Novel indicator for colonoscopy insertion difficulty and its application in predictive modelling: a multicentre prospective study protocol in China

Abstract

Background

Colonoscopy is a cornerstone for the screening and diagnosis of colorectal diseases. Colonoscopy insertion difficulty is associated with examination quality, procedural efficiency and patient comfort. Cecal intubation time (CIT) is commonly used as a surrogate measure of insertion difficulty; however, it is substantially influenced by endoscopist experience, equipment characteristics and institutional factors, limiting its objectivity and comparability. This study will develop and validate a standardised colonoscopy insertion time (SCIT) metric that better reflects patient-related procedural difficulty while minimising non-patient-related variability. Additionally, we will identify factors associated with insertion difficulty and develop predictive models using machine learning techniques.

Methods and analysis

This multicentre prospective observational study will enrol approximately 3000 adults undergoing sedated colonoscopy across one primary centre and five participating centres. CIT will be prospectively recorded and analysed in relation to patient characteristics, endoscopist experience, colonoscope type and study centre. Three standardisation approaches, including Z-score standardisation, median-based standardisation and min-max normalisation, will be applied to generate SCIT. The primary outcome will be SCIT, and the optimal standardisation method will be selected based on its ability to reduce variability attributable to operator-related, equipment-related and centre-related factors while preserving clinically meaningful associations with patient characteristics. Mixed-effects models will be used to account for clustering by endoscopist and study centre. Based on the selected SCIT metric, machine learning models will be developed using preprocedural clinical variables to predict insertion difficulty. Model performance will be evaluated using measures of discrimination, calibration and predictive accuracy.

Ethics and dissemination

The study received initial ethical approval from the Ethics Committee of the Second Hospital of Jilin University (Approval No. 2024-467-1) on 4 November 2024. Subsequent protocol amendments were approved on 25 March 2026. All participants will provide written informed consent prior to enrolment. Patient privacy and data confidentiality will be strictly maintained. The results will be disseminated through SCI-indexed journals and presentations at national and international academic conferences to inform future research and clinical practice related to colonoscopy performance evaluation.

Trial registration number

NCT07228715.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This multicentre prospective study enhances the generalisability of the findings by including colonoscopy procedures performed across multiple tertiary medical centres. Standardisation of colonoscopy insertion time will reduce operator-related and device-related variability.

• A relatively large and well-characterised cohort will be included, involving endoscopists with different levels of experience and multiple colonoscope systems.

• A stepwise analytical framework will be adopted, progressing from raw insertion time analysis to standardised metrics, followed by multivariable statistical analysis and machine learning-based modelling.

• The machine learning models developed in this study will be exploratory and based on available clinical variables; external validation in independent datasets will be required to confirm their generalisability and predictive performance.

Background

With the continuous improvement of medical technology and the widespread adoption of minimally invasive procedures, sedated colonoscopy has become an important approach for the screening, diagnosis and management of gastrointestinal diseases.¹ Accurate and efficient completion of colonoscopy is crucial for the detection of colorectal polyps, adenomas and early-stage colorectal cancer.^{2 3}

The overall quality of colonoscopy examination is closely related to the difficulty of colonoscope insertion. Patients with higher insertion difficulty often have poorer bowel preparation, require longer withdrawal time and may present a reduced adenoma detection rate (ADR), thereby negatively affecting the effectiveness of screening and diagnosis.^4–6

Procedural difficulty varies greatly among patients due to anatomical, physiological and clinical factors—such as waist-to-height ratio (WHtR), prior abdominal surgery, intestinal adhesions and abnormal colon morphology.^{7 8}

Beyond its impact on examination outcomes, colonoscopy insertion difficulty is also closely associated with patient experience. Greater difficulty often leads to prolonged procedure duration and increased patient discomfort. Therefore, accurate prediction and objective evaluation of colonoscopy insertion difficulty are clinically important.

Currently, the evaluation of colonoscopy insertion difficulty primarily depends on the subjective assessment and experience of the endoscopist,^{9 10} which lacks objectivity and may limit consistency across clinicians and institutions. The cecal intubation time (CIT)—the time required for the colonoscope to reach the cecum—has been proposed as an indirect quantitative measure of insertion difficulty. Previous studies have shown that CIT is associated with patient-related factors such as body mass index (BMI), prior abdominal surgery history and other baseline characteristics. However, existing research has often overlooked non-patient-related influences on CIT, such as endoscopist experience and colonoscope type or performance.^{5 7 8} This limits the objectivity and reproducibility of CIT as a reliable indicator, highlighting the need for a more standardised and scientific evaluation system for colonoscopy insertion difficulty.

To address these challenges, this study will introduce a novel approach to objectively assess colonoscopy insertion difficulty by applying three different standardisation methods—Z-score standardisation, median standardisation and min–max normalisation to each set of CIT data (one group containing approximately 200 cases, completed by one doctor using the corresponding equipment), respectively. If a predefined standardisation subgroup contains fewer than 100 observations, adjacent subgroups with comparable operator and equipment characteristics will be combined to ensure stability of standardisation estimates. These methods help eliminate distributional bias, reduce the impact of outliers and minimise variability arising from operator experience or equipment differences. This approach aims to enable standardised colonoscopy insertion time (SCIT) to more accurately reflect true procedural difficulty, independent of subjective or external influences.

Furthermore, this study will not only focus on standardisation but also explore the integration of machine learning techniques into the evaluation process. By collecting comprehensive patient information, including demographic characteristics, medical history and bowel preparation quality, subsequent analyses will use machine-learning algorithms to develop predictive models for colonoscopy insertion difficulty. These models are expected to enable preprocedural prediction of SCIT and insertion complexity, allowing clinicians to anticipate procedural challenges, allocate cases appropriately and tailor procedural strategies for individual patients.

The primary objective of this study is to develop and validate a SCIT metric that more accurately reflects patient-related insertion difficulty while minimising variability attributable to endoscopist experience, equipment characteristics and centre-related factors. The secondary objective is to identify clinical and demographic factors associated with SCIT. Finally, based on the validated SCIT metric, machine learning models will be developed to predict insertion difficulty using preprocedural variables. Through this stepwise framework, the study will aim to establish a standardised and reproducible approach for assessing colonoscopy insertion difficulty and to provide a foundation for individualised procedural planning and quality improvement in endoscopy practice.

Methods and analysis

Study design

This study is designed as a multicentre prospective observational study aimed at developing and validating a standardised indicator for assessing colonoscopy insertion difficulty (figure 1). A total of 3000 patients undergoing colonoscopy will be consecutively enrolled across multiple participating centres, including a main centre and 5 subcentres. 15 endoscopists will participate in the study (10 from the main centre and 5 from subcentres), representing different levels of clinical experience. Colonoscopy examinations will be performed using commonly used endoscope systems, including Olympus CF-HQ290I, PCF-Q260JI, as well as other devices available in subcentres (eg, SonoScape EC-550 and Fujifilm EC-760R-V/M), allowing evaluation of equipment-related variability.

Figure 1. Study flow diagram. ANN, artificial neural network; AUC, area under the curve; BBPS, Boston Bowel Preparation Scale; BMI, body mass index; CIT, cecal intubation time; CV, cross-validation; MAE, mean absolute error; RMSE, root mean square error; SHAP, Shapley Additive Explanations; SMOTE, Synthetic Minority Over-sampling Technique; SVM, support vector machine; WHtR, waist-to-height ratio.

Among the enrolled population, approximately 200 patients from the main centre will undergo repeat colonoscopy within 3 months, performed by the same endoscopist using different colonoscope models (CF-HQ290I vs PCF-Q260JI), enabling within-subject comparison of device-related effects on colonoscopy insertion time.

For the repeat-colonoscopy subgroup, the order of the two colonoscope models (CF-HQ290I vs PCF-Q260JI) will be randomised using a computer-generated sequence with a 1:1 allocation ratio. The 3-month interval between procedures was chosen to minimise carry-over effects while maintaining clinical feasibility. Bowel preparation quality will be assessed using the Boston Bowel Preparation Scale for both procedures and included as a covariate in the paired analysis to control for potential differences in preparation quality between sessions.

Clinical and procedural data will be prospectively collected for all patients, including CIT, demographic characteristics (age, sex, height), and relevant clinical variables such as surgical history. The study will first analyse raw CIT to evaluate its associations with patient-related factors, as well as its variability across physician experience levels, endoscope types, and participating centres. Procedural variables including bowel preparation quality, sedation status, assistant involvement, patient position changes during insertion, and abdominal pressure application will be recorded whenever available.

Subsequently, CIT will be standardised using multiple statistical approaches to generate SCIT. The performance of SCIT will then be evaluated by reassessing its associations with clinical variables and by determining whether standardisation reduces variability attributable to operator experience, equipment, and centre-related factors.

Finally, based on the optimised standardised metric, machine learning-based predictive models will be developed using clinical variables to estimate SCIT and colonoscopy insertion difficulty. These models aim to enable preprocedural prediction of procedural complexity and support individualised clinical decision-making.

Patient populations

This multicentre prospective study will compare the CIT required for colonoscope insertion to the cecum among physicians with different levels of experience. The study will be conducted at the Second Hospital of Jilin University and five other participating centres from 1 July 2025 to 31 August 2026. Adult patients undergoing colonoscopy at these centres will be considered for inclusion.

Sample size considerations

As this study is designed to develop and validate a standardised metric and predictive models rather than to test a specific hypothesis, sample size was determined based on methodological requirements for regression modelling and machine learning.

First, for multivariable mixed-effects regression, we anticipate including approximately 15–20 candidate predictors. Following the events-per-variable rule of thumb, a minimum of 10–20 events per predictor is recommended for stable estimation, which would require 150–400 complete procedures. However, considering the clustered structure (multiple endoscopists and centres), anticipated missing data (up to 10%–15%), and the need for subgroup analyses (eg, by centre and endoscopist experience), we conservatively target approximately 3000 participants. This sample size also allows robust estimation of random-effects variance components.

Second, for machine learning model development, current guidelines recommend a minimum sample size of at least several hundred to thousands of observations to reduce overfitting and improve generalisability. A cohort of approximately 3000 patients provides adequate power for training (70%, ~2100), validation (15%, ~450) and testing (15%, ~450) sets, enabling stable model development and internal validation. Third, the repeat-colonoscopy subgroup (~200 patients) is designed to provide sufficient power (≥80%) to detect a moderate effect size (Cohen’s d=0.3) in paired comparisons. Based on these considerations, a target sample size of 3000 participants was deemed appropriate.

Definition of CIT

CIT is defined as the time interval from insertion of the colonoscope tip through the anal verge to successful visualisation of the cecal landmarks, specifically the appendiceal orifice and the ileocecal valve.

Timing will be recorded automatically by the endoscopy reporting system (ERS) whenever available. For centres where ERS-based automatic timing is not available, trained study personnel will record the time using a standardised stopwatch protocol. To ensure accuracy, a validation substudy will be conducted in the first 100 patients at each centre to compare manual and automatic timing, with discrepancies >10 s triggering retraining.

For patients in whom cecal intubation is not achieved, CIT will be recorded as missing for the primary analysis, but the procedure will be documented as incomplete, and the patient’s clinical characteristics will be described separately.

Inclusion criteria

• Adult patients (aged 18 years and above) who are scheduled to undergo a sedated colonoscopy at the Second Hospital of Jilin University and the other five participating centres.

• Patients who voluntarily agree to participate in the study and provide informed consent.

• Patients who are eligible for sedated colonoscopy procedures.

• Patients undergoing scheduled sedated colonoscopy.

Exclusion criteria

• Patients who are under 18 years of age or do not meet the eligibility criteria for sedated colonoscopy.

• Patients with incomplete colonoscopy will not be included in the primary SCIT analysis because CIT cannot be reliably measured. However, these cases will be recorded separately, and their clinical characteristics will be described descriptively.

• Patients with a history of severe cardiovascular, respiratory or other systemic diseases that would contraindicate colonoscopy.¹¹

• Patients with known or suspected colorectal malignancies or other conditions that may affect colonoscopy insertion (eg, significant bowel obstruction, severe inflammatory bowel disease).

• Pregnant women or breastfeeding mothers who are not suitable for the procedure.

• Patients who are unable or unwilling to provide informed consent to participate in the study.

Primary outcome

The primary outcome of this study is the SCIT derived from raw CIT using one of three candidate standardisation approaches. SCIT will be treated as a continuous variable in the primary analyses. The optimal standardisation method will be selected based on its ability to: (1) reduce between-endoscopist, between-equipment and between-centre variance as quantified by intraclass correlation coefficient (ICC); (2) preserve or improve the explanatory power of patient-related factors (assessed by adjusted R² in mixed-effects models) and (3) demonstrate stability across standardisation subgroups through cross-validation. The selected SCIT metric will serve as the primary endpoint for subsequent association analyses and for machine learning-based prediction of continuous insertion time.

Secondary outcomes

• Raw CIT.

• Identification of patient-related factors associated with SCIT.

• Assessment of the effects of endoscopist experience, colonoscope type and study centre on both CIT and SCIT.

• Development and validation of machine learning models for predicting SCIT using preprocedural clinical variables.

• Exploratory classification of difficult colonoscopy, defined a priori as SCIT values above the 75th percentile of the study population.¹²

Statistical analysis

All statistical analyses will be performed using SPSS 31.0 and R software. Statistical significance will be defined as a two-sided p<0.05. Descriptive statistics will be used to summarise baseline characteristics. Continuous variables will be reported as mean±SD or median (IQR), depending on data distribution, while categorical variables will be presented as frequencies and percentages. Normality will be assessed using the Shapiro-Wilk test and graphical methods. Missing data will be evaluated before analysis. If the proportion of missing data is less than 5%, complete-case analysis will be performed. If the proportion exceeds 5%, multiple imputation methods will be considered.

If multiple imputation is applied, the imputation model will include all variables used in the primary analysis (including CIT/SCIT, patient demographics, anthropometric measures, surgical history, bowel preparation score and procedural variables) as well as auxiliary variables (eg, centre, endoscopist experience level, equipment type). Imputation will be performed using the Multivariate Imputation by Chained Equations algorithm with 20 imputed datasets. Results across imputed datasets will be pooled using Rubin’s rules.

The primary outcome of this study is SCIT. Three candidate standardisation methods, including Z-score standardisation, median-based standardisation and min–max normalisation, will be applied to CIT. The optimal standardisation method will be selected according to its ability to reduce variability attributable to endoscopist experience, colonoscope type and study centre while preserving clinically meaningful associations with patient-related characteristics (table 1).

Table 1. Comparison of common standardisation and normalisation methods.

Method	Core formula/concept	Output range	Robustness to outliers	Computational complexity	Applicable scenarios
Z-score standardisation	z =(X−μ)/σ	Mean=0, SD=1	Moderate	High	Suitable for most machine learning algorithms; effective for approximately normal data but sensitive to outliers.
Median standardisation	z =(X−XMedian)/MAD	No fixed range	High	Moderate	For non-normal distributions or in the presence of severe outliers
Min–Max normalisation	x=(X−XMin)/(XMax−XMin)	(0, 1) or (−1, 1)	Low	Low	For data requiring a fixed range and without outliers

x: the original data value, μ: the mean of the dataset, σ: the SD of the dataset. XMin and XMax: the minimum and maximum values of the dataset. XMedian: the median of the dataset, and MAD, which is calculated as: MAD=XMedian(∣Xi−XMedian∣).

MAD, median absolute deviation.

To account for the multicentre design and clustering of observations, linear mixed-effects models will be used. Study centre and endoscopist will be treated as random effects, whereas patient-related variables will be treated as fixed effects. The effects of endoscopist experience, colonoscope type and study centre on both CIT and SCIT will be evaluated within this framework.

Associations between SCIT and clinical variables will first be explored using univariable analyses. Variables with p<0.10 in univariable analyses, together with clinically relevant variables identified from previous literature, will be entered into multivariable mixed-effects regression models. Results will be reported as regression coefficients with 95% CIs. For the repeat-colonoscopy subgroup, paired analyses and mixed-effects models accounting for repeated measurements will be performed to assess differences between colonoscope types within the same individual.

Machine learning models will be developed using preprocedural clinical variables to predict SCIT. Candidate preprocedural predictors will include age, sex, BMI, WHtR, history of abdominal surgery, history of pelvic surgery, history of constipation and colonoscopy indication (screening, surveillance or diagnostic). Additional variables will be considered based on univariable screening (p<0.10). For machine learning analyses, all standardisation parameters used to derive SCIT will be estimated using the training dataset only and subsequently applied to validation and testing datasets to avoid information leakage.

Two prediction targets will be explored: (1) prediction of continuous SCIT values; (2) classification of difficult colonoscopy, defined as SCIT above the 75th percentile. Candidate algorithms will include multiple linear regression, random forest, support vector machine and artificial neural network models. The dataset will be randomly divided into training (70%), validation (15%) and testing (15%) datasets. Fivefold cross-validation will be used during model development. Hyperparameters will be optimised using grid-search procedures.

For the classification task, class imbalance will be addressed using the Synthetic Minority Over-sampling Technique applied exclusively to the training set to avoid data leakage. Model performance for classification will be evaluated using: (1) discrimination metrics including area under the receiver operating characteristic curve, sensitivity, specificity, positive predictive value and negative predictive value and (2) calibration metrics including the Brier score and calibration plots with Hosmer-Lemeshow goodness-of-fit test. Calibration-in-the-large and calibration slope will also be reported.

Model performance will be evaluated using multiple complementary metrics. For regression models, predictive performance will be assessed using the coefficient of determination (R²), root mean square error and mean absolute error. Calibration performance will also be assessed using calibration plots. Feature importance analyses, including permutation importance and Shapley Additive Explanations values, will be performed to improve model interpretability.

Sensitivity analyses

To address potential selection bias due to exclusion of incomplete colonoscopy procedures, two sensitivity analyses will be performed: (1) baseline characteristics of patients with incomplete procedures will be compared with those with complete procedures using standardised differences and (2) a worst-case scenario analysis will be conducted by assigning the maximum observed CIT value to incomplete procedures and re-evaluating the primary associations in mixed-effects models. Results of these sensitivity analyses will be reported alongside the primary analyses to assess the robustness of findings.

Where multiple comparisons are performed, appropriate corrections (eg, Bonferroni adjustment) will be applied.

Patient and public involvement

Patients and/or the public were not involved in the design, conduct, reporting or dissemination plans of this study protocol. The research question and outcome measures were developed based on clinical evidence and methodological considerations rather than patient priorities. The study involves human participants, and results will be disseminated through peer-reviewed publications and academic conferences.

Selection of the optimal standardisation method

The optimal standardisation method will be selected using a three-criterion framework: Reduction of non-patient-related variance: Linear mixed-effects models with random intercepts for endoscopist and centre will be fitted for raw CIT and for each SCIT variant. The proportion of total variance attributable to endoscopist and centre (ICC) will be compared. The method yielding the greatest reduction in ICC will be favoured.

Preservation of patient-related associations: For each standardisation method, the adjusted R² of the fixed patient-related effects (eg, age, sex, BMI/WHtR, surgical history) in the mixed-effects model will be compared. The method that maintains or improves explanatory power relative to raw CIT will be preferred.

Stability across subgroups: The standardisation method’s performance will be assessed via leave-one-subgroup-out cross-validation to ensure that results are not overly influenced by any single endoscopist-equipment group.

The final selection will be based on a composite assessment of these three criteria, with primary emphasis on criterion 1 (variance reduction).

Ethics and dissemination

This study will be conducted in accordance with the principles of the Declaration of Helsinki and relevant national ethical guidelines for medical research involving human participants. The original study protocol received ethical approval from the Ethics Committee of the Second Hospital of Jilin University (Approval No. 2024-467-1) on 4 November 2024.

Following the initial approval, amendments were submitted to further refine and expand the study protocol. The main amendments included the expansion of the study from a single-centre design to a multicentre prospective study involving five additional participating centres, incorporation of the repeat-colonoscopy subgroup for assessment of equipment-related variability, and refinement of the statistical analysis plan including the development of the SCIT metric and machine-learning-based predictive models. The amended protocol received additional ethical approval from the Ethics Committee of the Second Hospital of Jilin University on 25 March 2026.

Ethical approval was also obtained from the institutional review boards or ethics committees of all participating centres before study initiation. The study was registered on ClinicalTrials.gov (NCT07228715) in October 2025. All participants will provide written informed consent before enrolment. Patient privacy and confidentiality will be strictly protected. All collected data will be anonymised and stored securely in password-protected databases accessible only to authorised members of the research team.

Protocol version

This manuscript describes the amended protocol V.1.1, dated 25 March 2026.

Data monitoring

Data monitoring in this study will be implemented to ensure the accuracy, completeness, and integrity of all collected data throughout the research process. The monitoring procedures will follow Good Clinical Practice guidelines and institutional standards.

Regular audits will be conducted by the study coordinator to identify and correct discrepancies or missing data. Data consistency checks and validation rules will be programmed into the database to minimise entry errors. Any corrections made to the data will be fully documented and traceable.

Interim data reviews will be conducted periodically to assess the progress of patient recruitment, data completeness and adherence to the study protocol. Any protocol deviations or inconsistencies identified will be addressed promptly and reported to the principal investigator and the ethics committee, if necessary.

All study data will be anonymised and stored securely. Only authorised personnel will have access to identifiable patient information. Backup copies of all data will be maintained on encrypted drives to prevent data loss. Before statistical analysis, a final data validation process will be performed to confirm data completeness and accuracy. Any outliers or inconsistencies will be reviewed and resolved according to the predefined data management plan.

Discussion

With the continuous advancement of medical technology, colonoscopy performed under sedation has become widely adopted as a key approach for the screening and early detection of intestinal diseases.^{13 14} Accurate assessment of colonoscope insertion difficulty is clinically important, as increased difficulty is associated with prolonged procedure duration, greater patient discomfort and reduced examination quality, ultimately affecting diagnostic yield. To enhance the generalisability and clinical applicability of findings, the present study is designed as a multicentre prospective study, enabling data collection across diverse institutions and clinical environments.

This multicentre prospective study introduces multiple standardisation approaches to adjust CIT, aiming to establish a reproducible and comparable metric for colonoscopy insertion difficulty. Specifically, three standardisation methods—Z-score standardisation, median-based standardisation and min–max normalisation—will be applied to CIT data to reduce distributional bias, minimise the influence of operator experience and equipment variability, and mitigate the impact of outliers. The resulting standardised metric, referred to as SCIT, is expected to more accurately reflect intrinsic procedural difficulty attributable to patient-related characteristics such as WHtR, colonic anatomy, intestinal adhesions and structural variations.

Using prospectively collected multicentre data, this study will evaluate whether SCIT effectively reduces variability caused by non-patient-related factors while preserving meaningful associations with patient characteristics. This approach aims to establish an objective and standardised framework for quantifying colonoscopy insertion difficulty.

Clinically, SCIT may support preprocedural risk assessment and individualised patient management. By analysing associations between SCIT and pre-procedural symptoms (eg, abdominal pain, bloating, constipation) as well as anthropometric parameters such as WHtR and abdominal wall compliance, clinicians may be able to identify patients at increased risk of difficult colonoscope insertion. This may facilitate tailored procedural strategies, including optimised patient positioning, sedation planning and the use of auxiliary techniques.

In addition, SCIT may serve as an objective indicator of procedural difficulty that enables more accurate evaluation of physician performance. By accounting for case complexity, this metric could support structured training assessment, optimise workload distribution and improve comparability across endoscopy centres.

Colonoscopy insertion difficulty is closely associated with ADR, a key quality indicator of colonoscopy performance.^15–18 Increased difficulty may lead to incomplete examinations and suboptimal mucosal visualisation, potentially reducing lesion detection.^{19 20} The use of SCIT may allow for more rational case allocation and improved procedural efficiency, thereby indirectly contributing to improved ADR and reduced risk of missed lesions.

An important extension of this work is the integration of machine learning techniques. Based on the standardised metric, predictive models will be developed using preprocedural clinical variables to estimate SCIT and insertion difficulty. These models may enable early identification of complex cases, facilitate resource allocation and support data-driven clinical decision-making.

Limitations

Several limitations of this study should be acknowledged. First, although this is a multicentre study, all participating centres are located within a limited geographic region, which may affect the generalisability of the findings to other populations and healthcare systems. External validation using data from more diverse clinical settings will be necessary. Second, as an observational study, residual confounding cannot be completely excluded. Although standardisation methods are applied to minimise the effects of operator experience and equipment variability, other unmeasured procedural factors—such as insertion techniques, assistant involvement and patient positioning—may still influence colonoscopy insertion difficulty. Third, the machine learning models developed in this study are exploratory and dependent on the quality and completeness of the collected clinical variables. Prospective validation and further refinement will be required before these models can be implemented in routine clinical practice. Finally, although SCIT is designed to provide a more objective measure of insertion difficulty, its clinical utility will depend on further validation and its integration with existing quality indicators in endoscopy practice. Exclusion of incomplete procedures may introduce selection bias because these cases may represent the most technically challenging examinations.

The funder had no role in the study design, data collection, data analysis, interpretation of data, manuscript preparation, or the decision to submit the manuscript for publication.