Abstract
Aims
To compare the risk of gastrointestinal adverse events in new users of glucagon‐like peptide‐1 receptor agonists (GLP‐1RA), sodium‐glucose cotransporter‐2 inhibitors (SGLT2i) and insulin glargine.
Materials and Methods
We conducted an active comparator, new user design study in veterans with type 2 diabetes who initiated one of these drug classes between 1 January 2018 and 31 December 2021 (N = 141 080). Inverse probability weighted Cox regression models were used to relate drug class to outcomes of gastroparesis, intestinal obstruction, gallstones, acute cholecystitis, acute pancreatitis and all‐cause death.
Results
There were 19 765 (14.0%) veterans initiated on GLP‐1RA, 75 058 (53.2%) on SGLT2i and 46 257 (32.8%) on insulin glargine. Compared to SGLT2i, GLP‐1RA had a higher hazard of gastroparesis (HR 1.65, 95% CI 1.33–2.05) but a similar mortality hazard. Compared to insulin glargine, GLP‐1RA had a higher hazard of gastroparesis (HR 1.24, 95% CI 1.02, 1.52), but a lower hazard of all‐cause death (HR 0.62, 95% CI 0.58, 0.66). Compared to SGLT2i, insulin glargine had a higher hazard of gastroparesis (HR 1.29, 95% CI 1.07, 1.56), intestinal obstruction (HR 1.26, 95% CI 1.11, 1.43) and all‐cause death (HR 1.58, 95% CI 1.50, 1.65). Risks of gallstones, acute cholecystitis and pancreatitis were similar across the classes.
Conclusions
In patients with type 2 diabetes at risk for gastroparesis, SGLT2i might be the preferred agent. In patients for whom SGLT2i is not an option or another agent is needed, patients and providers might need to weigh the higher risk of death with insulin glargine against the higher risk of gastroparesis with GLP‐1RA.
Keywords: antidiabetic drug, database research, GLP‐1, insulin glargine, SGLT2 inhibitor
1. INTRODUCTION
In recent landmark trials, glucagon‐like peptide‐1 receptor agonists (GLP‐1RA) reduced the risk of cardiovascular events, chronic kidney disease (CKD) progression and all‐cause mortality compared to placebo. 1 , 2 , 3 , 4 The 2025 American Diabetes Association guidelines recommend the use of GLP‐1RA in adults with type 2 diabetes (T2D) and any of the following: established or high risk of atherosclerotic cardiovascular disease, chronic kidney disease (CKD), symptomatic heart failure with preserved ejection fraction in obese individuals and metabolic dysfunction‐associated steatotic liver disease (MASLD) in overweight or obese individuals. 5
While gastrointestinal (GI) symptoms such as nausea, vomiting, diarrhoea and constipation are the most frequent adverse events of GLP‐1RA therapy, 6 more serious adverse events such as gastroparesis, 7 , 8 intestinal obstruction, 9 , 10 gallstones 11 , 12 , 13 cholecystitis 14 and pancreatitis 15 have been reported. In patients with or without T2D, those initiated on semaglutide or liraglutide for weight loss had a higher risk of pancreatitis, bowel obstruction or gastroparesis compared to those initiated on bupropion‐naltrexone. 7 For T2D patients already on metformin, alternatives to GLP‐1RA in glycaemic management are sodium‐glucose cotransporter‐2 inhibitors (SGLT2i) or long‐acting insulins. As gastroparesis and intestinal obstruction are complications of T2D per se, the comparative risks of GLP‐1RA and alternative T2D agents on GI complications might be different. While there is evidence that SGLT2i have a lower risk of GI adverse events compared to GLP‐1RA, 16 it is unclear whether GLP‐1RA have a higher risk of GI adverse events compared to insulin glargine, a commonly used long‐acting insulin.
The active comparator, new user (ACNU) design is a causal inference approach that minimises confounding and can be used to emulate randomised controlled trials for comparative effectiveness using real‐world, large‐scale electronic health record (EHR) observational data. 17 , 18 In the current study, we used the ACNU design and examined the comparative effects of GLP‐1RA versus insulin glargine on GI adverse events using SGLT2ias a negative control in a national cohort of veterans with T2D already on metformin.
2. MATERIALS AND METHODS
2.1. Data sources
We accessed the Veterans Affairs (VA) databases through the VA Informatics and Computing Infrastructure (VINCI), which hosts available data from the Corporate Data Warehouse (CDW), in addition to other sources. 19 This included assessments, diagnoses, interventions, laboratory results, imaging, medication orders and consultations from 144 VA Medical Centers and 1221 outpatient clinics since 1st October 1999. 20 , 21 We obtained approval for the current study by the institutional review board at the University of Utah.
2.2. Study design and study cohort
In a national cohort of veterans with T2D defined by Internationqal Classification of Diseases, Ninth Revision (ICD‐9) and Tenth Revision (ICD‐10) codes (N = 2 462 623), we identified 605 756 veterans initiated on GLP‐1RA, SGLT2i or insulin glargine for the first time between 1 January 2018 and 31 December 2021 (Figure S1). The index date was defined as the date of the first prescription fill for one of the study medications during that time period. We excluded 307 323 veterans who were not on metformin, 13 727 veterans who had a prescription filled at an outpatient VA pharmacy for any of the three classes of interest from 1 January 2008 until the index date. We excluded one veteran whose first prescription was a combination of the study drugs and 286 veterans who were initiated on exendin‐based GLP‐1RA (exenatide, lixisenatide) or albiglutide due to a low prescription rate. In the main analysis, we excluded 20 039 veterans with a baseline history of any of the GI adverse events of interest. Thus, the cohort for the main analysis included 141 080 veterans.
2.3. Baseline variables
Age, sex, race, ethnicity, income and education data were obtained from the VA CDW or Veterans Eligibility Trends and Statistics (USVETS) datafiles. 19 VA Inpatient and Outpatient Medical Statistical Analysis System Datasets 22 were used to identify T2D and comorbid conditions (Table S1). Blood pressure, height, weight and laboratory values such as haemoglobin A1C (HbA1C), urine albumin to creatinine ratio (UACR) and estimated glomerular filtration rate (eGFR) were defined as in Table S2. Baseline medications were obtained from outpatient VA pharmacies prescription fill data. This included other anti‐diabetic medications, anti‐hypertensive medications and statins that had prescriptions filled within 180 days before the index date (Table S3).
2.4. Study outcomes
We considered time‐to‐event outcomes and followed each veteran from their index date to the earliest occurrence of the event of interest, last follow‐up, or until 31 March 2023. Outcomes of interest were the adverse events of gastroparesis, intestinal obstruction, gallstones, acute cholecystitis and acute pancreatitis events as identified by ICD‐10 codes (Table S4) from VA inpatient and outpatient records as well as fee‐for‐service data for care provided outside the VA. All‐cause death was identified by data from the VA Vital Status database, which includes information from the Centers for Medicaid and Medicare Services, Social Security Administration and the VA Austin Information Technology Center. 23
2.5. Statistical analysis
Multinomial logistic regression models with an extensive list of variables that could potentially affect the treatment assignment (Table S5) were used to estimate generalized propensity scores for the probability of treatment assignment to the three medication arms and construct inverse probability of treatment weights. Missing values for baseline covariates were imputed using K‐nearest neighbour single imputation. 24 , 25 Covariate balance was assessed using the maximum of the pairwise absolute standardized mean differences across the three medication groups before and after inverse propensity score weighting. 26 Cox models with inverse probability weighting were used to estimate cause‐specific hazard ratios to compare the occurrence of specific gastrointestinal adverse events between each pair of treatment classes. For each model, the proportional hazards (PH) assumption was evaluated both graphically using Kaplan–Meier plots and analytically by including interaction terms with time. Inverse probability weighted Cox models also compared all‐cause death between the three treatment classes. We use the shorthand terminology ‘higher risk’ and ‘lower risk’ to describe comparisons in which the hazard ratios were greater than or less than 1, respectively. The 95% confidence intervals for the hazard ratios were calculated for each drug pair and GI adverse event using standard errors accounting for inverse probability of treatment survey weights. All analyses followed an intention‐to‐treat approach, so that treatment comparisons evaluate the effect of initial treatment with the respective drug classes, irrespective of subsequent changes in treatment. We further conducted a subgroup analysis in women. We also conducted a sensitivity analysis in which the analytic cohort was expanded to include veterans with baseline GI history while retaining the remaining entry criteria (N = 161 119). Baseline GI history, in addition to the same list of variables that were included in the main analysis, was used to estimate generalized propensity scores and construct inverse probability of treatment weights. We again applied inverse probability weighted Cox regression models to compare the previously described GI adverse events between the three medication classes. In subgroup analysis for each GI outcome based on the presence or absence of that GI condition at baseline, we conducted pair‐wise comparisons of the study drugs. Then, we compared the regression coefficients of each pairwise comparison in those with and without the GI condition at baseline. All analyses were completed using R, version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria).
Our diagnostic evaluations of the PH assumptions suggested that some treatment comparisons mildly attenuated over time, but the deviations from PH were too small to impact interpretation. Hence, treatment effects are summarised using single overall hazard ratios for the full follow‐up period. Inverse probability weighted cumulative incidence function curves were used to visually compare GI adverse events between the treatments, using the Kaplan–Meier method for all‐cause death and a competing risk framework for each of the GI adverse events and death before the first GI adverse event.
3. RESULTS
A total of 141 080 veterans were included in the main analysis (Figure S1). A majority of patients were initiated on SGLT2i (N = 75 058, 53.2%), followed by insulin glargine (N = 46 257, 32.8%) and GLP‐1RA (N = 19 765, 14.0%). A majority of SGLT2i users were started on empagliflozin (99.4%) at 74 628, 408 users on dapagliflozin (0.5%) and 22 users on canagliflozin (<0.1%). A majority of GLP‐1RA users were started on semaglutide (49.5%) at 9775, 5077 users on dulaglutide (25.7%) and 4913 users on liraglutide (24.8%). In unweighted comparisons, those initiated on GLP‐1RA were more likely to be women and have a higher body mass index. Those initiated on insulin glargine were more likely to be African American or Black and have a higher HbA1C, and those initiated on SGLT2i were older with a longer duration of T2D and had a higher prevalence of cardiovascular comorbidities such as coronary artery disease, heart failure and hypertension (Table 1). These baseline differences were attenuated after applying inverse probability weighting (Figure 1, panels A–D).
TABLE 1.
Baseline characteristics of the study cohort by study drug classes (N = 141 080).
| Unweighted | Inverse probability weighted | |||||||
|---|---|---|---|---|---|---|---|---|
| GLP‐1RA, n = 19 765 a | SGLT2i, n = 75 058 a | Insulin glargine, n = 46 257 a | Max pairwise SMD b | GLP‐1RA a | SGLT2i a | Insulin glargine a | Max pairwise SMD b | |
| Demographics | ||||||||
| Age, years | 63.3 (10.9) | 65.4 (10.3) | 64.2 (11.5) | 0.190 | 64.6 (10.7) | 64.7 (10.5) | 64.9 (11.2) | 0.030 |
| Female | 1714 (8.7%) | 3014 (4.0%) | 2495 (5.4%) | 0.192 | 5.5% | 5.0% | 5.0% | 0.019 |
| Race | 0.143 | 0.045 | ||||||
| African American or Black | 3298 (16.7%) | 12 933 (17.2%) | 10 214 (22.1%) | 18.0% | 18.8% | 18.9% | ||
| White | 15 354 (77.7%) | 57 964 (77.2%) | 33 805 (73.1%) | 76.7% | 75.9% | 75.7% | ||
| Other or mixed races | 1113 (5.6%) | 4161 (5.5%) | 2238 (4.8%) | 5.4% | 5.3% | 5.4% | ||
| Hispanic/Latino | 1991 (10.1%) | 8583 (11.4%) | 5009 (10.8%) | 0.044 | 10.9% | 11.1% | 11.3% | 0.012 |
| Completed High School | 8028 (40.6%) | 30 299 (40.4%) | 18 365 (39.7%) | 0.019 | 40.5% | 40.2% | 40.1% | 0.010 |
| Household income <$50 000 | 8987 (45.5%) | 34 808 (46.4%) | 23 429 (50.6%) | 0.104 | 47.1% | 47.7% | 48.1% | 0.021 |
| Clinical characteristics | ||||||||
| Duration of T2D c , years | 8.2 (5.6) | 8.8 (5.6) | 8.3 (6.0) | 0.107 | 8.5 (5.7) | 8.6 (5.6) | 8.7 (6.0) | 0.025 |
| Coronary artery disease | 6520 (33.0%) | 31 215 (41.6%) | 15 746 (34.0%) | 0.179 | 37.4% | 38.2% | 38.2% | 0.017 |
| Heart failure | 2387 (12.1%) | 12 278 (16.4%) | 6289 (13.6%) | 0.123 | 14.2% | 15.2% | 15.0% | 0.029 |
| Cerebrovascular disease | 2723 (13.8%) | 12 182 (16.2%) | 7773 (16.8%) | 0.084 | 15.7% | 16.3% | 16.5% | 0.022 |
| Cardiomyopathies | 2657 (13.4%) | 13 758 (18.3%) | 6758 (14.6%) | 0.134 | 15.5% | 16.5% | 16.4% | 0.027 |
| Peripheral vascular disease | 3864 (19.5%) | 16 867 (22.5%) | 9359 (20.2%) | 0.072 | 21.0% | 21.4% | 21.3% | 0.008 |
| Hypertension | 17 845 (90.3%) | 68 583 (91.4%) | 41 038 (88.7%) | 0.089 | 90.5% | 90.5% | 90.5% | 0.002 |
| Tobacco abuse | 5400 (27.3%) | 22 769 (30.3%) | 15 399 (33.3%) | 0.130 | 29.9% | 30.8% | 30.6% | 0.021 |
| Alcohol abuse | 2432 (12.3%) | 10 262 (13.7%) | 7413 (16.0%) | 0.107 | 13.4% | 14.1% | 14.0% | 0.020 |
| Diabetic retinopathy | 4868 (24.6%) | 18 677 (24.9%) | 12 111 (26.2%) | 0.036 | 25.1% | 25.8% | 26.3% | 0.029 |
| Diabetic neuropathy | 7056 (35.7%) | 23 803 (31.7%) | 15 611 (33.7%) | 0.084 | 33.9% | 33.4% | 33.9% | 0.010 |
| Physical measurements | ||||||||
| Body mass index, kg/m2 | 36 (7) | 33 (6) | 33 (6) | 0.421 | 34 (6) | 33 (6) | 33 (7) | 0.054 |
| Laboratory measurements | ||||||||
| Haemoglobin A1C, % | 8.8 (1.7) | 8.6 (1.5) | 9.9 (2.3) | 0.670 | 9.0 (1.8) | 9.0 (1.9) | 9.0 (2.1) | 0.019 |
| Haemoglobin A1C, % | 0.659 | 0.223 | ||||||
| <7 | 1380 (7.0%) | 4243 (5.7%) | 1847 (4.0%) | 6.2% | 4.2% | 8.1% | ||
| 7 to <8.5 | 8177 (41.4%) | 38 159 (50.8%) | 12 750 (27.6%) | 38.9% | 43.0% | 38.3% | ||
| 8.5 to <10 | 4191 (21.2%) | 15 823 (21.1%) | 7571 (16.4%) | 20.7% | 20.7% | 17.4% | ||
| ≥10 | 6017 (30.4%) | 16 833 (22.4%) | 24 089 (52.1%) | 34.2% | 32.1% | 36.1% | ||
| eGFR, mL/min/1.73 m2 d | 81 (20) | 81 (18) | 81 (21) | 0.020 | 81 (20) | 81 (19) | 80 (20) | 0.10 |
| eGFR, mL/min/1.73 m2 d | 0.133 | 0.138 | ||||||
| <30 | 75 (0.4%) | 56 (0.1%) | 277 (0.6%) | 0.4% | 0.1% | 0.6% | ||
| 30 to <45 | 849 (4.3%) | 1800 (2.4%) | 1862 (4.0%) | 4.5% | 2.4% | 4.0% | ||
| 45 to <60 | 2410 (12.2%) | 9374 (12.5%) | 6067 (13.1%) | 12.2% | 12.9% | 12.9% | ||
| ≥60 | 16 431 (83.1%) | 63 828 (85.0%) | 38 051 (82.3%) | 82.8% | 84.6% | 82.5% | ||
| Urine albumin to creatinine ratio, mg/g | 0.119 | 0.023 | ||||||
| <30 | 6872 (34.8%) | 26 134 (34.8%) | 13 794 (29.8%) | 33.2% | 33.0% | 32.4% | ||
| 30 to <100 | 2392 (12.1%) | 9684 (12.9%) | 5795 (12.5%) | 12.7% | 12.6% | 12.7% | ||
| 100 to <300 | 1149 (5.8%) | 4537 (6.0%) | 2786 (6.0%) | 6.1% | 6.0% | 6.1% | ||
| ≥300 | 788 (4.0%) | 3165 (4.2%) | 2035 (4.4%) | 4.4% | 4.3% | 4.2% | ||
| Not available | 8564 (43.3%) | 31 538 (42.0%) | 21 847 (47.2%) | 43.6% | 44.1% | 44.5% | ||
| Medications | ||||||||
| Dipeptidyl peptidase IV inhibitors | 5391 (27.3%) | 22 698 (30.2%) | 10 605 (22.9%) | 0.166 | 27.8% | 27.5% | 27.4% | 0.008 |
| Sulfonylureas | 10 680 (54.0%) | 40 129 (53.5%) | 23 278 (50.3%) | 0.074 | 53.8% | 52.2% | 52.0% | 0.036 |
| Thiazolidinediones | 2014 (10.2%) | 6060 (8.1%) | 3091 (6.7%) | 0.126 | 8.4% | 8.0% | 7.9% | 0.020 |
| Other insulins | 4251 (21.5%) | 7505 (10.0%) | 15 635 (33.8%) | 0.601 | 20.0% | 20.2% | 20.7% | 0.015 |
| Angiotensin converting enzyme inhibitors/angiotensin II receptor blockers | 13 796 (69.8%) | 52 961 (70.6%) | 29 957 (64.8%) | 0.124 | 69.1% | 68.7% | 68.8% | 0.009 |
| Statin | 15 876 (80.3%) | 61 714 (82.2%) | 34 665 (74.9%) | 0.178 | 80.2% | 79.8% | 79.5% | 0.016 |
All continuous covariates are summarised as mean (SD); all categorical covariates are summarised as n (%).
Maximum pairwise standardized mean difference.
Type 2 diabetes.
Estimated glomerular filtration rate.
FIGURE 1.
Balance plots for overall comparisons in main analysis (A–D).
Adherence, measured by prescription fill in the 6 months before censoring at administrative censor date of 03/31/23, death or loss to follow‐up, was 66.7% for insulin glargine, 74.1% for GLP‐1RA and 76.1% for SGLT2i. The corresponding numbers for the 24 months after index date were 63.2%, 71.5% and 71.2%, respectively.
3.1. Cumulative incidences of GI adverse events and all‐cause death by study drug classes
Unweighted and weighted event rates for the GI adverse events are summarised in Table 2. There were 852 gastroparesis events over 390 056 person‐years of follow‐up and 12 507 all‐cause deaths over 391 425 person‐years of follow‐up (Table 2). Inverse probability weighted cumulative incidence curves showed the highest incidence of all‐cause deaths (irrespective of gastroparesis events) in those using insulin glargine (Figure 2, panel A), with similar incidence of all‐cause death in the GLP‐1RA and SGLT2i groups. There was clear evidence of the competing risk of death for gastroparesis events as shown by the higher incidence of all‐cause death before gastroparesis events in the insulin glargine group compared to GLP‐1RA or SGLT2i groups (Figure 2, panel B). For a composite of all‐cause death/gastroparesis events, insulin glargine had the highest incidence (Figure 2, panel C). This trend was seen with all GI adverse events (Figures S3–S6).
TABLE 2.
Pairwise hazard ratios and event rates of gastrointestinal adverse events, by drug class (N = 141 080).
| Gastrointestinal adverse events | Unweighted and weighted event rates a | Weighted pairwise hazard ratios (95% confidence intervals) | ||||
|---|---|---|---|---|---|---|
| GLP‐1RA (n = 19 765) | SGLT2i (n = 75 058) | Insulin glargine (n = 46 257) | GLP‐1RA vs. insulin glargine | Insulin glargine vs. SGLT2i | GLP‐1RA vs. SGLT2i | |
| Gastroparesis | ||||||
| Unweighted | 0.31 (181/58956) | 0.15 (288/187926) | 0.27 (383/143174) | 1.24 (1.02, 1.52) | 1.29 (1.07, 1.56) | 1.65 (1.33, 2.05) |
| Weighted | 0.30 | 0.18 | 0.25 | |||
| Intestinal obstruction | ||||||
| Unweighted | 0.45 (267/58875) | 0.37 (700/187467) | 0.54 (773/142666) | 0.91 (0.78, 1.07) | 1.26 (1.11, 1.43) | 1.16 (0.99, 1.36) |
| Weighted | 0.48 | 0.40 | 0.52 | |||
| Gallstones | ||||||
| Unweighted | 0.74 (434/58531) | 0.64 (1193/186636) | 0.71 (1006/142120) | 1.05 (0.92, 1.19) | 1.06 (0.96, 1.17) | 1.12 (0.99, 1.27) |
| Weighted | 0.74 | 0.66 | 0.70 | |||
| Acute cholecystitis | ||||||
| Unweighted | 0.23 (137/59050) | 0.20 (370/187822) | 0.24 (350/143263) | 1.01 (0.81, 1.26) | 1.09 (0.92, 1.31) | 1.13 (0.90, 1.40) |
| Weighted | 0.24 | 0.21 | 0.24 | |||
| Acute pancreatitis | ||||||
| Unweighted | 0.30 (179/58949) | 0.23 (434/187707) | 0.33 (473/143034) | 1.02 (0.85, 1.24) | 1.13 (0.96, 1.33) | 1.17 (0.96, 1.42) |
| Weighted | 0.30 | 0.26 | 0.29 | |||
| All‐cause death | ||||||
| Unweighted | 2.34 (1386/59269) | 2.46 (4635/188328) | 4.51 (6486/143828) | 0.62 (0.58, 0.66) | 1.58 (1.50, 1.65) | 0.96 (0.90, 1.03) |
| Weighted | 2.64 | 2.65 | 4.31 | |||
Event rates presented as events per 100 person years (N events/years follow‐up).
FIGURE 2.
Inverse probability weighted cumulative incidence of all‐cause death irrespective of the occurrence of gastroparesis events (panel A), gastroparesis events and all‐cause death under competing risk framework (panel B) and a composite of gastroparesis events or all‐cause death (panel C) by each of the study drug classes.
3.2. GLP‐1RA versus insulin glargine
Patients who initiated treatment with GLP‐1RA had a higher risk of gastroparesis (HR 1.24, 95% CI 1.02, 1.52) and a lower risk of all‐cause death (HR 0.62, 95% CI 0.58, 0.66) compared to insulin glargine (Table 2). There were no discernible differences in the risks of intestinal obstruction, gallstones, acute cholecystitis and acute pancreatitis between those initiating GLP‐1RA and insulin glargine.
3.3. Insulin glargine versus SGLT2i
When compared to those who initiated treatment with SGLT2i, those initiating treatment with insulin glargine had a significantly higher risk of gastroparesis (HR 1.29, 95% CI 1.07, 1.56), intestinal obstruction (HR 1.26, 95% CI 1.11, 1.43), as well as all‐cause death (HR 1.58, 95% CI 1.50, 1.65).
3.4. GLP‐1RA versus SGLT2i
GLP‐1RA initiators had a higher risk of gastroparesis (HR 1.65, 95% CI 1.33, 2.05) than SGLT2i initiators, but no discernible difference in the risk of intestinal obstruction. Gallstones (HR 1.12, 95% CI 0.99, 1.27), acute cholecystitis (HR 1.13, 95% CI 0.90, 1.40) and acute pancreatitis (HR 1.17, 95% CI 0.96, 1.42) risks were similar between the two groups. All‐cause death (HR 0.96, 95% CI 0.90, 1.03) was also similar between the two groups.
3.4.1. Women subgroup analysis
In the women subgroup (n = 7223), SGLT2i and GLP‐1RA had similar hazards of GI adverse events, and insulin glargine had a higher hazard of acute cholecystitis and death compared to either SGLT2i or GLP‐1RA (Table S6).
3.5. Sensitivity analysis
A total of 161 119 participants with baseline GI history were included in the sensitivity analysis (Table S7). The risk of gastroparesis was no longer different between GLP‐1RA and insulin glargine initiators, but the risk of intestinal obstruction (HR 0.85, 95% CI 0.75, 0.99) was lower with GLP‐1RA initiators compared to insulin glargine initiators (Table 3). As with the main analysis, treatment initiation with GLP‐1RA continued to be associated with a lower risk of all‐cause death when compared to insulin glargine. Compared to SGLT2i initiators, treatment initiation with insulin glargine had a higher risk of all GI adverse events as in the main analysis. Compared to SGLT2i initiators, GLP‐1RA initiators continued to have a higher risk of gastroparesis; however, there was a lower risk of acute pancreatitis (HR 0.79, 95% CI 0.64, 0.91) and all‐cause death (HR 0.94, 95% CI 0.88, 1.00).
TABLE 3.
Pairwise hazard ratios and event rates of gastrointestinal adverse events, by drug class those with baseline history of gastrointestinal conditions (N = 161 119).
| Gastrointestinal adverse events | Unweighted and weighted event rates a | Weighted pairwise hazard ratios (95% confidence intervals) | ||||
|---|---|---|---|---|---|---|
| GLP‐1RA (n = 21 859) | SGLT2i (n = 84 885) | Insulin glargine (n = 54 375) | GLP‐1RA vs. insulin glargine | Insulin glargine vs. SGLT2i | GLP‐1RA vs. SGLT2i | |
| Gastroparesis | ||||||
| Unweighted | 0.37 (241/64931) | 0.26 (541/211735) | 0.41 (677/166145) | 1.05 (0.90, 1.23) | 1.17 (1.01, 1.34) | 1.21 (1.02, 1.44) |
| Weighted | 0.38 | 0.32 | 0.36 | |||
| Intestinal obstruction | ||||||
| Unweighted | 0.60 (386/64699) | 0.51 (1086/211165) | 0.81 (1341/165165) | 0.85 (0.75, 0.99) | 1.25 (1.13, 1.38) | 1.06 (0.93, 1.22) |
| Weighted | 0.62 | 0.59 | 0.72 | |||
| Gallstones | ||||||
| Unweighted | 0.92 (591/64307) | 0.91 (1918/209718) | 1.05 (1719/164289) | 0.94 (0.84, 1.04) | 1.08 (1.00, 1.17) | 1.01 (0.91, 1.12) |
| Weighted | 0.93 | 0.94 | 0.98 | |||
| Acute cholecystitis | ||||||
| Unweighted | 0.26 (171/65058) | 0.24 (515/211853) | 0.33 (542/166519) | 0.92 (0.76, 1.12) | 1.14 (0.98, 1.32) | 1.05 (0.86, 1.27) |
| Weighted | 0.28 | 0.27 | 0.30 | |||
| Acute pancreatitis | ||||||
| Unweighted | 0.33 (214/64951) | 0.37 (791/211348) | 0.72 (1193/164998) | 0.64 (0.55, 0.75) | 1.27 (1.08, 1.36) | 0.79 (0.64, 0.91) |
| Weighted | 0.34 | 0.46 | 0.52 | |||
| All‐cause death | ||||||
| Unweighted | 2.40 (1570/65333) | 2.58 (5476/212566) | 4.89 (8189/167440) | 0.60 (0.57, 0.64) | 1.59 (1.53, 1.66) | 0.94 (0.88, 1.00) |
| Weighted | 2.73 | 2.80 | 4.58 | |||
Event rates presented as events per 100 person years (N events/years follow‐up).
Subgroup analyses by those with and without each of baseline gastroparesis, intestinal obstruction, gallstones, acute cholecystitis and acute pancreatitis are presented in Tables S8a–e. In general, the number of patients and events was smaller but the event rates were higher in the subgroups with each of the GI conditions at baseline. While no interactions for gallstones, acute cholecystitis and acute pancreatitis were observed, paradoxically, in those with a baseline history of gastroparesis, GLP‐1RA had a lower hazard of gastroparesis compared to insulin glargine or SGLT2i; in those with a baseline history of intestinal obstruction, both insulin glargine and GLP‐1RA had a higher hazard of intestinal obstruction compared to SGLT2i.
4. DISCUSSION
The main finding of this study in a national cohort of veterans with T2D on metformin without baseline history of GI conditions is that, compared to insulin glargine initiation, initiation with GLP‐1RA had a higher risk of gastroparesis but a lower risk of all‐cause death. On the other hand, compared to initiation of SGLT2i, initiation of insulin glargine had a higher risk of gastroparesis, intestinal obstruction and all‐cause death. The risks of gallstones, acute cholecystitis and pancreatitis were similar across the three classes. Thus, while SGLT2i had a lower risk of GI adverse events and GLP‐1RA had a higher risk of gastroparesis, initiation of insulin glargine carried a higher risk of all‐cause death.
Delayed gastric emptying is one of the mechanisms by which GLP‐1RA suppresses food intake and promotes weight loss. Previous reports using 13C breath test, 27 paracetamol absorption technique 28 and gastric residue in an esophagogastroduodenoscopy 29 have shown that gastric emptying was delayed with GLP‐1RA. Reports of clinical diagnoses of gastroparesis have emerged with the widespread use of GLP‐1RA in treating obesity, drawing significant public attention. 30 A retrospective study evaluating over 10 000 American adults with new use of GLP‐1RA within the prior month and diagnosed T2D or obesity found that gastroparesis occurred in 5.1% of participants. 8 In addition, as mentioned above, a recent ACNU design study reported a higher risk of gastroparesis compared to bupropion‐naltrexone for weight loss. 7
However, to our knowledge, the risks of gastroparesis with GLP‐1RA compared to SGLT2i or insulin glargine in the management of T2D have not been reported. In addition, there are no currently well‐accepted risk scoring systems for gastroparesis in T2D to assist in the prediction of those at higher risk of gastroparesis. The results of the current study show that the risk of gastroparesis was higher with GLP‐1RA compared to SGLT2i. Thus, in T2D patients at higher risk of gastroparesis, SGLT2i should be preferred over GLP‐1RA to minimise the risk of this complication. Nonetheless, in those on metformin and SGLT2i and at higher risk of gastroparesis, if another agent is needed for glycaemic control, GLP‐1RA might still be more beneficial than insulin glargine as insulin glargine is associated with a higher risk of all‐cause death compared to GLP‐1RA.
Furthermore, when those with pre‐existing gastrointestinal conditions were included in the analysis, there were no differences between insulin glargine and GLP‐1RA for the risk of gastroparesis. This is biologically plausible. While it is less well recognised, insulin treatment was associated with delayed gastric emptying as evidenced by gastric residue after overnight fasting for esophagogastroduodenoscopy. 31
There is conflicting evidence on the risk of intestinal obstruction with GLP‐1RA compared to SGLT2i. A previous ACNU design study from the United Kingdom reported a higher risk of intestinal obstruction with GLP‐1RA compared to SGLT2i (HR 1.69, 95% CI 1.04–2.74). 10 However, similar to the current study results, a Scandinavian ACNU design study did not find a higher risk of intestinal obstruction with GLP‐1RA compared to SGLT2i (HR 0.83, 95% CI 0.69–1.01). 32 The reasons for the discrepant findings are unclear. A potential possibility is the differences in the specific drugs within each drug class that were used in these different countries, and this needs to be explored further. However, in the current study, compared to the initiation of SGLT2i, initiation of insulin glargine had a higher risk of intestinal obstruction.
As a result of suppression of cholecystokinin secretion by GLP‐1RA 33 leading to decreased gallbladder emptying and marked weight loss in some of the patients 10 treated with GLP‐1RA, these medications might increase the risk for gallstones. Indeed, this is supported by evidence from placebo‐controlled randomized clinical trials. 11 , 12 , 13 However, in the current analysis that compared GLP‐1RA to insulin glargine or SGLT2i, there was no evidence that GLP‐1RA had a higher risk of gallstones or cholecystitis. These discrepant findings between the data from placebo‐controlled randomized trials and the current ACNU study are supported by the findings of a recent meta‐analysis. 14 In that study that included both placebo and active comparator randomized controlled trials of GLP‐1RA, randomization to GLP‐1RA use resulted in a higher risk of gallbladder or biliary diseases in T2D (HR 1.27, 95% CI 1.14–1.43). Even though the interaction p‐value for subgroup analyses by placebo versus active comparator studies was not significant, the higher risk was observed only in placebo‐controlled trials (HR 1.30, 95% CI 1.15–1.47) but not in active comparator trials (HR 1.02, 95% CI 0.73–1.44).
There have been concerns about the risk of pancreatitis with GLP‐1RA use, 15 and it currently has a warning on the drug label. 34 However, meta‐analyses of randomized controlled trials that compared GLP‐1RA use to placebo did not find such a risk. 35 , 36 In the current study, in those without baseline history of GI conditions, all three drug classes had a similar risk of pancreatitis. However, in the sensitivity analysis that included those with preexisting gastrointestinal conditions, GLP‐1RA had a lower risk of acute pancreatitis than either insulin glargine or SGLT2i.
Further studies are warranted to understand the mechanisms by which SGLT2i have lower GI adverse effects. A possible explanation is that SGLT2 is either absent or has a negligible presence in the GI tract.
There are several strengths to the current analysis. The analysis is based on a large national cohort of veterans with T2D. Unlike randomized controlled trials that primarily compared GLP‐1RA against placebo, the current study compared this class in a real‐world setting against two other active drugs, namely insulin glargine, an old staple, and SGLT2i, a newer class in the treatment of T2D. This is of clinical relevance, as placebo is often not a therapeutic option, and the practitioner in the management of T2D must choose from the different classes of agents. Furthermore, randomized comparisons against placebo do not provide direct information on how one class of drugs compares to other classes of drugs.
There are also several limitations to the current study. Residual confounding, such as the impact of patients' lifestyles and genetic factors, might exist despite the use of an active‐control new user design. There have been associations between variations in specific genes and gastrointestinal outcomes; however, we do not evaluate genetic testing in this study. Even though we used a large national cohort, the study population is predominantly male, which limits generalisability, although we did conduct a subgroup analysis based on sex. This study depended solely on provider‐generated ICD‐10 codes, which may miss some diagnoses. We did not use any laboratory data to identify outcomes. We defined drug initiation based on pharmacy prescription fill data, and it is possible that patients filled their prescriptions; however, they did not adhere to taking medications. In addition, we did not examine specific drugs, dosages or duration of drug therapy. Although SGLT2i users were mostly comprised of empagliflozin, GLP‐1RA users included semaglutide, dulaglutide and liraglutide. It is possible that there is heterogeneity in the occurrence of GI adverse events among these three drugs, and by combining them, differences may be masked. Future randomised controlled trials are warranted to further explore the impacts of these drugs, including evaluation of individual GLP‐1RA as well as the use of a combination of medications, on adverse events of the GI system.
In conclusion, while initiation of GLP‐1RA had the highest risk of gastroparesis, initiation with insulin glargine was associated with a higher risk of all‐cause death. Initiation with SGLT2i was associated with a lower risk of gastroparesis but a similar all‐cause death risk compared to initiation with GLP‐1RA. Thus, SGLT2i might be preferred T2D agents in those at risk for gastroparesis. If SGLT2i is not an option, patients and providers might need to weigh the higher risk of death with insulin glargine as opposed to the higher risk of gastroparesis with GLP‐1RA. Developing and validating a risk score for gastroparesis should be a research priority to optimise the use of these agents in T2D.
CONFLICT OF INTEREST STATEMENT
SB received research support from Bayer, Novo Nordisk and Boehringer Ingelheim. SD is a consultant with Abbott and receives research support from Novartis.
PEER REVIEW
The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1111/dom.16642.
Supporting information
Data S1. Supporting Information.
ACKNOWLEDGEMENTS
The research reported in this publication was supported by grants from the National Institute of Diabetes and Digestive and Kidney Diseases R01 DK128640 (SB), T35 DK103596 (SB) National Institute of Aging R01 AG074592 (SB) and the Veterans Affairs Office of Rural Health (SB). This project was also supported by an NHLBI‐funded fellowship T32 HL007576 (SEH). In addition, the research reported in this publication was supported in part by the National Center for Advancing Translational Sciences of the National Institutes of Health under Grant UM1TR004409 (formerly 5UL1TR001067‐05, 8UL1TR000105 and UL1RR025764). The content is solely the responsibility of the authors.
Sarwal A, Singh R, Wei G, et al. Glycemic therapies and the risk of gastrointestinal adverse events in veterans with type 2 diabetes. Diabetes Obes Metab. 2025;27(10):5865‐5877. doi: 10.1111/dom.16642
DATA AVAILABILITY STATEMENT
VA data is available through https://www.research.va.gov/programs/vinci/.
REFERENCES
- 1. Marso SP, Bain SC, Consoli A, et al. Semaglutide and cardiovascular outcomes in patients with type 2 diabetes. N Engl J Med. 2016;375(19):1834‐1844. [DOI] [PubMed] [Google Scholar]
- 2. Marso SP, Daniels GH, Brown‐Frandsen K, et al. Liraglutide and cardiovascular outcomes in type 2 diabetes. N Engl J Med. 2016;375(4):311‐322. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Gerstein HC, Colhoun HM, Dagenais GR, et al. Dulaglutide and cardiovascular outcomes in type 2 diabetes (REWIND): a double‐blind, randomised placebo‐controlled trial. Lancet. 2019;394(10193):121‐130. [DOI] [PubMed] [Google Scholar]
- 4. Perkovic V, Tuttle KR, Rossing P, et al. Effects of semaglutide on chronic kidney disease in patients with type 2 diabetes. N Engl J Med. 2024;391(2):109‐121. [DOI] [PubMed] [Google Scholar]
- 5. American Diabetes Association Professional Practice C . 9. Pharmacologic approaches to glycemic treatment: standards of care in diabetes‐2025. Diabetes Care. 2025;48(Suppl 1):S181‐S206. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Gorgojo‐Martinez JJ, Mezquita‐Raya P, Carretero‐Gomez J, et al. Clinical recommendations to manage gastrointestinal adverse events in patients treated with Glp‐1 receptor agonists: a multidisciplinary expert consensus. J Clin Med. 2022;12(1):145. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Sodhi M, Rezaeianzadeh R, Kezouh A, Etminan M. Risk of gastrointestinal adverse events associated with glucagon‐like peptide‐1 receptor agonists for weight loss. JAMA. 2023;330(18):1795‐1797. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Aldhaleei WA, Abegaz TM, Bhagavathula AS. Glucagon‐like peptide‐1 receptor agonists associated gastrointestinal adverse events: a cross‐sectional analysis of the National Institutes of Health all of us cohort. Pharmaceuticals (Basel). 2024;17(2):199. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Faillie JL, Yin H, Yu OHY, et al. Incretin‐based drugs and risk of intestinal obstruction among patients with type 2 diabetes. Clin Pharmacol Ther. 2022;111(1):272‐282. [DOI] [PubMed] [Google Scholar]
- 10. Faillie JL, Yu OH, Yin H, Hillaire‐Buys D, Barkun A, Azoulay L. Association of bile duct and gallbladder diseases with the use of incretin‐based drugs in patients with type 2 diabetes mellitus. JAMA Intern Med. 2016;176(10):1474‐1481. [DOI] [PubMed] [Google Scholar]
- 11. Wadden TA, Bailey TS, Billings LK, et al. Effect of subcutaneous semaglutide vs placebo as an adjunct to intensive behavioral therapy on body weight in adults with overweight or obesity: the STEP 3 randomized clinical trial. JAMA. 2021;325(14):1403‐1413. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Wilding JPH, Batterham RL, Calanna S, et al. Once‐weekly semaglutide in adults with overweight or obesity. N Engl J Med. 2021;384(11):989‐1002. [DOI] [PubMed] [Google Scholar]
- 13. Lundgren JR, Janus C, Jensen SBK, et al. Healthy weight loss maintenance with exercise, liraglutide, or both combined. N Engl J Med. 2021;384(18):1719‐1730. [DOI] [PubMed] [Google Scholar]
- 14. He L, Wang J, Ping F, et al. Association of glucagon‐like peptide‐1 receptor agonist use with risk of gallbladder and biliary diseases: a systematic review and meta‐analysis of randomized clinical trials. JAMA Intern Med. 2022;182(5):513‐519. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Franks AS, Lee PH, George CM. Pancreatitis: a potential complication of liraglutide? Ann Pharmacother. 2012;46(11):1547‐1553. [DOI] [PubMed] [Google Scholar]
- 16. Vale C, Lourenco IM, Jordan G, et al. Early combination therapy with SGLT2i and GLP‐1 RA or dual GIP/GLP‐1 RA in type 2 diabetes. Diabetes Obes Metab. 2025;27(2):468‐481. [DOI] [PubMed] [Google Scholar]
- 17. Yoshida K, Solomon DH, Kim SC. Active‐comparator design and new‐user design in observational studies. Nat Rev Rheumatol. 2015;11(7):437‐441. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Hernan MA, Robins JM. Using Big Data to Emulate a Target Trial When a Randomized Trial Is Not Available. Am J Epidemiol. 2016;183(8):758‐764. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. U.S. Department of Veterans Affairs . Veterans affairs informatics and computing infrastructure (VINCI). Accessed July 27, 2024. https://www.hsrd.research.va.gov/for_researchers/vinci/
- 20. U.S. Department of Veterans Affairs . National Center for Veterans Analysis and Statistics. Accessed July 30, 2024. https://www.va.gov/vetdata/
- 21. U.S. Department of Veterans Affairs . VIReC research user guides. Accessed July 27, 2024. https://www.virec.research.va.gov/Resources/RUGs.asp
- 22. U.S. Department of Veterans Affairs . VHA medical SAD inpatient datasets FY2006. Accessed July 27, 2024. https://vaww.virec.research.va.gov/RUGs/MedSAS/Archive/IP/RUG-MedSAS-IP-FY06-RA.pdf.
- 23. U.S. Department of Veterans Affairs . VHA death ascertainment file (DAF) and other data sources for mortality ascertainment. Accessed July 30, 2024. https://www.hsrd.research.va.gov/for_researchers/cyber_seminars/archives/video_archive.cfm?SessionID=6303
- 24. Hastie T, Tibshirani R, Sherlock G, Eisen M, Brown P, Botstein D. Imputing missing data for gene expression arrays. 1999. https://hastie.su.domains/Papers/missing.pdf
- 25. Troyanskaya O, Cantor M, Sherlock G, et al. Missing value estimation methods for DNA microarrays. Bioinformatics. 2001;17(6):520‐525. doi: 10.1093/bioinformatics/17.6.520 [DOI] [PubMed] [Google Scholar]
- 26. McCaffrey DF, Griffin BA, Almirall D, Slaughter ME, Ramchand R, Burgette LF. A tutorial on propensity score estimation for multiple treatments using generalized boosted models. Stat Med. 2013;32(19):3388‐3414. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Kuwata H, Yabe D, Murotani K, et al. Effects of glucagon‐like peptide‐1 receptor agonists on secretions of insulin and glucagon and gastric emptying in Japanese individuals with type 2 diabetes: a prospective, observational study. J Diabetes Investig. 2021;12(12):2162‐2171. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Dahl K, Brooks A, Almazedi F, Hoff ST, Boschini C, Baekdal TA. Oral semaglutide improves postprandial glucose and lipid metabolism, and delays gastric emptying, in subjects with type 2 diabetes. Diabetes Obes Metab. 2021;23(7):1594‐1603. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Kobori T, Onishi Y, Yoshida Y, et al. Association of glucagon‐like peptide‐1 receptor agonist treatment with gastric residue in an esophagogastroduodenoscopy. J Diabetes Investig. 2023;14(6):767‐773. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Goodman B. They took blockbuster drugs for weight loss and diabetes. Now their stomachs are paralyzed. Accessed July 30, 2024. https://www.cnn.com/2023/07/25/health/weight-loss-diabetes-drugs-gastroparesis/index.html
- 31. Kobori T, Onishi Y, Iwamoto M, et al. Association of insulin treatment with gastric residue during an esophagogastroduodenoscopy. J Diabetes Investig. 2022;13(3):501‐504. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Ueda P, Wintzell V, Melbye M, et al. Use of DPP4 inhibitors and GLP‐1 receptor agonists and risk of intestinal obstruction: Scandinavian Cohort Study. Clin Gastroenterol Hepatol. 2024;22(6):1226‐1237.e14. [DOI] [PubMed] [Google Scholar]
- 33. Nexoe‐Larsen CC, Sorensen PH, Hausner H, et al. Effects of liraglutide on gallbladder emptying: a randomized, placebo‐controlled trial in adults with overweight or obesity. Diabetes Obes Metab. 2018;20(11):2557‐2564. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Food and Drug Administration . Highlights of prescribing information. Accessed July 27, 2024. https://www.accessdata.fda.gov/drugsatfda_docs/label/2022/209637Orig1s009lbl.pdf
- 35. Bethel MA, Patel RA, Merrill P, et al. Cardiovascular outcomes with glucagon‐like peptide‐1 receptor agonists in patients with type 2 diabetes: a meta‐analysis. Lancet Diabetes Endocrinol. 2018;6(2):105‐113. [DOI] [PubMed] [Google Scholar]
- 36. Storgaard H, Cold F, Gluud LL, Vilsboll T, Knop FK. Glucagon‐like peptide‐1 receptor agonists and risk of acute pancreatitis in patients with type 2 diabetes. Diabetes Obes Metab. 2017;19(6):906‐908. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data S1. Supporting Information.
Data Availability Statement
VA data is available through https://www.research.va.gov/programs/vinci/.