Risk of Intestinal Blockage Varies by Type and Dose of GLP-1 and SGLT2 Diabetes Drugs: A Comparison of Clinical Trials

The study selection process is summarized in Figure 1. After screening and full-text review, 110 articles were excluded for reasons such as ineligible design, absence of relevant outcomes, or duplicate reporting (Table S3) [6,7,8,9,10,11,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125]. Ultimately, 47 publications describing 50 distinct RCTs were included in the NMA (Table S4) [12,13,14,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159,160,161,162,163,164,165,166,167,168,169]. Across these trials, 192,359 participants were randomized. The mean age was 62.9 years (range 44.9–71.9 years), and women comprised a mean of 38.0% of participants (range 23.4–81.0%). Mean trial duration was 132.2 weeks, with follow-up ranging from 13 to 281 weeks.

GLP-1 receptor agonists evaluated included tirzepatide, efpeglenatide, liraglutide, albiglutide, dulaglutide, exenatide, semaglutide, and lixisenatide. SGLT2 inhibitors assessed were bexagliflozin, canagliflozin, empagliflozin, ertugliflozin, dapagliflozin, and sotagliflozin. Trials varied in background therapy and baseline cardiovascular risk, but collectively represented contemporary clinical use of these agents in type 2 diabetes and cardiometabolic populations.

In the primary NMA, canagliflozin use was associated with a statistically significant increase in intestinal obstruction compared with pooled control treatments (OR 2.56, 95% CI 1.01–6.49). This corresponds to an absolute risk difference of 0.15% and a number needed to harm of approximately 658. Among all regimens, canagliflozin occupied the highest rank in terms of obstruction risk (Figure 2A and Figure 3A, Figure S3A, and Table 1). As addressed in the introduction, patients with diabetes mellitus often have multiple risk factors for intestine obstruction (for example, autonomic nervous system dysfunction and obesity). Although we intended to perform subgroup analysis focusing on RCTs with specific comorbidity (such as autonomic nervous system dysfunction and obesity), the analytic process could not be performed due to the sparse numbers of RCTs focusing such specific comorbidity.

Liraglutide, in contrast, was the only regimen associated with a statistically significant reduction in intestinal obstruction. Compared with controls, liraglutide yielded an OR of 0.44 (95% CI 0.24–0.81), translating into an absolute risk difference of −0.34% and a number needed to treat of 294.7. For all other GLP-1 receptor agonists and SGLT2 inhibitors, point estimates for obstruction were generally close to unity and did not reach conventional thresholds for statistical significance.

Subgroup analyses stratified by dosing tier highlighted a pronounced dose–response signal for canagliflozin. High-dose canagliflozin (300 mg/day) was associated with a significantly increased risk of intestinal obstruction relative to control (OR 3.42, 95% CI 1.08–10.76), with an absolute risk difference of 0.11% and a number needed to harm of 933.39. Across all evaluated dose nodes, high-dose canagliflozin consistently ranked as the regimen with the greatest obstruction risk (Figure 2B and Figure 3B, Figure S3B, and Table 2). In contrast, lower canagliflozin doses and dose tiers for other agents did not demonstrate comparable elevations in risk.

We used all-cause study withdrawal as a global indicator of treatment acceptability. Several agents exhibited significantly lower discontinuation rates than their respective control arms. These regimens included tirzepatide (OR 0.44, 95% CI 0.35–0.55), oral semaglutide (OR 0.40, 95% CI 0.19–0.85), a secondary tirzepatide comparison (OR 0.56, 95% CI 0.42–0.74), canagliflozin (OR 0.66, 95% CI 0.55–0.80), albiglutide (OR 0.71, 95% CI 0.58–0.86), liraglutide (OR 0.76, 95% CI 0.66–0.89), dapagliflozin (OR 0.79, 95% CI 0.62–0.99), injectable semaglutide (OR 0.81, 95% CI 0.70–0.94), sotagliflozin (OR 0.86, 95% CI 0.74–0.99), and empagliflozin (OR 0.87, 95% CI 0.79–0.95). Overall, tirzepatide ranked most favorably in terms of retention and global acceptability (). Figures S1, S2 and S3C, and Table S5

Bayesian NMA models produced effect estimates and treatment rankings that were highly consistent with those obtained from frequentist analyses (). SUCRA-based rankings derived from Bayesian models are presented in. Deviation-model diagnostics did not reveal major incoherence or instability across treatment comparisons, suggesting that the network structure was well supported by the underlying trial data (). Figure S4A–C Table S6A–C and Figure S5A–F Figure S6A–I

Across all included trials, 81.4% (285 of 350 items) of risk-of-bias judgments were categorized as low risk, 15.4% (54 items) as unclear, and 3.2% (11 items) as high risk (). Network-wide consistency checks did not identify statistically significant discrepancies between direct and indirect estimates for key comparisons (). According to GRADE assessments adapted for NMA, the certainty of evidence for most agent-level comparisons ranged from moderate to high, supporting the reliability of the main conclusions (). Figure S7A,B Table S7A–C Table S8A–C

This study has several strengths that enhance confidence in the conclusions. First, the network meta-analytic framework allowed us to integrate data from multiple active and placebo comparators and to compare individual agents and dose tiers within a single coherent model. This design is particularly valuable in the context of modern antidiabetic therapy, where head-to-head trials of all clinically relevant regimens are not feasible. Second, we restricted analyses to RCTs and excluded participants with pre-existing obstruction, thereby focusing on incident events and improving internal validity. Third, dose-stratified modeling enabled us to separate class effects from dose-specific pharmacodynamic signals, an important consideration for agents such as canagliflozin where high-dose regimens may have distinct intestinal exposures compared with lower doses. Fourth, we corroborated frequentist results using Bayesian NMA and SUCRA-based rankings, which yielded concordant patterns and support the stability of our findings.

Several limitations should also be acknowledged. First, we could not consistently distinguish small bowel from large bowel obstruction, or mechanical obstruction from functional ileus, because such distinctions were rarely reported in the source trials. This lack of granularity limited our ability to examine specific pathophysiological subtypes. Second, by focusing on RCTs, we may have excluded valuable information from large observational cohorts that capture longer exposure durations and broader patient populations. Nonetheless, RCT-based estimates provide an essential anchor for causality and reduce confounding by indication. Third, heterogeneity in diagnostic criteria, imaging practices, and adverse-event coding across international trials may have introduced variability in outcome ascertainment and contributed to imprecision in some estimates. As addressed in the methodology section, the current NMA did not make distinction between mechanical obstruction and functional ileus, nor between small-bowel and large-bowel obstruction. In fact, the most frequently recorded "intestine obstruction events" among the included RCTs were just remarked as "intestine obstruction" without any other specifiers. Therefore, it would be practically difficult to distinguish them. We surely recognized that the process of merging these site-specific and mechanism-specific intestine obstruction into one group would impose potential bias in the main result. However, meta-analyses of large datasets remain a cornerstone for safety signal detection and are frequently used to generate hypotheses for regulatory and pharmacoepidemiologic follow-up [,,,]. Fourth, concomitant medications that influence gastrointestinal motility (e.g., opioids, anticholinergics) were not systematically accounted for in the trial reports. This limitation was the natural result that the most RCTs were not initially designed to evaluate intestine obstruction, which needed specific information regarding baseline medication affecting gastrointestinal motility. Although randomization should mitigate systematic imbalances, residual confounding cannot be completely excluded. Therefore, future large-scale RCTs specifically designed for exploring gastrointestinal motility related to such medications should be warranted. Fifth, since some of the intestine obstruction events of the included RCTs came from spontaneous reporting systems and none of the included RCTs were primarily designed to evaluate intestinal obstruction or ileus, the possibility of underreporting and inconsistent capture of rare events might raise the possibility of underestimation of intestine obstruction. In fact, the lack of large-scale RCTs directly investigating intestine obstruction related to such medication prescription might limit the certainty of the evidence. Specifically, evidence from RCTs with direct investigation of intestine obstruction adverse event data might provide more reliable evidence than post-marketing reports. Therefore, dedicated prospective studies, ideally with predefined gastrointestinal safety endpoints and standardized adjudication, will be necessary to fully characterize the mechanistic links between these agents and obstructive complications. In addition, there was still a gap for the linkage between animal study and statistical evidence in our NMA so that clinicians should pay special attention when applied our result in clinical practice. Finally, as addressed in the introduction, patients with diabetes mellitus often have multiple risk factors for intestine obstruction (for example, autonomic nervous system dysfunction and obesity). Despite our intention to perform subgroup analysis to exclude the potential confounding effects, the analytic process could not be performed due to the sparse data. Therefore, future large-scale RCTs focusing specific comorbidity should be warranted. ^{15 16 183 184}

A comprehensive search strategy was implemented across eight electronic databases: PubMed, Embase, ClinicalKey, Cochrane CENTRAL, ProQuest, ScienceDirect, Web of Science, and ClinicalTrials.gov, from inception through 21 January 2025 (Table S2). Search terms combined controlled vocabulary and free-text terms related to GLP-1 receptor agonists, SGLT2 inhibitors, randomized trials, and intestinal obstruction or ileus. Two investigators (PT Tseng and BY Zeng) independently screened titles and abstracts, followed by full-text review of potentially eligible articles. Discrepancies at any stage were resolved through discussion and, when necessary, consultation with a third reviewer. Reference lists of relevant narrative reviews and meta-analyses were screened manually to identify additional RCTs not captured in the initial database search [9,10,22]. No language restrictions were imposed.

Eligibility criteria were predefined using the Population–Intervention–Comparator–Outcome–Study design (PICOS) framework. Eligible populations were adults without baseline intestinal obstruction. Interventions included any GLP-1 receptor agonist or SGLT2 inhibitor administered at a therapeutic dose approved or under investigation for type 2 diabetes or related cardiometabolic indications. Comparators comprised placebo, standard-of-care therapy, or alternative active treatments. The primary outcome was incident intestinal obstruction (small or large bowel) reported during trial follow-up. Only randomized controlled trials with at least one GLP-1 receptor agonist or SGLT2 inhibitor arm and an eligible comparator arm were included.

To minimize selective reporting and under-ascertainment of harms, we restricted inclusion to RCTs that incorporated systematic adverse-event monitoring or explicitly listed intestinal obstruction, ileus, or related terms as part of safety reporting [187]. Trials exclusively enrolling patients with pre-existing obstruction were excluded to maintain focus on incident or drug-attributable events. Additional exclusion criteria were non-randomized designs, animal or preclinical studies, trials without relevant comparator arms, and studies lacking sufficient data on intestinal obstruction to contribute to the network.

Two reviewers independently evaluated risk of bias for each included RCT using the original Cochrane Risk of Bias Tool (version 1.0) [188]. The domains assessed included random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, completeness of outcome data, selective reporting, and other potential sources of bias. Any disagreements in domain-level judgments were resolved through consensus with a third senior investigator.

For the purposes of this NMA, "intestinal obstruction" encompassed both small bowel obstruction and large bowel obstruction (including colonic and rectal involvement), as recorded in individual trial reports. Given that these clinical entities commonly form a continuum and are often aggregated in adverse-event tables [189], we pooled them into a composite endpoint. Most source trials did not clearly distinguish mechanical obstruction from functional ileus, nor did they provide consistent radiologic or operative confirmation; therefore, these subtypes were not analyzed separately. All-cause study withdrawal, regardless of reason, was extracted as a secondary safety endpoint reflecting overall treatment acceptability.

To explore potential dose–response relationships, we categorized treatment arms according to dosing tiers defined in the original trials and aligned with typical clinical use. Dose nodes were specified a priori as follows: Canagliflozin: low 100 mg; high 300 mg. Efpeglenatide: low 2 mg; medium 4 mg; high 6 mg. Ertugliflozin: low 5 mg; high 15 mg. Dapagliflozin: low 2.5 mg; medium 5 mg; high 10 mg. Injectable semaglutide: low 0.5 mg; medium 1.0 mg; high 2.4 mg. Empagliflozin: low 1–10 mg; high 25–50 mg.

These dose categories were modeled as separate nodes in the network to allow explicit assessment of dose stratification within each molecule.

Data extraction was performed independently by two investigators (PT Tseng and BY Zeng), who used a standardized electronic form to record trial characteristics (design, phase, follow-up duration), participant demographics, baseline risk factors, treatment regimens (agent, dose, route, and frequency), and outcome data for intestinal obstruction and study discontinuation. When necessary information (e.g., number of obstruction events or exact dose) was missing from published reports, attempts were made to contact corresponding authors. Extraction procedures followed recommendations from the Cochrane Handbook for Systematic Reviews of Interventions and best practices in evidence-based pharmacotherapy research [190].

We conducted a frequentist random-effects NMA to integrate direct and indirect evidence across all eligible comparisons, treating each drug–dose combination as an individual treatment node [191]. Analyses were performed using MetaInsight (version 4.0.2, Complex Reviews Support Unit, National Institute for Health Research, London, UK), which implements the netmeta package (version 4.0.2) in R for graph-theoretical NMA estimation [192].

For dichotomous outcomes, continuity corrections were applied to trials with zero events in a single arm, whereas studies with zero events in all arms were excluded from the primary analysis to avoid biased effect estimates [193,194]. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated and displayed using forest plots for pairwise and network estimates [195]. Treatment ranking probabilities and relative safety hierarchies were derived from the network model. Consistency between direct and indirect evidence was examined using node-splitting approaches, which are especially useful when synthesizing multi-arm pharmacologic trials [192,196]. A two-sided p value < 0.05 was considered statistically significant.

To test robustness of the primary findings, we repeated the analyses using Bayesian random-effects NMA models with non-informative priors. From these models, we derived SUCRA (surface under the cumulative ranking) values, presented using Litmus Rank-O-Gram and radial SUCRA visualizations, to summarize the probability that each regimen occupies a given rank in terms of intestinal obstruction risk and treatment acceptability [197]. Additional model diagnostics, including deviation-based ranking analyses, were used to identify any influential nodes or incoherent loops in the network [198]. Overall certainty of evidence for each comparison was graded using the GRADE framework adapted for NMA, incorporating risk of bias, indirectness, inconsistency, imprecision, and publication bias [199].