Endoscopy International Open Predicting procedure duration of colorectal endoscopic submucosal dissection at Western endoscopy centers

: BACKGROUND&STUDY-AIMS:Overcoming logistical obstacles for the implementation of colorectal endoscopic submucosal dissection (ESD) requires accurate prediction of procedure times. We aimed to evaluate existing and new prediction models for ESD duration.


Introduction
Colorectal endoscopic submucosal dissection (ESD) is an endoscopic resection technique which is widely used in Eastern countries. However, it has been slow to be taken up by Western endoscopists [1,2]. This is mainly due to a lack of local experts and opportunities to acquire sufficient skill in ESD, which makes the procedure complex and time-consuming for them. Several studies demonstrated that for equivalent-sized lesions, the procedural time for ESD may exceed the time needed for piecemeal EMR (pEMR) by 3-4 times [3,4]. This introduces logistical problems in Western endoscopy centers because the shortage of resources do not allow for the routine use of such timeconsuming procedures that are difficult to learn. Nevertheless, colorectal ESD is gradually gaining ground in the West because it enables accurate histological staging and decreases the risk of recurrence [5], thereby also reducing the need for intensive follow-up programs and additional endoscopic or surgical treatment after the initial endoscopic resection.
To overcome the logistical obstacles for implementing colorectal ESD, it is essential to anticipate possible long procedure durations by accurate time scheduling. In this study, we evaluated the performance of current time planning practice and a previous Eastern prediction model [6] for colorectal ESD duration, and developed a new model with procedure time as continuous prediction outcome.

Methods
This study is reported in accordance with the TRIPOD statement (Supplementary Checklist) and was approved by the Medical Ethical Committee of Leiden University Medical Center (reference number G18.097). The variable definitions, ESD endoscopists/procedures and statistical analyses are detailed in the Supplementary Methods.

Patients
The analysis cohort consisted of consecutive patients who underwent ESD for colorectal neoplasms before 1 January 2020 at 3 Dutch tertiary endoscopy centers (Leiden University Medical Center, University Medical Center Utrecht, Erasmus MC Cancer Institute in Rotterdam). We included all cases starting from the first ESD ever performed, with the aim of making our findings also generalizable to beginning Western ESD endoscopists, the target group where the greatest benefit can be attained [7].
We excluded hybrid ESDs and procedures with ≥2 simultaneous ESDs. In addition, resections of subepithelial tumors (e.g. gastrointestinal stroma cell tumors, neuro-endocrine tumors, lipomas), anal intraepithelial neoplasms and post-polypectomy scar resections by ESD were also excluded.
Lastly, we excluded intended ESD procedures which were terminated prematurely without (complete) lesion removal. This was because in all these cases (n=43) the reason for premature ESD termination turned out to be deep tumor invasion (i.e. suspected invasion into the muscularis propria during the procedure), which made it impossible or futile to continue the resection. The vast majority of deeply invasive tumors could be anticipated pre-procedurally by optical diagnosis [8]. As we were interested in procedure durations resulting from the complexity of the ESD (and not from patient selection-related factors), we reasoned that including prematurely terminated ESDs would distort the analyses (e.g. relatively short procedure durations for very large but deeply invasive tumors). We chose not to exclude intended ESDs which were converted to piecemeal resection, as endoscopists are in general reluctant to do so without first trying their utmost to achieve en bloc resection by ESD. The analysis cohort was used for descriptive statistics, evaluation of ESD durationrelated outcomes and development of a prediction model for colorectal ESD duration.
External validation of the developed prediction model was performed in an independent cohort of consecutive colorectal ESDs from a Swedish tertiary endoscopy center (Karolinska Institutet, Stockholm). Parts of this cohort have been described previously [9,10]. Single, non-hybrid ESDs were selected using the above inclusion and exclusion criteria. This article is protected by copyright. All rights reserved.

Time planning analyses
The amount of time scheduled was decided for each case individually by or in consultation with the endoscopist performing the ESD. Procedure time was not scheduled in a standardized manner.
Actual procedure duration was defined as the time between first introduction and final removal of the endoscope (the total "scope time": i.e. including possible cleaning of the lesion and marking of its perimeter, dissection and retrieval of the resected specimen; excluding induction and recovery time of propofol sedation). When the actual ESD duration was >1 hour longer or shorter than the amount of time scheduled, endoscopy reports were extensively reviewed for possible explanations for the difference between the scheduled and actual time.
We also evaluated the performance of a previously published Eastern prediction model for colorectal ESD duration [6] (see Supplementary Methods for details). The predicted outcome of the model is ESD completion <60 minutes (yes/no). For sensitivity analyses, we also varied the dichotomization cut-off of the outcome with 1-minute increments over the entire range of ESD durations in our analysis cohort. For each cut-off, the model's performance was evaluated.

Model development
As the abovementioned Eastern model [6] used a dichotomized outcome, we built a new prediction model, the cESD-TIME formula, with colorectal ESD duration as continuous outcome. The initial formula consisted of the 4 variables in the Eastern model (tumor size, circumference, location and morphology). Besides, we also considered other pre-procedural variables that have been reported to be associated with the presence of submucosal fibrosis, a factor that could considerably increase ESD difficulty and duration [11][12][13][14]. These pre-procedural variables included the type of lesion (recurrent or naïve lesion to be resected) [15,16], presence of a depressed area [17,18], suspected invasive cancer [16], prior biopsies taken [16,19], and inflammatory bowel disease [20,21]. Lastly, we included the consecutive colorectal ESD number for each endoscopist in the model as a proxy for ESD experience, an important determinant of procedure duration [22][23][24]. The adopted ESD method was not considered in the model development, because a randomized trial found no significant difference in dissection speed between the 2 ESD methods (conventional and pocket-creation method) that were used in the analysis cohort [25].
As exploratory analysis, we also included endoscopic maneuverability in the cESD-TIME formula and evaluated how the model performance would change. Maneuverability is generally referred to as the degree of having difficulty with obtaining and maintaining an optimal dissection plane through the submucosa, e.g. due to paradoxical movement of the endoscope [7,26,27]. However, there are currently no standardized criteria available for determining or classifying endoscopic

Performance of a previous Eastern prediction model
A prediction model for colorectal ESD duration (i.e. completion <60 minutes yes/no) was previously developed by Eastern endoscopists [6], with a c-statistic of 0.70 in the development cohort and 0.69 in an Eastern external validation cohort. We evaluated the performance of this Eastern model in our analysis cohort and found a c-statistic of 0.70 (95%-CI: 0.62-0.77; Supplementary Fig.2). Notably, the proportion of ESDs completed <60 minutes was much lower in our Western cohort (14%) as compared to the Eastern cohort (54%).
When changing the dichotomization cut-off over the entire range of ESD durations (15-720 minutes) in our analysis cohort, we found that both the proportion of ESDs completed within the cut-off time and the c-statistic proportionally increased with the cut-off value (Supplementary Fig.3A-C). For example, when dichotomizing ESD duration into completion <120 minutes yes/no, the c-statistic was 0.75 (95%-CI: 0.69-0.80) and the proportion of ESDs completed <120 minutes was 51.0%. When using a dichotomization cut-off of 180 or 240 minutes, the c-statistic was 0.79 (95%-CI: 0.72-0.84) or 0.85 (95%-CI: 0.75-0.91) and the proportion of ESDs completed within the cut-off time was 77% or 91%, respectively.

cESD-TIME formula development
Next, we developed a prediction model (the cESD-TIME formula) which included ESD duration as continuous outcome. A transformed version of the Eastern model yielded an R 2 of 61% (Supplementary Table 3 and Supplementary Results). To evaluate whether the model could be simplified or its performance further improved, we applied regression-based backward selection with the 4 variables of the Eastern model and other pre-procedural factors which are associated with submucosal fibrosis or ESD experience (unadjusted associations between all candidate predictors and the outcome in Supplementary Table 4). This resulted in the cESD-TIME formula, a 6-variable model which includes tumor size, luminal circumference, morphology (as defined in the Eastern model [6]), depressed area, inflammatory bowel disease and consecutive number of colorectal ESD performed (Table 4). Notably, tumor location (as defined in the Eastern model or categorized into rectum, left and right hemicolon) was not selected in the final cESD-TIME formula, as well as the type of lesion (naïve or recurrent lesion to be resected), whether or not prior biopsies were taken and whether or not invasive cancer was suspected (Supplementary Table 5). The final cESD-TIME formula explained 63% of the variance of the ESD durations (95%-CI: 57-69%; Fig.1B), with 79% of ESDs completed ±1 hour of the predicted time, 12% completed >1 hour ahead of the predicted time and 9.2% exceeding the predicted time >1 hour. An example calculation in the online calculator (https://cesdtimeformula.shinyapps.io/calculator/) is provided in Supplementary Fig.4.
The exploratory analyses with endoscopic maneuverability as potential predictor are detailed in the Supplementary Results.  Fig.1C). Based on these findings, we included a customizable slope for the linear predictor in the online calculator. As users acquire more information on their ESD performance level, the recalibration slope can be adjusted if necessary.

Discussion
This study provides 2 useful tools for predicting colorectal ESD duration at Western endoscopy centers. First, we successfully validated a previous Eastern model [6] for ESD duration in a large Dutch cohort. However, this model had a dichotomized outcome and turned out to have more discriminative power when using higher dichotomization cut-offs. Secondly, we developed the new, easy-to-use cESD-TIME formula (https://cesdtimeformula.shinyapps.io/calculator/), which predicted procedure duration as continuous outcome. This formula increased the current planning accuracy more than two-fold (R 2 : 61% vs. 27%) and performed well (R 2 =66%) in a cohort with relatively higher ESD performance levels after recalibration. These prediction tools could facilitate individualized time scheduling of colorectal ESD at Western centers.
To our knowledge, this is the first study to provide empirical data on time planning outcomes of colorectal ESDs in Western practice, where logistical issues concerning ESD implementation stand out more strikingly due to a lack of experience and ESD experts [7]. This is illustrated by our finding that current unstandardized time planning was inaccurate, with a substantial proportion (~30%) of procedures running ahead or exceeding the allotted time. An >1 hour exceeding was also associated with significantly lower en bloc and R0 resection rates, which could probably be attributed to the increased fatigue and time pressure experienced by the endoscopist. Reasons for ESD completion ahead of or behind the scheduled time were rarely reported. Only for a few procedures that were completed >1 hour later, we could deduce that unanticipated difficulties (submucosal fibrosis, intraprocedural bleedings) were the main causes of time exceeding. It should be kept in mind that besides unanticipated peri-procedural events or findings, suboptimal time allocation may also explain why a procedure was completed ahead of or behind the scheduled time. To illustrate this, most ESDs in center #2 (73%) appeared to be scheduled for a standard 2-hour time slot, but a substantial proportion of these procedures turned out not to require 2 hours (120±0min: 9.3%, 120±30min: 25%, 120±60min: 64%). Altogether, our findings emphasize the necessity of systematic and more accurate time planning of colorectal ESDs.
To improve planning accuracy, we first evaluated the performance of a previously published Eastern prediction model for colorectal ESD duration [6]. Interestingly, the Eastern prediction model was also valid in a Western setting, given the similar c-statistic values (0.70) in our cohort and the original development cohort. This may have come as a surprise, considering the large difference in performance levels between Eastern and Western ESD practice [2,28]. It appears that there may be certain "universal" pre-procedural factors which determine ESD duration. However, despite successful external validation, the Eastern model was still of limited use for Western practice because of the relatively low proportion of procedures completed within the 60-minute cut-off (14%). When increasing this cut-off value, both the c-statistic and the proportion of procedures completed within the cut-off time increased considerably. This indicates that the Eastern model seems more useful for determining whether an ESD is very likely or unlikely to take much time in Western ESD practice.
We decided not to further update the Eastern model, as the dichotomized outcome does not provide information on how much longer or shorter a procedure would take than the cut-off time. Instead, we developed a new formula which predicted ESD duration as continuous outcome. Using the 4 predictors from the Eastern model and some simple transformations, we were already able to construct a model with satisfactory performance. Strikingly, a relatively low β (+2.2 minutes) and high p-value (p=0.52) were found for tumor location, and this remained so after including different locations as separate variables or redefining the categories. Besides, tumor location was also omitted from the final cESD-TIME formula after backward selection, suggesting that the influence of this factor on ESD duration was limited. This was unexpected because anatomical location is in general closely correlated to endoscopic maneuverability, a crucial determinant of ESD difficulty [7,26].
However, a recent editorial has already pointed out that performing ESD in certain situations (e.g. patients with severe abdominal adhesions or a long and flexible colon) can always be difficult, regardless of tumor location [7]. Besides, selection bias of lesions in supposedly difficult locations may also explain the limited predictive value of anatomical location. In any case, it appears that location does not always faithfully recapitulate endoscopic maneuverability, and thus the complexity and duration of colorectal ESD.
Despite the superior performance of the cESD-TIME formula over current time planning practice, predicting exact procedure times remains quite challenging. This is for instance reflected in the wide prediction intervals and the need for recalibration in the Swedish validation cohort, which was reasonable given the between-cohort difference in ESD performance. Although achieving 100% accuracy is probably utopian[29], as we can learn from prediction modelling studies for surgical procedure times (highest R 2 values around 80%[30-34]), we acknowledge that there is still room for improvement. As proposed by several ESD experts [7,26], endoscopic maneuverability could be key to increasing the predictive performance. This is substantiated by a Japanese study[27] and our finding that maneuverability (as subjectively evaluated during the ESD) was a significant predictor of procedure duration (Supplementary Results). Unfortunately, no objective and standardized criteria are currently available for assessing endoscopic maneuverability. Jacques et al. recently proposed the Size, Maneuverability, Site, History (SMSH) score[35], the "ESD-equivalent" of the SMSAclassification, for predicting R0 resection without perforation. However, we think that the Maneuverability-category of this classification lacks sufficient detail for the prediction of ESD durations, as it categorizes maneuverability only into "good" and "bad" without further specifications on the scoring criteria. Therefore, we propose some objective and more precise criteria for systematically classifying endoscopic maneuverability for colorectal ESD ( Table 5). These criteria can be integrated in a classification system with specific weighting factors or multiplication coefficients.
Of course, such a system should preferably be developed by a large expert panel of Western and Eastern endoscopists.
While awaiting further improvements and validations of the abovementioned ESD time prediction models, we suggest the following tips for time scheduling in current Western ESD practice. As a first screening, we recommend using the Eastern model to determine which procedure are very likely or unlikely to be completed within 120 minutes. The 120-minute cut-off is proposed because in this case the lowest ("Easy") and highest ("Very difficult") categories both have a >80% chance of ESD completion within and outside the cut-off time, respectively. Thus, "Easy" procedures should be scheduled for less than 2 hours, and "Very difficult" procedures for 2 hours at least. To determine how much longer or shorter a procedure would take than the cut-off time, and for the "Intermediate" or "Difficult" procedures, we recommend using the cESD-TIME formula as main guidance. In our opinion, the predicted duration from the formula can be used to determine the optimal amount of scheduled time.
The main limitations of our study are related to the retrospective design. As a result, some relevant data (e.g. reasoning behind current time scheduling practice, size of the margin of normal tissue taken along for each ESD) and potentially important pre-procedural predictors (e.g. flexibility of the colon, maneuverability-related factors mentioned in Table 5) were unavailable for analysis. The latter issue particularly impacted the clinical value of the cESD-TIME formula, as we were only able to develop a formula with moderate performance (R 2 of ~60%) based on the data that was available.
Moreover, it was unclear how accurate certain variables were assessed that can be prone to interobserver variability (e.g. tumor size, circumference). This may have decreased the model's predictive accuracy but increased its generalizability as our data reflects daily practice in which inaccurate assessments may sometimes occur. Thirdly, the findings of our study may not be fully applicable to right-sided lesions due to the relatively small proportion of such lesions in our cohort (7.8%).
However, it is important to note that these lesions are typically not recommended as initial cases for novice ESD endoscopists (the primary target audience of the cESD-TIME formula) because the risk of perforation during right-sided ESDs in European practice can be relatively high (up to ~20%) [22].
Lastly, no traction device-assisted ESD techniques were used in our cohort. This is relevant because these techniques may markedly change the nature of a procedure [26], and thereby also the factors determining ESD duration. Therefore, our findings may not be valid for traction device-assisted ESD procedures.
To conclude, by validating a previous Eastern prediction model and developing the new cESD-TIME formula, we established 2 useful prediction tools that could aid time scheduling of colorectal ESDs in Western settings. These tools have the potential to considerably improve time planning accuracy, as well as facilitating more widespread implementation of colorectal ESD at Western endoscopy centers.  Table 7 Values are n (%) unless otherwise defined.
CRC: colorectal cancer, ESD: endoscopic submucosal dissection, SD: standard deviation Table 2. ESD-related outcomes (n=435 procedures) *Numbers of missing values per center are shown in Supplementary Table 7 Values are n (%) unless otherwise defined.
FU: follow-up, SD: standard deviation Table 4. cESD-TIME formula to predict ESD duration in minutes ESD: endoscopic submucosal dissection, LST-NG: laterally spreading tumor with a non-granular surface pattern Table 5. Proposed assessment criteria for classifying endoscopic maneuverability for colorectal ESD. These criteria can be integrated in a scoring system with specific weighting factors or multiplication coefficients. ESD: endoscopic submucosal dissection

Clinical parameters and definitions
The following parameters were retrospectively retrieved from electronical medical records: Only explicit mentions of parameters were recorded. In case a certain parameter was missing, we classified it as such and did not consider it to be "absent", "negative" or equivalent. Unless there were clear signs of missingness not at random (MNAR; e.g. certain categories are completely missing, only extreme values are being reported), we assumed that missing values were not MNAR.

ESD procedure
In the Dutch ESD cohort, all assessments and ESDs were performed by 6 experienced interventional endoscopists (WdG, ADK, PD, LMG, JCH, JJB) who followed extensive training in ESD (tutorial courses and animal in vivo training). None of the endoscopists had prior experience in ESD before. In the early phase of the study period (2011 till ~2014-2015), ESD was performed according to the conventional method as described previously 8 . Thereafter, ESD was mostly carried out according to the pocket-creation 9 or tunneling method 10 . The perimeter of the lesion was not routinely marked before starting the dissection. The adopted ESD method was not routinely recorded in the endoscopy reports.
In Predicted probabilities could not be calculated because the intercept of the Eastern prediction model was not reported in the original paper. Therefore, calibration of the model also could not be evaluated by a calibration plot (which presents the predicted probability against the observed risk).

Statistical analyses
All statistical analyses were performed using R v4.1.2. Nominal and ordinal variables were expressed as frequencies and percentages, and continuous variables as means and standard deviations (SD).
Pearson's chi-square test was used to compare categorical data. Continuous variables were compared using a one-way analysis of variance. A p-value of <0.05 was considered statistically significant.
Multiple imputation by chained equations (mice package, 10 datasets) was used to address missing data while respecting the correlation structure. Details on the extent of missing data per variable are provided in Supplementary Table 7.
After imputation, the performance of current time planning practice was quantified by calculating the proportion of explained variance (R 2 ). Pooling of the R 2 was done according to Rubin's Rules 12 using the pool.r.squared function. The performance of the Eastern prediction model for ESD For the development of the cESD-TIME formula, no formal sample size calculation was performed.
This was because the number of subjects per predictor (SPP) was a priori expected to be much larger (>20-30 SPP) than rules of thumb (2 SPP) which have been proposed for adequate estimation of regression coefficients 13,14 .
The cESD-TIME formula was developed using the rms package. Continuous variables were winsorized at the 1 st and 99 th percentile before imputation 15 . The cESD-TIME formula was built using lower than the optimism-corrected R 2 , simple recalibration methods such as calibration-in-the-large and slope recalibration of the linear predictor were considered first before using more advanced methods such as model revision and extension.
The online calculator was built using the shiny package, and Bayesian estimates and prediction intervals were created using the brms package and the predict function. All model parameters were set as non-informative priors.

Transformations of a previous Eastern prediction model
Linear regression showed that the original Eastern model explained 45% of the variance of the ESD durations (95%-CI: 38-52%). After transformation of the dichotomized quantitative variables of the Eastern model to continuous variables and non-linear fitting of these variables, the model's performance increased to 61% (full model with regression coefficients shown in Supplementary   Table 3). Including all locations from the Eastern model (cecum, dentate line, flexure) as 3 separate yes/no variables did not further increase the R 2 (61%, 95%-CI: 54-68%). Moreover, the performance remained unchanged when categorizing tumor location into rectum, left and right hemicolon (R 2 =61%, 95%-CI: 54-67%). Lastly, ungrouping the combined item "morphology" into its 2 separate components (gross morphology: protruding, sessile or flat; granularity: granular or non-granular surface) also did not considerably improve the model's performance (R 2 =62%, 95%-CI: 55-67%).
Endoscopic maneuverability as potential predictor in the cESD-TIME formula Endoscopic maneuverability has been proposed as a crucial determinant of ESD complexity and duration [17][18][19] . In the analysis cohort, maneuverability was subjectively evaluated and reported in 49% of cases. After imputation and backward selection, maneuverability was included as predictor in the resulting model (β=19, p<0.001; Supplementary Table 8), together with the other 6 variables of the cESD-TIME formula. The R 2 of this exploratory model was 65% (95%-CI: 59-71%).

Supplementary Tables
Supplementary Table 1 Recurrence to be resected 7 (3.5)    *Internal-external cross-validation was not performed for endoscopist #4 because the number of procedures was relatively low (n=8), which resulted in unreliable performance estimates.    Table 7 Values are n (%) unless otherwise defined.

Supplementary Figure Legends
ESD: endoscopic submucosal dissection, SD: standard deviation  ESD: endoscopic submucosal dissection, LST-NG: laterally spreading tumor with a non-granular surface pattern Table 5. Proposed assessment criteria for classifying endoscopic maneuverability for colorectal ESD. These criteria can be integrated in a scoring system with specific weighting factors or multiplication coefficients.