WHAT IS ALREADY KNOWN ON THIS TOPIC

WHAT THIS STUDY ADDS

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

We conducted a randomised controlled unblinded trial at 15 academic and non-academic referral hospitals in Germany (online supplemental figure S1). The primary objective of this trial was to determine the impact of prophylactic application of a haemostatic gel to the resection field following EMR of large flat polyps in the duodenum or colorectum on delayed bleeding rate.

Patients with large flat polyps of the duodenum (largest diameter ≥10 mm) or colorectum (largest diameter ≥20 mm) and an indication for hot snare EMR according to current guidelines3 18 were eligible to participate in this study. Exclusion criteria included dual antiplatelet therapy (ie, two antiplatelet agents taken simultaneously), antiplatelet therapy in combination with an inhibitor of plasmatic coagulation, significant pre-existing coagulopathy and planned resection with a technique other than hot snare EMR such as cold snare EMR, endoscopic submucosal dissection (ESD) or full thickness resection. Of note, patients taking either a single antiplatelet agent or a single inhibitor of plasmatic coagulation were allowed to be included in the study. Online supplemental table S1 shows full inclusion and exclusion criteria. Patients with multiple polyps were allowed to participate if the total cumulative of the largest diameters of all lesions resected did not exceed 100 mm. Randomisation, evaluation of the primary endpoint and most other analyses were carried out on a per-patient level. Per polyp analyses are provided in addition and are indicated as such.

Hot snare EMR including peri-interventional management of anticoagulant medications was performed according to the centres’ standard procedure. To treat intraprocedural bleeding, all haemostatic tools except for topical haemostatic agents were allowed. If use of such an agent was deemed necessary to address intraprocedural bleeding, the patient was excluded from the trial prior to randomisation. Within the trial, the use of haemostatic clips was regulated as follows: ‘clip closure’, that is, the use of clips in order to adapt the margins of the resection field thereby partially or fully closing it was not allowed. If partial or full clip closure of the resection field was deemed necessary for any reason, the patient was excluded from the trial prior to randomisation. ‘Selective clipping’, that is, the application of one or more clips within the resection field without moving the margins of the resection field was allowed, but only prior to randomisation and only if the resection field remained open (ie, no adaption of its margins). Specifically, selective clipping prior to randomisation could be performed in order (1) to treat intraprocedural bleeding, (2) to treat intraprocedural deep mural injury or perforation and (3) to prevent delayed bleeding (eg, by selectively clipping visible vessels within the resection field). All of these measures were documented in the case report form. Again, if clipping resulted in partial or complete closure of the resection field, the patient was excluded from the trial prior to randomisation. Thus, only patients where the resection field remained open were randomised and subsequently included in the analysis. Like selective clipping, coagulation within the resection field could be applied to treat intraprocedural bleeding or to seal non-bleeding visible vessels within the resection field; this also had to be documented in the case report form. Margin coagulation in order to prevent adenoma recurrence was allowed but not obligatory in the trial. When margin coagulation was applied, it had to be done prior to randomisation and to be documented in the case report form. After randomisation, patients allocated to the gel group had haemostatic gel applied to the resection field; apart from this study intervention, no other measures to prevent bleeding were allowed after randomisation.

After complete resection, randomisation was performed using opaque envelopes and patients were thus allocated to the control (no prophylaxis) or intervention (haemostatic gel) group. The group assignment was done by a computer-generated randomisation list with stratification for colorectal and duodenal polyp localisation. Each patient was randomised only once. In the intervention group, the haemostatic gel was applied according to the manufacturer’s instructions aiming to cover the entire resection field with gel (online supplemental figure S2). In patients with multiple lesions allocated to the intervention group, it was mandatory to apply gel to the resection fields of all lesions larger than 5 mm. There were two scheduled follow-up visits: (1) a telephone interview 30±2 days after the study procedure and (2) a repeat endoscopy 63±7 days after the study procedure. The timing of the repeat endoscopy was chosen in order to conform to the German guideline for follow-up after piecemeal resection of colorectal lesions.

The primary outcome of this trial was the rate of CSDB from the resection site within 30 days after hot snare EMR. CSDB was defined as clinical evidence of delayed bleeding directly related to the EMR procedure and resulting in one or more of the following: (1) transfusion of blood products, (2) unplanned endoscopy, (3) radiological intervention, (4) surgical intervention, (5) prolongation of hospital stay, (6) emergency room visit or (7) admission to hospital care. In a prespecified subgroup analysis, we investigated the possible effect of polyp location (ie, proximal vs distal colon; colon vs duodenum), anticoagulant medication, size of the resection field and pre-existing conditions that predispose to bleeding. A per patient analysis was carried out to assess the primary outcome. Even though patients with multiple polyps were permitted, we opted against a per polyp analysis for the primary outcome because this would have required re-endoscopy after every bleeding event to identify the culprit lesion.

Secondary outcomes included adverse events other than delayed bleeding, wound healing and rate of residual/recurrent polyps at the time of follow-up endoscopy. For the evaluation of wound healing, the Sakita-Miwa Classification was used (online supplemental table S2);19 briefly, this classification divides ulcer healing into three stages (active, healing and scarring) with two substages per stage thus translating the wound healing process into categorical data.

For the power analysis, we assumed 30% duodenal and 70% colorectal polyps with expected CSDB rates of 18% and 8% without gel prophylaxis, respectively. Randomisation was stratified to guarantee a balanced allocation of duodenal and colorectal lesions to the two groups, but we did not prespecify that 30% duodenal and 70% colorectal must be reached nor did we implement a recruiting mechanism to guarantee this. Assuming that gel prophylaxis would be as effective as clip closure,4 7 13 a reduction to 2% significant postprocedural bleeding in the haemostatic gel arm was assumed.4 14 As outlined above, these assumptions applied to a population where ‘selective clipping’ within the resection field as deemed necessary was allowed prior to randomisation, but ‘clip closure’ (ie, adaption of the resection margins) was not allowed.

We planned for the inclusion of 248 patients, randomised 1:1 to haemostatic gel prophylaxis (intervention) versus no haemostatic gel prophylaxis (control). This was intended to detect differences between the treatment groups with a power of 80% at a significance level of 5% using a two-sided Fisher’s exact test.

There were significant uncertainties regarding the expected bleeding rates and the number of lesions per patient. Furthermore, the rate of recruitment of patients with lesions in sections of the bowel with higher or lower bleeding risk over the planned recruitment phase was difficult to predict. Thus, we planned an interim analysis after 200 patients had reached the 3-month follow-up. We chose this late time point assuming a relatively low event rate and given that the effect size to be expected was difficult to estimate given the limited data available. Depending on the observed effect size (ie, difference in postprocedural bleeding rates between the groups at the time of the interim analysis), the study protocol allowed for (1) discontinuation of the study (if effect size non-existent or larger than expected), (2) upwards adjustment of the number of patients to be included, that is, including more cases than initially planned (if effect present but effect size smaller than expected) or (3) continuation of the study as planned (effect size as expected; inclusion of planned case number). Of note, this does not correspond to a statistically based formal stopping rule,20 but represented a prespecified pragmatic approach where the observed effect size dictated how to proceed with the trial. Alternatively, group sequential designs with interim analyses could have been considered. The sequential probability ratio test could have been used to examine futility and early proof of efficacy. However, the recruitment of patients lesions in intestinal segments with a low risk of bleeding during the course of the study could have impaired the consistency of the results, so we opted against it. No toxicities have been observed with haemostatic gel tested here during years of clinical application, so the risk of exposing patients to toxicity was considered very low in this trial. For this reason, we considered a non-rigorous stopping rule acceptable.

Data were analysed with R Core Team 2024. Categorical data are expressed as count (percentage) and continuous data are expressed as medians (quartiles) and means (SD). The primary objective was examined using a two-sided Fisher’s exact test. A logistic regression model was applied, linking both treatment group and localisation to delayed bleeding. Delayed bleeding for selected risk factors was examined using the difference in proportions (95% CIs). Furthermore, bivariable logistic regression models linking each of the risk factors and the treatment group to CSDB were calculated. A p<0.05 was considered significant.

Patients or the public were not directly involved in this study, including but not limited to the trial design, patient recruitment, conduct of the trial, analysis of data and manuscript preparation.

Between May 2022 and January 2024, 234 patients were randomised into the control (n=114) or haemostatic gel (n=120) group. At this time, the recruitment was stopped due to futility based on the results of a prespecified interim analysis. Thus, the study was terminated early, but the number of cases randomised reached 94.4% (234 out of 248) of the initially planned number.

The primary endpoint could be evaluated in 232 out of 234 cases (online supplemental figure S3). Two patients were excluded due to death from a cardiac cause shortly after the procedure (n=1) and small polyp size (n=1). Baseline characteristics were comparable between the groups (table 1, online supplemental table S3 and S4). When only main lesions were considered, there was no size difference in the colorectum (median diameter 30 mm for both gel and control group); the duodenal lesions were somewhat larger in the gel group (median diameter 20 mm vs 15 mm), but this was not significant (table 2).

Histopathology showed about half of the lesions resected to be low-grade adenomas. Malignancy was detected in 2.2% of lesions resected and 3.4% of patients included in the final analysis. Procedural information is shown in online supplemental table S5. In total, 368 lesions were resected (178 in the haemostatic gel and 190 in the control group). Polyp characteristics of all lesions resected can be found in online supplemental table S6 and S7.

Selective clipping and coagulation (snare tip, coagulation forceps or argon beamer) were allowed during the procedure to treat intraprocedural bleeding and after completed resection but before randomisation to prevent bleeding from visible vessels. Selective clipping was also allowed to treat deep mural injury or perforation. Any clips and/or coagulation methods for bleeding prevention prior to randomisation were used in 51.7% of patients in the haemostatic gel group and 52.2% of patients in the control group (table 3). 57.6% of patients (55.8% in colorectal and 73.1% in duodenal) did not obtain focal clipping prior to randomisation.

In the intervention group, a mean of 3.3±1.4 mL of haemostatic gel was applied. In two cases, two packages (5 mL) of haemostatic gel were used. Haemostatic gel application was technically successful in all (100%) of cases.

CSDB occurred in 14 cases (11.7%; 95% CI 7.1% to 18.6%) in the haemostatic gel group and 7 cases (6.3%; 95% CI 3.1% to 12.3%) in the control group (table 4; figure 1A). This difference was not statistically significant (p=0.277). In the subgroup with colorectal lesions, the delayed bleeding rates were 9.3% (95% CI 5.2% to 16.4%) and 5.1% (95% CI 2.2% to 11.3%) in the haemostatic gel and control group, respectively. In the subgroup with duodenal lesions, the CSDB rates were 30.8% (95% CI 12.7% to 57.6%) and 15.4% (95% CI 4.3% to 42.2%) in the haemostatic gel and control group, respectively. In the post hoc analysis of patients where no selective clipping prior to randomisation was performed for any reason, there were also numerically more bleeding events in the gel group, but again the difference was not statistically significant (table 5).

In the overall cohort, the median time between the intervention and the diagnosis of bleeding was comparable between the groups (2 vs 3 days, respectively) with half of bleeding events occurring within 48 hours after the procedure (figure 1B). Of note, in the subgroup of colorectal lesions, the median time until postprocedural bleeding was 1 day (range 0–7 days) in the control group compared with 5 days (range 1–13 days) in the gel group. Subgroup analysis evaluating additional factors potentially associated with risk of postprocedural bleeding did not show any subgroup where gel application might be advantageous (online supplemental figure S4)

There were no statistically significant differences in terms of prespecified secondary outcomes (transfusion requirement, unplanned endoscopy, readmission to hospital care) between the groups. An endoscopic assessment of wound healing conforming to the prespecified time window was available for 290 lesions in 190 patients. The majority of resection fields were fully healed at this time point and there was no statistically significant difference between the groups (online supplemental table S8).

Besides the main outcome events for delayed bleeding detailed above, there was one duodenal perforation in the gel group that, after a protracted course eventually resulted in the death of the patient. One patient in the control group died of a cardiac event within 24 hours of the intervention; this was considered potentially related to the study procedure. There were three instances of postpolypectomy syndrome (one in the gel groups and two in the control group). There were no additional adverse events considered related to the study procedures, and no adverse events were related to the haemostatic gel used. Of note, three more patients died during the study period from causes deemed unrelated to the study procedure. This included progression of a pre-existing malignant disease (n=1), newly diagnosed malignant disease (n=1) and trauma resulting from a fall 27 days after the trial procedure (n=1).

Not applicable.

The clinical study protocol was approved by the Ethics Committee (EC) of the Hamburg chamber of physicians (registration number 2021-100711_1-BO-ff) as well as at the local ECs of participating centres. Further, the study was registered at the Deutsches Register Klinischer Studien (Study ID: DRKS00028119). All patients participating in this clinical trial signed the informed consent prior to randomisation.