Introduction
Endoscopic submucosal dissection (ESD) allows en bloc resection of large superficial colorectal lesions. ESD is the gold standard in Japan for this indication and is used increasingly frequently in the Western world because of its advantages in term of recurrence and oncological outcomes, compared with piecemeal endoscopic mucosal resection (EMR).
Clinically significant delayed bleeding (CSDB) is one of the most frequent complications after colorectal ESD (incidence 1%–9% [1 ]
[2 ]
[3 ]
[4 ]). Delayed bleeding increases the length of stay or induces new hospitalization, and can require blood transfusion or new hemostatic colonoscopy, thereby increasing costs. Few data are available on risk prediction and risk factors of post-ESD delayed bleeding; those that are available are from observational Asian studies. A large Korean study developed a risk prediction model based on three risk factors: use of antiplatelet agents, tumor size >30 mm, and location in the rectosigmoid area [5 ]. This enabled differentiation of a low risk group (1.9% bleeding rate) and a high risk group (4.9%). However, the study population did not reflect Western populations, with small lesions (median 28 mm), low rate of anticoagulant use (3%), and few comorbidities (9.2% American Society of Anesthesiologists [ASA] class III or IV) [5 ]. Moreover, planned hybrid ESD was performed in 27.7% of the cases.
We performed a large Western cohort study to identify risk factors of post-colorectal ESD bleeding, tested the Korean model in a Western population, and developed a new scoring system to accurately predict bleeding after colorectal ESD.
Methods
Study design
We performed a post hoc analysis of a single-center prospectively maintained consecutive database (Institutional Review Board approval #87RI20–0021_FECCo). The Ethics Committee of Limoges University Hospital approved the study, and a No Opposition to Data Use form (mandated by French legislation on RIPH3-type studies) was sent to all patients, who then provided written informed consent for the procedures.
The patients were contacted (face to face or by telephone) after 1 month for histopathological results and verification of complications (including post-ESD delayed bleeding). Details of the CSDB (date of bleeding, characteristics of the bleeding, clinical outcome of the bleeding) were collected retrospectively.
Definitions
Laterally spreading tumors (LSTs) were defined according to the Japanese definition as lesions larger than 10 mm with width greater than height and as granular or nongranular according to their morphology. Lesions that were larger than 10 mm (except sessile serrated lesions) with height greater than width were defined as protruding lesions.
CSDB was defined as a post-ESD bleeding event (hematochezia) needing a prolongation of hospitalization, readmission, new endoscopy (or surgery or angiography), or a blood transfusion, and occurring at least 6 hours after, and within 30 days after the procedure.
Periprocedural perforation was defined as exposure of the peritoneal space as a result of a muscular defect (Sidney classification IV–V [6 ]). R0 resection (histologically complete) was defined as neoplasia-free vertical and lateral resection margins on histology examination. Curative resection was defined according to the European Society of Gastrointestinal Endoscopy (ESGE) guidelines [7 ] as R0 resection and the absence of high risk features on pathology (submucosal invasion >1000 µm, lymphovascular invasion, poor differentiation, significant budding [8 ]) and requiring secondary surgery.
Inclusion criteria
All epithelial colonic lesions treated by ESD were included prospectively and consecutively from January 2013 to March 2022. We stopped inclusion in March 2022 because our center was involved in a multicenter randomized trial of the efficacy of a self-assembling peptide in preventing post-colorectal ESD delayed bleeding (NCT05031325). All adult patients (≥18 years)were included from the time at which ESD was initiated, irrespective of whether ESD was discontinued because of technical difficulty.
Exclusion criteria
Lesions exhibiting signs of deep submucosal carcinoma by optical examination (Japan Narrow-band imaging Expert Team [JNET] III, Sano IIIB, Kudo Vn, Colorectal Neoplasia Classification to Choose the Treatment [CONECCT] III) were excluded. Submucosal lesions, neuroendocrine tumors, and non-neoplastic lesions were also excluded. Patients with two lesions resected by ESD during the same colonoscopy were excluded because of the difficulty in identifying the lesion responsible for the delayed bleeding.
ESD procedure
In our unit we have performed ESD for all large (>25 mm) LSTs of the rectum since 2013 and for the colon since 2016.
ESD procedures were performed by five operators (J.J., R.L., M.D., J.A., and H.L.), all of whom had extensive experience of ESD in animal models prior to study commencement and who were trained in France.
Single-channel high-resolution endoscopes were used for ESD, with patients under general anesthesia. After detailed optical diagnosis confirming the resectability of the lesion, a complete or partial incision was performed using an endoscopic knife. The strategy for submucosal dissection was at the discretion of the operators, but our team has developed and frequently uses a double-clip traction strategy with clips and rubber bands.
VIO200D and VIO 3 (since 2018) electrosurgical units (Erbe Elektromedizin, Tübingen, Germany) were used. EndoCUT I was used for mucosal incision, and swiftCOAG, preciseSECT, or endoCUT I was used for submucosal dissection. At the end of the procedure, visible vessels were systematically prophylactically coagulated using monopolar coagulation forceps. We did not include adrenalin in the injection solution and did not routinely close mucosal defects after ESD.
Owing to geographical constraints, many patients were hospitalized the day before the procedure. With the exception of patients with small lesions who lived less than 20 minutes from the hospital (outpatient treatment), all patients remained in hospital for 1 night after ESD. Only patients who experienced adverse events had prolonged hospitalization of more than 1 night after ESD.
Management of patients on anticoagulants or antiplatelets
Anticoagulants were managed according to the ESGE guidelines [9 ]. Warfarin was stopped 5 days before the procedure and restarted 1 day after the procedure. Heparin bridge therapy was performed only in cases at high risk of thromboembolic events. Direct oral anticoagulants were stopped 2 days before the procedure and restarted 2 days after the procedure. Regarding antiplatelet agents, aspirin was not stopped for the ESD procedures. All other antiplatelets were stopped 5 days before the procedure and replaced by aspirin monotherapy, which continued for 5 days after the procedure. On Day 6, aspirin was replaced with the previous antiplatelet agent.
Statistical analysis
The characteristics of patients were described using descriptive statistics, with frequency (percentage) for categorical variables, and mean (SD) or median (interquartile range) for continuous variables, depending on their distribution. The chi-squared test (or Fisher’s exact test) was used for between-group comparisons, and a value of P <0.05 was taken to indicate statistical significance.
The incidence of CSDB was estimated using sample proportions with 95%CIs. We evaluated the performance of an Asian risk-scoring model for the prediction of delayed bleeding after colorectal ESD [5 ]. Model discrimination was described by the area under the receiver operating characteristic (ROC) curve (AUC) with 95%CIs. Univariate logistic regression models were used to analyze a set of variables considered to be related to delayed bleeding based on clinical criteria (sex, age, polyp size, ASA, polyp location, anticoagulant or antiplatelet use, en bloc resection, clip closure, chronic kidney disease, and high blood pressure); those variables with P <0.2 were included in a multivariate model and only those with P <0.2 in the multivariate model were retained to develop the CSDB risk model. The risk model odds ratios (ORs) and 95%CIs are presented.
We assessed model calibration using the Hosmer–Lemeshow goodness-of-fit test, together with calibration plots of predicted risk against actual risk for each decile. The model was validated internally using the resampling validation method for logistic models with 250 bootstrap re-samples, and the optimism-corrected ROC and discrimination slope (difference in mean of predictions between patients with or without CSDB) with their 95%CIs were estimated. Complementarily, leave-one-out cross-validation was performed to calculate the accuracy of the model (measured as the percentage of corrected classified patients) when the probability threshold was the observed incidence of CSDB. We developed a risk score by assigning a weight to each predictor based on the β parameter from the multivariate final model. These β coefficients were divided by the smallest β coefficient and rounded to the nearest integer as described previously [10 ]
[11 ]
[12 ]
[13 ]. We divided the 9-point score into two risk categories for simple interpretation in a clinical setting using the Youden index or the incidence of CSDB (giving similar results). The weighted averaged probabilities of CSDB were estimated for each risk group. Finally, a decision curve was plotted relating the net benefit of using the score compared with the treat-all and treat-none approaches for a range of plausible risk thresholds.
Statistical analysis was performed using IBM Statistical Package for the Social Sciences (SPSS) v. 25.0 (IBM Corp., Armonk, New York, USA) and R v. 4.0.0 (R Foundation for Statistical Computing, Vienna, Austria). This study complies with the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis Or Diagnosis guidelines (see the online-only Supplementary material) [14 ].
Results
General characteristics of the patients and lesions
A total of 940 ESDs of large colorectal lesions were performed consecutively in 940 patients ([Fig. 1 ], [Table 1 ]). There were 514 male patients (54.7%), and the mean age of the cohort was 69 years (SD 10.1 years; range 63–75 years). The ASA score was III or IV in 340 patients (36.2%). A total of 267 patients (28.4%) had some form of anticoagulant/antiplatelet therapy prescribed. The mean lesion size was 60.6 (SD 25.4) mm (range 45–70 mm). Lesion location was the rectum in 369 patients (39.3%) and the proximal colon in 339 patients (36.1%).
Fig. 1 Flow chart of the study. EMR, endoscopic mucosal resection; ESD, endoscopic submucosal dissection; JNET, Japan Narrow-band imaging Expert Team; NET, neuroendocrine tumor.
Table 1 Patient characteristics.
Characteristic
Total1
No CSDB1
CSDB1
P value2
CSDB, clinically significant delayed bleeding; ASA, American Society of Anesthesiologists; LST, laterally spreading tumor; LGD, low grade dysplasia; HGD, high grade dysplasia;
1 Information in the column “Total” is presented as frequencies (%) by column whereas, information in “No CSDB” and “CSDB” columns is presented as frequencies (%) by row.2
P <0.05 was considered significant.3 Fisher’s exact test.
Patients, n (%)
940
865 (92.0)
75 (8.0)
Demographic characteristics
Sex, n (%)
514 (54.7)
474 (92.2)
40 (7.8)
0.81
426 (45.3)
391 (91.8)
35 (8.2)
Age, mean (SD), years
69 (10)
69 (10)
73 (10)
<0.001
668 (71.1)
630 (94.3)
38 (5.7)
<0.001
272 (28.9)
235 (86.4)
37 (13.6)
Clinical characteristics
ASA classification, n (%)
600 (63.8)
572 (95.3)
28 (4.7)
<0.001
340 (36.2)
293 (86.2)
47 (13.8)
Antiplatelet/anticoagulant therapy, n (%)
267 (28.4)
231 (86.5)
36 (13.5)
<0.001
97 (10.3)
76 (78.4)
21 (21.6)
<0.001
182 (19.4)
161 (88.5)
21 (11.5)
0.07
59 (6.3)
50 (84.7)
9 (15.3)
0.043
458 (48.7)
411 (89.7)
47 (10.3)
0.02
Location, n (%)
369 (39.3)
331 (89.7)
38 (10.3)
0.07
165 (17.6)
153 (92.7)
12 (7.3)
67 (7.1)
66 (98.5)
1 (1.5)
339 (36.1)
315 (92.9)
24 (7.1)
Lesion size, mean (SD), mm
61 (25)
59 (23)
82 (37)
<0.001
412 (43.8)
399 (96.8)
13 (3.2)
<0.001
528 (56.2)
466 (88.3)
62 (11.7)
Type of lesion, n (%)
138 (14.7)
130 (94.2)
8 (5.8)
0.02
613 (65.2)
552 (90.0)
61 (10.0)
167 (17.8)
161 (96.4)
6 (3.6)
22 (2.3)
22 (100)
0 (0)
Pathological analysis, n (%)
1 (0.1)
1 (100)
0 (0)
0.87
409 (43.5)
382 (93.4)
27 (6.6)
225 (23.9)
202 (89.8)
23 (10.2)
187 (19.9)
171 (91.4)
16 (8.6)
42 (4.5)
38 (90.5)
4 (9.5)
43 (4.6)
40 (93.0)
3 (7.0)
13 (1.4)
12 (92.3)
1 (7.7)
20 (2.1)
19 (95.0)
1 (5.0)
ESD characteristics
The resection was en bloc in 899 patients (95.6%), R0 in 798 (84.9%), and curative in 755 (80.3%) ([Table 2 ]). Lateral margins were free of adenoma for 820 lesions (87.2%) and deep margins were free of cancer in 899 lesions (95.6%). The mean duration of the procedure was 85.8 (SD 71.5) minutes (range 6–570 min). The mean speed of the procedure was 35.7 (SD 19) mm2 /min. After ESD, 111 defects (11.8%) were closed completely with clips.
Table 2 Procedure characteristics.
Characteristic
Total1
No CSDB1
CSDB1
P value2
CSDB, clinically significant delayed bleeding; ESD, endoscopic submucosal dissection.
1 Information in the column “Total” is presented as frequencies (%) by column whereas, information in “No CSDB” and “CSDB” columns is presented as frequencies (%) by row.2
P <0.05 was considered significant.3 Fisher’s exact test.4 Data available for 916 patients (97.4%).
ESD characteristics
Resection, n (%)
899 (95.6)
828 (92.1)
71 (7.9)
0.563
798 (84.9)
737 (92.4)
61 (7.6)
0.47
755 (80.3)
698 (92.5)
57 (7.5)
0.41
86 (71)
82 (66)
133 (109)
<0.001
111 (11.8)
105 (94.6)
6 (5.4)
0.38
Adverse events, n (%)
Perforation
845 (89.9)
781 (92.4)
64 (7.6)
0.24
95 (10.1)
84 (88.4)
11 (11.6)
12 (1.3)
1 (8.3)
11 (91.7)
<0.0013
40 (4.3)
0 (0)
40 (100)
<0.0013
19 (2.0)
0 (0)
19 (100)
<0.0013
Secondary surgery and reason
72 (7.7)
68 (94.4)
4 (5.6)
0.57
54 (5.7)
50 (92.6)
4 (7.4)
0.80
13 (1.4)
13 (100)
0 (0)
4 (0.4)
4 (100)
0 (0)
1 (0.1)
1 (100)
0 (0)
Adverse events
CSDB occurred in 75 patients (8.0%; 95%CI 6.36–9.95) and perforation in 95 patients (10.1%). A blood transfusion was necessary in 12 patients (1.3%). Significant bleeding led to prolonged hospitalization in 19 patients (2.0%) and new admission in 54 patients (5.7%). Overall, 40 patients (4.3%) required new endoscopic treatment for hemostasis, and 72 patients (7.7%) needed secondary surgery (4 [0.4%] for post-procedural perforation, 1 [0.1%] for appendicitis, 13 [1.4%] for ESD failure, and 54 [5.7%] for high risk noncurative adenocarcinoma) ([Table 2 ]). One patient (0.1%) died 1 day post-procedure of a sudden unexplained cardiac arrest while returning home.
Performance of the Asian model
The scores of the independent variables according to the Asian predictive model are described in Seo et al. [5 ]. Table 1s shows a univariate comparison using the Asian model CSDB predictors in the current Limoges cohort. Two of the CSDB predictors of the Asian model (tumor size and antiplatelet agents) were not associated with CSDB in the Limoges cohort. When the Asian score model was applied to the Limoges cohort, an AUC of 0.567 (95%CI 0.508–0.626) was obtained, suggesting low discriminative ability (AUC <0.70).
Development and performance of our model
[Table 3 ] shows univariate and multivariate analyses of factors independently predictive of CSDB after ESD. The factors found to be predictive in the final model ([Table 4 ]) were age ≥75 years (OR 1.63; 95%CI 0.97–2.73; P =0.06), antithrombotic agent (anticoagulant or antiplatelet) use during ESD (OR 1.72; 95%CI 1.01–2.94; P =0.046), location in the rectum (OR 1.51; 95%CI 0.92–2.48; P =0.10), ASA class III or IV (OR 2.26; 95%CI 1.32–3.92; P =0.003), and lesion size >50 mm (OR 3.67; 95%CI 2.02–7.14; P <0.001).
Table 3 Univariate and multivariate analyses of factors independently predictive of clinically significant delayed bleeding after endoscopic submucosal dissection (N=940).
Characteristic
Univariate analysis
Multivariate analysis
OR (95%CI)
P value
OR (95%CI)
P value
OR, odds ratio; ASA, American Society of Anesthesiologists.
Female sex
1.06 (0.66–1.70)
0.81
Age ≥75 years
2.61 (1.62–4.21)
<0.001
1.60 (0.94–2.69)
0.08
Size >50 mm
4.08 (2.29–7.86)
<0.001
3.67 (2.03–7.15)
<0.001
ASA III–IV (vs. ASA I–II)
3.28 (2.02–5.40)
<0.001
2.20 (1.27–3.86)
0.01
Rectal location
1.66 (1.03–2.66)
0.04
1.52 (0.93–2.50)
0.10
Proximal location
0.82 (0.49–1.35)
0.44
Anticoagulant/antiplatelet use
2.53 (1.57–4.09)
<0.001
1.67 (0.96–2.89)
0.07
Chronic kidney disease
2.22 (0.99–4.52)
0.05
1.09 (0.46–2.37)
0.83
High blood pressure
1.85 (1.15–3.05)
0.01
1.12 (0.65–1.95)
0.69
Not en bloc resection
1.26 (0.37–3.26)
0.68
Not complete closure
1.59 (0.73–4.18)
0.26
Table 4 Final clinically significant delayed bleeding score after endoscopic submucosal dissection
Predictors
Delayed bleeding (final model)
Score
OR (95%CI)
P value1
OR, odds ratio.1
P <0.05 was considered significant.
(Intercept)
0.01 (0.01–0.03)
<0.001
Age ≥75 years
1.63 (0.97–2.73)
0.06
1
Size >50 mm
3.67 (2.02–7.14)
<0.001
3
ASA III–IV
2.26 (1.32–3.92)
0.003
2
Location in rectum
1.51 (0.92–2.48)
0.10
1
Anticoagulants/antiplatelet
1.72 (1.01–2.94)
0.046
1
The multivariate model had an AUC of 0.751 (95%CI 0.690–0.812), showing good discrimination (Fig. 1s ). The bias-corrected index after internal bootstrap validation was 0.739 (95%CI 0.671–0.805). Calibration of the model was fair according to calibration curves ([Fig. 2 ]) and to the Hosmer–Lemeshow test results (chi-squared 7.652; P =0.47). The discrimination slope (difference in average predictions for those with and without the outcome) was 0.07 (95%CI 0.04–0.13). Additionally, leave-one-out cross-validation was performed to calculate the accuracy of the model (measured as the percentage of corrected classified patients), obtaining an accuracy equal to 74.2%.
Fig. 2 Calibration curve for the multivariate regression.
The constructed CSDB score ranged from 0 (lowest risk) to 8 (highest risk) (Table 2s ). The formula and illustration of the CSDB score and risk calculator are shown in Table 3s . The probability of CSDB according to the score is shown in [Table 5 ]. The Youden index and the observed incidence of CSDB were used to categorize the score into two risk groups (similar for both approaches), those with low–medium CSDB risk (0 to 4 points), and those with high CSDB risk (5 to 8 points), with respective bleeding rates of 4.1% and 17.5%. Applied to our population, 269 patients (28.6%) were in the high risk category (Table 4s ). The decision curve showed the net benefit, across a range of cutoffs, of using the score compared with the treat-all and treat-none approaches (Fig. 2s ). The net benefit for the 9% delayed bleeding risk threshold calculated using the Youden index (corresponding to a 1:10 harm:benefit ratio) was 0.393. When a 14% risk threshold was used (corresponding to a 1:6 harm:benefit ratio), the net benefit was equal to 0.252. In this more restrictive situation, the high risk group would include only patients with 7–8 points in the score (corresponding to 20% of the sample and a weighted average risk of delayed bleeding equal to 20.3%).
Table 5 Probability of clinically significant delayed bleeding according to score.
Score
Delayed bleeding risk, % (95%CI)
Delayed bleeding risk category, %
0
1.3 (0.7–2.5)
Low–medium risk: 4.1
1
2.0 (1.2–3.5)
2
3.1 (2.0–4.8)
3
4.7 (3.4–6.6)
4
7.2 (5.6–9.3)
5
10.7 (8.6–13.3)
High risk: 17.5
6
15.8 (12.5–19.8)
7
22.7 (17.2–29.3)
8
31.3 (22.6–41.6)
Discussion
CSDB occurred in 8% of patients in this large series of colorectal ESD cases. Rectal location, size >50 mm, ASA score III/IV, treatment with antithrombotic agents, and age >75 years were significant risk factors in a multivariate analysis, and these factors were used to develop a score to predict CSDB. The score was based on pre-procedural data, and the scoring system is easy to use. The risk of CSDB increased as the risk score increased, with a 50% increase in risk for each 1-point increase in score.
A CSDB prediction model has been developed previously in Korea [5 ]. This score was not predictive of CSDB in our cohort (AUC 0.567, compared with AUC 0.751 for our score). Differences in the lesions and patients included in the Asian study are likely to explain the ineffectiveness of this model in our population. Indeed, the median lesion size was 28 mm (60 mm in our cohort), only 2.3% of polyps were <30 mm (none of which bled), only 3% of the population were on anticoagulants (10% in our cohort), and patients had fewer comorbidities (9.2% ASA III/IV vs. 36.2% ASA III/IV in our cohort). Moreover, 27.7% of the lesions were resected in a planned hybrid ESD procedure using a snare, which is more like an EMR than an ESD. This is important because our findings confirm a difference in bleeding risk factors between these two resection strategies – rectal location for ESD and right colon location for EMR [11 ]
[15 ]. This difference of included lesions also explains the higher CSDB rate (8%) in the current study, which is similar to that in our recent prospective randomized trial of ESD and piecemeal EMR for large LSTs of the colon [16 ]. By contrast to observational retrospective studies, Asian randomized studies have reported a higher CSDB rate, similar to that found in the current study [17 ].
In our cohort, patients with CSDB were significantly older, with more comorbidity, and larger lesions more frequently located in the rectum, and were more frequently treated with antithrombotic agents compared with patients without CSDB. However, no difference was identified according to the closure rate between patients with and without CSDB.
Rectal location has previously been described as a risk factor in several Asian studies [1 ]
[5 ]
[18 ]
[19 ]
[20 ]. Right colon location is an important risk factor for post-EMR bleeding [11 ]
[15 ]
[21 ]. This discrepancy between the two techniques is difficult to explain but suggests a different pathophysiology of post-resection bleeding between EMR and ESD. This is important and precludes extrapolation of the post-EMR bleeding risk factors to colorectal ESD procedures.
Lesion size was the most important risk factor, and size is a risk factor for CSDB after both ESD and EMR [1 ]
[5 ]
[11 ]
[15 ]
[18 ]
[21 ], which is likely to be because a larger lesion will leave a larger scar, meaning a greater number of injured submucosal vessels.
Other post-ESD CSDB risk factors include comorbidities (ASA III/IV), anticoagulants, and age. Further larger studies should focus on CSDB risk according to type of anticoagulant (vitamin K antagonist, direct oral anticoagulants) or type of antiplatelet. We were unable to perform such an analysis because separating patients according to anticoagulant or antiplatelet would have weakened the statistical power of the analysis.
The strengths of this study are the use of a prospective database with very few missing data, and in-person or telephone contact to evaluate delayed bleeding. Additionally, the nonuse of new preventive hemostatic strategies (gels or powders) prevented important bias of contamination of our results.
The main limitations of the study are the single-center design, the limited number of cases in the sample (75 CSDB cases), and the absence of external validation of the model. This means that the results need to be independently validated in Western and Asian populations. Moreover, the study was conducted over a long period (10 years); nevertheless, bleeding prevention, risk factors, and incidence have not changed much during this period. Stopping enrollment at the outset of a randomized study evaluating a new hemostatic agent prevented the introduction of bias into the results. Additionally, we did not analyze the importance of periprocedural bleeding as a potential risk factor of CSDB. We also did not analyze the number of times the coagulation forceps were used as a risk factor for bleeding; however, as the aim was to assess the risk of bleeding prior to the procedure, we focused on pre-procedure risk factors. Ideally, a greater number of events would have increased the power of our analysis and the precision of our estimates; nevertheless, with 75 events, this study is the study with the largest number of events on this topic to date and the first from Western countries.
The 12% clip closure rate could be considered a limitation. However, there was no difference in CSDB rate between the clip closure and nonclosure groups. Clip closure reduced the rate of right colon bleeding after large EMRs in three high-level randomized controlled trials [22 ]
[23 ]
[24 ]. Clip closure after right colon EMR was effective only if the defect was completely closed. However, data on the effectiveness of clip closure after colorectal ESD to prevent CSDB are scarce, retrospective, and of low quality. The different risk factors for CSDB after ESD and EMR (i.e. rectum [5 ] and right colon locations, respectively) indicate that the two procedures must be considered distinct in terms of post-procedural CSDB. Therefore, the results for EMRs may not be directly applicable to ESDs. The complete clip closure rate was only 70% in a recent randomized trial of clip closure to prevent CSDB after EMR of lesions with mean size 30 mm [23 ]. In the current study, because more than 50% of the lesions were larger than 50 mm, systematic clip closure would have been impossible. However, new suturing systems are becoming increasingly available [25 ]
[26 ], and our data will enable the selection of patients at high risk of CSDB, which could be prevented by closing the scar.
What is the potential impact of such a prediction model?
First, because the model was based on pre-procedural data, it could provide individualized information on the risk of CSDB and optimize the care of patients undergoing colorectal ESD. The score enabled identification of lesions at low–medium risk (CSDB score 0–4; 4.1% risk of bleeding) and high risk (CSDB score 5–8; 17.5% risk of bleeding) of delayed bleeding. This categorization is important from a clinical point of view as patients in the high-risk group might need close monitoring and use of prevention strategies. Ingestion of antithrombotic agents could be influenced by delaying their resumption in high risk patients. In this sense, the decision curve showed a higher net benefit of using the score compared with a treat-all or treat-none approach across a range of plausible cutoff thresholds, suggesting potential clinical usefulness.
The effect on CSDB of new topical agents and closing systems is not supported by a high level of evidence [25 ]
[26 ]
[27 ]
[28 ]. Our results will help the design of a randomized trial investigating the efficacy of these new devices, one that enrolls patients at high risk of CSDB. Concerning the low–medium risk population, specific data on this population including cost-effectiveness analysis is mandatory to identify the best strategy in this subgroup of patients.
Moreover, after being independently validated, our prediction model should be used in further studies of CSDB risk, to enable comparison of post-ESD CSDB rates.
Finally, we created an excel calculator allowing individual estimation of the CSDB before the procedure to provide individual information and optimal management organization. (The calculator is available from the author on request.)
In conclusion, lesion size >50 mm, antithrombotic use, age >75 years, ASA score III/IV, and rectal location were independent risk factors for CSDB after ESD. Our easy-to-use risk score based on these factors enabled prediction of the risk of CSDB after colorectal ESD. After being independently validated, the CSDB score will allow provision of pre-procedural individualized information to patients, optimization of their care, and identification of patients who will benefit from new post-ESD CSDB prevention strategies.
