risk: a systematic review and meta-analysis. JAMA Neurol 2025;82:450–460 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Tang B, Sjölander A, Wastesson JW, et al. Comparative effectiveness of glucagon-like peptide-1 agonists, dipeptidyl peptidase-4 inhibitors, and sulfonylureas on the risk of dementia in older individuals with type 2 diabetes in Sweden: an emulated trial study. EClinicalMedicine 2024;73:102689. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. University of Oxford . Effects of Single-Dose Semaglutide on Cognition and Energy in Healthy Volunteers. In: ClinicalTrials.Gov. Bethesda, MD, National Library of Medicine, 2024. Accessed 13 April 2025. NLM Identifier: NCT06363487. Available from https://clinicaltrials.gov/study/NCT06363487 [Google Scholar]
- 9. Novo Nordisk A/S . Randomised Double-Blind Placebo-Controlled Clinical Trial Investigating the Effect and Safety of Oral Semaglutide in Subjects With Early Alzheimer′s Disease (EVOKE Plus). In: ClinicalTrials.Gov. Bethesda, MD, National Library of Medicine, 2025. NLM Identifier: NCT04777409. Accessed 13 April 2025. Available from https://clinicaltrials.gov/study/NCT04777409 [Google Scholar]
- 10. Novo Nordisk A/S . A Randomised Double-Blind Placebo-Controlled Clinical Trial Investigating the Effect and Safety of Oral Semaglutide in Subjects With Early Alzheimer′s Disease (EVOKE). In: ClinicalTrials.Gov. Bethesda, MD, National Library of Medicine, 2025. NLM Identifier: NCT04777396. Accessed 13 April 2025. Available from https://clinicaltrials.gov/study/NCT04777396 [Google Scholar]
- 11. U.S. Renal Data System . 2021 USRDS Annual Data Report: Epidemiology of Kidney Disease in the United States. Bethesda, MD, National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases, 2021 [Google Scholar]
- 12. Klein KR, Pate V, Assimon MM, Stürmer T, Buse JB, Flythe JE. Antihyperglycemic medication use among U.S. adults with hemodialysis-dependent kidney disease and comorbid diabetes. Diabetes Care 2022;45:126–128 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. U.S. Food and Drug Administration . October – December 2019 | Potential Signals of Serious Risks/New Safety Information Identified by the FDA Adverse Event Reporting System (FAERS). Accessed 17 April 2025. Available from https://www.fda.gov/drugs/fdas-adverse-event-reporting-system-faers/october-december-2019-potential-signals-serious-risksnew-safety-information-identified-fda-adverse [Google Scholar]
- 14. Medicines & Healthcare products Regulatory Agency . GLP-1 receptor agonists: reports of diabetic ketoacidosis when concomitant insulin was rapidly reduced or discontinued. Accessed 17 April 2025. Available from https://www.gov.uk/drug-safety-update/glp-1-receptor-agonists-reports-of-diabetic-ketoacidosis-when-concomitant-insulin-was-rapidly-reduced-or-discontinued
- 15. Chen J-J, Wu C-Y, Jenq C-C, et al. Association of glucagon-like peptide-1 receptor agonist vs dipeptidyl peptidase-4 inhibitor use with mortality among patients with type 2 diabetes and advanced chronic kidney disease. JAMA Netw Open 2022;5:e221169. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Orandi BJ, Chen Y, Li Y, et al. GLP-1 receptor agonist outcomes, safety, and body mass index change in a national cohort of patients on dialysis. Clin J Am Soc Nephrol 2025;20:1100–1110 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Shin J-I, Grams ME. Trial emulation methods. Am J Kidney Dis 2024;83:264–267 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Liu J, Huang Z, Gilbertson DT, Foley RN, Collins AJ. An improved comorbidity index for outcome analyses among dialysis patients. Kidney Int 2010;77:141–151 [DOI] [PubMed] [Google Scholar]
- 19. Chesnaye NC, Stel VS, Tripepi G, et al. An introduction to inverse probability of treatment weighting in observational research. Clin Kidney J 2022;15:14–20 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Stuart EA, Lee BK, Leacy FP. Prognostic score-based balance measures can be a useful diagnostic for propensity score methods in comparative effectiveness research. J Clin Epidemiol 2013;66(Suppl.):S84–S90.e1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Austin PC, Fine JP. Practical recommendations for reporting Fine-Gray model analyses for competing risk data. Stat Med 2017;36:4391–4400 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. R Core Team. R: a Language And Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, 2020 [Google Scholar]
- 23. StataCorp . Stata Statistical Software: Release 17. College Station, TX, StataCorp LLC, 2021 [Google Scholar]
- 24. VanderWeele TJ, Ding P. Sensitivity analysis in observational research: introducing the E-value. Ann Intern Med 2017;167:268–274 [DOI] [PubMed] [Google Scholar]
- 25. Marso SP, Daniels GH, Brown-Frandsen K, et al.; LEADER Trial Investigators . Liraglutide and cardiovascular outcomes in type 2 diabetes. N Engl J Med 2016;375:311–322 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Frías JP, Davies MJ, Rosenstock J, et al.; SURPASS-2 Investigators . Tirzepatide versus semaglutide once weekly in patients with type 2 diabetes. N Engl J Med 2021;385:503–515 [DOI] [PubMed] [Google Scholar]
- 27. Pfeffer MA, Claggett B, Diaz R, et al.; ELIXA Investigators . Lixisenatide in patients with type 2 diabetes and acute coronary syndrome. N Engl J Med 2015;373:2247–2257 [DOI] [PubMed] [Google Scholar]
- 28. De Giorgi R, Ghenciulescu A, Dziwisz O, et al. An analysis on the role of glucagon-like peptide-1 receptor agonists in cognitive and mental health disorders. Nat Mental Health 2025;3:354–373 [Google Scholar]
- 29. 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:e70161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Luchsinger JA, Rosin SP, Kazemi EJ, et al.; GRADE Research Group . Glucose-lowering medications, glycemia, and cognitive outcomes: the GRADE randomized clinical trial. JAMA Intern Med 2025;185:778–787 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Sun F, Wu S, Wang J, et al. Effect of glucagon-like peptide-1 receptor agonists on lipid profiles among type 2 diabetes: a systematic review and network meta-analysis. Clin Ther 2015;37:225–241.e8 [DOI] [PubMed] [Google Scholar]
- 32. Wanner C, Krane V, März W, et al.; German Diabetes and Dialysis Study Investigators . Atorvastatin in patients with type 2 diabetes mellitus undergoing hemodialysis. N Engl J Med 2005;353:238–248 [DOI] [PubMed] [Google Scholar]
- 33. Rawlings AM, Sharrett AR, Albert MS, et al. The association of late-life diabetes status and hyperglycemia with incident mild cognitive impairment and dementia: the ARIC study. Diabetes Care 2019;42:1248–1254 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Fondation Hôpital Saint-Joseph . Efficacy and Tolerance of Liraglutide for Weight Loss in Obese Type 2 Diabetic Hemodialysis Patients. In: ClinicalTrials.Gov. Bethesda, MD, National Library of Medicine, 2023. NLM Identifier: NCT04529278. Accessed 29 March 2025. Available from https://clinicaltrials.gov/study/NCT04529278 [Google Scholar]
- 35. Lingvay I. Semaglutide for Dialysis-Treated Patients - a Glucose Time in Range Study- DIALYSIS-TIR Study. In: ClinicalTrials.Gov. Bethesda, MD, National Library of Medicine, 2024. NLM Identifier: NCT06042153. Accessed 28 January 2025. Available from https://clinicaltrials.gov/study/NCT06042153 [Google Scholar]
- 36. de Boer IH, Anderson LD, Ashford NK, et al. Glycemia assessed by continuous glucose monitoring among people treated with maintenance dialysis. J Am Soc Nephrol 2025;36:1798–1810 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Kim WJ, Lee SJ, Lee E, Lee EY, Han K. Risk of incident dementia according to glycemic status and comorbidities of hyperglycemia: a nationwide population-based cohort study. Diabetes Care 2022;45:134–141 [DOI] [PubMed] [Google Scholar]
- 38. MacKnight C, Rockwood K, Awalt E, McDowell I. Diabetes mellitus and the risk of dementia, Alzheimer’s disease and vascular cognitive impairment in the Canadian Study of Health and Aging. Dement Geriatr Cogn Disord 2002;14:77–83 [DOI] [PubMed] [Google Scholar]
- 39. Inoue K, Saliba D, Gotanda H, et al. Glucagon-like peptide-1 receptor agonists and incidence of dementia among older adults with type 2 diabetes: a target trial emulation. Ann Intern Med 2025;178:1258–1267 [DOI] [PubMed] [Google Scholar]
- 40. Fujiyoshi A, Jacobs DR, Alonso A, Luchsinger JA, Rapp SR, Duprez DA. Validity of death certificate and hospital discharge ICD codes for dementia diagnosis: the multi-ethnic study of atherosclerosis. Alzheimer Dis Assoc Disord 2017;31:168–172 [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.