Endoscopic hemostasis combined with vascular interventional therapy for acute nonvariceal upper gastrointestinal bleeding: A meta-analysis

Abstract

BACKGROUND

Acute nonvariceal upper gastrointestinal bleeding (NVUGIB) is a life-threatening emergency. Endoscopic hemostasis and vascular interventional therapy are two major minimally invasive treatment strategies. Although these modalities are widely used, their comparative efficacy and safety across different patient populations and anatomical sites remain controversial.

AIM

To evaluate the clinical outcomes and safety of endoscopic hemostasis combined with vascular interventional therapy for NVUGIB.

METHODS

A systematic search was done on PubMed, EMBASE, Cochrane Library, and Web of Science (from database establishment to April 2025). Randomized controlled trial (RCT) quality was assessed via Cochrane RoB 2.0, and observational studies via the Critical Appraisal Skills Program. RevMan 5.4 was used for quantitative analysis; fixed/random-effects models were chosen through I²-assessed heterogeneity. Publication bias was checked using funnel plots and sensitivity analysis via model switching.

RESULTS

Twenty-one studies (3 RCTs, 12 single-group studies, and 6 retrospective cohort studies) with good quality were included. For single-group data, combined therapy had risk differences of 0.70 (clinical success), 0.24 (mortality), and 0.22 (rebleeding; all P < 0.00001, I² = 0). Moreover, the ≥ 60-year rebleeding risk difference was 0.43. Reintervention was found to differ by approach (Z = 3.03, P = 0.002, inter-subgroup I² = 99%). In the RCT and cohort studies, combined vs standard therapy had similar initial hemostasis (Z = 0.04, P = 0.97) and mortality (Z = 1.56, P = 0.12) but lower rebleeding (Z = 3.26/P = 0.001; Z = 2.95/P = 0.03). Symmetrical funnel plots and robust sensitivity analysis showed no publication bias.

CONCLUSION

Combined endoscopic hemostasis and vascular interventional therapy for acute NVUGIB can significantly reduce rebleeding, without differences in mortality. Age and vascular interventional methods may influence the therapeutic efficacy.

Key Words: Endoscopic hemostasis; Vascular intervention; Acute; Nonvariceal; Upper gastrointestinal bleeding; Rebleeding; Clinical outcomes

Core Tip: This meta-analysis demonstrates that combining endoscopic hemostasis with vascular interventional therapy is effective for acute nonvariceal upper gastrointestinal bleeding, significantly reducing rebleeding rates versus standard therapy. While it shows no significant differences in initial hemostasis success or mortality, mortality findings require further validation. These results aid clinical treatment optimization, though subgroup effects related to age and intervention methods need deeper investigation.

INTRODUCTION

Acute nonvariceal upper gastrointestinal bleeding (NVUGIB) is a life-threatening emergency in gastroenterology, characterized by high mortality rates that pose a severe threat to patient health. With the continuous advancement of endoscopic and vascular interventional techniques, endoscopic hemostasis and vascular interventional therapy have emerged as critical modalities for treating acute NVUGIB. However, controversy persists regarding the superiority of these methods in terms of clinical efficacy and safety. Epidemiological data indicate an increasing global incidence of NVUGIB. According to recent studies, the annual incidence of NVUGIB is approximately 54.4 cases per 100000 individuals, with an overall mortality rate as high as 13.5%[1]. Currently, endoscopic hemostasis is the first-line treatment for NVUGIB, owing to its minimally invasive nature and operational convenience[2]. Nevertheless, vascular interventional therapy has a unique therapeutic value for patients with failed endoscopic treatment, recurrent bleeding, or complex anatomical structures[3]. Most existing studies have focused on single-center, small-sample comparisons of the two treatment methods and lack a comprehensive, systematic evaluation. Although some studies have compared endoscopic hemostasis with vascular interventional therapy, their conclusions vary owing to factors such as sample size, study design, and baseline patient characteristics. Therefore, an urgent need exists to integrate relevant evidence through meta-analyses to provide reliable support for clinical decision-making.

We systematically collected high-quality research worldwide, applied strict inclusion and exclusion criteria, and used advanced meta-analytical methods to comprehensively evaluate the clinical efficacy and safety of endoscopic hemostasis and vascular interventional therapies for acute NVUGIB. This study aimed to systematically compare the two treatment methods in terms of hemostasis success, rebleeding, complications, and mortality rates, thereby clarifying their advantages and limitations in clinical application. Through a rigorous scientific meta-analysis, this study provides clinicians with more objective and comprehensive evidence-based medical insights for selecting treatment protocols for acute NVUGIB, optimizing treatment strategies, improving patient survival rates, and enhancing outcomes, thus offering significant guidance for its clinical management.

MATERIALS AND METHODS

Search strategy

Computer-based searches were conducted in PubMed, Web of Science, EMBASE, and the Cochrane Library from inception to May 2025, focusing on English-language literature. For instance, the search strategy for PubMed was as follows: {[nonvariceal(Title/Abstract)] AND [upper gastrointestinal bleeding(Title/Abstract)] OR [acute(Title/Abstract)] AND [vascular intervention(Title/Abstract)] OR [endoscopy hemostasis(Title/Abstract)]}.

This strategy integrated keywords related to the disease (“nonvariceal”, “upper gastrointestinal bleeding”, “acute”) and treatment modalities (“vascular intervention”, “endoscopic hemostasis”), with systematic retrieval of titles and abstracts to ensure comprehensive capture of relevant studies. Similar search strategies were adopted for other databases using subject headings (e.g., Medical Subject Headings terms) and free-text keywords to maintain consistency and comparability across databases. This study has been registered in the International Prospective Register of Systematic Reviews (No. CRD420251077603).

Literature screening criteria

Inclusion criteria: (1) Study design: Single-arm intervention studies, retrospective cohort studies, and randomized controlled trials (RCTs) in English; (2) Study population: Patients diagnosed with acute NVUGIB via gastroscopy, imaging, or other clinical examinations[4], regardless of age, sex, or race; (3) Interventions: Experimental group - endoscopic hemostasis (including injection hemostasis, thermal coagulation, and mechanical hemostasis) or vascular interventional therapy (e.g., transarterial embolization); control group - standard treatment[5] or conventional therapy (e.g., pharmacological hemostasis and conservative treatment); and (4) Outcome measures: Successful initial hemostasis rate, clinical success rate, rebleeding rate, mortality rate, reintervention rate, and complication rate (at least one of them).

Exclusion criteria: (1) Duplicate publications or overlapping datasets; (2) Studies with missing primary outcome data that cannot be supplemented by contacting the authors; (3) Studies that included patients with variceal upper gastrointestinal bleeding or comorbidities that severely affect outcomes (e.g., terminal cancer or multi-organ failure); (4) Abstract-only articles, conference abstracts, reviews, case reports, or lecture notes; and (5) Studies without full-text access or that failed quality assessment (high risk of bias in RCTs and severe methodological flaws in nonrandomized studies).

Literature screening and data extraction: Two independent reviewers strictly followed predefined inclusion and exclusion criteria to conduct literature screening, data extraction, and cross-verification. The screening process was performed in two stages. First, titles and abstracts were reviewed to exclude studies that did not meet the inclusion criteria or were duplicate publications. Second, the full texts of the remaining studies were evaluated to determine the final eligibility. The extracted key data included author information, publication year, study design, sample size, detailed interventions, follow-up duration, and outcome metrics. In cases of disagreement between the reviewers, a third reviewer was consulted to reach a consensus, ensuring the accuracy and reliability of literature screening and data extraction.

Literature quality assessment: To ensure the reliability of this meta-analysis, two reviewers independently evaluated the methodological quality of the included studies using targeted tools based on internationally recognized evidence-based medicine criteria. For RCTs, the Cochrane Risk of Bias tool (RoB 2.0)[6] was applied for systematic evaluation. This assessment covered dimensions including random sequence generation, allocation concealment, implementation of blinding, completeness of outcome data, selective reporting, and other potential sources of bias, aiming to comprehensively identify risks of bias in the study design and conduct. For single-group intervention studies and retrospective cohort studies, the Critical Appraisal Skills Program checklist[7] was used. The key evaluation criteria included sample representativeness, accuracy of intervention descriptions, objectivity of outcome measures, completeness of follow-up, and control of confounding factors to systematically assess the quality of the research evidence. After the two reviewers independently performed quality assessments, cross-verification was performed to ensure consistency. Discrepancies were initially resolved through discussion and, when necessary, adjudicated by a third reviewer. Finally, the quality assessment results were summarized by study type to provide a scientific basis for subsequent stratified meta-analyses.

Statistical analysis

The meta-analysis was performed using RevMan 5.4 software. Heterogeneity across studies was assessed using the I² statistic with the following criteria: Low heterogeneity was defined as P ≥ 0.10 and I² < 50%, warranting the use of the fixed-effect model. Significant heterogeneity was defined as P < 0.10 and I² ≥ 50%, requiring the random-effects model. In cases where P ≥ 0.10 but I² ≥ 50%, the I² statistic served as the primary determinant.

Sensitivity analysis was performed by switching between the fixed-effects model and the random-effects model. Specifically, the pooled effect sizes of the included studies were calculated based on the two models, and the magnitude of numerical change in pooled effect sizes and consistency of effect directions under different models were compared. If the difference in pooled effect sizes obtained from the two models was small and the effect directions were consistent, it indicated that the results of the meta-analysis were less affected by model selection and had good overall stability. In contrast, significant differences in effect sizes or changes in effect directions suggested poor result stability, requiring further analysis of the potential impact of heterogeneity sources on the conclusions.

Funnel plots were used to assess the publication bias of the included studies. The horizontal axis of the funnel plot represented the study effect sizes (e.g., mean difference, odds ratio), while the vertical axis represented the sample size (reflecting study precision). A symmetric “inverted funnel” shape of the funnel plot - where studies with effect sizes close to the true value were evenly distributed on both sides of the central line, with small-sample studies scattered at the bottom and large-sample studies concentrated at the top - indicated a low risk of publication bias. Conversely, asymmetric patterns such as an overabundance of studies on one side or a gap at the bottom suggested the potential existence of publication bias.

RESULTS

Literature screening process and results

A total of 178 records were retrieved, with no additional studies identified from other sources. After the removal of 36 duplicate records, 37 irrelevant studies were excluded, leaving 75 full-text articles for evaluation. Ultimately, 21 studies[8-28] were included in the qualitative analysis. The literature screening process is illustrated in Figure 1.

Figure 1 The literature screening process. One hundred and seventy-eight records were retrieved, with no additional studies from other sources. After removing 34 duplicate records, 37 irrelevant studies were excluded, leaving 78 full-text articles for evaluation. Ultimately, 24 studies were included for qualitative synthesis.

Study characteristics and quality

The included studies were single-group intervention studies, retrospective cohort studies, and RCTs. The baseline characteristics are summarized in Table 1. All RCTs achieved a quality assessment score of ≥ 7 points. Most single-group intervention studies demonstrated “fairly good” performance in intervention description, objectivity of outcome measures, and completeness of follow-up, while sample representativeness and control of confounding bias were rated “average” - primarily in small-sample studies. Overall, the methodological quality of the included studies is acceptable. The quality assessment results are presented in Table 2.

Table 1 Baseline characteristics of included studies.

Ref.	Country	Study type1	Participants	Male	Median age	Vascular intervention	Follow-up	Outcome2
McGraw et al[8], 2023	Pennsylvania	C	269	-	-	TAE	6 years	(3), (4)
Lau et al[9], 2019	Thailand	C	241	-	-	AE	30 days	(4), (5)
Nykänen et al[10], 2017	Finland	B	85			TAE	30 days	(3), (4), (6)
Manta et al[11], 2013	Italy	A	30	-	-	OTSC	1-month	(3), (5)
Aina et al[12], 2001	Canada	A	132	-	63	AE	30 days	(2)
Arrayeh et al[13], 2012	United States	B	115	46	61	AE	30 days	(1)
Defreyne et al[14], 2001	Belgium	A	91			AE	-	(1)
Dixon et al[15], 2013	United Kingdom	B	40	32	71	AE	-	(1)
Holme et al[16], 2006	Denmark	A	40	21	70	AE	-	(3), (4)
Kaminskis et al[17], 2019	Latvia	B	738	222	69	AE	-	(3), (4), (5)
Laursen et al[18], 2014	Denmark	C	105	-	-	STAE	30 days	(3)
Muhammad et al[19], 2019	Pakistan	A	32	0.687	56	TAE	30 days	(2), (4)
Padia et al[20], 2009	United States	A	108	66	66	Transcatheter embolization	30 days	(2), (4), (5)
Poultsides et al[21], 2008	United States	A	57	38	65	Embolization procedures	-	(2), (5), (6)
Song et al[22], 2011	Korea	A	16	11	59	TAE	4 weeks	(2), (3), (4), (5)
Spiliopoulos et al[23], 2018	Italy	A	44	44	74	TAE	3.5 years	(3), (4), (6)
Ephraim et al[24], 2022	United States	A	74	-	-	TAE	30 days	(2), (3), (4)
Ang et al[25], 2012	Singapore	B	93	64	67	TAE	30 days	(1), (4), (6)
Lee et al[26], 2015	Korea	A	66	42	60	TAE or AE	30 days	(3), (4)
Huang et al[27], 2014	Taiwan, China	A	49	31	67	TAE	30 days	(2), (3), (4), (5), (6)
Hur et al[28], 2017	Korea	A	152	109	66	AE	1-month	(2), (4), (6)

¹A: Single-group intervention study; B: Retrospective cohort study; C: Randomized controlled trial.

²(1) Successful initial hemostasis; (2) Clinical success; (3) Rebleeding; (4) Mortality; (5) Reinterventions; and (6) Complications. TAE: Transarterial embolization; OTSC: Over-the-scope clip; STAE: Supplementary transcatheter arterial embolization; AE: Angiographic embolization.

Table 2 Quality assessment results.

Ref.	Study type1	Cochrane	CASP
			Representativeness of the sample	Clarity of intervention description	Objectivity of outcome measures	Completeness of follow-up	Control of confounding bias
McGraw et al[8], 2023	C	8	-	-	-	-	-
Lau et al[9], 2019	C	7	-	-	-	-	-
Nykänen et al[10], 2017	B	-	Fairly good	Fairly good	Fairly good	Fairly good	Fairly good
Manta et al[11], 2013	A	-	Average	Fairly good	Fairly good	Fairly good	Average
Aina et al[12], 2001	A	-	Fairly good	Fairly good	Fairly good	Fairly good	Average
Arrayeh et al[13], 2012	B	-	Fairly good	Fairly good	Fairly good	Fairly good	Average
Defreyne et al[14], 2001	A	-	Fairly good	Fairly good	Fairly good	Average	Average
Dixon et al[15], 2013	B	-	Average	Fairly good	Fairly good	Average	Fairly good
Holme et al[16], 2006	A	-	Average	Fairly good	Fairly good	Average	Average
Kaminskis et al[17], 2019	B	-	Fairly good	Fairly good	Fairly good	Average	Fairly good
Laursen et al[18], 2014	C	8	-	-	-	Fairly good	-
Muhammad et al[19], 2019	A	-	Average	Fairly good	Fairly good	Fairly good	Average
Padia et al[20], 2009	A	-	Fairly good	Fairly good	Fairly good	Fairly good	Average
Poultsides et al[21], 2008	A	-	Fairly good	Fairly good	Fairly good	Average	Fairly good
Song et al[22], 2011	A	-	Fairly good	Fairly good	Fairly good	Fairly good	Fairly good
Spiliopoulos et al[23], 2018	A	-	Average	Fairly good	Fairly good	Fairly good	Average
Ephraim et al[24], 2022	A	-	Fairly good	Fairly good	Fairly good	Fairly good	Fairly good
Ang et al[25], 2012	B	-	Fairly good	Fairly good	Fairly good	Fairly good	Average
Lee et al[26], 2015	A	-	Fairly good	Fairly good	Fairly good	Fairly good	Fairly good
Huang et al[27], 2014	A	-	Fairly good	Fairly good	Fairly good	Fairly good	Fairly good
Hur et al[28], 2017	A	-	Fairly good	Fairly good	Fairly good	Fairly good	Fairly good

¹A: Single-group intervention study; B: Retrospective cohort study; C: Randomized controlled trial.

CASP: Critical appraisal skills program.

Pooled results of efficacy outcome measures

Single-group intervention studies: (1) Clinical success and mortality: Several studies[12,14,19,27,28] reported the clinical success rate. The heterogeneity was I² = 0 (P = 0.59), and the risk difference (RD) of clinical success before and after intervention was 0.70 [95% confidence interval (CI): 0.65-0.75] (P < 0.00001) (Figure 2A). Several studies[8,12,14,16,19,20,23,27,28] reported the mortality of endoscopic hemostasis combined with vascular interventional therapy. The heterogeneity was I² = 0 (P = 0.57), and the RD of mortality before and after intervention was 0.24 (95%CI: 0.21-0.28) (P < 0.00001) (Figure 2B); (2) Rebleeding: Two studies[14,24] reported the rebleeding rate in patients of all ages who underwent endoscopic hemostasis combined with vascular interventional therapy. Analyzed using a fixed-effects model, the RD of rebleeding before and after intervention was 0.14 (95%CI: 0.08-0.21). Two studies[26,27] reported the rebleeding rate in patients aged ≥ 60 years who received the same combined therapy. The RD of rebleeding before and after intervention was 0.43 (95%CI: 0.32-0.54). The overall rebleeding RD for all age groups was 0.22 (95%CI: 0.16-0.27) (Z = 7.75, P < 0.00001) (Figure 2C). After switching to a random-effects model, the overall rebleeding RD was 0.28 (95%CI: 0.12-0.43) (Z = 3.47, P < 0.00001) (Figure 2D). These results indicate that there are differences in the overall rebleeding rate between different effect models, and factors such as age may have an impact on the rebleeding rate; and (3) Reinterventions: Two studies[20,21] reported the reintervention rate in patients who underwent transcatheter embolization/embolization procedures. Analyzed using a fixed-effects model, the RD of reinterventions before and after intervention was 0.22 (95%CI: 0.16-0.28). Three studies[11,22,27] reported the reintervention rate in patients who received over-the-scope clip (OTSC)/endoscopic treatment/transcatheter arterial embolization. Analyzed using a fixed-effects model, the RD of reinterventions was -2.64 (95%CI: -2.94 to -2.34). The overall effect test showed a statistically significant difference (Z = 3.03, P = 0.002), suggesting a significant difference in the reintervention rate between different vascular interventional methods. This suggests that the choice of interventional method affects the probability of subsequent reinterventions. In addition, the subgroup heterogeneity test result (I² = 0) suggested no significant heterogeneity among the included studies within subgroups, while there was significant heterogeneity between the two subgroups (I² = 99%, P < 0.05), indicating that the heterogeneity may originate from the difference between embolization and non-embolization interventional methods (Figure 2E).

Figure 2 Forest plots. A and B: Clinical success and mortality based on single-group intervention studies (A: Clinical success; B: Mortality); C and D: Rebleeding based on single-group intervention studies (C: Fixed-effects model; D: Random-effects model); E and F: Reinterventions based on single-group intervention studies (E: Embolization-related interventional methods category: Transcatheter embolization/embolization procedures; F: Non-embolization-dominated interventional methods category: Over-the-scope clip/endoscopic treatment/transcatheter arterial embolization); G and H: Successful initial hemostasis and mortality (G: Successful initial hemostasis; H: Mortality); I and J: Rebleeding based on randomized controlled trials and retrospective cohort studies (I: Fixed-effects model; J: Random-effects model).

Analysis after switching to a random-effects model showed that the intra-subgroup analysis results were consistent with those of the aforementioned fixed-effects model, while the overall RD between subgroups was inconsistent with that of the fixed-effects model. When there is significant inter-subgroup heterogeneity (heterogeneity between embolization and non-embolization interventional methods), the fixed-effects model may produce bias due to the assumption of no substantial heterogeneity between studies. In contrast, the random-effects model better accounts for variations between studies; the stability and reliability of the results are more affected by heterogeneity, which also suggests that there are true differences in the effects of different interventional methods (Figure 2F).

RCTs and retrospective cohort studies: (1) Successful initial hemostasis and mortality: Two studies[13,15] reported the rate of successful initial hemostasis. There was no significant difference in the overall rate of successful initial hemostasis between the endoscopic hemostasis combined with vascular interventional therapy group and the standard treatment group (Z = 0.04, P = 0.97) (Figure 2G). Four retrospective cohort studies[10,13,17,25] reported that there was no significant difference in the overall mortality between the endoscopic hemostasis combined with vascular interventional therapy group and the standard treatment group (Z = 1.56, P = 0.12) (Figure 2H); (2) Rebleeding: Three RCTs[8,9,18] and one retrospective cohort study[17] reported that the overall rebleeding rate in the endoscopic hemostasis combined with vascular interventional therapy group was lower than that in the standard treatment group (Z = 3.26, P = 0.001). After switching to the random-effects model, the result was consistent with that of the aforementioned fixed-effects model (Z = 2.95, P = 0.03) (Figure 2I and J). This suggests that the regimen of endoscopic hemostasis combined with vascular interventional therapy has more advantages in reducing rebleeding and may be a better clinical treatment option; the study results are less affected by heterogeneity and, thus, relatively robust.

Publication bias assessment

Publication bias assessment was performed for the extensively reported clinical success and mortality indicators from single-group intervention studies, as well as successful initial hemostasis and mortality indicators from RCTs or retrospective cohort studies. Funnel plots were used for visualization, and the results showed that the funnel plots of clinical success and mortality from single-group intervention studies, along with successful initial hemostasis and mortality from RCTs or retrospective cohort studies, were basically symmetrical (Figure 3). This indicates that there was no significant publication bias among the included studies, and the results are reliable.

Figure 3 Funnel plots for publication bias assessment. A and B: Represent clinical success and mortality indicators from single-group intervention studies, respectively; C and D: Represent successful initial hemostasis and mortality from randomized controlled trials or retrospective cohort studies, respectively. SE: Standard error; RD: Risk difference; OR: Odds ratio.

DISCUSSION

Initial hemostasis success rate and clinical success rate

This study demonstrated no significant difference in the initial hemostasis success rate between combined endoscopic-vascular interventional therapy and standard treatment, whereas the overall clinical success rate reached 70% (95%CI: 65%-75%). This discrepancy may be attributed to the definition of “clinical success”, which encompasses long-term efficacy (e.g., absence of rebleeding within 48-72 hours post-procedure), whereas initial hemostasis only reflects immediate effectiveness. A potential mechanism is that vascular interventional therapy (e.g., transarterial embolization) plays a critical role in controlling persistent or recurrent bleeding, particularly for bleeding sites inaccessible to endoscopy (e.g., branches of the short gastric artery) or patients with coagulopathy. Data from one study[29] showed that combined therapy achieved higher clinical success rates for Dieulafoy lesions than endoscopic therapy alone, supporting the complementary value of vascular intervention in complex bleeding cases. Heterogeneity in clinical success rates across the included studies may stem from differences in intervention timing; early combined therapy (within 24 hours of bleeding) appears superior to delayed intervention (> 48 hours)[30].

Study design-dependent variations in rebleeding rates

Single-group intervention studies showed that the rebleeding rate of the combined therapy group (endoscopic hemostasis combined with vascular interventional therapy) was 22%, and high heterogeneity might be mainly attributed to age. In RCTs and retrospective cohort studies, there was a significant difference in the rebleeding rate between the combined therapy and standard treatment groups (the former had a lower rebleeding rate). The lack of a control group in single-group studies may overestimate complication rates, as these studies often include patients with more complex conditions (e.g., advanced age and multi-organ failure). In controlled studies, the consistent finding of reduced rebleeding risk with combined therapy may be attributed to precise patient selection (e.g., screening high-risk patients based on the Forrest classification) and technological advancements (e.g., novel hemostatic clips combined with coil embolization)[31].

Heterogeneity and reliability of mortality outcomes

Single-group intervention studies reported a mortality rate of 24% (95%CI: 21%-28%), whereas retrospective cohort studies showed no significant difference between combined therapy and standard treatment. This discrepancy may arise from a case-mix bias in single-group studies (e.g., inclusion of more patients with end-stage liver disease or cancer), whereas cohort studies minimize confounding through baseline matching[32]. Mortality, a difficult endpoint, requires high-quality evidence to reach robust conclusions. In this study, only four cohort studies reported mortality data with small sample sizes, potentially leading to insufficient statistical power.

Prognostic indicators: Reinterventions

In the combined therapy group, the reintervention rate was significantly higher in the subgroup using embolization-related interventional methods than in the control group; in contrast, the reintervention rate was significantly lower in the subgroup adopting non-embolization-dominated interventional methods than in the control group. These findings indicate that the reintervention rate of combined therapy is closely associated with the application of embolization-related interventional methods - embolization-related interventions tend to increase the reintervention rate, while non-embolization-dominated interventions contribute to a reduction in this rate.

From the perspective of clinical practice and technical characteristics, the high reintervention rate of embolization-related interventions can also be attributed to their “non-reparative” effect on lesions. Specifically, these interventions only block blood flow without addressing the primary pathological lesions, such as ulcers or vascular malformations. When embolic materials (e.g., gelatin sponges) are absorbed or collateral circulation is established as a compensatory mechanism, recurrent bleeding is likely to occur, necessitating secondary interventions[33]. Additionally, embolization may damage surrounding normal blood vessels, leading to local tissue ischemia and necrosis, which further induces new bleeding or perforation and consequently increases the need for reinterventions.

On the other hand, the low reintervention rate of non-embolization-dominated interventions (e.g., OTSC) lies in their “precise lesion repair” capability. For instance, OTSCs directly occlude bleeding points, and tissue adhesives seal vascular breaches, thereby eliminating the causes of bleeding at the source[2]. Meanwhile, these interventions are confined to the mucosal layer, without affecting the main vascular trunks or collateral circulation, and thus, do not cause ischemic complications. The stable postoperative recovery process significantly reduces the probability of reinterventions due to complications or inadequate hemostasis, further highlighting the decisive role of intervention method selection in determining the reintervention rate.

Underlying mechanisms and unresolved questions

Immediate hemostasis and precise embolization: Endoscopic injection of epinephrine or placement of hemostatic clips under direct vision controls visible bleeding, whereas angiography localizes microvascular injuries (e.g., arterial branches < 2 mm in diameter) for superselective embolization using microcoils or gelatin sponges, minimizing damage to normal tissue[34].

Reducing hemodynamic instability risk: In patients with hypotension (systolic blood pressure < 90 mmHg), combination therapy shortens the hemostasis time and mitigates coagulopathy caused by repeated hemostasis failures[35]. Despite existing research exploring the clinical efficacy and safety of endoscopic and vascular interventional therapies for acute NVUGIB, several critical unresolved questions remain.

Optimal treatment sequence for bleeding at different anatomical sites (e.g., gastric fundus and duodenum): Universal evidence across subtypes is lacking.

Biomarker-based treatment stratification guidelines: Guidelines based on biomarkers such as platelet count and hemoglobin decline rate have not been established. The precise value of biomarkers in treatment selection - including threshold definitions and dynamic monitoring - remains unclear.

Treatment response and long-term prognosis in patients with low-risk forrest grade III: Differences in treatment response (e.g., necessity of combined therapy and preference for single-modality treatment) and their associations with long-term outcomes require further validation and refinement through large-sample studies.

Study limitations

Heterogeneity and bias risks: Despite efforts to control heterogeneity through sensitivity and subgroup analyses, significant heterogeneity persisted in certain outcomes (e.g., mortality), likely stemming from differences in study design (single-group vs controlled studies) and variability in interventions (different endoscopic techniques and interventional materials).

Insufficient high-quality evidence: Only three RCTs with small sample sizes were included, leading to inadequate statistical power for the subgroup analyses.

Lack of long-term follow-up data: Over 80% of studies had a follow-up period ≤ 30 days, precluding assessment of long-term outcomes (e.g., 1-year rebleeding rate, quality of life).

Potential impact of publication bias: Although Begg’s and Egger’s tests did not indicate a significant bias, the notable change in results after excluding studies at the funnel plot margins suggests a possible undetected small-sample bias.

Implications for individualized treatment: For patients with high risk (e.g., Forrest Ia and hemodynamic instability), early combined therapy may improve clinical success rates. For patients with low risk (e.g., Forrest IIc-III), primary endoscopic hemostasis is recommended to avoid overtreatment.

Importance of multidisciplinary collaboration: Establishing hemorrhage management teams comprising gastroenterologists, interventional radiologists, and intensivists to develop standardized protocols and reduce decision-making time is critical.

Future research directions: Large-scale multicenter RCTs are urgently required to address this issue: (1) Comparison between combined therapy and endoscopic/interventional monotherapy; (2) Development of artificial intelligence-based risk prediction models to guide treatment selection; and (3) Safety and efficacy of novel bioabsorbable embolization materials.

CONCLUSION

This meta-analysis demonstrated that combined endoscopic hemostasis and vascular interventional therapy for acute NVUGIB achieved high clinical success rates and reduced the risk of rebleeding compared with standard treatment. However, mortality outcomes are significantly influenced by study design, which limits their reliability. Current evidence supports the use of combination therapy in patients at high risk of bleeding; however, bias risks in single-group studies require cautious interpretation. Future research should include more high-quality RCTs and long-term follow-up data to clarify optimal treatment strategies and target populations.