Glucagon-like peptide-1 receptor agonists for major cardiovascular and kidney outcomes in type 1 diabetes

Abstract

There is substantial interest in the use of glucagon-like peptide-1 receptor agonists (GLP-1RAs) in type 1 diabetes, but data on the long-term clinical outcomes are lacking. Using national electronic health record data from 174,678 patients with type 1 diabetes, we conducted a sequential target trial emulation from January 2013 to March 2024. After propensity score weighting, GLP-1RA initiation was associated with significantly lower risks of major adverse cardiovascular events (5-year risk: 4.3% versus 5.0%; risk difference: −0.7% (95% confidence interval (CI): −1.2% to −0.2%); hazard ratio (HR): 0.85 (0.77–0.95)) and end-stage kidney disease (5-year risk: 1.6% versus 1.9%; risk difference: −0.3% (−0.6% to 0%); HR: 0.81 (0.69–0.95)). No increased risks of hospitalization for diabetic ketoacidosis or severe hypoglycemia were observed. These findings suggest that GLP-1RAs may be beneficial against major adverse cardiorenal events in patients with type 1 diabetes, without compromising safety.

A trial emulation study using electronic health record data from 174,678 patients with type 1 diabetes revealed that GLP-1RA initiation is associated with reduced risk of major cardiovascular events and end-stage kidney disease.

Main

Individuals with type 1 diabetes (T1D) have higher risks of cardiovascular disease and chronic kidney disease compared with individuals without diabetes¹. Despite advances in glycemic management, 31% and 7% of patients with T1D develop major adverse cardiovascular events (MACEs) and end-stage kidney disease (ESKD) by middle age, respectively^2,3. Management of traditional cardiorenal risk factors in T1D remains suboptimal; for example, only 20–30% of patients in the USA achieve the recommended glycemic control targets^4,5. New strategies are urgently needed to prevent cardiorenal events in this high-risk population.

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) have demonstrated clinically significant cardiovascular and kidney benefits in individuals with type 2 diabetes (T2D) and cardiovascular benefits in those with obesity and atherosclerotic cardiovascular disease at baseline^6,7. However, landmark cardiovascular and kidney outcome trials of GLP-1RAs excluded T1D populations^6,7. Evidence in T1D is limited to small, short-term, randomized clinical trials (RCTs) and a few observational studies focusing on surrogate outcomes such as glycemic control and weight loss^8–12. Two early trials (ADJUNCT ONE and TWO) showed increased risks of symptomatic hypoglycemia and hyperglycemia with ketosis associated with liraglutide^8,9. However, a later trial reported improved safety with continuous glucose monitoring¹⁰ and a meta-analysis of 13 RCTs found no increased risk of severe hypoglycemia¹³. Liraglutide has been the most extensively studied GLP-1RA in T1D, but data on newer agents (semaglutide and tirzepatide) remain limited.

Conducting RCTs evaluating hard cardiorenal outcomes in T1D is challenging due to the young age of this population, with low event rates requiring extended follow-up. Indeed, no RCT has evaluated therapies for the prevention of MACEs or ESKD in T1D. Target trial emulation applies RCT design principles to observational data and can yield conclusions comparable to those of RCTs¹⁴. This approach enables timely evaluation of GLP-1RA’s potential effects on major cardiovascular and kidney outcomes in T1D.

Therefore, we conducted a target trial emulation using de-identified electronic health record (EHR) data from Optum Labs Data Warehouse (OLDW) to evaluate the long-term effectiveness and safety of GLP-1RA initiation in T1D. OLDW is a national dataset including ~300 million patients from >60 health systems. The primary outcomes were MACEs (composite of myocardial infarction, stroke or all-cause mortality) and ESKD (dialysis or kidney transplantation). Secondary outcomes included hospitalization for heart failure (HF), major adverse liver events (composite of decompensated cirrhosis, hepatocellular carcinoma or liver transplantation) and weight loss. We included HF and liver outcomes because emerging evidence suggests that GLP-1RAs may improve these outcomes in T2D, partly through weight loss^15,16. Similar effects may be plausible in T1D, but data are limited. Safety outcomes were hospitalization for diabetic ketoacidosis (DKA), hospitalization for severe hypoglycemia and gastrointestinal events (composite of biliary disease, pancreatitis, bowel obstruction or gastroparesis).

Consistent with prior sequential target trial emulation studies¹⁷, we sequentially emulated 135 target trials constructed every consecutive month between January 2013 and March 2024 (Extended Data Fig. 1). Individuals initiating GLP-1RA within the month of enrollment were classified as initiators; all others were noninitiators. Both groups could use other noninsulin glucose-lowering agents within a year before the baseline. The study population included 174,678 individuals with T1D who contributed 6,092,537 person-trials (mean age 43 years, 47% female). Of these, 14,488 person-trials initiated GLP-1RA during a median follow-up of 38 (interquartile range: 18–64) months. Extended Data Table 1 shows baseline characteristics of all person-trials before and after propensity score weighting. After weighting, baseline characteristics were well balanced between groups (all standardized mean differences <10%).

Extended Data Fig. 1.

Study design and enrollment.

Extended Data Table 1.

Baseline characteristics of person-trials among 174,678 patients with type 1 diabetes by GLP-1RA initiation before and after weighting

Note: The study population was pooled from 135 trials and one individual may contribute to multiple trials. * Variables measured within the one year prior. Abbreviations: ACEi, angiotensin-converting enzyme inhibitor; ACR, albumin-to-creatinine ratio; ARB, angiotensin receptor blocker; CCB, calcium channel blocker; DPP4i, dipeptidyl peptidase 4 inhibitor; eGFR, estimated glomerular filtration rate; GLP-1RA, glucagon-like peptide-1 receptor agonist; MACE, major adverse cardiovascular event; SBP, systolic blood pressure; SD, standard deviation; SMD, standardized mean difference; SGLT2i, sodium-glucose cotransporter 2 inhibitor.

In intention-to-treat analysis of the primary outcomes, the risk of MACEs was lower among GLP-1RA initiators compared with noninitiators, with Kaplan–Meier curves separating after 1.5 years and remaining divergent (Fig. 1a). The 5-year risk of MACEs was 4.3% in the GLP-1RA initiation group and 5.0% in the noninitiation group (risk difference: −0.7% (95% CI: −1.2% to −0.2%)). The overall HR across the study period was 0.85 (95% CI: 0.77–0.95). GLP-1RA initiation was associated with a lower risk of ESKD (Fig. 1b). The risk was 1.6% with GLP-1RA versus 1.9% without at 5 years (risk difference: −0.3% (95% CI: −0.6% to 0.0%)), with an overall HR of 0.81 (95% CI: 0.69–0.95).

Fig. 1. Weighted cumulative incidence of primary outcomes by GLP-1RA initiation among patients with T1D.

a,b, Kaplan–Meier curves showing cumulative incidence of MACEs (a) and ESKD (b) among patients with T1D initiating GLP-1RA (blue) and not initiating GLP-1RA (red). For MACEs, person-trials with prior outcome events were excluded (unweighted n = 5,300,986).

In analyses of secondary outcomes, risk of hospitalization for HF was lower with GLP-1RA initiation (HR: 0.82 (95% CI: 0.71–0.94)). GLP-1RA initiation was associated with a lower risk of major adverse liver events (HR: 0.72 (95% CI: 0.60–0.85)). Patients who initiated GLP-1RA were more likely to achieve weight loss of 5%, 10% and 15% (Table 1), with Kaplan–Meier curves diverging within 1 year and remaining separated throughout follow-up (Extended Data Fig. 2c–e).

Table 1.

Risk of effectiveness and safety outcomes comparing GLP-1RA initiation versus noninitiation among patients with T1D

Outcome	Unweighted no. of person-trials		Unweighted no. of events		Weighted 5-year risks, % (95% CI)		Weighted 5-year risk difference, % (95% CI)	Weighted HR (95% CI)
	GLP-1RA initiation	Noninitiation	GLP-1RA initiation	Noninitiation	GLP-1RA initiation	Noninitiation
Primary outcomes
MACEsa	11,873	5,289,113	344	153,312	4.3 (3.8, 4.8)	5.0 (4.9, 5.1)	−0.7 (−1.2, −0.2)	0.85 (0.77, 0.95)
Myocardial infarctiona	11,873	5,289,113	115	51,808	1.6 (1.3, 1.9)	1.8 (1.8, 1.8)	−0.2 (−0.5, 0.1)	0.79 (0.66, 0.95)
Strokea	11,873	5,289,113	131	48,842	1.7 (1.3, 2.0)	1.7 (1.7, 1.8)	−0.1 (−0.4, 0.3)	0.93 (0.78, 1.10)
All-cause mortalitya	11,873	5,289,113	128	66,965	1.5 (1.2, 1.9)	1.9 (1.8, 1.9)	−0.4 (−0.7, 0.0)	0.84 (0.71, 1.00)
ESKD	14,488	6,078,049	148	73,902	1.6 (1.3, 1.9)	1.9 (1.9, 1.9)	–0.3 (–0.6, 0.0)	0.81 (0.69, 0.95)
Secondary outcomes
HFb	13,545	5,820,496	208	74,088	2.4 (2.0, 2.8)	2.7 (2.6, 2.7)	−0.3 (−0.7, 0.1)	0.82 (0.71, 0.94)
Major adverse liver eventsc	14,413	6,053,034	124	55,597	1.4 (1.1, 1.7)	1.8 (1.7, 1.8)	−0.4 (−0.7, −0.1)	0.72 (0.60, 0.85)
Weight loss 5%d	13,691	5,487,154	7,724	2,165,362	68.9 (67.9, 69.9)	63.3 (63.2, 63.4)	5.6 (4.6, 6.7)	1.25 (1.23, 1.28)
Weight loss 10%d	13,691	5,487,154	4,629	1,141,236	45.1 (44.0, 46.3)	40.2 (40.0, 40.3)	5.0 (3.7, 6.0)	1.22 (1.19, 1.26)
Weight loss 15%d	13,691	5,487,154	2,653	621,081	27.7 (26.6, 28.7)	24.9 (24.8, 25.0)	2.8 (1.7, 3.8)	1.14 (1.09, 1.18)
Safety outcomes
DKA	14,488	6,078,049	596	445,095	6.0 (5.5, 6.6)	7.1 (7.0, 7.1)	−1.1 (−1.6, −0.5)	0.83 (0.76, 0.90)
Severe hypoglycemia	14,488	6,078,049	210	123,234	2.2 (1.8, 2.5)	2.6 (2.5, 2.6)	−0.4 (−0.7, 0.0)	0.82 (0.72, 0.94)
Gastrointestinal eventse	12,840	5,434,028	884	268,475	9.9 (9.1, 10.6)	9.7 (9.6, 9.8)	0.2 (−0.6, 1.0)	1.05 (0.99, 1.13)

^aFor the cardiovascular events analyses, person-trials with a history of MACEs were excluded; the unweighted total was 5,300,986.

^bFor the HF events analysis, person-trials with a history of HF were excluded; the unweighted total was 5,834,041.

^cFor the major adverse liver events analysis, person-trials with a history of cirrhosis, liver transplantation or liver cancer were excluded; the unweighted total was 6,067,447.

^dFor the weight loss analyses, person-trials without follow-up weight data were excluded; the unweighted total was 5,500,845.

^eFor the gastrointestinal events analysis, person-trials with a history of gastrointestinal disease were excluded; the unweighted total was 5,446,868.

Extended Data Fig. 2. Weighted cumulative incidence of secondary outcomes by GLP-1RA initiation among patients with type 1 diabetes.

The panel figure shows cumulative incidence estimated using the Kaplan-Meier method of hospitalization for heart failure (A), major adverse liver events (B), and body weight decline of 5% (C), 10% (D) and 15% (E) among patients with type 1 diabetes initiating GLP-1RA (blue) and not initiating GLP-1RA (red). For hospitalization for heart failure analysis, person-trials with a history of outcome events were excluded (unweighted N = 5,834,041). For major adverse liver events analysis, person-trials with a history of outcome events were excluded (unweighted N = 6,067,447). For the weight loss analysis, person-trials without follow-up weight data were excluded (unweighted N = 5,500,845). Abbreviations: GLP-1RA, glucagon-like peptide-1 receptor agonist.

There was no increased risk of hospitalization for DKA (HR: 0.83 (95% CI: 0.76–0.90)) or severe hypoglycemia (HR: 0.82 (95% CI: 0.72–0.94)) comparing GLP-1RA initiation versus noninitiation. Gastrointestinal events were more frequent in GLP-1RA initiators than in noninitiators, although the risk difference was not statistically significant (Table 1).

In exploratory subgroup analysis, associations between GLP-1RA initiation and lower risks of MACEs, ESKD and major adverse liver events, as well as greater weight loss, were consistent across age (<45 years versus ≥45 years; Extended Data Table 2) and baseline glycated hemoglobin (HbA1c) levels (<8% versus ≥8%; Extended Data Table 3).

Extended Data Table 2.

Weighted hazard ratio of outcomes associated with GLP-1RA initiations within age subgroups among patients with type 1 diabetes

Subgroup hazard ratios with 95% confidence intervals were estimated using Cox regression and adjusted for baseline covariates via overlap weighting; interaction P values were derived from two-sided Wald tests. Abbreviations: GLP-1RA, glucagon-like peptide-1 receptor agonist.

Extended Data Table 3.

Weighted hazard ratio of outcomes associated with GLP-1RA initiations within HbA1c subgroups among patients with type 1 diabetes

Subgroup hazard ratios with 95% confidence intervals were estimated using Cox regression and adjusted for baseline covariates via overlap weighting; interaction P values were derived from two-sided Wald tests. Abbreviations: GLP-1RA, glucagon-like peptide-1 receptor agonist.

In the per-protocol analysis censoring deviation of treatment strategy (median follow-up of 7 months), the results were less precise but consistent with the intention-to-treat results (Extended Data Table 4). The results were similar in sensitivity analyses that applied additional exclusions (no prior hospitalization for outcomes of interest), modified outcome ascertainment (including events with a nonprimary diagnosis), different censoring criteria (censoring at the last body weight measurement for weight changes) and a stricter lab-based T1D definition (Extended Data Table 4). In the sensitivity analysis applying mean imputation for missing covariates, results were consistent (Extended Data Table 4). Using traffic accident as a negative control outcome to assess residual or unmeasured confounding, we found no significant difference between GLP-1RA initiators and noninitiators (HR: 0.96 (95% CI: 0.77–1.20)). E-values were 1.63 for MACEs and 1.77 for ESKD.

Extended Data Table 4.

Sensitivity analyses of outcomes associated with GLP-1RA initiations

* For the per-protocol analysis, the median follow-up was 7 months (interquartile range: 6-11). Abbreviations: GLP-1RA, glucagon-like peptide-1 receptor agonist; MACE, major adverse cardiovascular event.

In this large, target trial emulation study of youth and adults with T1D, GLP-1RA initiation was associated with significantly lower risks of MACEs and ESKD. Importantly, GLP-1RA initiation was not associated with increased risks of DKA or severe hypoglycemia. These associations were consistent across age and HbA1c subgroups.

Prior GLP-1RA trials in T1D demonstrated short-term effects on HbA1c reduction (0.2–0.5%) and weight loss (3.9–8.8 kg) compared with placebo, but lacked power for long-term outcomes^11–13. We observed significant risk reductions with GLP-1RA for MACEs (15%) and ESKD (19%). The effect sizes are comparable to the risk reductions observed in T2D trials (13–14% in MACEs and 16% in ESKD)^6,7. GLP-1RAs may protect heart and kidneys by reducing inflammation, improving insulin sensitivity and metabolic control, enhancing endothelial function and decreasing platelet aggregation, independent of weight loss^18,19. This may explain the similar cardiorenal benefits observed in our study compared with T2D trials, despite the smaller effect size for weight loss.

Our safety findings showed no increased risk of hospitalization for DKA or severe hypoglycemia. These results are consistent with contemporary trials that used continuous glucose monitoring and automated insulin delivery systems^10–12. Observational studies in T1D also report no increased risk with semaglutide and tirzepatide^20,21. Although the glucose-lowering effect of GLP-1RAs may require insulin dose reductions that theoretically increase DKA risk, we did not observe such safety concerns, which may reflect advances in diabetes care. Increased use of diabetes technologies and an evolved understanding of necessary insulin dose adjustments support safer implementation of GLP-1RA therapy in T1D⁴. Careful patient selection and close monitoring in real-world practice may have also contributed to mitigating safety concerns identified in early trials.

Exploratory subgroup analyses showed generally consistent findings across age groups and HbA1c levels. The observed effectiveness in younger patients is particularly meaningful, because they represent a large portion of the T1D population and face long-term cumulative risk of cardiorenal events. Similarly, consistent cardiorenal benefits across HbA1c levels align with evidence from T2D trials, which demonstrated no treatment heterogeneity by glycemic control^6,22. Current guidelines recommend GLP-1RAs for cardiovascular prevention in T2D, regardless of HbA1c levels²³. Our results support the consideration of similar therapeutic strategies for all patients with T1D, regardless of age or glycemic status.

GLP-1RA initiation was associated with a 18% lower risk of hospitalization for HF, comparable to the 11–13% reduction seen in T2D trials^6,7. This is an important finding in T1D, where HF risk emerges early due to diabetic cardiomyopathy. We observed a 28% lower risk of major liver disease with GLP-1RA initiation. Given that liver disease affects at least 20% of patients with T1D²⁴, these hepatoprotective effects represent an important potential benefit of GLP-1RAs. We also found that patients with T1D who initiated GLP-1RAs were 14–25% more likely to achieve clinically meaningful weight loss. This highlights the potential of GLP-1RAs to enhance obesity management—a need that has grown rapidly in both youth and adults with T1D⁵.

This study has several strengths. We leveraged a large national cohort of patients with T1D with extensive follow-up and applied target trial emulation methodology rigorously to evaluate hard clinical endpoints. We minimized confounding by including numerous patient characteristics and achieving good covariate balance through propensity score weighting. Consistent results across sensitivity analyses and the null association with a negative control outcome (traffic accidents) speak to the robustness of our findings.

This study also has limitations. First, residual and unmeasured confounding is possible. The E-values for MACEs and ESKD suggest that an unmeasured confounder would need to be moderately strong to explain away the observed associations. Second, misclassification of T1D is possible despite using a previously validated algorithm with a high positive predictive value of 88%. However, sensitivity analysis restricted to lab-confirmed T1D cases yielded similar results. Third, hypoglycemia and DKA were identified from hospitalization records, likely capturing only severe cases. However, this ascertainment approach should affect both groups similarly. Fourth, some observed weight loss, especially for 5% weight loss, may have been unintentional (for example, due to illness), which could not be distinguished from intentional weight reduction in EHR data. We did not examine whether observed weight loss was sustainable. Fifth, information on insulin dosing was unavailable, limiting our ability to assess the insulin dose adjustments. Last, although this is one of the largest comparative effectiveness studies of GLP-1RA use in T1D to date, the sample size precluded examination of individual GLP-1RA agents separately or head-to-head comparisons with specific treatments. Future studies are needed to identify the optimal GLP-1RA or dosage for T1D and establish comparative effectiveness against other noninsulin diabetes treatments through active comparator designs.

In this national cohort of patients with T1D, GLP-1RA initiation was associated with lower risks of MACEs and ESKD, without increasing the risks of DKA or severe hypoglycemia. Given the early onset and lifelong burden of T1D, these findings suggest that GLP-1RAs may offer an important strategy to prevent long-term cardiovascular and kidney disease. Large-scale RCTs in people with T1D are warranted to confirm these findings.

Methods

Data source

This study used longitudinal de-identified EHR data from >60 health systems enrolled in the OLDW²⁵. Data are extracted and updated quarterly from participating health systems and harmonized into a standardized format. OLDW includes demographic information, vital signs, prescription medication orders, laboratory test results and healthcare encounters. The database contains data from 300 million enrollees and patients, representing a mixture of ages and geographical regions across the USA.

The study was approved by the Johns Hopkins Bloomberg School of Public Health Institutional Review Board. Informed consent was waived because this study used de-identified data and presented no more than minimal risk of harm to participants. The study follows the Transparent Reporting of Observational Studies Emulating a Target Trial (TARGET) guidelines for reporting²⁶.

Creating a sequence of target trials

We emulated a sequence of 135 target trials between January 2013 and March 2024. In each calendar month, eligible individuals were identified and followed using the same trial protocol. This sequential approach serves two purposes: first, to eliminate immortal time bias by properly aligning treatment initiation with follow-up start (T0) and, second, to maximize statistical power by allowing individuals to contribute data across multiple trials. The final study population included 174,678 individuals with T1D, contributing 6,092,537 person-trials across the 135 trials (Extended Data Fig. 1).

Emulating a target trial

We designed a target trial comparing the effectiveness and safety of GLP-1RA initiation versus noninitiation among individuals with T1D. The key design elements are summarized in Extended Data Table 5 and detailed below.

Extended Data Table 5.

Specification and emulation of a target trial of GLP-1RA initiation versus non-initiation for effectiveness and safety outcomes in persons with type 1 diabetes

Eligibility criteria and study population

We first identified individuals with T1D using a previously validated algorithm²⁷. Eligibility was then assessed in January 2013, which was the baseline (T0) for the first target trial. Individuals with T1D were eligible if they met the following criteria: (1) on insulin therapy, (2) no prior use of GLP-1RA, (3) available measurements of body mass index (BMI) and HbA1c measurements at T0 and (4) at least 12 months of engagement with the health system before T0. The 12-month prior engagement was required to ensure availability of baseline information and to increase the probability of continued follow-up. Individuals were excluded if they met any of the following before T0: contraindications to GLP-1RAs (thyroid cancer or multiple endocrine neoplasia syndrome type II), ESKD or pregnancy (Extended Data Table 6).

Extended Data Table 6.

Definition of inclusion and exclusion criteria and outcome events

Note: Table references correspond to Methods-only references and are numbered independently from the main text.

Treatment strategies

We defined treatment strategies based on prescription for GLP-1RA in January 2013 (baseline, T0 for the first target trial): patients who initiated any GLP-1RAs (albiglutide, dulaglutide, exenatide, liraglutide, lixisenatide, semaglutide and tirzepatide) were classified as initiators and all others were classified as noninitiators.

Treatment assignment: emulation of randomization by propensity score-based overlap weighting

To balance baseline characteristics between GLP-1RA initiators and noninitiators, we applied propensity score-based overlap weighting, which emphasizes the population with clinical equipoise²⁸. Propensity score (that is, the probability of initiating GLP-1RA versus not) was estimated using a multivariable logistic regression that included all measured baseline covariates and indicators for each trial²⁹. Covariate balance between groups was evaluated using the standardized mean difference, with an absolute standardized mean difference <10% indicating adequate balance³⁰.

Baseline characteristics were extracted from EHR records, including age, sex, race or ethnicity, health insurance coverage, smoking status, BMI, systolic blood pressure and laboratory measurements (for example, serum creatinine and HbA1c). Obesity status was categorized as normal, overweight and obesity class I–III, based on age-specific and sex-specific percentiles for youth (2–19 years) and World Health Organization cut points for adults (≥20 years)^31–33. The estimated glomerular filtration rate (eGFR) was calculated based on the 2021 Chronic Kidney Disease Epidemiology Collaboration creatinine equation using the most recent outpatient serum creatinine measurement³⁴. The eGFR was categorized as <60, 60–89, ≥90 ml min⁻¹ 1.73 m⁻², or unknown. Albuminuria quantification was based on the most recent urine albumin-to-creatinine ratio (ACR) within the prior 2 years; if unavailable, the urine protein-to-creatine ratio or dipstick protein result was converted to the ACR using the adjusted conversion equation³⁵. Urine ACR was categorized as <30, 30–300, >300 mg g⁻¹, or unknown. Diabetes duration was calculated as the interval between the first algorithm-identified T1D diagnosis and the index date²⁷. Comorbidities were identified by relevant codes in at least two outpatient records within 2 years, one inpatient record or the problem list before T0. Charlson’s comorbidity index was calculated to capture the overall comorbidity burden³⁶. We also measured health utilization, including the number of hospitalizations and outpatient visits in the past year. Use of continuous glucose monitoring in the past year was ascertained based on outpatient prescription records and related procedure codes. Data on the use of insulin pump and medications were extracted from outpatient prescription records in the past year. Medications included other noninsulin, non-GLP-1RA glucose-lowering agents, antihypertensive drugs, statins, non-GLP-1RA anti-obesity medications (that is, orlistat, phentermine–topiramate, naltrexone–bupropion, diethylpropion, phentermine and benzphetamine), corticosteroids, antipsychotics and antidepressants.

Outcomes and follow-up

Primary outcomes were (1) MACEs (composite of acute myocardial infarction, stroke and all-cause mortality) and (2) ESKD (that is, dialysis and kidney transplantation using a validated algorithm)³⁷.

Secondary outcomes were: (1) hospitalization for HF; (2) major adverse liver events including cirrhosis and related complications (for example, ascites, varices, portal hypertension and hepatorenal syndrome), hepatocellular carcinoma and liver transplantation^38,39; and (3) first weight loss reaching clinically meaningful thresholds of 5%, 10% and 15% from baseline. We included HF outcome because several recent trials in patients with HF and obesity or T2D have demonstrated improvements in symptoms and physical function¹⁵. Liver outcomes were assessed given the evidence that GLP-1RAs improve liver health through weight loss and metabolic improvements in T2D^40,41. Similar benefits may be plausible in T1D, but supporting data are limited.

Safety outcomes were: (1) hospitalization with primary diagnosis of DKA⁴²; (2) hospitalization with primary diagnosis of hypoglycemia^43,44; and (3) gastrointestinal adverse events, including biliary disease, pancreatitis, bowel obstruction and gastroparesis⁴⁵.

Incident events were identified through diagnosis and procedures (Extended Data Table 6). For the outcomes of MACEs, major adverse liver events and gastrointestinal adverse events, we further excluded individuals with a history of these conditions before T0. For weight loss outcomes, we excluded patients with no follow-up weight data.

We also used traffic accident as a negative control outcome⁴⁶. For intention-to-treat analysis, follow-up started at treatment assignment (T0) and ended at the occurrence of outcomes, death or last encounter (to 30 June 2024).

Statistical analysis

We pooled participants from all target trials and summarized their baseline characteristics at the person-trial level. Continuous variables were presented as means with s.d. and categorical variables as percentages. Therefore, the unit for Extended Data Table 1 was person-trial.

We used weighted Kaplan–Meier estimator curves to estimate the cumulative incidence for each outcome. We calculated 95% CI for absolute risk difference between GLP-1RA initiators and noninitiators using bootstrapping of 500 samples. Cox’s proportional regression model stratified by trial was used to estimate the HR with 95% CI using robust estimation for the s.e.

In exploratory analyses, we examined whether associations differed by baseline age (median age 45 years) by including an interaction term between treatment indicator and age subgroup indicator. We also conducted subgroup analyses stratified by median baseline HbA1c (<8% versus ≥8%). To estimate per-protocol effects, we additionally censored GLP-1RA initiators after treatment discontinuation (defined as a gap of >90 d between the end of the last prescription and a new prescription) and censored noninitiators on subsequent GLP-1RA initiation⁴⁷.

We also conducted several sensitivity analyses. First, we further excluded patients hospitalized for DKA, severe hypoglycemia or cirrhosis-related complications within the previous year, because they were at high risk of recurrence. Second, we included events with a nonprimary diagnosis for DKA, severe hypoglycemia, MACEs and individual cardiovascular events. Third, for weight outcome analysis, we additionally censored at the last body weight measurement. Fourth, to address potential misclassification of diabetes type, we repeated the analysis restricted to patients who had positive diabetes autoantibodies or negative C-peptide results. Last, to assess the impact of missing data, we included individuals who had been excluded due to missing BMI or HbA1c at baseline and applied mean imputation for BMI, HbA1c, systolic blood pressure, eGFR and ACR within each trial, then repeated analyses for cardiovascular, kidney, liver and safety outcomes. Two-tailed tests with a significance level of 0.05 were used. All analyses were performed using Stata (v16.1, StataCorp LLC).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41591-026-04274-0.

Author contributions

Y.X., N.D.M. and J.-I.S. had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Y.X. and J.-I.S. conceptualized and designed the study. Y.X., M.F. and J.-I.S. acquired, analyzed or interpreted the data. Y.X. and N.D.M. carried out the statistical analysis. J.-I.S. and M.E.G. provided administrative, technical or material support. M.F. and J.-I.S. supervised the study. Y.X. drafted the manuscript. All authors critically revised the manuscript for important intellectual content.

Peer review

Peer review information

Nature Medicine thanks Irene Caruso, Michael Nauck and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Liam Messin, in collaboration with the Nature Medicine team.

Data availability

Data were obtained from Optum Labs and consist of de-identified EHR data. These third-party proprietary data are not publicly available under the Optum Labs data use agreement. The authors do not have permission to share the data directly. Qualified researchers may request access from Optum Labs, subject to approval of a data use agreement and compliance with applicable requirements (https://www.optumlabs.com).

Code availability

No customized code or software was developed for data collection or analysis in this study; analyses were conducted using standard, publicly available statistical software.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s41591-026-04274-0.

Supplementary information

The online version contains supplementary material available at 10.1038/s41591-026-04274-0.