Study subjects

Consecutive patients referred to the Hospital Universitario 12 de Octubre, a tertiary referral center in Madrid, who underwent ESD for the treatment of esophageal, gastric or CR neoplasms between September 2008 and December 2015 were recruited into the study. Data were analyzed retrospectively using a prospectively collected database from the first lesion removed via ESD.

Written informed consent was obtained from all patients. The study was approved by the Institutional Review Board of the Hospital Universitario 12 de Octubre (CEI 14/384; 29/9/2015). The study was performed in accordance with the Declaration of Helsinki and was conducted following the STROBE statement for reporting cross-sectional studies.

Endoscopic examination

All lesions were assessed before the ESD procedure using a magnifying endoscope (EG 590 ZW for upper gastrointestinal lesions and EC 590 ZW or EC 600 ZW for CR lesions; Fujifilm, Tokyo, Japan). With regard to CR ESD, the proposed indications by the Japanese Colon ESD Standardization Implementation Working Group were considered ¹. Curative resections in early gastric cancer were considered when the guideline or expanded criteria were met ⁷.

ESD procedure

ESD procedures were performed by two teams with two therapeutic endoscopists in each team (team 1: JCMG and JDT; and team 2: SRM and AJdPG). The lateral margins were marked using a Flush knife or Flush knife BT (Fujifilm, Tokyo, Japan) after targeted chromoendoscopy for esophageal and gastric lesions. ICC(r) 200 or VIO(r) 300D (ERBE, Tübingen, Germany) electrosurgical generators were used. Carbon dioxide insufflation was used in all cases (Olympus UCR, Olympus, Tokyo, Japan). The following electrosurgical knives were used for the dissection at the discretion of the endoscopist: Flush knife or Flush knife BT (Fujifilm, Tokyo, Japan) and IT knife 2(tm) or IT nano(tm) (Olympus, Tokyo, Japan). When unresolved technical difficulties occurred (poor maneuverability or a perpendicular approach to the muscle layer with a high risk of perforation) or when the ESD procedure was excessively long (usually over 180 minutes), a salvage knife-assisted snare resection (KAR) was performed. There was never a planned KAR in any case. Although the original intention was to achieve an en bloc resection, even for KARs, this could not be achieved in some circumstances.

Histopathological evaluation

The endoscopic morphology of the lesion was classified according to the Paris classification ⁸ and the proposed classification by Kudo et al. ⁹ was followed for laterally spreading tumors (LSTs). Tumor size was measured with a scaled ruler in the case of en bloc resections after being stretched with pins on a piece of foam. With regard to piecemeal resections, tumor size was estimated during colonoscopy by reference to an open 20 mm polypectomy snare. The histopathological diagnosis was based on the Vienna classification ¹⁰.

Outcome parameters

An en bloc resection was defined as a resection in a single piece. When a resection in more than one fragment was achieved, the technique was considered as a piecemeal resection. KAR was considered when the submucosal layer was partially dissected, although a snare was finally used to complete the resection. Curative resection was considered as an en bloc resection in a well or moderately differentiated carcinoma when the lateral and vertical margins were free of neoplasia, without lymphovascular involvement as well as the absence of submucosal invasion deeper than sm1 ¹¹ (< 200 microns for esophageal squamous cell carcinoma, < 500 microns for gastric carcinoma and < 1,000 microns for colorectal carcinoma). The expanded criteria were used to assess curability for gastric neoplasms ^7,12,13. Scarred lesions were defined as local recurrences after prior endoscopic resections.

Statistical analysis

Data collection was performed using MS-Access (Microsoft Corp., Redmond, WA, USA). The IBM SPSS Statistics software version 21.0 for Windows (IBM Corp., Armonk, NY, USA) was used for statistical analysis. Continuous patient variables were described with mean (SD) values when they followed a normal distribution. Otherwise, median and range were obtained. Categorical variables were presented as absolute and relative frequencies. Data were stratified for en bloc resections and p-KARs and their distributions were compared with the Student's t test, Wilcoxon test or Chi-square statistics. Some potential risk factors related to the need for p-KAR were analyzed: the number of procedures performed, the team that performed the resection, any previous electrosurgery applied to the tumor, evidence of a non-curative submucosal invasion, tumor location, a depressed morphology of the tumor, the time it took to perform the procedure and the lesion size. Experience was categorized as ≤ 50 and > 50 procedures according to Saito et al. ². Furthermore, the maximum diameter of the tumor was categorized as ≤ 30 mm and > 30 mm, as the recommended size for beginners is under 3 cm ¹⁴. A p value less than 0.05 was considered as statistically significant. The meaningful variables with a p value < 0.1 according to the univariate analysis were included in the multivariate binary logistic model. Odds ratios (OR) and 95% confidence intervals (CI) were calculated. The model developed to predict the risk of p-KAR was simplified as an easy-to-use risk scoring system. In order to establish the clinical risk score for predicting a piecemeal resection, weighted points proportional to  regression coefficient values were assigned for the factors determined in the multivariate analysis ¹⁵. The  coefficients were divided by the smallest absolute value of the regression coefficient and rounded to the nearest integer. The performance of the predictive model and risk score were evaluated using receiver operating characteristic (ROC) curve analysis. The optimal cut-off point was calculated as the maximum value of the sum of sensitivity and specificity. Sensitivity, specificity, predictive values, likelihood ratios and diagnostic accuracy were calculated for an optimal cutoff point using a 2 x 2 table analysis.

Procedural outcomes of ESD

The outcome of the ESD procedures are presented in Table 2. The procedure had to be changed to KAR in 64 cases (47%). The specimen was resected in a piecemeal fashion in 44 KARs (32.3%). Thus, en bloc resections were achieved in 72 ESDs (53%) and 20 KARs (14.7%).

Table 2 Procedure outcomes of the initially planned endoscopic submucosal dissections (ESD) 

SD: standard deviation; ESD: endoscopic submucosal dissection; p-KAR: piecemeal knife-assisted snare resection. *En bloc resections include ESD as well as KARs in which a single specimen was obtained.

Risk factors for the need for p-KAR

Experience with the technique of ≤ 50 ESD procedures, lesion size > 30 mm, CR location and a depressed morphology were all significantly associated with a higher frequency of piecemeal resections, whereas the team that performed the procedure, a scarred lesion and the presence of deep submucosal invasion were not associated with piecemeal resections (Table 3).

Table 3 Risk factors related to the need for a p-KAR procedure (univariate analysis) 

ESD: endoscopic submucosal dissection; p-KAR: piecemeal knife-assisted snare resection.

Statistically significant differences with regard to lesion size of gastric and colorectal locations (stomach vs rectum: 29.3 ± 13.2 vs 38.6 ± 18.3, p = 0.009; colon vs rectum: 28.9 ± 13.7 vs 38.6 ± 18.3, p = 0.02) were found. An interaction between a colorectal location and a diameter of the tumor of > 30 mm was observed when their association with p-KAR was assessed (overall OR = 3.33 [95% CI: 1.6-7.1], CR location OR = 5.53 [95% CI: 2.1-14.6] and non-CR location OR = 0.26 [0.03-2.4]). There were no statistically significant differences with regard to the percentage of colorectal cases by year (p = 0.45). There were also no statistically significant differences with regard to the maximum diameter of the lesions by the year of recruitment (Fig. 2).

Fig. 2 Maximum diameter of the lesions according to the year of recruitment. 

Although the overall perforation rate was higher in the p-KAR group (22.7% vs 12%), the difference was not statistically significant (p = 0.11). In fact, there were no statistically significant differences in the intra-procedural perforation rate between both groups (n = 8: 8.7% in the en bloc group vs n = 8: 18.2% in the p-KAR group; p = 0.11).

According to the multivariate logistic regression analysis, experience with the technique of ≤ 50 procedures and the combination of CR location and lesion size (measured in cm) were significant factors for predicting a piecemeal resection during the procedure (Table 4). Although the simple variables of the interaction term (CR location and size > 30 mm) were not statistically significant according to the multivariate analysis, both were retained in the model as recommended.

Table 4 Multivariate logistic regression model and independent predictors for a piecemeal KAR 

OR: odds ratio; CI: confidence interval; CR: colorectal; KAR: knife-assisted snare resection. The assignment of points to risk factors was based on a linear transformation of the corresponding  regression coefficient. The coefficient of each variable was divided by 0.43 (the lowest positive  value, corresponding to CR location) and rounded to the nearest integer.

Predictive model for p-KAR

The model to predict the end of the procedure with a piecemeal resection was as follows (Table 4): logit (p) = -2.56-0.8 (size > 30 mm) + 1.70 x (≤ 50 procedures) + 0.43 (CR location) + 2.82 x (CR location and size > 30 mm). This model calibrated well with the Hosmer-Lemeshow goodness-of-fit test (2 = 1.25, df = 5, p = 0.94). The value of the area under the ROC curve for the predictive model was 0.82 (95% CI: 0.74-0.89), with a good discrimination ability.

The final predictive score system for a p-KAR was as follows: 1 x (colorectal location [0 = no; 1 = yes]) - 2 x (size > 30 mm [0 = no; 1 = yes]) + 4 x (number of the procedure < 50 [0 = no; 1 = yes]) + 6 x (size > 30 mm and colorectal location [0 = no; 1 = yes]). The score ranged from -2 (best prognosis to achieve an en bloc resection) to 9 (worst prognosis). For example, a patient with a colorectal tumor size smaller than 30 mm and an operator experience of under 50 cases would have an estimated risk of 39% of a p-KAR and would receive a score of five points. The value of the area under the ROC curve for the score (Fig. 3) was 0.81 (95% CI: 0.74-0.89). The diagnostic accuracy of the score for a cut-off point ≥ 5 had a sensitivity of 0.79 (95% CI: 0.66-0.93), a specificity of 0.71 (95% CI: 0.61-0.80), a positive predictive value of 0.56 (95% CI: 0.43-0.70) and a negative predictive value of 0.88 (95% CI: 0.80-0.96).

Fig. 3 ROC curve of the predictive score system for a piecemeal knife-assisted snare resection (KAR) and cut-off points. 

DISCUSSION

ESD training is a major issue in the European setting. Thus, suboptimal en bloc resection rates have been reported from the outset ¹⁶. Nevertheless, performing ESD in Western countries is challenging. It is recommended to start the learning curve of this procedure in the context of early gastric cancer. However, this condition is less prevalent in Europe. This is not a minor matter due to the fact that before moving to the CR area, Japanese experts recommended a minimum experience of 20 ESD procedures in gastric locations ¹⁷. After gaining experience with gastric cases, LST-Gs of 2-3 cm in size that are in the rectum are considered to be the most appropriate lesions for CR ESD beginners ¹⁴. Furthermore, in order to obtain a successful R0 resection rate of colonic lesions with few complications, some professionals believe that experience in up to 80-100 procedures is required ¹⁸.

We acknowledge that our case-mixed series (although mainly gastric and colorectal) could reduce the ability to compare our data with other published studies that focus on a specific organ. However, we provide data of the ESD procedure that is performed in a real-life scenario with Western endoscopists. In this setting, technically easy gastric lesions are difficult to find and cases are selected when trying to avoid surgery, even for difficult colorectal tumors. Therefore, our results can be generalized to patient populations in other tertiary referral centers that are performing their first procedures in a Western European setting. In our study, an en bloc resection could not be achieved in almost one third of the cases during the early learning curve. We were forced to conclude the procedure by changing to the KAR technique in some cases in order to successfully complete the resection.

Our study demonstrates the association between some preoperative modifiable factors and the possibility to complete the procedure as a salvage piecemeal resection in a European setting. Experience with human ESD cases, as well as size and location of the tumor, are all predictive factors that can be easily determined preoperatively. This is undoubtedly the main strength of our study. Since this model has been conducted by a group of European endoscopists who were ESD beginners, our results may have special significance for other non-expert endoscopists when they encounter their first human cases. This study has demonstrated the good performance of the predictive model.

The need to detect factors that predict technically difficult ESD has been addressed in the recent literature, mainly for CR lesions. Isomoto et al. ¹⁹ indicated that right-sided tumors with fibrosis have a high risk of an incomplete resection. Furthermore, they demonstrated that only fibrosis was significantly associated with the need for a piecemeal resection by multivariate analysis. However, the presence of submucosal fibrosis cannot be easily assessed preoperatively. With regard to failure of en bloc resection, Takeuchi et al. ²⁰ noted that a low case-load, snare use and poor lifting after submucosal injection were associated with technical difficulties. Poor endoscopic maneuverability and deep submucosal invasion were identified as risk factors for an incomplete CR ESD in the study conducted by Hayashi et al. ²¹. Hori et al. ²² observed that scarring or a locally recurrent tumor and a tumor located at a flexure were all risk factors for a piecemeal resection. Finally, Imai et al. ²³ have identified some independent predictors of en bloc resection failure in a large series of CR ESD procedures: fold convergence, a protruding morphology, poor endoscope operability, colonic location vs rectal lesions, an underlying semilunar fold and a less-experienced endoscopist (< 40 cases of CR ESD).

Our study has several limitations. First, there was a relatively small sample size. However, only four variables were used to predict more than 40 events in order to avoid the risk of overfitting the logistic regression model. Future studies with a pre-determined sample size will be needed in order to confirm our results. Second, a statistically significant difference with regard to size between the ESD and the p-KAR group was observed. The lesions that were finally removed via p-KAR were larger. However, we believe that a small difference in the mean size between both groups (only 1 cm) does not explain a selection bias. Third, although the four endoscopists who performed the procedures were not experts in the ESD technique, all of them had extensive experience with EMR with large colonic lesions. Therefore, our results may not be generalized to the general practice endoscopists. Fourth, a selection bias may not be fully excluded due to the lack of experience of the endoscopists. They could potentially bias the results by selecting the type of lesions with a greater chance of a successful en bloc resection a priori. However, the relatively high percentage of piecemeal resections, the percentage of lesions over 30 mm in size and the prevalence of scarred lesions in the study (approximately 6%) makes this scenario rather unlikely. Fifth, aborted procedures were not included in the final analysis. This decision might add a selection bias. However, only six of 142 procedures (4%) were aborted. Thus, the impact of removing them from the analysis should be negligible.

Finally, the main weakness of this study is the absence of internal and external validation groups. Further prospective multicenter studies are needed to confirm our results and to develop a more complete model.