Target Trial of Glucagon-Like Peptide-1 Agonists versus Dipeptidyl Peptidase-4 Inhibitors and Mortality in Hemodialysis Authors: Dustin Le, Mark Kilpatrick, Walter K. Kraft, Morgan E. Grams, Bernard G. Jaar, Jung-Im Shin Published: 2026 ABSTRACT IntroductionThe comparative effectiveness of glucagon-like peptide-1 (GLP-1) agonists (GLP-1s) versus dipeptidyl peptidase-4 (DPP-4) inhibitors (DPP-4s) in hemodialysis is unknown.MethodsUsing 2011 to 2021 US Renal Data System and Medicare claims data, we emulated a target trial comparing new users of GLP-1s versus DPP-4s with patients who survived ≥ 90 days on in-center hemodialysis before medication initiation. We used inverse probability of treatment weighting (IPTW) and Cox proportional hazards models to estimate hazard ratios (HRs).ResultsWe identified 3671 GLP-1 and 12,039 DPP-4 users. After IPTW, mean age was 63 years, 52% male, 63% White, and mean body mass index was 31 kg/m2. The median (interquartile interval) follow-up was 1.5 (0.6–2.9) years. In intention-to-treat analysis, the all-cause mortality HR for GLP-1s (vs. DPP-4s) was 0.98 (95% confidence interval [CI]: 0.89–1.07); cardiovascular mortality HR was 1.06 (95% CI: 0.92–1.23); infection mortality HR: 0.83 (95% CI: 0.54–1.26). In per-protocol analysis with censoring at medication discontinuation, the all-cause mortality HR was 0.81 (95% CI: 0.68–0.96), and the 1-year cumulative incidence of discontinuation was 53% for GLP-1s versus 42% for DPP-4s. The cardiovascular mortality HR was 0.89 (95% CI: 0.69–1.13); infection mortality HR 0.65 (95% CI: 0.37–1.14). Finally, the gastrointestinal symptom HR was 1.11 (95% CI: 1.02–1.20).ConclusionGLP-1 (vs. DPP-4) initiation was not associated with reduced mortality in intention-to-treat analysis, but continuous use was associated with improved survival. The therapeutic potential of GLP-1s among patients receiving hemodialysis may be limited by medication discontinuation. More than 800,000 Americans live with kidney failure, and diabetes is a common comorbidity with 500,000 living with both conditions.1 Both diseases are life long and associated with significant morbidity and mortality. Diabetes treatment options for patients with kidney failure receiving dialysis are limited, with most patients being prescribed insulin.1 This contrasts with the nondialysis population where GLP-1s have gained rapid adoption.2 Importantly, patients receiving dialysis were excluded from landmark GLP-1 clinical trials,3, 4, 5, 6, 7 so studies using observational data are needed to quantify the effectiveness and safety of these medications. For patients receiving dialysis, GLP-1s are the de facto medical therapy recommended by the American Diabetes Association and Kidney Disease: Improving Global Outcomes joint recommendations, as both metformin and sodium-glucose cotransporter 2 inhibitors are not recommended because of impaired renal function.8 However, data from the US Renal Data System (USRDS) show that GLP-1 uptake remains less than 5% in patients receiving hemodialysis in 2022 (compared with 36% on insulin).9 The low utilization of GLP-1s in patients with both diabetes and kidney failure receiving dialysis likely stems from treatment inertia, exclusion of similar patients from clinical trials,10,11 and lack of real-world evidence supporting their use.12 In the absence of clinical trial evidence, studies evaluating the real-world effectiveness of GLP-1s in this high-risk population are needed, and comparative effectiveness studies of different diabetes management strategies may help inform clinical practice. Given guideline recommendations to utilize GLP-1s over other diabetes medications and limited data in patients receiving hemodialysis, we emulated a target trial using observational (nonrandomized) data13 to evaluate the comparative effectiveness and safety of GLP-1s versus DPP-4s, which are a more commonly used diabetes medication. Methods Data Source We used the USRDS database, which is a national surveillance system monitoring most patients in the US receiving dialysis or kidney transplant.1 We combined USRDS data with Medicare enrollment history and Medicare Part A, B, and D claims. Medicare claims are administrative data used by the Centers for Medicare and Medicaid Services for reimbursement and payment of service for Medicare beneficiaries. Medicare Part A contains hospitalizations/billing diagnoses; Medicare Part B contains physicians claims/billing diagnoses; and Medicare Part D contains medication dispenses. We evaluated claims from January 1, 2011 to December 31, 2021. Our dataset was assumed to be complete given linkage between USRDS data and Medicare claims data. The Jefferson Health Institutional Review Board approved the study and waived the need for informed consent. We followed the Transparent Reporting of Observational studies Emulating a Target Trial (TARGET) guideline.14 This study protocol was not prospectively registered, and the study design is comparable to our previous target trial emulation in patients receiving hemodialysis.15 Emulating a Target Trial: Active Comparator, New User Design Eligibility Criteria and Study Population We emulated a hypothetical target trial with enrollment starting on January 1, 2011 evaluating new users of GLP-1s versus new users of DPP-4s among patients receiving in-center hemodialysis (Supplementary Table S1). Patients were eligible for enrollment at medication initiation (i.e., baseline) if they were aged ≥ 18 years, survived ≥ 90 days on in-center hemodialysis (per USRDS recommendations),1 and did not have a dispense for a GLP-1 or DPP-4 in the previous 180 days. Exclusion criteria included the following: simultaneous dispense for GLP-1 and DPP-4 at study start, no longer on in-center hemodialysis (e.g., kidney transplant, recovery off dialysis, transition to other dialysis modality, or unconfirmed dialysis modality), < 180 days of continuous Medicare Part A, B, and D primary payer, missing information on the Centers for Medicare and Medicaid Services Medical Evidence Form 2728 (Figure 1), and absence of type 2 diabetes mellitus.Figure 1Study flow diagram. BMI, body-mass-index; CMS, Centers for Medicare & Medicaid services; DPP4, dipeptidyl peptidase 4 inhibitors; GLP1, glucagon like peptide-1 receptor agonists; sHR, subdistribution hazard ratio. KT/V, (quantitative measurement of dialysis adequacy). Study flow diagram. BMI, body-mass-index; CMS, Centers for Medicare & Medicaid services; DPP4, dipeptidyl peptidase 4 inhibitors; GLP1, glucagon like peptide-1 receptor agonists; sHR, subdistribution hazard ratio. KT/V, (quantitative measurement of dialysis adequacy). Treatment Strategies We compared new users of GLP-1s with new users of DPP-4s between January 1, 2011, and December 31, 2021 (Supplementary Table S2). Covariate Ascertainment We used Centers for Medicare and Medicaid Services Medical Evidence Form 2728 which contains patient demographics, initial dialysis access, comorbidities at dialysis initiation (i.e., atherosclerotic heart disease, cancer, chronic pulmonary disease, cerebrovascular disease, diabetes mellitus, dysrhythmia, heart failure, hypertension, myocardial infarction, other cardiac disease, and peripheral vascular disease), smoking status, end-stage renal disease network region (i.e., geographic region), and etiology of kidney failure (diabetes or not). We additionally used 1 inpatient International Classification of Disease (ICD) code or 2 outpatient ICD claims in the previous 180 days to evaluate history of atherosclerotic heart disease, cerebrovascular disease, heart failure, peripheral vascular disease, other cardiac disease, respiratory disease, liver disease, dysrhythmia, cancer, and diabetes mellitus (Supplementary Table S3).16 We additionally evaluated number of hospitalizations in the previous 180 days (including hypoglycemia- and hyperglycemia-specific hospitalizations from Medicare Part A), history of endocrinology and primary care visits (from Medicare Part B with primary care defined as “Internal Medicine,” “Family Practice,” “Geriatric Medicine,” or “General Medicine”), medication dispenses in the previous 90 days (Supplementary Table S2), and Medicare Part D subsidy status (from Medicare Part D). Finally, we assessed the most recent body-mass-index (BMI), KT/V (measure of dialysis adequacy), and erythropoietin stimulating agent status within the previous 90 days from monthly dialysis claims (with 92% having updated information in the previous 30 days before study initiation and 100% at 90 days previous). We additionally noted dialysis vintage (baseline minus dialysis start date) and baseline calendar year. Treatment Assignment We used the stabilized IPTW method to account for the observational, nonrandomized study design (i.e., emulation of randomization). We specifically estimated the probability of new GLP-1 versus new DPP-4 use for each patient using a logistic regression model including all covariates in Table 1 (i.e., propensity score). We then multiplied the inverse of this value by the probability of GLP-1 use (i.e., stabilized weight) to create an IPTW-population where covariates should be balanced between treatment groups. This emulates the randomization of intervention in a clinical trial.Table 1Baseline characteristics of new GLP-1 versus new DPP-4 users pre- and post-IPTWBaseline characteristicsPre-IPTWSMDPost-IPTWSMDGLP-1DPP-4GLP-1DPP-4n = 3671n = 12,039n = 3517n = 12,155Age, SD60 (12)65 (12)0.3963 (12)63 (13)0.03Male1853 (50)6374 (53%)0.051815 (52%)6338 (52%)0.01Race0.140.08White2382 (65%)7282 (60%)2277 (65%)7523 (62%)Black/African American1035 (28%)3646 (30%)990 (28%)3596 (30%)Asian122 (3%)705 (6%)140 (4%)628 (5%)Native Hawaiian or Pacific Islander64 (2%)205 (2%)55 (2%)205 (2%)American Indian or Alaska Native50sa (2%)150 (1%)40a (1%)159 (1%)Other or Multiracial< 11a (0%)51 (51%)< 11a (0%)45 (0%)Year of medication initiation0.690.082011< 11a (0%)86 (1%)22 (1%)71 (1%)201220sa (1%)342 (3%)92 (3%)283 (2%)201353 (1%)576 (5%)109 (3%)484 (4%)201487 (2%)925 (8%)206 (6%)776 (6%)2015151 (4%)1290 (11%)310 (9%)1105 (9%)2016205 (6%)1497 (12%)399 (11%)1309 (11%)2017362 (10%)1683 (14%)402 (11%)1561 (13%)2018542 (15%)1682 (14%)511 (15%)1724 (14%)2019873 (24%)1712 (14%)623 (18%)2006 (17%)2020850 (23%)1532 (13%)556 (16%)1823 (15%)2021517 (14%)714 (6%)287 (8%)1012 (8%)Dialysis vintage2.3 [1.2–2.6]1.9 [0.9–3.5]0.192.0 [1.0–3.7]2.0 [0.9–3.6]0.02KTV, median [IQR]1.6 [1.4–1.7]1.6 [1.4–1.8]0.011.6 [1.4–1.8]1.6 [1.4–1.8]0.01Smoking status at dialysis initiation183 (5%)585 (5%)0.01191 (5%)601 (5%)0.02Medicare Part D low-income subsidy2953 (80%)9518 (79%)0.032765 (79%)9671 (80%)0.02ESKD from diabetes2943 (80%)9106 (76%)0.112735 (78%)9339 (77%)0.02USRDS network region0.100.03West985 (27%)3042 (25%)910 (26%)3190 (26%)Northeast687 (19%)2740 (23%)733 (21%)2615 (22%)South1307 (36%)4128 (34%)1218 (35%)4183 (34%)Midwest692 (19%)2129 (18%)657 (19%)2167 (18%)Nephrology care before dialysis initiation0.140.03None1198 (33%)4694 (39%)1298 (37%)4505 (37%)< 6 mos557 (15%)1821 (15%)531 (15%)1840 (15%)> 12 mos1089 (30%)3,137 (26%)979 (28%)3264 (27%)6–12 mos827 (23%)2387 (20%)709 (20%)2546 (21%)Initial dialysis access via catheter2748 (75%)9231 (77%)0.042709 (77%)9308 (77%)0.01BMI, median [IQR]34 [29–40]29 [25–34]0.6531 [26–36]30 [25–36]0.01Arrhythmia702 (19%)2894 (24%)0.12789 (22%)2823 (23%)0.02Atherosclerosis896 (24%)4029 (33%)0.201080 (31%)3770 (31%)0.01Cancer213 (6%)928 (8%)0.08270 (8%)869 (7%)0.02Congestive heart failure1146 (31%)4751 (39%)0.171333 (38%)4537 (37%)0.01Dementia33 (1%)432 (4%)0.18101 (3%)357 (3%)< 0.01Diabetic retinopathy1140 (31%)2789 (23%)0.18924 (26%)3068 (25%)0.02Gastrointestinal bleed157 (4%)889 (7%)0.13217 (6%)798 (7%)0.02Hypertension3313 (90%)10,917 (91%)0.013203 (91%)11,026 (91%)0.01Nonambulatory status209 (6%)769 (6%)0.03238 (7%)762 (6%)0.02Liver disease217 (6%)1434 (12%)0.21344 (10%)1269 (10%)0.02Other cardiac disease472 (13%)2092 (17%)0.13562 (16%)1962 (16%)< 0.01Peripheral vascular disease1372 (37%)5861 (49%)0.231549 (44%)5544 (46%)0.03Respiratory disease450 (12%)1634 (14%)0.04494 (14%)1607 (13%)0.02Cerebrovascular disease645 (18%)2720 (23%)0.13743 (21%)2586 (21%)< 0.01Number of hospitalizations, < 6 mos0.400.05 02274 (62%)5356 (44%)1725 (49%)5954 (49%) 1775 (21%)2916 (24%)808 (23%)2832 (23%) 2330 (9%)1731 (14%)434 (12%)1583 (13%) 3150 (4%)952 (8%)244 (7%)843 (7%) 4+145 (4%)1095 (9%)312 (9%)953 (8%)Outpatient primary care visit, ≥ 12175 (59%)6835 (57%)0.052011 (57%)6948 (57%)< 0.01Outpatient endocrinology visit, ≥ 1929 (25%)2882 (24%)0.03806 (23%)2924 (24%)0.03Hyperglycemia hospitalization, ≥ 1298 (8%)1226 (10%)0.07365 (10%)1174 (10%)0.02Hypoglycemia hospitalization, ≥ 189 (2%)695 (6%)0.17181 (5%)604 (5%)0.01Number of Unique Medications, < 90 d11 [8–15]10 [7–14]0.1411 [8–14]11 [8–14]0.04ACE/ARB1115 (30%)4321 (36%)0.121277 (36%)4190 (34%)0.04Aldosterone46 (1%)146 (1%)< 0.0139 (1%)142 (1%)< 0.01Antiplatelet694 (19%)2458 (20%)0.04732 (21%)2485 (20%)0.01Alpha glucosidase inhibitors< 11a (0%)18 (0%)0.01< 11a (0%)21 (0%)0.01Beta blocker2028 (55%)7146 (59%)0.082123 (60%)7087 (58%)0.04Calcium channel blocker1520 (41%)5777 (48%)0.131575 (45%)5623 (46%)0.03Diuretics1172 (32%)3137 (26%)0.131007 (29%)3334 (27%)0.03DOAC260 (7%)683 (6%)0.06220 (6%)795 (7%)0.01History of any Epo use2948 (80%)9025 (75%)0.132685 (76%)9268 (76%)< 0.01Insulin2591 (71%)4975 (41%)0.621777 (51%)5900 (49%)0.04Phosphate binders2300 (63%)7306 (61%)0.042119 (60%)7374 (61%)0.01SGLT218 (0%)20 (0%)0.06< 11a31 (0%)< 0.01Sulfonylurea247 (7%)1558 (13%)0.21378 (11%)1389 (11%)0.02Thiazolidinediones84 (2%)358 (3%)0.04131 (4%)348 (3%)0.05Warfarin217 (6%)834 (7%)0.04217 (6%)808 (7%)0.02ACE/ARB, angiotensin-converting enzyme inhibitor/angiotensin receptor blocker; BMI, body mass index; DOAC, direct oral anticoagulant; DPP-4, dipeptidyl peptidase-4 inhibitor; Epo, erythropoietin; ESKD, end-stage kidney disease; GLP-1, glucagon-like peptide-1 receptor agonist; Kt/V, fractional urea clearance; SD, standard deviation; SGLT2, sodium-glucose cotransporter-2 inhibitor; SMD, absolute standardized mean difference; USRDS, United States Renal Data System.Data reported as mean (SD), number (percent), or median (interquartile interval).aCells with < 11 were censored per USRDS reporting policy or rounded down (if multiple categories). Table 1Baseline characteristics of new GLP-1 versus new DPP-4 users pre- and post-IPTWBaseline characteristicsPre-IPTWSMDPost-IPTWSMDGLP-1DPP-4GLP-1DPP-4n = 3671n = 12,039n = 3517n = 12,155Age, SD60 (12)65 (12)0.3963 (12)63 (13)0.03Male1853 (50)6374 (53%)0.051815 (52%)6338 (52%)0.01Race0.140.08White2382 (65%)7282 (60%)2277 (65%)7523 (62%)Black/African American1035 (28%)3646 (30%)990 (28%)3596 (30%)Asian122 (3%)705 (6%)140 (4%)628 (5%)Native Hawaiian or Pacific Islander64 (2%)205 (2%)55 (2%)205 (2%)American Indian or Alaska Native50sa (2%)150 (1%)40a (1%)159 (1%)Other or Multiracial< 11a (0%)51 (51%)< 11a (0%)45 (0%)Year of medication initiation0.690.082011< 11a (0%)86 (1%)22 (1%)71 (1%)201220sa (1%)342 (3%)92 (3%)283 (2%)201353 (1%)576 (5%)109 (3%)484 (4%)201487 (2%)925 (8%)206 (6%)776 (6%)2015151 (4%)1290 (11%)310 (9%)1105 (9%)2016205 (6%)1497 (12%)399 (11%)1309 (11%)2017362 (10%)1683 (14%)402 (11%)1561 (13%)2018542 (15%)1682 (14%)511 (15%)1724 (14%)2019873 (24%)1712 (14%)623 (18%)2006 (17%)2020850 (23%)1532 (13%)556 (16%)1823 (15%)2021517 (14%)714 (6%)287 (8%)1012 (8%)Dialysis vintage2.3 [1.2–2.6]1.9 [0.9–3.5]0.192.0 [1.0–3.7]2.0 [0.9–3.6]0.02KTV, median [IQR]1.6 [1.4–1.7]1.6 [1.4–1.8]0.011.6 [1.4–1.8]1.6 [1.4–1.8]0.01Smoking status at dialysis initiation183 (5%)585 (5%)0.01191 (5%)601 (5%)0.02Medicare Part D low-income subsidy2953 (80%)9518 (79%)0.032765 (79%)9671 (80%)0.02ESKD from diabetes2943 (80%)9106 (76%)0.112735 (78%)9339 (77%)0.02USRDS network region0.100.03West985 (27%)3042 (25%)910 (26%)3190 (26%)Northeast687 (19%)2740 (23%)733 (21%)2615 (22%)South1307 (36%)4128 (34%)1218 (35%)4183 (34%)Midwest692 (19%)2129 (18%)657 (19%)2167 (18%)Nephrology care before dialysis initiation0.140.03None1198 (33%)4694 (39%)1298 (37%)4505 (37%)< 6 mos557 (15%)1821 (15%)531 (15%)1840 (15%)> 12 mos1089 (30%)3,137 (26%)979 (28%)3264 (27%)6–12 mos827 (23%)2387 (20%)709 (20%)2546 (21%)Initial dialysis access via catheter2748 (75%)9231 (77%)0.042709 (77%)9308 (77%)0.01BMI, median [IQR]34 [29–40]29 [25–34]0.6531 [26–36]30 [25–36]0.01Arrhythmia702 (19%)2894 (24%)0.12789 (22%)2823 (23%)0.02Atherosclerosis896 (24%)4029 (33%)0.201080 (31%)3770 (31%)0.01Cancer213 (6%)928 (8%)0.08270 (8%)869 (7%)0.02Congestive heart failure1146 (31%)4751 (39%)0.171333 (38%)4537 (37%)0.01Dementia33 (1%)432 (4%)0.18101 (3%)357 (3%)< 0.01Diabetic retinopathy1140 (31%)2789 (23%)0.18924 (26%)3068 (25%)0.02Gastrointestinal bleed157 (4%)889 (7%)0.13217 (6%)798 (7%)0.02Hypertension3313 (90%)10,917 (91%)0.013203 (91%)11,026 (91%)0.01Nonambulatory status209 (6%)769 (6%)0.03238 (7%)762 (6%)0.02Liver disease217 (6%)1434 (12%)0.21344 (10%)1269 (10%)0.02Other cardiac disease472 (13%)2092 (17%)0.13562 (16%)1962 (16%)< 0.01Peripheral vascular disease1372 (37%)5861 (49%)0.231549 (44%)5544 (46%)0.03Respiratory disease450 (12%)1634 (14%)0.04494 (14%)1607 (13%)0.02Cerebrovascular disease645 (18%)2720 (23%)0.13743 (21%)2586 (21%)< 0.01Number of hospitalizations, < 6 mos0.400.05 02274 (62%)5356 (44%)1725 (49%)5954 (49%) 1775 (21%)2916 (24%)808 (23%)2832 (23%) 2330 (9%)1731 (14%)434 (12%)1583 (13%) 3150 (4%)952 (8%)244 (7%)843 (7%) 4+145 (4%)1095 (9%)312 (9%)953 (8%)Outpatient primary care visit, ≥ 12175 (59%)6835 (57%)0.052011 (57%)6948 (57%)< 0.01Outpatient endocrinology visit, ≥ 1929 (25%)2882 (24%)0.03806 (23%)2924 (24%)0.03Hyperglycemia hospitalization, ≥ 1298 (8%)1226 (10%)0.07365 (10%)1174 (10%)0.02Hypoglycemia hospitalization, ≥ 189 (2%)695 (6%)0.17181 (5%)604 (5%)0.01Number of Unique Medications, < 90 d11 [8–15]10 [7–14]0.1411 [8–14]11 [8–14]0.04ACE/ARB1115 (30%)4321 (36%)0.121277 (36%)4190 (34%)0.04Aldosterone46 (1%)146 (1%)< 0.0139 (1%)142 (1%)< 0.01Antiplatelet694 (19%)2458 (20%)0.04732 (21%)2485 (20%)0.01Alpha glucosidase inhibitors< 11a (0%)18 (0%)0.01< 11a (0%)21 (0%)0.01Beta blocker2028 (55%)7146 (59%)0.082123 (60%)7087 (58%)0.04Calcium channel blocker1520 (41%)5777 (48%)0.131575 (45%)5623 (46%)0.03Diuretics1172 (32%)3137 (26%)0.131007 (29%)3334 (27%)0.03DOAC260 (7%)683 (6%)0.06220 (6%)795 (7%)0.01History of any Epo use2948 (80%)9025 (75%)0.132685 (76%)9268 (76%)< 0.01Insulin2591 (71%)4975 (41%)0.621777 (51%)5900 (49%)0.04Phosphate binders2300 (63%)7306 (61%)0.042119 (60%)7374 (61%)0.01SGLT218 (0%)20 (0%)0.06< 11a31 (0%)< 0.01Sulfonylurea247 (7%)1558 (13%)0.21378 (11%)1389 (11%)0.02Thiazolidinediones84 (2%)358 (3%)0.04131 (4%)348 (3%)0.05Warfarin217 (6%)834 (7%)0.04217 (6%)808 (7%)0.02ACE/ARB, angiotensin-converting enzyme inhibitor/angiotensin receptor blocker; BMI, body mass index; DOAC, direct oral anticoagulant; DPP-4, dipeptidyl peptidase-4 inhibitor; Epo, erythropoietin; ESKD, end-stage kidney disease; GLP-1, glucagon-like peptide-1 receptor agonist; Kt/V, fractional urea clearance; SD, standard deviation; SGLT2, sodium-glucose cotransporter-2 inhibitor; SMD, absolute standardized mean difference; USRDS, United States Renal Data System.Data reported as mean (SD), number (percent), or median (interquartile interval).aCells with < 11 were censored per USRDS reporting policy or rounded down (if multiple categories). Baseline characteristics of new GLP-1 versus new DPP-4 users pre- and post-IPTW ACE/ARB, angiotensin-converting enzyme inhibitor/angiotensin receptor blocker; BMI, body mass index; DOAC, direct oral anticoagulant; DPP-4, dipeptidyl peptidase-4 inhibitor; Epo, erythropoietin; ESKD, end-stage kidney disease; GLP-1, glucagon-like peptide-1 receptor agonist; Kt/V, fractional urea clearance; SD, standard deviation; SGLT2, sodium-glucose cotransporter-2 inhibitor; SMD, absolute standardized mean difference; USRDS, United States Renal Data System. Data reported as mean (SD), number (percent), or median (interquartile interval). Cells with < 11 were censored per 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 followed until study outcome, kidney transplant, renal recovery off dialysis, loss of Medicare primary payer status, death, or December 31, 2021, whichever came first. The primary study outcome was all-cause mortality, and secondary outcomes included cholecystitis,17 hyperkalemia,18 hyperglycemia, hypoglycemia, pancreatitis,19 and diabetic retinopathy.20 We additionally evaluated the cumulative incidence of medication discontinuation during study follow-up. Date and cause of death were determined by the USRDS.1 We categorized cardiovascular mortality as those that the USRDS determined as arrhythmia, atherosclerosis, cardiac arrest, cardiomyopathy, congestive heart failure, myocardial, pericardial, or valvular cause of death (Centers for Medicare and Medicaid Services Form 2746). We estimated both intention-to-treat and per-protocol effects. In per-protocol 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). For secondary outcomes (i.e., cholecystitis, hyperkalemia, hyperglycemia, hypoglycemia, pancreatitis, and diabetic retinopathy), we defined the outcome of interest using the first hospitalization or outpatient claim (Supplementary Table S4). To increase specificity of outpatient claims, we required at least 2 outpatient claims, and we used the earlier date to define the study outcome date. We then repeated analyses looking at hospitalization only outcomes. Statistical Analysis We compared baseline characteristics by GLP-1 versus DPP-4 use before and after IPTW with good balance defined as an absolute standardized mean difference < 0.10.21 We used Cox proportional hazards regression on the IPTW-weighted cohort to assess GLP-1 use and risk of all-cause mortality and study outcomes. For each outcome, we estimated the HR. Kaplan-Meier plots are shown for all-cause mortality using the intention-to-treat and per-protocol approach. We used cause-specific Cox proportional hazards regression where follow-up was censored at the occurrence of competing events (i.e., noncause specific mortality and kidney transplant). We repeated analyses with Fine-Gray models22 to address a potential violation of independent censoring by competing risk. To estimate incidence of medication discontinuation, we used Fine-Gray models (accounting for the competing risks of mortality and kidney transplantation) on the IPTW-weighted cohort. For secondary outcomes, we excluded individuals who had a claim for the specific outcome at any time before study initiation. These exclusions were assessed independently for each outcome, and IPTW weights were recalculated for each analysis. Analyses were performed in R (version 4.2.2, R Foundation for Statistical Computing, Vienna, Austria),23 and Stata (version 17, StataCorp LLC, College Station, TX).24 Significant results were determined by a 2-tailed P-value < 0.05. Sensitivity Analyses We looked at subgroups by age (≥ 65), physician reported sex on CMS form 2728, race, BMI (by quartile), heart failure, insulin use, and Medicare Part D low-income subsidy for all-cause mortality using the intention-to-treat and per-protocol approach. We recalculated IPTW weights within each subgroup and included an interaction term between the treatment variable and subgroup indicator variable in the Cox regression model. We repeated our primary and secondary outcome analysis using a 60-day (instead of 30-day) refill window to define medication discontinuation. We then estimated the E-value,25 which is the minimum relative risk that an unmeasured confounder would need to have with both the exposure and the outcome to explain away the observed association between GLP-1 use and all-cause mortality. To provide context for the E-value, we reported the top 3 HR of study covariates for all-cause mortality using multivariable Cox proportional hazards models. Finally, we conducted 3 additional sensitivity analyses. First, we truncated study follow-up on December 31, 2020 (instead of 2021), to mitigate censoring related to the 21st Century Cures Act,26 which allowed patients receiving hemodialysis to switch from Medicare primary payer to Medicare Advantage starting January 1, 2021. Second, we conducted a subgroup analysis with treatment year (before or after January 1, 2018) to account for changes in medication utilization following semaglutide approval; and third, we utilized inverse probability of censoring weighting (IPCW) to adjust for baseline predictors of treatment discontinuation in the per-protocol analysis.27 Positive and Negative Control Outcomes We tested our target trial emulation approach by using positive and negative control outcomes. We used gastrointestinal distress (R11, R12, R14, K59.0, and K30) as a positive control, and we used lower respiratory diseases (J40–47), hearing loss (H90, H91), and lens disorders (H25–28) as negative controls.28 We used Fine-Gray models to estimate subdistribution HRs (sHR) accounting for the competing risks of all-cause mortality and kidney transplantation. Results Patient Characteristics We identified 3671 GLP-1 and 12,039 DPP-4 new users who met eligibility criteria (Figure 1). Most GLP-1 users were prescribed dulaglutide (41%), liraglutide (34%), or semaglutide (18%). Before weighting, GLP-1 users (vs. DPP-4) were more likely to be younger, White, have higher dialysis vintage, have nondiabetic cause of kidney failure, had seen a nephrologist before dialysis initiation, have a higher BMI, and use insulin. They were less likely to have arrhythmia, atherosclerosis, congestive heart failure, dementia, diabetic retinopathy, history of GI bleeding, liver disease, peripheral vascular disease, cerebrovascular disease, recent hospitalizations, hypoglycemia hospitalization, and use sulfonylureas. After IPTW, all variables were balanced with standardized mean differences < 0.10 (Table 1, Supplementary Figure S1). In the weighted cohort, there were 3517 GLP-1 and 12,155 DPP-4 users. The mean age of the weighted study population was 63 years, 63% were White, and 52% were male. Median (interquartile interval) dialysis vintage was 2.0 (1.0–3.6) years, and 77% had kidney failure from diabetes (Table 1). Intention to Treat Results The median (interquartile interval) follow-up was 1.5 (0.6–2.9) years, and in the unweighted cohort, the primary outcome of all-cause mortality occurred in 1158 individuals on GLP-1s and 6014 individuals on DPP-4s. Other individuals were censored for no longer having Medicare primary payer (n = 4658), surviving until the end of follow-up (n = 2782), receiving a kidney transplant (n = 748), and having renal recovery off hemodialysis (n = 350). Notably, of patients censored for loss of Medicare primary payer, 83% occurred in 2021, which was the last year of study follow-up. In the weighted cohort, the cumulative incidence of mortality was 18% at 1 year and 35% at 2 years for GLP-1s versus 19% and 34%, respectively, for DPP-4s (Figure 2). All-cause mortality (HR: 0.98, 95% CI: 0.89–1.07), cardiovascular disease mortality (sHR: 1.06, 95% CI: 0.92–1.23), or infection mortality (sHR: 0.83, 95% CI: 0.54–1.26, Table 2) were not significantly different between GLP-1 users and DPP-4 users. Kidney transplantation was also not significantly different (Supplementary Table S5).Figure 2IPTW-KM of all-cause mortality by GLP-1 (vs. DPP-4) use. Intention-to-treat (left) versus per-protocol (right). DPP-4, dipeptidyl peptidase 4 inhibitors; GLP-1, glucagon like peptide - 1 receptor agonists; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio.Table 2HR of mortality by GLP-1 (vs. DPP-4) use in Medicare beneficiaries receiving hemodialysisCauses of mortalityIPTW-No. of eventsEffect estimate—hazard ratiosIntention to treatGLP-1N = 3517DPP-4 N = 12,155IPTW-HRIPTW-sHRAll-cause153155880.98 (0.89–1.07)Cardiac67722761.06 (0.92–1.23)1.07 (0.93–1.23)Infection903870.83 (0.54–1.26)0.83 (0.54–1.26)Per-protocolAll-cause36119620.81 (0.68–0.96)Cardiac1748620.89 (0.69–1.13)0.87 (0.68–1.11)Infection181210.65 (0.37–1.14)0.65 (0.37–1.14)DPP-4, dipeptidyl peptidase-4 inhibitors; GLP-1, glucagon-like peptide-1 receptor agonists; HR, hazard ratio; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio.IPTW-HRs were estimated using Cox proportional hazards regression, and IPTW-sHRs were estimated using Fine-Gray models accounting for competing risk of mortality and kidney transplantation. IPTW-KM of all-cause mortality by GLP-1 (vs. DPP-4) use. Intention-to-treat (left) versus per-protocol (right). DPP-4, dipeptidyl peptidase 4 inhibitors; GLP-1, glucagon like peptide - 1 receptor agonists; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio. Table 2HR of mortality by GLP-1 (vs. DPP-4) use in Medicare beneficiaries receiving hemodialysisCauses of mortalityIPTW-No. of eventsEffect estimate—hazard ratiosIntention to treatGLP-1N = 3517DPP-4 N = 12,155IPTW-HRIPTW-sHRAll-cause153155880.98 (0.89–1.07)Cardiac67722761.06 (0.92–1.23)1.07 (0.93–1.23)Infection903870.83 (0.54–1.26)0.83 (0.54–1.26)Per-protocolAll-cause36119620.81 (0.68–0.96)Cardiac1748620.89 (0.69–1.13)0.87 (0.68–1.11)Infection181210.65 (0.37–1.14)0.65 (0.37–1.14)DPP-4, dipeptidyl peptidase-4 inhibitors; GLP-1, glucagon-like peptide-1 receptor agonists; HR, hazard ratio; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio.IPTW-HRs were estimated using Cox proportional hazards regression, and IPTW-sHRs were estimated using Fine-Gray models accounting for competing risk of mortality and kidney transplantation. HR of mortality by GLP-1 (vs. DPP-4) use in Medicare beneficiaries receiving hemodialysis DPP-4, dipeptidyl peptidase-4 inhibitors; GLP-1, glucagon-like peptide-1 receptor agonists; HR, hazard ratio; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio. IPTW-HRs were estimated using Cox proportional hazards regression, and IPTW-sHRs were estimated using Fine-Gray models accounting for competing risk of mortality and kidney transplantation. Per-protocol Results In the unweighted cohort, censoring because of medication discontinuation was high with 1970 (54%) stopping GLP-1s after a median (interquartile interval) 120 (60–270) days, and 6826 (57%) stopping DPP-4s after 163 (60–448) days. The sHR of mediation discontinuation for GLP-1s versus DPP-4s was 1.36 (95% CI: 1.26–1.45) with a cumulative incidence of mediation discontinuation at 0.25, 0.5, and 1-year of 27%, 43%, and 53%, respectively, for GLP-1 use, versus 17%, 30%, and 42%, respectively, for DPP-4 use (Figure 3). GLP-1 use was significantly associated with a reduction in all-cause mortality (HR: 0.81, 95% CI: 0.68–0.96) but not cardiovascular disease (HR: 0.89, 95% CI: 0.69–1.13) or infection-related mortality (HR: 0.65, 95% CI: 0.37–1.14, Table 2). The cumulative incidence of all-cause mortality at 1 year was 11% versus 13%, and at 2-years, it was 19% versus 22%. In the unweighted cohort, there were 343 deaths among those on GLP-1s and 2067 on DPP-4s. Other individuals were censored for no longer having Medicare primary payer (n = 2396), survival until end of follow-up (n = 1081), undergoing kidney transplantation (n = 444; Supplementary Table S5), initiation of opposing medication class (n = 394), and having renal recovery (n = 189).Figure 3IPTW-cumulative incidence curve of GLP-1 versus DPP-4 discontinuation in per-protocol analysis. Estimated via Fine-Gray models accounting for competing risk of all-cause mortality and kidney transplantation. DPP4, dipeptidyl peptidase 4 inhibitors; GLP-1, glucagon like peptide - 1 receptor agonists; IPTW, inverse probability of treatment weighting. IPTW-cumulative incidence curve of GLP-1 versus DPP-4 discontinuation in per-protocol analysis. Estimated via Fine-Gray models accounting for competing risk of all-cause mortality and kidney transplantation. DPP4, dipeptidyl peptidase 4 inhibitors; GLP-1, glucagon like peptide - 1 receptor agonists; IPTW, inverse probability of treatment weighting. Secondary Outcomes GLP-1s were not associated with cholecystitis, hyperkalemia, hypoglycemia, pancreatitis, or diabetic retinopathy in intention-to-treat or per-protocol analyses. GLP-1s were associated with hyperglycemia in per-protocol analysis (sHR: 1.22, 95% CI: 1.02–1.46) but not intention-to-treat analysis (sHR: 1.06, 95% CI: 0.92–1.23, Table 3). When we used inpatient claims only to ascertain secondary outcomes, GLP-1s were not associated with any secondary outcome in intention to treat or per-protocol analysis (Supplementary Table S6).Table 3HR of secondary outcomes associated with GLP-1 use (vs. DPP-4s) in Medicare beneficiaries receiving hemodialysisSecondary outcomesIPTW-No. of eventsEffect estimate—hazard ratiosIntention to treatEvents/IPTW-GLP-1Events/IPTW-DPP-4IPTW-HRIPTW-sHRCholecystitis50/3448185/11,9991.02 (0.68–1.53)0.99 (0.66–1.49)Hyperkalemia644/29442400/96120.92 (0.91–1.04)0.91 (0.80–1.02)Hyperglycemia419/14642284/78451.10 (0.95–1.27)1.06 (0.92–1.23)Hypoglycemia310/28571222/10,1820.98 (0.81–1.18)0.95 (0.80–1.15)Pancreatitis68/3425281/11,8300.91 (0.64–1.29)0.89 (0.62–1.26)Retinopathy382/15361981/77101.04 (0.90–1.20)1.02 (0.89–1.19)Per-protocolCholecystitis23/344898/11,9991.04 (0.88–1.24)1.01 (0.54–2.12)Hyperkalemia334/29441419/96120.95 (0.81–1.11)0.94 (0.81–1.11)Hyperglycemia236/14641405/78451.27 (1.06–1.52)1.22 (1.02–1.46)Hypoglycemia111/2857617/10,1820.81 (0.64–1.02)0.81 (0.64–1.03)Pancreatitis38/3425161/11,8301.03 (0.66–1.61)1.02 (0.65–1.60)Retinopathy208/15361536/77101.04 (0.88–1.24)1.06 (0.89–1.26)DPP-4, Dipeptidyl peptidase 4 inhibitors; GLP-1, Glucagon Like Peptide - 1 Receptor Agonists; HR, hazard ratio; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio.For each outcome, we excluded individuals who had a claim for the specific outcome at any time before study initiation. IPTW-HRs were estimated using Cox proportional hazards regression, and IPTW-sHRs were estimated using Fine-Gray models accounting for competing risk of mortality and kidney transplantation. Table 3HR of secondary outcomes associated with GLP-1 use (vs. DPP-4s) in Medicare beneficiaries receiving hemodialysisSecondary outcomesIPTW-No. of eventsEffect estimate—hazard ratiosIntention to treatEvents/IPTW-GLP-1Events/IPTW-DPP-4IPTW-HRIPTW-sHRCholecystitis50/3448185/11,9991.02 (0.68–1.53)0.99 (0.66–1.49)Hyperkalemia644/29442400/96120.92 (0.91–1.04)0.91 (0.80–1.02)Hyperglycemia419/14642284/78451.10 (0.95–1.27)1.06 (0.92–1.23)Hypoglycemia310/28571222/10,1820.98 (0.81–1.18)0.95 (0.80–1.15)Pancreatitis68/3425281/11,8300.91 (0.64–1.29)0.89 (0.62–1.26)Retinopathy382/15361981/77101.04 (0.90–1.20)1.02 (0.89–1.19)Per-protocolCholecystitis23/344898/11,9991.04 (0.88–1.24)1.01 (0.54–2.12)Hyperkalemia334/29441419/96120.95 (0.81–1.11)0.94 (0.81–1.11)Hyperglycemia236/14641405/78451.27 (1.06–1.52)1.22 (1.02–1.46)Hypoglycemia111/2857617/10,1820.81 (0.64–1.02)0.81 (0.64–1.03)Pancreatitis38/3425161/11,8301.03 (0.66–1.61)1.02 (0.65–1.60)Retinopathy208/15361536/77101.04 (0.88–1.24)1.06 (0.89–1.26)DPP-4, Dipeptidyl peptidase 4 inhibitors; GLP-1, Glucagon Like Peptide - 1 Receptor Agonists; HR, hazard ratio; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio.For each outcome, we excluded individuals who had a claim for the specific outcome at any time before study initiation. IPTW-HRs were estimated using Cox proportional hazards regression, and IPTW-sHRs were estimated using Fine-Gray models accounting for competing risk of mortality and kidney transplantation. HR of secondary outcomes associated with GLP-1 use (vs. DPP-4s) in Medicare beneficiaries receiving hemodialysis DPP-4, Dipeptidyl peptidase 4 inhibitors; GLP-1, Glucagon Like Peptide - 1 Receptor Agonists; HR, hazard ratio; IPTW, inverse probability of treatment weighting; sHR, subdistribution hazard ratio. For each outcome, we excluded individuals who had a claim for the specific outcome at any time before study initiation. IPTW-HRs were estimated using Cox proportional hazards regression, and IPTW-sHRs were estimated using Fine-Gray models accounting for competing risk of mortality and kidney transplantation. Sensitivity Analysis Covariates were well balanced across all sub-groups with standardized mean differences < 0.10, and we did not observe significant interaction across intention-to-treat and per-protocol analysis (Supplementary Figures S2 and S3). When we used a 60-day refill window (instead of 30-day refill window), per-protocol results remained consistent. We did not estimate the E-value for the intention-to-treat analysis, as GLP-1 use (vs. DPP4) was not associated with all-cause mortality. For the per-protocol analysis, the E-value was 1.61. In comparison, the measured covariates with the strongest associations with all-cause mortality were smoking at dialysis initiation (HR: 1.39), congestive heart failure (HR: 1.28), and nonambulatory status (HR: 1.21). When we ended study follow-up on December 31, 2020, both intention-to-treat and per-protocol analyses were unchanged. Similarly, we did not observe significant interaction by medication initiation before or after January 1, 2018. After further applying IPCW for medication discontinuation in per-protocol analysis, the IPCW-IPTW-HR was 0.79 (95% CI: 0.63–0.99), consistent with the primary analysis (Supplementary Figure S4). Positive or Negative Control Outcomes GLP-1s were associated with increased risk of gastrointestinal symptoms (sHR: 1.11, 95% CI: 1.02–1.20]), and there was no increased risk for pneumonia (sHR: 1.06, 95% CI: 0.96–1.18), hearing loss (sHR: 1.03, 95 CI: 0.84–1.26), or lens disease (sHR: 1.05, 95% CI: 0.92–1.19). Average treatment estimates were similar (data not shown). Discussion In this IPTW retrospective target trial emulation study among patients receiving hemodialysis, GLP-1 (vs. DPP-4) use was not associated with a significant reduction in all-cause mortality when using the intention-to-treat approach. The cumulative incidence of medication discontinuation at 1-year was 53% for GLP-1s versus 42% for DPP-4s. In both intention-to-treat and as-treated analyses, GLP-1 use was not associated with cholecystitis, hyperkalemia, hypoglycemia, pancreatitis, or retinopathy. Our results suggest that GLP-1s versus DPP-4s in patients receiving hemodialysis is not associated with improvements in mortality, but the high rate of medication discontinuation may limit the causal implications of our primary findings. Our negative mortality results using the intention-to-treat approach are disappointing given landmark clinical trials showing efficacy of GLP-1 in patients with and without moderate kidney disease.3, 4, 5, 6, 7,10 The null association between GLP-1 use and all-cause mortality in our study may be related to increased morbidity and mortality in this population, which may not be amenable to a single medication intervention. Hydroxymethylglutaryl-coenzymeA reductase inhibitors (i.e., statins) and angiotensin receptor blockers, which have both been shown to reduce mortality in the general population, showed no benefit in randomized clinical trials of patients receiving hemodialysis29, 30, 31—with limited efficacy attributed to high comorbidity and late-stage disease. We did see increased risks of gastrointestinal symptoms (i.e., positive control outcome), consistent with findings from clinical trials4, 5, 6, 7,10 and observational studies of GLP-1 use,32 which supports the validity of our trial emulation approach. Gastrointestinal symptoms occur early after medication initiation,32 which may explain these positive findings despite a null result for all-cause mortality in intention-to-treat analysis. Three previous studies have also evaluated the effectiveness of GLP-1s on mortality in patients receiving hemodialysis, and GLP-1 use was associated with reductions in all-cause mortality in all 3.12,33,34 However, the first study included patients with advanced kidney disease without dialysis dependence and had only 208 GLP-1 users receiving dialysis compared with 3671 in our study. The 2 remaining studies used USRDS data, but they employed different methodologic frameworks.14 One used a new-user versus nonuser design, which can be more susceptible to confounding by indication and immortal time bias33 than the active comparator, new-user design used in this study. Finally, the last study conditioned on subsequent GLP-1 dispensing,34 which would not account for the high rate of GLP-1 discontinuation that we observed. Thus, the high rate of medication discontinuation may have limited GLP-1 effectiveness. Randomized trials of GLP-1s had 1-year medication adherence rates above 80% to 90%,4,6,7,10 which was significantly higher than our 1-year adherence of ∼ 50%. When we accounted for medication continuation (which was similar to other US observational studies reporting ∼ 50% 1-year adherence),35, 36, 37 we saw a statistically significant 20% reduction in all-cause mortality with GLP-1s (vs. DPP-4s), compared with 12% seen in meta-analysis of randomized trials.38 Patients who remained on therapy may represent those who tolerated therapy or saw beneficial effects on weight and/or glycemic control. Alternatively, long-term use of GLP-1s may decrease the proinflammatory state39 seen in dialysis dependent kidney disease.40 We did not observe a significant protective association with cardiovascular disease mortality or infection-related mortality12 in intention-to-treat or per-protocol analyses, but USRDS defined cause-specific mortality may be susceptible to misclassification.41 The DPP-4 adherence rate was also low at 60% at 1 year, but this is similar to the general population.42 Consistent with the “obesity paradox,” where obesity is associated with improved survival,43 we observed the highest mortality in the BMI < 25 kg/m2 group, but we did not observe a significant interaction between GLP-1 use and BMI subgroups in intention-to-treat or per-protocol analysis. Future longitudinal studies assessing weight and nutrition status are needed to contextualize the risks and benefits of GLP-1 associated weight loss.33,44 GLP-1s were generally safe without increased risk of cholecystitis or pancreatitis. There was no difference in diabetic retinopathy,45 but less than 50% of patients receiving hemodialysis receive an annual diabetes eye exam.9 For hyperkalemia, where GLP-1s have been associated with decreased risk in patients with nondialysis dependent kidney disease,18 we did not observe a difference in risk. GLP-1s also improve glycemic control, but if these benefits extend to patients receiving dialysis, who have high rates of hypo- and hyperglycemia,46 require further investigation. We saw no difference in hypoglycemia, and there was increased risk with hyperglycemia in per-protocol analysis. However, the accuracy and clinical severity of our ICD outcome ascertainment is unclear, as GLP-1s have greater reductions in hemoglobin A1c in the general population.47 Ongoing studies evaluating the effect of GLP-1s on glucose time-in-range in patients receiving hemodialysis will help address this question.48,49 Further studies are needed to replicate our safety findings, and studies evaluating other outcomes (e.g., obstructive sleep apnea,50 steatohepatitis,51 alcohol use disorder,52 mental health,53 and ischemic optic neuropathy)54 will further clarify the risks and benefits of GLP-1 use in this population. Our study has several limitations. Given the observational design with claims data and lack of randomization, there is likely unmeasured and residual confounding despite the rigorous IPTW method. Examples include hemoglobin A1c, weight, and frailty status, which may affect choice of medication and all-cause mortality. Similarly, these variables may act as time-varying confounders after medication initiation. Although we applied IPCW to address differential medication discontinuation, our per-protocol estimates should be interpreted cautiously, because we were only able to use baseline covariates to derive IPCW given limited data availability. To address residual confounding, we analyzed positive and negative control outcomes, which were consistent with findings seen in the general population. Moreover, using the E-value method in per-protocol analysis, we estimated an unmeasured confounder would need a risk ratio of ≥ 1.61 for both exposure and outcome to explain away the observed benefit, but the combination of unmeasured confounders could exceed this value. We could not evaluate indications for prescription (e.g., diabetes or weight loss). We were also unable to assess dosing patterns, which may affect clinical outcomes, or specific GLP-1s medications. Most GLP-1 users were on dulaglutide or liraglutide, which may confer decreased weight loss and/or cardiovascular benefit compared with newer GLP-1s like semaglutide or tirzepatide.10,50,55, 56, 57 Similarly, we did not evaluate longitudinal changes in weight or glycemic control which may have clinical significance for those who remain on treatment given our different results from the intention-to-treat versus per-protocol analysis. The generalizability of our findings may also be limited as all patients had Medicare as their primary payer. Access to GLP-1s may differ from that of other populations with private insurance, Medicare Advantage, or those in international health systems. Finally, there may be misclassification given ascertainment of covariates and outcomes using ICD codes, and the power for subgroup interactions may have been limited by low event size (i.e., cholecystitis or pancreatitis). Despite these limitations, strengths include the target trial emulation approach, the use of medication dispensing records rather than prescription records, and the strict requirement of Medicare primary payer status before and after study initiation to ensure accurate ascertainment of covariates, exposures, and secondary outcomes. In conclusion, for patients receiving in-center hemodialysis, GLP-1 (vs. DPP-4) initiation was not associated with a reduction in mortality, but in per-protocol analysis, where GLP-1 discontinuation was over 50% at 1-year, continuous GLP-1 use was associated with a reduction in mortality. Disclosure DL 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. The data reported here have been supplied by the US Renal Data System (USRDS). The interpretation and reporting of these data are the responsibility of the author(s) and in no way should be seen as an official policy or interpretation of the US government. All the other authors declared no competing interests.