Introduction
Endoscopic submucosal dissection (ESD) allows an en bloc resection irrespective of
the size and morphology of the lesion. This is essential for a precise pathological
evaluation and for its lower risk of recurrence compared with piecemeal resection
by endoscopic mucosal resection (EMR) [1 ]. In the last decades, ESD has been applied with very good outcomes, not only in
stomach but also in the esophagus, colon, and rectum [2 ]
[3 ]
[4 ]. Even though the majority of colorectal lesions can still be managed by EMR [5 ], the role of ESD in this context has been growing in the West [6 ]
[7 ], particularly in lesions with a higher risk of harboring malignancy, such as nongranular
laterally spreading tumors (LSTs) [8 ] and large granular mixed-type LSTs [9 ]
[10 ], as well as all those that present with other high risk features on endoscopic evaluation
[8 ]. Nevertheless, ESD is technically challenging, with a long learning curve [11 ]. Depending on the experience of the endoscopist and lesion characteristics, it is
expected that a significant number of ESDs will not complete all the requisites for
being considered curative [12 ].
The European Society of Gastrointestinal Endoscopy (ESGE) renamed curative resection
as “low risk resection”, and redefined noncurative ESD (NC-ESD) as “local risk resection”
(LocRR) or “high risk resection” (HRR), depending on the presence of classic low or
high risk features, respectively [13 ]. For LocRR procedures, endoscopic follow-up may be sufficient; in HRR, complementary
treatment is usually warranted owing to the risk of lymph node metastasis (LNM). Nonetheless,
the best strategy for NC-ESDs is still under debate, as the different criteria for
HRR may not carry the same independent risk for LNM or residual lesions. This is of
critical importance because an inaccurate selection of patients may lead to a significant
proportion of unnecessary surgical procedures, with consequent morbimortality.
The aim of this multicenter project was to evaluate all of the consecutive NC-ESDs
performed in several Western reference centers, assessing the rate of residual lesions
in the surgical specimen or during endoscopic follow-up in malignant lesions, in order
to search for predictors of local residual disease or locoregional LNM.
Methods
Study design and patient selection
We performed a retrospective multicenter analysis of prospectively collected registries
of consecutive patients undergoing colorectal ESD between November 2009 and June 2021. A
total of 15 centers in Portugal, Spain, France, Belgium, Italy, Austria, and Australia
agreed to participate. All of the endoscopists had performed at least 100 ESDs at
the time of data collection; however, in some cases, this series included all of the
procedures that were performed since the beginning of ESD practice in a center, so
were early on the learning curve. The development of this endoscopic technique at
each center allowed all of the specimens to be analyzed by expert gastrointestinal
pathologists. Although the histological report could not be standardized owing to
the retrospective design of the study, a systematic approach was followed in each
center to document an accurate histopathological diagnosis, agreed upon by pathologists
and expert interventional endoscopists, according to international standards. Missing
data from the initial database led to direct contact with study centers to provide
further details if available.
The indication for ESD was the presence of a colorectal neoplastic lesion without
endoscopic features of deep invasive (> sm1) adenocarcinoma [13 ]
[14 ]. Only patients with an NC-ESD (LocRR or HRR) done for colorectal cancer (at least
T1 cancer) who underwent surgery or had at least one follow-up endoscopy were selected
for further analysis.
The Ethics Board approved the study (reference number 255/2020).
Definitions and outcomes
An en bloc resection was recorded if the target lesion had been retrieved in a single
specimen, or piecemeal resection if it was removed in more than one fragment. An R0
resection was recorded when pathological evaluation showed free horizontal and vertical
margins in an en bloc resected specimen.
Colorectal curative resections were R0 lesions, with low or high grade dysplasia,
or well to moderately differentiated mucosal or superficial submucosal (sm1; < 1000-μm
vertical submucosal invasion) adenocarcinoma, without lymphovascular invasion. All
other resections were considered NC-ESDs, as well as those that presented with tumor
budding [15 ].
Metachronous neoplasia in locations other than the ESD site, local recurrence or metastatic
disease in the long-term, and differences between curative and NC-ESDs were not assessed
(outside the scope of this study).
The major outcome parameters were: the rate of residual dysplastic lesions in the
scar after an NC-ESD, observed in the follow-up endoscopies (confirmed by pathological
analysis) or in the surgical specimens, as well as the rate of LNM in patients who
underwent surgery. Our goal was then to explore the risk factors and weigh these in
order to create predictive scores for local residual disease and for locoregional
LNM.
Statistical analysis
Categorical variables were described as absolute (n) and relative frequencies (%).
Mean and SD, or median and interquartile range (IQR) were used for continuous variables
as appropriate. Normal assumptions were verified to ensure correct test selection.
Continuous variables were then compared using either the Student’s t test or Mann–Whitney test, while the chi-squared test or Fisher’s exact test were
used for categorical variables, as appropriate.
First, univariate bivariable analyses were conducted in order to identify potential
predictors for LNM or residual lesions at the ESD site. Those predictors significantly
associated with the outcome were then considered for stepwise backward binomial logistic
regression, with either LNM or a residual lesion at the ESD site as the dependent
variable. A P value of < 0.10 was defined as the cutoff for inclusion of the assessed factors in
the final risk model.
The relative weights of the predictors (as described by β regression coefficient values,
rounded to the nearest integer) were used to create a clinical score. No interactions
with statistical significance were verified between the variables within the models.
The performance of the score was evaluated using a receiver operating characteristic
(ROC) curve. We assessed the calibration of the model with the Hosmer–Lemeshow goodness-of-fit
test. The model was validated internally using the resampling validation method for
logistic models with 100 bootstrap resamples and the c statistic was used to evaluate the discrimination of the models. Bias-corrected and
accelerated (BCa) 95 %CIs for the c statistic were then calculated.
Statistical analyses were conducted using the Statistical Package for the Social Sciences
(SPSS) v.25 and R version 4.1.2.
Results
Sample and procedures description
Among 2255 colorectal ESDs, 381 (17 %) were identified as being noncurative. A total
of 352 ESDs were performed in epithelial lesions that had available follow-up data,
with at least one endoscopy or surgery being performed. Most of the lesions were T1
cancers (n = 135; 38 %), Tis (n = 36; 10 %), or high grade dysplasia (n = 97; 28 %).
The remainder were low grade dysplasia (n = 76; 22 %), or serrated (n = 6; 2 %) and
anorectal squamous cell cancer lesions (n = 2; 1 %). The two anorectal squamous cell
cancers were further excluded from the analysis owing to their rarity and because
they represented a separate tumor entity that was not part of the indications for
colorectal ESD [13 ]
[14 ]. Therefore, 135 lesions were at least T1 adenocarcinomas and were included in this
ongoing study ([Table 1 ]; Fig.1 s , see online-only Supplementary material).
Table 1
Baseline characteristics of the cohort of 135 patients with at least T1 adenocarcinoma
who were included in this study, their lesions, and the procedures they underwent.
Patient characteristics
Sex, n (%)
84 (62)
51 (38)
Age, mean (SD), years
66 (10)
Lesion characteristics
Lesion size, median (IQR), mm
40 (26–60)
Paris classification, n (%)
31 (23)
30 (22)
40 (30)
29 (21)
5 (4)
Location, n (%)
60 (44)
75 (56)
Laterally spreading tumor (LST) morphology, n (%)
13 (10)
53 (39)
12 (9)
18 (13)
39 (29)
Procedure characteristics
ESD duration, median (IQR), minutes
100 (60–168)
Complementary techniques
93 (69)
28 (21)
2 (1)
10 (7)
2 (1)
Resection type
123 (91)
12 (9)
Histopathological analysis
Endoscopic submucosal dissection staging
15 (11)
120 (89)
Differentiation
45 (33)
73 (54)
17 (13)
Lymphatic permeation
17 (13)
118 (87)
Venous permeation
13 (12)
99 (88)
23
Horizontal margin
24 (18)
111 (82)
Vertical margin
71 (53)
64 (47)
Tumor budding
51 (39)
79 (61)
5
IQR, interquartile range.
The majority of the patients were male (62 %), with a mean (SD) age of 66 (10) years.
There were 60 colonic lesions (44 %) and 75 rectal lesions (56 %). The median lesion
and specimen size were 40 mm (IQR 26–60 mm) and 41 mm (IQR 29–66 mm), respectively.
The median duration of the ESD procedures was 100 minutes (IQR 60–168 minutes). Most
of the cases were performed in granular mixed nodular LSTs (39 %). Three patients
(2 %) needed surgery because of ESD adverse events. The ESD mortality rate was 0 %.
Follow-up of patients with noncurative lesions
There were eight patients who refused surgery, and 15 that had criteria for surgery
but, after multidisciplinary conference discussions, were only followed-up by endoscopy
owing to their age and co-morbidities. From the remaining patients, we report a total
of 96 (71 %) patients who underwent surgery for an NC-ESD. Among these, 16 (18 %)
developed serious adverse events of surgery and one patient died. From the patients
who did not undergo surgery, the median endoscopic follow-up time was 12 months, and
this was similar for the patients who presented with or without a residual lesion
(11 vs. 12 months, respectively; P = 0.28).
Presence of residual lesion after an NC-ESD for colorectal cancer
Among the 135 cancer lesions, 15 (11 %) had sm1 submucosal invasion, with the remainder
having deeper invasion. The proportion of patients with > sm1 invasion was similar
in the colon and the rectum (93 % and 85 %, respectively; P = 0.14).
A total of 24 patients (18 %) showed a residual lesion (20 in the surgical specimen
and 4 during endoscopic follow-up). No differences were found between centers regarding
recurrence, either globally (P = 0.55) or when separately analyzing patients who were operated on (P = 0.30) and those who were not (P = 0.97). Risk factors were poor differentiation (P < 0.001), positive/indeterminate vertical resection margin (VM + /VMx; P = 0.02), lymphatic permeation (P = 0.007), and piecemeal resection (P = 0.002).
Overall, the rate of LNM was 14 % and the rate of residual lesions in the wall was
13 %. However, the rates of LNM or a residual lesion in the wall were 0 % after NC-ESD
of T1 sm1 cancers. For colorectal cancer with deeper submucosal invasion, 24 % had
residual lesions; however, in those lesions without any other risk factors, such as
poor differentiation, positive horizontal margins, VM + , lymphatic permeation, or
venous invasion, the risk of LNM or lesions in the wall was also 0 %.
Among patients who were not operated on (n = 39), four (10 %) showed a residual benign
lesion in the scar that was treated by endoscopic resection. The presence of LNM in
this group of patients was not considered for score calculation because we considered
only patients with histologically proven LNM (in the surgical specimen). Follow-up
imaging was available in 32 of these patients (31 with computed tomography scanning
and one with magnetic resonance imaging). Only one patient who did not initially undergo
surgery because of co-morbidities presented with metastatic disease and is currently
receiving chemotherapy. Overall survival was 90 % in this group after a median follow-up
time of 17 months (IQR 10–36 months); deaths reported in this group were unrelated
to the lesion that motivated the ESD.
Predictive score for the presence of LNM
Among patients who underwent surgery (n = 96), 20 (21 %) showed a residual lesion
in the wall, LNM, or both. An example case of an NC-ESD with LNM is illustrated in
[Fig. 1 ].
Fig. 1 An example case of lymph node metastasis (LNM) following a noncurative endoscopic
submucosal dissection (NC-ESD) showing: a–d endoscopic images of one of the few cases of LNM (1 positive ganglion out of 12)
in a patient with an NC-Lymph score of 0; e,f histological images of: e submucosal adenocarcinoma; f with the presence of tumor budding.
Risk factors for LNM were poor differentiation (P = 0.002), lymphatic permeation (P < 0.001), and venous permeation (P = 0.03) ([Table 2 ]). After logistic regression, only poor differentiation and lymphatic permeation
remained independently related to LNM and were therefore selected to create a predictive
score (NC-Lymph score; β regression coefficient values for poor differentiation, lymphatic
permeation, and model intercept of 1.474, 2.028, and −4.775, respectively); lymphatic
permeation was scored with 2 points and poor differentiation with 1 point (Hosmer–Lemeshow
goodness-of-fit χ2 = 0.190; P = 0.91). Patients scoring 0, 1, 2, or 3 had a 6 %, 25 %, 30 %, and 75 % chance of
LNM (P < 0.001). Owing to the absence of LNM among sm1 cancers, this score could be applied
only in cases with > sm1 invasion, with similar risk rates (Table 1 s ). Bootstrap resampling showed similar results in the model, with c statistic of 0.766 (BCa 95 %CI 0.623–0.903). Example histological images for the
different scores are shown in [Fig. 2 ].
Table 2
Risk factors for the presence of lymph node metastasis among the 96 patients who underwent
surgery after a noncurative endoscopic submucosal dissection (ESD) performed for the
treatment of colorectal cancer.
n
Lymph node metastasis
P value
Yes
No
Patient characteristics
Sex, n (%)
0.44
57
9 (16)
48 (84)
39
4 (10)
35 (90)
Age, mean (SD), years
96
65 (10)
65 (10)
0.97
Lesion characteristics
Lesion size, median (IQR), mm
96
50 (29–60)
40 (25–50)
0.25
Paris classification
0.90
25
4 (16)
21 (84)
20
3 (15)
17 (85)
27
5 (19)
22 (81)
21
2 (10)
19 (90)
3
0
3 (100)
Location
0.85
54
7 (13)
47 (87)
42
6 (14)
36 (86)
Laterally spreading tumor morphology
0.65
9
0
9 (100)
36
6 (17)
30 (83)
9
1 (11)
8 (89)
13
1 (8)
12 (92)
Procedure characteristics
ESD duration, median (IQR), minutes
96
120 (90–185)
100 (60–159)
0.21
Complementary techniques
0.50
24
2 (8)
22 (92)
2
0
2 (100)
7
0
7 (100)
2
0
2 (100)
61
11 (18)
50 (82)
Resection type
0.07
88
10 (11)
78 (89)
8
3 (38)
5 (62)
Histopathological analysis
ESD staging
0.35
11
0
11 (100)
85
13 (15)
72 (85)
Differentiation
0.002
80
7 (6)
73 (94)
16
6 (38)
10 (62)
Lymphatic permeation
< 0.001
14
6 (43)
8 (57)
82
7 (8)
75 (92)
Venous permeation
0.03
10
4 (40)
6 (60)
86
9 (11)
77 (89)
Horizontal margin
0.38
13
3 (23)
10 (77)
83
10 (12)
73 (88)
Vertical margin
0.37
53
9 (17)
44 (83)
43
4 (9)
39 (91)
Tumor budding
0.69
39
6 (15)
33 (85)
56
7 (13)
49 (86)
IQR, interquartile range.
Fig. 2 Histological examples of the different NC-Lymph scores showing for: a score 0, a massive submucosal invasive adenocarcinoma; b score 1, an adenocarcinoma with poorly differentiated component; c score 2, evidence of lymphatic permeation; d score 3, a poorly differentiated adenocarcinoma (left-hand image) and lymphatic invasion
(right-hand image).
After surgery, follow-up was available for 84 patients: 72 with endoscopic follow-up
and 79 with radiological follow-up (67 had both). No evidence of endoscopic recurrence
was found, but there are two patients with metastatic disease who are receiving treatment.
One patient with severe co-morbidities who showed LNM in the surgical specimen died
2 years after the ESD, with radiological signs of disease progression. Another two
patients in this group died because of adverse events from surgery. The overall survival
rate was 96 % in this group after a median time of 24 months (IQR 13–36 months).
Predictive score for the presence of a residual lesion in the colorectal wall
Among patients with malignant lesions (n = 135), risk factors for the presence of
a residual lesion in the wall were piecemeal resection (P < 0.001), poor differentiation (P = 0.03), and VM + /VMx (P = 0.002) ([Table 3 ]). On multivariate analysis, all of these remained independently associated, with
a relative weight of 1 for poor differentiation and 2 for piecemeal resection and
VM + /VMx (Hosmer–Lemeshow goodness-of-fit χ2 = 0.860; P = 0.84; β regression coefficient values of 1.773, 1.389, 1.718, and −6.057 for piecemeal
resection, poor differentiation, VM + /VMx, and model intercept, respectively). Bootstrap
resampling showed similar results in the model, with c statistic of 0.792 (BCa 95 %CI 0.680–0.876).
Table 3
Risk factors for the presence of a residual lesion in the wall (at the endoscopic
submucosal dissection [ESD] site) among the 135 patients with a noncurative ESD performed
for the treatment of colorectal cancer.
n
Residual lesion in wall
P value
Yes
No
Patient characteristics
Age, mean (SD), years
135
68 (9)
66 (11)
0.54
Sex, n (%)
0.20
84
13 (16)
71 (84)
51
4 (8)
47 (92)
Lesion characteristics
Lesion size, median (IQR), mm
135
42 (29–60)
40 (25–60)
0.77
Paris classification
0.56
31
6 (19)
25 (81)
30
4 (13)
26 (87)
40
3 (8)
37 (82)
29
3 (10)
26 (90)
5
1 (20)
4 (80)
Location
0.12
60
11 (18)
49 (82)
75
6 (8)
69 (92)
Laterally spreading tumor (LST) morphology
0.60
13
1 (8)
12 (92)
53
5 (9)
48 (91)
12
3 (25)
9 (75)
18
2 (11)
16 (89)
39
6 (15)
33 (75)
Procedure characteristics
ESD duration, median (IQR), minutes
105
110 (60–191)
100 (60–165)
0.88
Complementary techniques
0.05
28
7 (25)
21 (75)
2
1 (50)
1 (50)
10
0
10 (100)
2
0
2 (100)
93
9 (10)
85 (90)
Resection type
< 0.001
123
11 (9)
112 (91)
12
6 (50)
6 (50)
Histopathological analysis
ESD staging
0.12
15
0
15 (100)
120
17 (14)
103 (86)
Differentiation
0.03
118
12 (10)
106 (90)
17
5 (29)
12 (71)
Lymphatic permeation
0.450
17
3 (18)
14 (82)
118
14 (12)
104 (88)
Venous permeation
> 0.99
13
1 (8)
12 (92)
99
9 (14)
85 (86)
Horizontal margin
0.19
24
5 (21)
19 (79)
111
12 (11)
99 (89)
Vertical margin
0.002
71
15 (21)
56 (79)
64
2 (3)
62 (97)
Tumor Budding
0.69
51
7 (14)
44 (86)
79
9 (11)
70 (89)
IQR, interquartile range.
Patients with an NC-Wall score of 0 had a 2 % chance of a residual lesion in the colorectal
wall, raising to 100 % in those with an NC-Wall score of 5. Patients were considered
low risk with a score of 0 (negative predictive value of 98 %), moderate risk with
a score of 1–3 (risk of 18 %), and high risk with a score of > 3 (risk of 50 %; P < 0.001) (Table 2 s ). Among patients with a score of NC-Lymph 0, patients with an NC-Wall score of 0
had a 2 % chance of a lesion in the wall compared with 50 % with an NC-Wall score
of 4 (P = 0.002) ([Fig. 3 ]).
Fig. 3 Management of the patients according to the predictive scores.
Discussion
Our multicenter, multinational Western study on noncurative colorectal ESDs demonstrates
that 79 % of the surgical specimens obtained for complementary treatment were free
of neoplastic cells in the wall and in the lymph nodes. By studying the risk of residual
wall lesions and LNM on this large retrospective cohort, we derived from regression
analysis two predictive scores with the aim of better stratifying the risk of residual
neoplasia in patients after NC-ESD.
In the last two decades, ESD has been shown to be a very useful technique for the
treatment of colorectal neoplasia, with good outcomes, not only in Eastern countries
[16 ] but also in Western countries [17 ]
[18 ]. Compared with EMR, ESD offers a higher rate of complete resection and a lower rate
of recurrence [18 ]
[19 ]
[20 ], but it is very demanding, with longer procedural times and a greater risk of perforation
[19 ], and a high rate of NC-ESDs is expected. Therefore, we need predictors of the probability
of residual wall lesions and of LNM after NC-ESD, in order to decide on the indications
for oncological surgery versus endoscopic follow-up.
We analyzed a large multinational case series on NC-ESD with a median endoscopic follow-up
time of 1 year for those patients who did not undergo surgery. The majority (96 %)
of residual lesions in the scar will be detected in the first 6 months after the resection,
and 98 % in the first year according to a large metanalysis [20 ]. Therefore, the duration of endoscopic follow-up was adequate for detection of residual
wall lesions.
In the presence of at least one HRR factor, the ESGE guideline recommends complementary
surgery. Our multicenter retrospective case series confirmed lymphatic permeation,
VM + /VMx, and poor differentiation, as well as piecemeal resection, to be predictors
for a residual lesion among malignant cases. The same was not demonstrated for other
factors such as deep submucosal invasion. With regard to the presence of budding,
we did not find a statistically significant relationship with the presence of residual
disease; however, budding evaluation and scoring may not be homogeneous among the
different centers in our study. Therefore, further prospective trials with rigorous
evaluation of budding are needed for clarification of this feature as a risk factor
for LNM.
The decision as to whether the patient needs complementary surgery must consider the
risk of LNM. In our series, only lymphatic permeation and poor differentiation were
independently associated with LNM in the surgical specimen. A smaller series found
lymphatic permeation to be the only independent risk factor for the presence of LNM
[21 ]. In fact, this has been reproduced in other studies and it was clearly verified
in our cohort [22 ].
For LNM, little is known about the individual relevance of the other high risk factors
in HRRs. Some studies have not shown deep invasion to be an independent risk factor
on multivariate analysis [23 ], others found 2000 µm to be the cutoff for a higher risk of LNM [24 ]. A meta-analysis on retrospective studies showed that 2-mm deep submucosal invasion,
as compared with 1-mm deep, was related to a higher risk of LNM [25 ]. A large Japanese study on surgical specimens showed that none of the T1 cancers
with submucosal invasion < 1000 µm presented with LNM [26 ]. In our cohort, we also did not find any sm1 T1 cancers with LNM, regardless of
the presence of other risk factors. Similarly, the presence of deeper invasion as
a risk factor for LNM did not find statistical significance in our cohort. In fact,
> sm1 T1 cancers without any other risk factors (i. e. negative for lymphovascular
permeation, budding, VM + /VMx, piecemeal resection) had a 0 % chance of LNM or a
lesion in the wall. Three patients were identified with metastasis during follow-up
in our cohort, but late metastatic disease after an NC-ESD was not a primary end point
in this study and it should be evaluated in prospective trials with long-term follow-up.
Surgery after an NC-ESD in the colorectum is usually effective and safe [27 ]
[28 ]; however, colorectal surgery can lead to morbidity and a decreased quality of life,
so the criteria for referral to surgery must be refined. We suggest that an NC-Lymph
score > 0 warrants surgery, owing to the high risk of LNM (above 27 %). For patients
with an NC-Lymph score of 0, a rigorous assessment of deep invasion, vertical margin,
and budding characterization should be performed, and the decision must be individualized.
Among patients with malignant lesions that did not directly qualify for surgery (NC-Lymph
score 0), patients with an NC-Wall score > 0 should undergo close endoscopic follow-up
or surgery, and those with a score of 0 could probably be followed-up by endoscopy,
because of the low risk of a colorectal lesion at the scar.
The application of computer-aided models for predicting LNM might become a potential
option to help with validation of our two scores in big data [29 ]. A recent study reported that artificial intelligence significantly reduced unnecessary
additional surgery after endoscopic resection of colorectal T1 cancer in comparison
with the current guidelines [30 ].
Our study has strengths and limitations. The main strength is that it is a multinational
large series on NC-ESDs, with the participation of multiple referral endoscopists
and institutions, which allowed us to have a significant number of cases with and
without a residual lesion in the Western setting. We were able to establish a tool
to identify those patients who will benefit from surgery and those who would have
a high probability of being submitted to a futile intervention.
The main limitations relate to the observational nature of the study and the heterogeneity
of the ESD technique, which reflects daily clinical practice. In addition, follow-up
losses and the absence of complementary surgery owing to refusal or co-morbidities
may represent a selection bias. Another limitation is that, despite gathering data
from several centers, the rate of NC-ESD among the total number of ESDs is low; because
of this, the predictive scores were calculated using the entire cohort, which did
not allow us to split the sample in order to have external validation. We used the
resampling methodology to mitigate this issue, but subsequent clinical trials should
be performed for external validation and to eventually refine these scores in order
to incorporate them into clinical algorithms.
In conclusion, with this large multicenter retrospective Western case series of colorectal
NC-ESD, we found that none of the patients with an NC-ESD of an sm1 T1 colorectal
cancer had a residual lesion in the wall or in the lymph nodes, even with the presence
of other high risk criteria. Similarly, > sm1 T1 cancers without any other risk factors
did not present residual lesions during the follow-up. Among > sm1 T1 cancer patients,
lymphatic permeation and poor differentiation in the ESD specimen were the only independent
factors for LNM, and its presence should warrant surgery. Lesions removed by piecemeal
resection that are poorly differentiated and without free vertical margins have the
highest risk of a residual lesion in the wall. The developed scores could be helpful
to reduce the rate of unnecessary adjuvant surgery following colorectal NC-ESD.
