AI-Driven Qualitative Skill Assessment in Laparoscopic Training: A Prospective Observational Study

AI-Driven Qualitative Skill Assessment in Laparoscopic Training: A Prospective Observational Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article AI-Driven Qualitative Skill Assessment in Laparoscopic Training: A Prospective Observational Study Dimitrios Chatziisaak, Moritz Sparn, Pascal Burri, Daniel Krstic, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8801088/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 11 You are reading this latest preprint version Abstract Background: Qualitative assessment of surgical skills is essential for proficiency-based training but remains resource-intensive and difficult to scale due to reliance on expert observation. Automated simulator metrics, while objective, are largely limited to quantitative parameters and fail to capture qualitative aspects of surgical performance. Artificial intelligence–based video analysis may address this gap by enabling standardized and reproducible qualitative assessment. Methods: In this prospective observational study conducted at the 41st Annual Davos Surgical Course (2024), 50 first- and second-year surgical residents performed laparoscopic cholecystectomy on porcine simulation models. Procedures were video recorded, anonymized, segmented, and independently assessed by expert raters and an AI-based model using the Global Operative Assessment of Laparoscopic Skills (GOALS). AI test–retest reliability and agreement with expert ratings were evaluated. Results: AI demonstrated excellent test–retest reliability for the total GOALS score (ICC 0.91) and good reliability across most domains. Agreement between AI and expert raters was excellent for the total score (ICC 0.92) and good to excellent for individual domains, with minimal bias on Bland–Altman analysis. Conclusion: AI-based video analysis enables reliable, reproducible qualitative assessment of laparoscopic surgical skills in a simulation setting and may support scalable integration of qualitative evaluation into surgical training programs and other video-based clinical skill assessment domains. Biological sciences/Computational biology and bioinformatics Health sciences/Gastroenterology Health sciences/Health care Physical sciences/Mathematics and computing Health sciences/Medical research Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Surgical education is undergoing a significant transformation. While the traditional apprenticeship model of education in surgical training has long centered around hands-on experience in the operating room, there is a growing shift toward simulation- and proficiency-based training ( 1 – 3 ). This evolution is driven by the need to enhance patient safety and to provide structured, efficient training environments for surgical residents ( 4 ). With this shift, the importance of objective performance assessment has increased. Simulator-based metrics offer useful quantitative insights but are difficult to interpret within a meaningful clinical context ( 5 ). These systems predominantly rely on automated, quantitative parameters such as task completion time, instrument path length, motion economy, or predefined error detection. While such metrics allow standardized and scalable data collection, they primarily capture efficiency-related aspects of performance and fail to reflect the qualitative nuances of surgical skill, including tissue handling, depth perception, or autonomous decision-making. To address these limitations, structured qualitative assessments such as the Objective Structured Assessment of Technical Skills (OSATS) and the Global Operative Assessment of Laparoscopic Skills (GOALS) have emerged as comprehensive tools for evaluating surgical performance ( 6 , 7 ). These instruments enable a nuanced evaluation of technical and non-technical skills using standardized criteria that better reflect real operative competence. However, the implementation of qualitative skill assessment in routine training remains resource-intensive. It requires the continuous involvement of experienced surgical educators who must either directly observe procedures or retrospectively review operative videos. This process is time-consuming, costly, and inherently limited in scalability. Furthermore, interobserver variability and rater-dependent fluctuations related to experience, workload, or cognitive state introduce additional sources of subjectivity and inconsistency ( 8 , 9 ). Recent advancements in artificial intelligence (AI), particularly in video-based surgical analysis, may offer a promising solution. AI-driven evaluation has the potential to apply qualitative scoring systems such as GOALS in a standardized, unbiased, and highly reproducible manner. By combining automated scalability with clinically meaningful qualitative assessment, AI-based approaches may overcome the limitations of both purely quantitative simulator metrics and human-dependent evaluations ( 10 , 11 ). The objective of this study was to evaluate whether an AI-based model can perform qualitative assessments of surgical performance in accordance with the GOALS criteria. Furthermore, the study aimed to examine the reproducibility and consistency of AI-generated evaluations across laparoscopic simulation videos. Methods Study design This prospective observational study was conducted during the 41st Annual Davos Surgical Course in 2024 ( www.davoscourse.ch ). Demographic data and previous surgical experience were collected using a standardized questionnaire (Supplement 1). The dataset consisted of video recordings of laparoscopic cholecystectomies performed on non-perfused porcine liver models placed in a pelvic trainer. Trainees were assigned to pairs and each participant was allocated 60 minutes of operative time to complete one procedure. All procedures were video recorded, anonymized, and stored for subsequent analysis. All methods were carried out in accordance with relevant guidelines and regulations. We obtained written informed consent from all study participants. This research followed the guidelines and regulations of the WMA Declaration of Helsinki and the Swiss Federal Act on “Research involving Human Beings 810.30” about scientific research in human beings. The Swiss Ethics Committee "swissethics.ch" grants a general waiver for the use of purely anonymized data and an application with approval is not requested. Video segmentation and standardization Due to current limitations in processing long video sequences, each surgical recording was segmented into nine shorter clips. Segmentation was performed prior to any assessment and applied uniformly across all cases. To ensure methodological fairness and avoid contextual bias, the expert panel assessed the same segmented video clips that were provided to the AI model. This approach ensured that both human and AI evaluations were based on identical visual input, preventing advantages related to continuous procedural context. Human expert assessment All anonymized videos were independently assessed by a panel of experienced surgeons (DC, MS, PB, DK, SB) using the Global Operative Assessment of Laparoscopic Skills (GOALS) score ( 7 ). GOALS evaluates laparoscopic performance across five domains—depth perception, bimanual dexterity, efficiency, tissue handling, and autonomy—using a standardized qualitative framework. Raters were blinded to participant identity, training background, and AI-generated results. AI-based assessment Video-based assessments were conducted using a large language model (ChatGPT-4o-mini-high, OpenAI, San Francisco, USA). The model was instructed to apply a rule-based interpretation of the GOALS domains, mapping observable surgical behaviors to predefined qualitative anchors. Output generation was constrained to structured scoring without free-text improvisation. Each video segment was analyzed independently. For every case, the AI generated a structured report including GOALS domain scores and a qualitative performance summary. After each assessment run, the AI session was reset to prevent carry-over or memory effects and to ensure independence between evaluations. Sample size determination An a priori power analysis was performed using a two-way random-effects intraclass correlation coefficient (ICC) model to determine the number of video cases required for interrater reliability (IRR) assessment ( 12 ). Assuming an expected ICC of 0.80 and a minimum acceptable ICC of 0.60, a sample size of 20 videos was estimated to provide 80% power at a significance level of α = 0.05. To increase precision and enable secondary comparisons between AI and the expert panel, 50 videos were included in the final analysis (Supplement 2). Comparative analysis AI-generated GOALS scores were compared with scores assigned by the expert panel. Three reviewers evaluated each AI-generated report for completeness, conformity with the predefined scoring framework, and alignment with the study objectives. Any discrepancies were resolved through panel discussion. Reproducibility To evaluate test–retest reliability, a subset of 15 videos was reassessed by the AI model in a second independent evaluation run. Prior to reanalysis, videos were re-anonymized, and the AI session was reset to eliminate any potential memory effects. This approach enabled assessment of score stability under controlled conditions. Statistical Analysis Statistical analyses were performed in two sequential steps. First, test–retest reliability of AI-generated GOALS scores was assessed by comparing results from two independent evaluation runs. Second, agreement between AI-generated scores and expert panel ratings was analyzed. Intraclass correlation coefficients (ICCs) were calculated to quantify reliability and agreement in accordance with established methodological recommendations ( 12 , 13 ). AI test–retest reliability was assessed using a two-way mixed-effects model with absolute agreement, while agreement between AI-generated scores and expert panel ratings was evaluated using a two-way random-effects model. Bland–Altman analyses were performed to visualize agreement, estimate systematic bias, and assess limits of agreement across the score range. For paired comparisons of GOALS domain and total scores, Wilcoxon signed-rank tests (Z) were applied when normality assumptions were not met ( 14 ). Agreement on checklist items was assessed using Cohen’s kappa. To explore potential sources of systematic bias between AI and expert ratings, a multiple linear regression analysis was conducted using the total score difference as the dependent variable and participant characteristics as independent predictors. All statistical tests were two-sided, and a p-value < 0.05 was considered statistically significant. Results Participants’ demographics Fifty surgical residents participated in the study (Postgraduate Year PGY 1–2), of whom 63% were female. The majority of participants reported limited prior experience in minimally invasive surgery, with ≤ 5 completed cases in laparoscopic appendectomy (73%) and laparoscopic cholecystectomy (80%). No participant reported prior experience in laparoscopic colectomy or hernia repair. All participants completed a pre-course self-assessment questionnaire; six questionnaires were returned incomplete. Reliability of AI-calculated GOALS scores Test–retest reliability of the AI-generated GOALS scores was assessed using two independent evaluation runs of 15 anonymized video cases. The AI demonstrated excellent repeatability for the GOALS total score (ICC 0.91), indicating high overall stability across repeated assessments. Good reliability was observed for the domains of bimanual dexterity, efficiency, and autonomy, whereas depth perception and tissue handling showed lower agreement with wider confidence intervals (Table 1 ). Table 1 Intraclass Correlation Coefficient (ICCs) AI test-retest GOALS ICC 95% CI p-value Interpretation Depth Perception 0.507 [-0.19, 0.716] 0.099 Moderate (Unstable) Bimanual Dexterity 0.814 [0.286, 0.882] 0.002 Good Efficiency 0.726 [0.101, 0.831] 0.011 Moderate–Good Tissue Handling 0.417 [-0.269, 0.673] 0.162 Poor (Unstable) Autonomy 0.795 [0.241, 0.871] 0.003 Good GOALS Total 0.906 [0.562, 0.939] 0.90 = Excellent, ICC 0.75–0.90 = Good, ICC < 0.75 = Needs discussion Paired comparisons between the first and second AI assessment runs revealed no significant differences for most GOALS domains. A small but statistically significant decrease was observed for bimanual dexterity and the total GOALS score on re-evaluation. Bland–Altman analysis demonstrated minimal overall bias between runs and no systematic trend across the score range (Figs. 1 & 2 ). Reliability of human GOALS scores Interrater reliability within the expert panel was excellent across all GOALS domains and subdomains. All intraclass correlation coefficients were statistically significant (p < 0.001). These findings confirm a high level of consistency among expert raters and support the use of the panel scores as a robust reference standard for comparison with AI-generated assessments (Supplement 3). AI-calculated versus human GOALS scores Comparison of AI-generated and expert panel GOALS scores demonstrated strong agreement across domains and for the total score. Median total GOALS scores did not differ significantly between AI and expert ratings. No significant differences were observed for the domains of bimanual dexterity (Z = − 0.946, p = 0.344) and autonomy, whereas statistically significant differences were identified for depth perception (Z = − 3.444, p = 0.001), efficiency (Z = − 2.335, p = 0.020), and tissue handling (Z = − 3.146, p = 0.002), with AI scores tending to be slightly higher than expert ratings (Fig. 3 ). Although statistically significant, the observed domain-level differences corresponded to less than one GOALS point and are unlikely to be educationally meaningful in formative assessment contexts. Inter-method reliability analysis revealed excellent agreement between AI and expert panel assessments for the GOALS total score (ICC 0.92). Domain-level agreement was excellent for bimanual dexterity and autonomy and good to excellent for depth perception, efficiency, and tissue handling (Table 2 ). All ICC estimates were statistically significant (p < 0.001). Table 2 Intraclass Correlation Coefficients (ICCs) Between AI and Expert Panel for GOALS Domains Domain ICC (Average) 95% CI (ICC) F(df1, df2) p-value Depth Perception 0.832 [0.748, 0.895] 5.943 (49, 245) < .001 Bimanual Dexterity 0.903 [0.854, 0.939] 10.292 (49, 245) < .001 Efficiency 0.849 [0.773, 0.905] 6.614 (49, 245) < .001 Tissue Handling 0.88 [0.820, 0.925] 8.333 (49, 245) < .001 Autonomy 0.88 [0.819, 0.925] 8.310 (49, 245) < .001 GOALS Total 0.92 [0.880, 0.950] 12.476 (49, 245) < .001 All ICCs are based on a two-way random-effects model using consistency definition Bland–Altman analysis of the total GOALS score demonstrated minimal mean bias between AI and expert ratings difference (Mean Difference = -1.00) and narrow limits of agreement, with no evidence of proportional bias across the performance range. Domain-specific Bland–Altman analyses showed variable levels of agreement, with the highest consistency observed for autonomy and bimanual dexterity (Fig. 4 ). Predictors of bias between AI and expert ratings To explore potential sources of systematic bias between AI-generated and expert panel GOALS total scores, a multiple linear regression analysis was performed. Participant characteristics including postgraduate year, gender, handedness, training status, and self-reported overall laparoscopic skill level were entered as predictors. Among the evaluated variables, self-reported overall laparoscopic skill level was significantly associated with the magnitude of bias between AI and expert ratings, with higher self-assessed skill levels corresponding to larger negative score differences (β = −0.46, p = 0.008). No significant associations were observed for postgraduate year, gender, handedness, or training status (Table 3 ). The overall regression model explained a modest proportion of variance in score differences (adjusted R² = 0.114). Table 3 Multiple Linear Regression Predicting Bias in AI vs Expert Panel Total GOALS Scores Predictor B SE B β (Beta) t p 95% CI (Lower) 95% CI (Upper) (Constant) -2.39 4.1 — -0.58 0.563 -10.69 5.91 PGY 0.52 0.55 0.15 0.95 0.35 -0.59 1.64 Gender -0.26 1.32 -0.03 -0.2 0.843 -2.92 2.4 Handedness 2.71 3.05 0.14 0.89 0.38 -3.46 8.87 Training Status 2.47 1.75 0.2 1.41 0.167 -1.08 6.02 Overall Lap. Skills -2.71 0.98 -0.46 -2.78 .008* -4.68 -0.74 R² = .217, Adjusted R² = .114, F (5, 38) = 2.11, p = .086, DV = Bias in AI vs. Panel GOALS total score Discussion This study demonstrates that a large language model can reliably assess videos of minimally invasive cholecystectomies performed in a simulation setting using a validated qualitative scoring system. AI-generated GOALS scores showed excellent agreement with expert ratings and high test–retest stability, supporting the feasibility of AI-based qualitative skill assessment in surgical training environments. As surgical training shifts toward proficiency-based curricula, and because surgeons should be trained under meticulous, unbiased supervision and constructive feedback, scalable and meaningful assessment becomes increasingly important ( 2 , 15 ). Prior work has shown that constructive teaching approaches supported by newer technologies can improve training outcomes ( 16 , 17 ). In addition, simulation-based systems often provide automated metrics that are predominantly quantitative and may not reflect the qualitative nuances of operative skill ( 5 ). In this context, our results support the introduction of AI technologies into surgical curricula to facilitate a stepwise transition toward standardized, inclusive, and proficiency-based training while maintaining patient safety and accelerating skill acquisition ( 18 ). While expert raters demonstrated excellent internal consistency, human-centered evaluation remains susceptible to variability in the interpretation of borderline performances and to unconscious bias related to rater experience, workload, or cognitive factors ( 6 , 19 , 20 ). In our study, regression analysis revealed that self-reported overall laparoscopic skill level was significantly associated with the magnitude of disagreement between AI and expert ratings, whereas postgraduate year, gender, handedness, and training status were not. This finding suggests that discrepancies were not driven by formal training level or demographic factors, but may reflect differences in how performance thresholds are implicitly applied across experience strata. Importantly, this pattern does not indicate systematic bias across the overall performance range, but rather highlights the sensitivity of qualitative assessment at higher skill levels. The relative stability of AI-generated scores across repeated evaluations suggests that AI-based assessment may apply more consistent evaluative criteria, which could contribute to increased standardization and fairness in formative surgical assessment, particularly when evaluating advanced or borderline performances. AI-based assessments should be viewed as decision-support tools rather than autonomous evaluators, complementing human supervision especially in summative or high-stakes contexts. According to Connor et al., mentorship provides a safe space for reflection and support, enabling exploration of strengths and weaknesses, self-challenge, insight generation, and goal orientation ( 21 ). AI-supported mentorship programs may facilitate structured learning pathways, provide timely feedback, and contribute to continuous and fair evaluation ( 22 ). This assumes that such tools can help trainees adhere to rigorous standards, acquire requisite skills, and be treated fairly and inclusively ( 23 ). Several limitations should be acknowledged. Although the LLM appeared largely impartial with respect to most participant characteristics, it may undervalue performance in individuals with greater experience or higher self-reported proficiency. This may reflect limitations in recognizing nuanced skill cues or stricter implicit thresholds in higher-skill subgroups, underscoring the need for further calibration to ensure fairness and validity. In addition, a carefully prepared prompt is required to generate qualitative results, and fully continuous human-like video analysis is not yet feasible. Current workflows rely on manual clip segmentation, which may introduce human influence, and the analysis depends on automated timestamped logging of actions. Future studies should evaluate these tools in larger populations, refine standardized metric extraction, define appropriate benchmarks, and test performance across diverse skill levels and settings. Advancing these automated procedures - and strengthening collaboration between surgeons and software engineers - will be essential for robust, fair, and scalable deployment in surgical education. Conclusion This study demonstrates that a large language model can reliably assess laparoscopic surgical performance in a simulation setting with expert-level agreement and high test–retest stability. By enabling standardized qualitative skill assessment without continuous expert presence, AI-based video analysis addresses key limitations of current evaluation approaches. While further optimization is required for selected skill domains and validation in larger cohorts, this approach has the potential to support scalable, reproducible, and resource-efficient integration of qualitative assessment into proficiency-based surgical training programs. Declarations Competing interests: All authors declare no financial or non-financial competing interests. Ethics approval and consent to participate All methods were carried out in accordance with relevant guidelines and regulations. We obtained written informed consent from all study participants. This research followed the guidelines and regulations of the WMA Declaration of Helsinki and the Swiss Federal Act on “Research involving Human Beings 810.30” about scientific research in human beings. The Swiss Ethics Committee "Swissethics.ch" grants a general waiver for the use of purely anonymized data and an application with approval is not requested. Funding: This study was supported by the national "Proficiency" research project, funded by the Swiss Innovation Agency Innosuisse in 2021 as one of 15 flagship initiatives. https://surgicalproficiency.ch/ Author Contribution Conception and design of the work: DC, MS, DH, SBData acquisition and analysis: DC, MS, PB, DK, AV, BS, DH, SBInterpretation of data: DC, MS, PB, DK, AV, BS, DH, SBDrafted the manuscript: DC, SBSubstantively revised the manuscript: DC, MS, PB, DK, BS, DH, SB Acknowledgement The authors like to thank the congress organization of the "Davos Course" (davoscourse.ch) for the organizational support. Data Availability The datasets generated and/or analyzed during the current study are included in this published article and its supplementary information files or are available from the corresponding author on request. References Gallagher AG, Ritter EM, Champion H, Higgins G, Fried MP, Moses G, et al. Virtual reality simulation for the operating room: proficiency-based training as a paradigm shift in surgical skills training. Ann Surg. 2005;241(2):364–72. Shaker D. Cognitivism and psychomotor skills in surgical training: from theory to practice. Int J Med Educ. 2018;9:253–4. Shahrezaei A, Sohani M, Taherkhani S, Zarghami SY. The impact of surgical simulation and training technologies on general surgery education. BMC Med Educ. 2024;24(1):1297. Burri P, Chatziisaak D, Sparn M, Bischofberger S. [Learn playfully, operate seriously: The new era of surgical training]. Chirurgie (Heidelb). 2025;96(3):236–42. Seymour NE, Gallagher AG, Roman SA, O'Brien MK, Bansal VK, Andersen DK, et al. Virtual reality training improves operating room performance: results of a randomized, double-blinded study. Ann Surg. 2002;236(4):458–63; discussion 63 – 4. Martin JA, Regehr G, Reznick R, MacRae H, Murnaghan J, Hutchison C, et al. Objective structured assessment of technical skill (OSATS) for surgical residents. Br J Surg. 1997;84(2):273–8. Vassiliou MC, Feldman LS, Andrew CG, Bergman S, Leffondré K, Stanbridge D, et al. A global assessment tool for evaluation of intraoperative laparoscopic skills. Am J Surg. 2005;190(1):107–13. Faherty A, Counihan T, Kropmans T, Finn Y. Inter-rater reliability in clinical assessments: do examiner pairings influence candidate ratings? BMC Med Educ. 2020;20(1):147. Tavares W, Eva KW. Exploring the impact of mental workload on rater-based assessments. Adv Health Sci Educ Theory Pract. 2013;18(2):291–303. Hashimoto DA, Rosman G, Rus D, Meireles OR. Artificial Intelligence in Surgery: Promises and Perils. Ann Surg. 2018;268(1):70–6. Lillemoe HA, Geevarghese SK. Stopping the Progression of Moral Injury: A Priority During Surgical Training. Ann Surg. 2021;274(6):e643-e5. Walter SD, Eliasziw M, Donner A. Sample size and optimal designs for reliability studies. Stat Med. 1998;17(1):101–10. Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15(2):155–63. Miller E. The signed-rank (Wilcoxon)test. Lancet. 1969;1(7590):371. Chikwe J, de Souza AC, Pepper JR. No time to train the surgeons. Bmj. 2004;328(7437):418–9. Sparn MB, Teixeira H, Chatziisaak D, Schmied B, Hahnloser D, Bischofberger S. Virtual reality simulation training in laparoscopic surgery - does it really matter, what simulator to use? Results of a cross-sectional study. BMC Med Educ. 2024;24(1):589. Chatziisaak D, Sparn M, Krstic D, Bauci G, Warschkow R, Brunner W, et al. Be prepared! Impact of structured video-assisted coaching on performance in a simulated bleeding exercise during laparoscopic surgery. Surg Endosc. 2024;38(10):6120–7. Chatziisaak D, Burri P, Sparn M, Hahnloser D, Steffen T, Bischofberger S. Concordance of ChatGPT artificial intelligence decision-making in colorectal cancer multidisciplinary meetings: retrospective study. BJS Open. 2025;9(3). Marcelin JR, Siraj DS, Victor R, Kotadia S, Maldonado YA. The Impact of Unconscious Bias in Healthcare: How to Recognize and Mitigate It. J Infect Dis. 2019;220(220 Suppl 2):S62-s73. Sherman MD, Ricco J, Nelson SC, Nezhad SJ, Prasad S. Implicit Bias Training in a Residency Program: Aiming for Enduring Effects. Fam Med. 2019;51(8):677–81. Connor MP, Bynoe AG, Redfern N, Pokora J, Clarke J. Developing senior doctors as mentors: a form of continuing professional development. Report Of an initiative to develop a network of senior doctors as mentors: 1994-99. Med Educ. 2000;34(9):747–53. Bhattacharya S, Bhattacharya K, Bhattacharya N. Surgical Mentorship in the Era of Artificial Intelligence. Indian Journal of Surgery. 2024. Satapathy P, Hermis AH, Rustagi S, Pradhan KB, Padhi BK, Sah R. Artificial intelligence in surgical education and training: opportunities, challenges, and ethical considerations - correspondence. Int J Surg. 2023;109(5):1543–4. Additional Declarations No competing interests reported. Supplementary Files 260205AllSupplements.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 09 Mar, 2026 Reviews received at journal 08 Mar, 2026 Reviews received at journal 03 Mar, 2026 Reviews received at journal 23 Feb, 2026 Reviewers agreed at journal 13 Feb, 2026 Reviewers agreed at journal 12 Feb, 2026 Reviewers agreed at journal 12 Feb, 2026 Reviewers invited by journal 11 Feb, 2026 Editor assigned by journal 11 Feb, 2026 Submission checks completed at journal 11 Feb, 2026 First submitted to journal 05 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8801088","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":592039794,"identity":"f1083125-dcdd-4320-ba86-2b291e039547","order_by":0,"name":"Dimitrios Chatziisaak","email":"","orcid":"","institution":"Guy's \u0026 St Thomas's NHS Foundation Hospitals NHS Trust","correspondingAuthor":false,"prefix":"","firstName":"Dimitrios","middleName":"","lastName":"Chatziisaak","suffix":""},{"id":592039795,"identity":"c9e55975-156c-4c18-8ad2-adb3448264fc","order_by":1,"name":"Moritz Sparn","email":"","orcid":"","institution":"HOCH Health Ostschweiz","correspondingAuthor":false,"prefix":"","firstName":"Moritz","middleName":"","lastName":"Sparn","suffix":""},{"id":592039796,"identity":"5a69daa4-ddc9-4f48-ac58-e0bbcd177e8e","order_by":2,"name":"Pascal Burri","email":"","orcid":"","institution":"University Hospital Zurich","correspondingAuthor":false,"prefix":"","firstName":"Pascal","middleName":"","lastName":"Burri","suffix":""},{"id":592039797,"identity":"2bed5cfd-e3c1-49de-9e1c-9f54243fa8b7","order_by":3,"name":"Daniel Krstic","email":"","orcid":"","institution":"UKE Hamburg","correspondingAuthor":false,"prefix":"","firstName":"Daniel","middleName":"","lastName":"Krstic","suffix":""},{"id":592039798,"identity":"3bc5ed1a-9660-420b-a341-44dccea1949f","order_by":4,"name":"Angelos Vouris","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Angelos","middleName":"","lastName":"Vouris","suffix":""},{"id":592039799,"identity":"3308bf01-f010-4fc6-9431-0891c5014b23","order_by":5,"name":"Bruno Schmied","email":"","orcid":"","institution":"HOCH Health Ostschweiz","correspondingAuthor":false,"prefix":"","firstName":"Bruno","middleName":"","lastName":"Schmied","suffix":""},{"id":592039800,"identity":"623b3073-15ad-4378-807a-5fdc5b553c92","order_by":6,"name":"Dieter Hahnloser","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/ElEQVRIie3PMUvEMBTA8VeEm+J1TfFLvFIoCkc/S0PhplKHWxwjhXPRvYegX0EXJ4cnAbtUXTs4dHLObQ4FzZ2dlLSrQ/5LwiM/yANwuf5j6udIgYNHu5sPbJjYiScHAnsSyClCvwnSBJnXB3W3fYTCv74i0v3iNKpfno8YJIWNBGomzqsPWPH31/Rps16ePDTF0pBsZSOoWFgyAiHbHNWhVBi3LA4qICGtxN+WvSG3O8L6L4yqScK80uwt7vZkRoicxVyPELNLuLkkLu4NMbtkyJs8OtaYWcn8TXX6kxbips3DTvcJ+hdN2KZniZUM8T8fHn/vcrlcrvG+ASKfYBsdjmEMAAAAAElFTkSuQmCC","orcid":"","institution":"Centre Hôpitalier Universitaire Vaudois","correspondingAuthor":true,"prefix":"","firstName":"Dieter","middleName":"","lastName":"Hahnloser","suffix":""},{"id":592039801,"identity":"22e3bf47-fc19-4c30-9cd2-263ab4736737","order_by":7,"name":"Stephan Bischofberger","email":"","orcid":"","institution":"HOCH Health Ostschweiz","correspondingAuthor":false,"prefix":"","firstName":"Stephan","middleName":"","lastName":"Bischofberger","suffix":""}],"badges":[],"createdAt":"2026-02-05 22:38:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8801088/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8801088/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":102854556,"identity":"35512dd1-5e3e-4b4e-bf34-0bf1d8445849","added_by":"auto","created_at":"2026-02-17 14:50:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":9934,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eTest–Retest Comparison of AI-Generated GOALS Scores Between Run 1 and Run 2\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ep \u0026lt; .05 (*)\u003c/p\u003e","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8801088/v1/e3b5524d87be2d3e25cf07cc.png"},{"id":102854561,"identity":"e7172b17-9dc4-42b5-bd28-5a874fd55e4a","added_by":"auto","created_at":"2026-02-17 14:50:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":73956,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBland-Altman Plot for AI GOALS Total Score\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8801088/v1/a1986f213bf33b80e9efcb7f.png"},{"id":102854557,"identity":"b637a408-fb2e-4871-907a-449f96953ac1","added_by":"auto","created_at":"2026-02-17 14:50:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":91799,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMean GOALS values AI vs Expert Panel\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e*=statistically significant p\u0026lt;0.05, **=statistically significant p\u0026lt;0.01, ns = not significant\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8801088/v1/3a0699be710eb747a7252b5e.png"},{"id":102854558,"identity":"811f8b03-180d-48f3-9854-26e53989fe41","added_by":"auto","created_at":"2026-02-17 14:50:12","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":113384,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBland-Altman Plot for GOALS Total Score (AI vs Expert Panel)\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8801088/v1/bd9c45e215c3234c97be4974.jpeg"},{"id":104397177,"identity":"34b7ad3f-13df-44fe-8663-18e9e772d6e6","added_by":"auto","created_at":"2026-03-11 11:40:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1234470,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8801088/v1/6d75c7b1-0b92-46ef-8e19-436178db9edb.pdf"},{"id":102963262,"identity":"4bc1381e-4a92-4cfb-823f-dcb6c5b1d288","added_by":"auto","created_at":"2026-02-19 04:14:54","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":149618,"visible":true,"origin":"","legend":"","description":"","filename":"260205AllSupplements.docx","url":"https://assets-eu.researchsquare.com/files/rs-8801088/v1/9f93406dfed1870578e32a2c.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"AI-Driven Qualitative Skill Assessment in Laparoscopic Training: A Prospective Observational Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003eSurgical education is undergoing a significant transformation. While the traditional apprenticeship model of education in surgical training has long centered around hands-on experience in the operating room, there is a growing shift toward simulation- and proficiency-based training (\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). This evolution is driven by the need to enhance patient safety and to provide structured, efficient training environments for surgical residents (\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e). With this shift, the importance of objective performance assessment has increased.\u003c/p\u003e \u003cp\u003eSimulator-based metrics offer useful quantitative insights but are difficult to interpret within a meaningful clinical context (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). These systems predominantly rely on automated, quantitative parameters such as task completion time, instrument path length, motion economy, or predefined error detection. While such metrics allow standardized and scalable data collection, they primarily capture efficiency-related aspects of performance and fail to reflect the qualitative nuances of surgical skill, including tissue handling, depth perception, or autonomous decision-making.\u003c/p\u003e \u003cp\u003eTo address these limitations, structured qualitative assessments such as the Objective Structured Assessment of Technical Skills (OSATS) and the Global Operative Assessment of Laparoscopic Skills (GOALS) have emerged as comprehensive tools for evaluating surgical performance (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). These instruments enable a nuanced evaluation of technical and non-technical skills using standardized criteria that better reflect real operative competence.\u003c/p\u003e \u003cp\u003eHowever, the implementation of qualitative skill assessment in routine training remains resource-intensive. It requires the continuous involvement of experienced surgical educators who must either directly observe procedures or retrospectively review operative videos. This process is time-consuming, costly, and inherently limited in scalability. Furthermore, interobserver variability and rater-dependent fluctuations related to experience, workload, or cognitive state introduce additional sources of subjectivity and inconsistency (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRecent advancements in artificial intelligence (AI), particularly in video-based surgical analysis, may offer a promising solution. AI-driven evaluation has the potential to apply qualitative scoring systems such as GOALS in a standardized, unbiased, and highly reproducible manner. By combining automated scalability with clinically meaningful qualitative assessment, AI-based approaches may overcome the limitations of both purely quantitative simulator metrics and human-dependent evaluations (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe objective of this study was to evaluate whether an AI-based model can perform qualitative assessments of surgical performance in accordance with the GOALS criteria. Furthermore, the study aimed to examine the reproducibility and consistency of AI-generated evaluations across laparoscopic simulation videos.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy design\u003c/h2\u003e \u003cp\u003eThis prospective observational study was conducted during the 41st Annual Davos Surgical Course in 2024 (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.davoscourse.ch\u003c/span\u003e\u003cspan address=\"http://www.davoscourse.ch\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Demographic data and previous surgical experience were collected using a standardized questionnaire (Supplement 1).\u003c/p\u003e \u003cp\u003eThe dataset consisted of video recordings of laparoscopic cholecystectomies performed on non-perfused porcine liver models placed in a pelvic trainer. Trainees were assigned to pairs and each participant was allocated 60 minutes of operative time to complete one procedure. All procedures were video recorded, anonymized, and stored for subsequent analysis.\u003c/p\u003e \u003cp\u003e All methods were carried out in accordance with relevant guidelines and regulations. We obtained written informed consent from all study participants. This research followed the guidelines and regulations of the WMA Declaration of Helsinki and the Swiss Federal Act on \u0026ldquo;Research involving Human Beings 810.30\u0026rdquo; about scientific research in human beings. The Swiss Ethics Committee \"swissethics.ch\" grants a general waiver for the use of purely anonymized data and an application with approval is not requested.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eVideo segmentation and standardization\u003c/h3\u003e\n\u003cp\u003eDue to current limitations in processing long video sequences, each surgical recording was segmented into nine shorter clips. Segmentation was performed prior to any assessment and applied uniformly across all cases. To ensure methodological fairness and avoid contextual bias, the expert panel assessed the same segmented video clips that were provided to the AI model. This approach ensured that both human and AI evaluations were based on identical visual input, preventing advantages related to continuous procedural context.\u003c/p\u003e\n\u003ch3\u003eHuman expert assessment\u003c/h3\u003e\n\u003cp\u003eAll anonymized videos were independently assessed by a panel of experienced surgeons (DC, MS, PB, DK, SB) using the Global Operative Assessment of Laparoscopic Skills (GOALS) score (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). GOALS evaluates laparoscopic performance across five domains\u0026mdash;depth perception, bimanual dexterity, efficiency, tissue handling, and autonomy\u0026mdash;using a standardized qualitative framework. Raters were blinded to participant identity, training background, and AI-generated results.\u003c/p\u003e\n\u003ch3\u003eAI-based assessment\u003c/h3\u003e\n\u003cp\u003eVideo-based assessments were conducted using a large language model (ChatGPT-4o-mini-high, OpenAI, San Francisco, USA). The model was instructed to apply a rule-based interpretation of the GOALS domains, mapping observable surgical behaviors to predefined qualitative anchors. Output generation was constrained to structured scoring without free-text improvisation.\u003c/p\u003e \u003cp\u003eEach video segment was analyzed independently. For every case, the AI generated a structured report including GOALS domain scores and a qualitative performance summary. After each assessment run, the AI session was reset to prevent carry-over or memory effects and to ensure independence between evaluations.\u003c/p\u003e\n\u003ch3\u003eSample size determination\u003c/h3\u003e\n\u003cp\u003eAn a priori power analysis was performed using a two-way random-effects intraclass correlation coefficient (ICC) model to determine the number of video cases required for interrater reliability (IRR) assessment (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). Assuming an expected ICC of 0.80 and a minimum acceptable ICC of 0.60, a sample size of 20 videos was estimated to provide 80% power at a significance level of α\u0026thinsp;=\u0026thinsp;0.05. To increase precision and enable secondary comparisons between AI and the expert panel, 50 videos were included in the final analysis (Supplement 2).\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eComparative analysis\u003c/h2\u003e \u003cp\u003eAI-generated GOALS scores were compared with scores assigned by the expert panel. Three reviewers evaluated each AI-generated report for completeness, conformity with the predefined scoring framework, and alignment with the study objectives. Any discrepancies were resolved through panel discussion.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eReproducibility\u003c/h3\u003e\n\u003cp\u003eTo evaluate test\u0026ndash;retest reliability, a subset of 15 videos was reassessed by the AI model in a second independent evaluation run. Prior to reanalysis, videos were re-anonymized, and the AI session was reset to eliminate any potential memory effects. This approach enabled assessment of score stability under controlled conditions.\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses were performed in two sequential steps. First, test\u0026ndash;retest reliability of AI-generated GOALS scores was assessed by comparing results from two independent evaluation runs. Second, agreement between AI-generated scores and expert panel ratings was analyzed.\u003c/p\u003e \u003cp\u003eIntraclass correlation coefficients (ICCs) were calculated to quantify reliability and agreement in accordance with established methodological recommendations (\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e). AI test\u0026ndash;retest reliability was assessed using a two-way mixed-effects model with absolute agreement, while agreement between AI-generated scores and expert panel ratings was evaluated using a two-way random-effects model. Bland\u0026ndash;Altman analyses were performed to visualize agreement, estimate systematic bias, and assess limits of agreement across the score range.\u003c/p\u003e \u003cp\u003eFor paired comparisons of GOALS domain and total scores, Wilcoxon signed-rank tests (Z) were applied when normality assumptions were not met (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). Agreement on checklist items was assessed using Cohen\u0026rsquo;s kappa. To explore potential sources of systematic bias between AI and expert ratings, a multiple linear regression analysis was conducted using the total score difference as the dependent variable and participant characteristics as independent predictors.\u003c/p\u003e \u003cp\u003eAll statistical tests were two-sided, and a p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eParticipants\u0026rsquo; demographics\u003c/h2\u003e \u003cp\u003eFifty surgical residents participated in the study (Postgraduate Year PGY 1\u0026ndash;2), of whom 63% were female. The majority of participants reported limited prior experience in minimally invasive surgery, with \u0026le;\u0026thinsp;5 completed cases in laparoscopic appendectomy (73%) and laparoscopic cholecystectomy (80%). No participant reported prior experience in laparoscopic colectomy or hernia repair. All participants completed a pre-course self-assessment questionnaire; six questionnaires were returned incomplete.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eReliability of AI-calculated GOALS scores\u003c/h2\u003e \u003cp\u003eTest\u0026ndash;retest reliability of the AI-generated GOALS scores was assessed using two independent evaluation runs of 15 anonymized video cases. The AI demonstrated excellent repeatability for the GOALS total score (ICC 0.91), indicating high overall stability across repeated assessments. Good reliability was observed for the domains of bimanual dexterity, efficiency, and autonomy, whereas depth perception and tissue handling showed lower agreement with wider confidence intervals (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eIntraclass Correlation Coefficient (ICCs) AI test-retest\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGOALS\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eICC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e95% CI\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eInterpretation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDepth Perception\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.507\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[-0.19, 0.716]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.099\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eModerate (Unstable)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBimanual Dexterity\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.814\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[0.286, 0.882]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.002\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGood\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEfficiency\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.726\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[0.101, 0.831]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.011\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eModerate\u0026ndash;Good\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTissue Handling\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.417\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[-0.269, 0.673]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.162\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePoor (Unstable)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAutonomy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.795\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[0.241, 0.871]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.003\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eGood\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGOALS Total\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.906\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e[0.562, 0.939]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eExcellent\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eICC\u0026thinsp;\u0026gt;\u0026thinsp;0.90\u0026thinsp;=\u0026thinsp;Excellent, ICC 0.75\u0026ndash;0.90\u0026thinsp;=\u0026thinsp;Good, ICC\u0026thinsp;\u0026lt;\u0026thinsp;0.75\u0026thinsp;=\u0026thinsp;Needs discussion\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003ePaired comparisons between the first and second AI assessment runs revealed no significant differences for most GOALS domains. A small but statistically significant decrease was observed for bimanual dexterity and the total GOALS score on re-evaluation. Bland\u0026ndash;Altman analysis demonstrated minimal overall bias between runs and no systematic trend across the score range (Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e \u0026amp; \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eReliability of human GOALS scores\u003c/h2\u003e \u003cp\u003eInterrater reliability within the expert panel was excellent across all GOALS domains and subdomains. All intraclass correlation coefficients were statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). These findings confirm a high level of consistency among expert raters and support the use of the panel scores as a robust reference standard for comparison with AI-generated assessments (Supplement 3).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eAI-calculated versus human GOALS scores\u003c/h2\u003e \u003cp\u003eComparison of AI-generated and expert panel GOALS scores demonstrated strong agreement across domains and for the total score. Median total GOALS scores did not differ significantly between AI and expert ratings. No significant differences were observed for the domains of bimanual dexterity (Z\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;0.946, p\u0026thinsp;=\u0026thinsp;0.344) and autonomy, whereas statistically significant differences were identified for depth perception (Z\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;3.444, p\u0026thinsp;=\u0026thinsp;0.001), efficiency (Z\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;2.335, p\u0026thinsp;=\u0026thinsp;0.020), and tissue handling (Z\u0026thinsp;=\u0026thinsp;\u0026minus;\u0026thinsp;3.146, p\u0026thinsp;=\u0026thinsp;0.002), with AI scores tending to be slightly higher than expert ratings (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Although statistically significant, the observed domain-level differences corresponded to less than one GOALS point and are unlikely to be educationally meaningful in formative assessment contexts.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eInter-method reliability analysis revealed excellent agreement between AI and expert panel assessments for the GOALS total score (ICC 0.92). Domain-level agreement was excellent for bimanual dexterity and autonomy and good to excellent for depth perception, efficiency, and tissue handling (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). All ICC estimates were statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eIntraclass Correlation Coefficients (ICCs) Between AI and Expert Panel for GOALS Domains\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDomain\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eICC (Average)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e95% CI (ICC)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eF(df1, df2)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDepth Perception\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.832\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[0.748, 0.895]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e5.943 (49, 245)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBimanual Dexterity\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.903\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[0.854, 0.939]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e10.292 (49, 245)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEfficiency\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.849\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[0.773, 0.905]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6.614 (49, 245)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTissue Handling\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[0.820, 0.925]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e8.333 (49, 245)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAutonomy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[0.819, 0.925]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e8.310 (49, 245)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGOALS Total\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e[0.880, 0.950]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12.476 (49, 245)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eAll ICCs are based on a two-way random-effects model using consistency definition\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e Bland\u0026ndash;Altman analysis of the total GOALS score demonstrated minimal mean bias between AI and expert ratings difference (Mean Difference = -1.00) and narrow limits of agreement, with no evidence of proportional bias across the performance range. Domain-specific Bland\u0026ndash;Altman analyses showed variable levels of agreement, with the highest consistency observed for autonomy and bimanual dexterity (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003ePredictors of bias between AI and expert ratings\u003c/h2\u003e \u003cp\u003eTo explore potential sources of systematic bias between AI-generated and expert panel GOALS total scores, a multiple linear regression analysis was performed. Participant characteristics including postgraduate year, gender, handedness, training status, and self-reported overall laparoscopic skill level were entered as predictors.\u003c/p\u003e \u003cp\u003eAmong the evaluated variables, self-reported overall laparoscopic skill level was significantly associated with the magnitude of bias between AI and expert ratings, with higher self-assessed skill levels corresponding to larger negative score differences (β = \u0026minus;0.46, p\u0026thinsp;=\u0026thinsp;0.008). No significant associations were observed for postgraduate year, gender, handedness, or training status (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The overall regression model explained a modest proportion of variance in score differences (adjusted R\u0026sup2; = 0.114).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eMultiple Linear Regression Predicting Bias in AI vs Expert Panel Total GOALS Scores\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePredictor\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eB\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSE B\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eβ (Beta)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003et\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003e95% CI (Lower)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003e95% CI (Upper)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003e(Constant)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-2.39\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026mdash;\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.58\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.563\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-10.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e5.91\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePGY\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.52\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.35\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-0.59\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e1.64\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGender\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-0.26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.03\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.843\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-2.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e2.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHandedness\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.14\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-3.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e8.87\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTraining Status\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.47\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e1.41\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.167\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-1.08\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e6.02\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eOverall Lap. Skills\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e-2.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-0.46\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e-2.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e.008*\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e-4.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e-0.74\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"8\"\u003e\u003cem\u003eR\u0026sup2; = .217, Adjusted R\u0026sup2; = .114, F (5, 38)\u0026thinsp;=\u0026thinsp;2.11, p = .086, DV\u0026thinsp;=\u0026thinsp;Bias in AI vs. Panel GOALS total score\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study demonstrates that a large language model can reliably assess videos of minimally invasive cholecystectomies performed in a simulation setting using a validated qualitative scoring system. AI-generated GOALS scores showed excellent agreement with expert ratings and high test\u0026ndash;retest stability, supporting the feasibility of AI-based qualitative skill assessment in surgical training environments.\u003c/p\u003e \u003cp\u003eAs surgical training shifts toward proficiency-based curricula, and because surgeons should be trained under meticulous, unbiased supervision and constructive feedback, scalable and meaningful assessment becomes increasingly important (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). Prior work has shown that constructive teaching approaches supported by newer technologies can improve training outcomes (\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e). In addition, simulation-based systems often provide automated metrics that are predominantly quantitative and may not reflect the qualitative nuances of operative skill (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e). In this context, our results support the introduction of AI technologies into surgical curricula to facilitate a stepwise transition toward standardized, inclusive, and proficiency-based training while maintaining patient safety and accelerating skill acquisition (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eWhile expert raters demonstrated excellent internal consistency, human-centered evaluation remains susceptible to variability in the interpretation of borderline performances and to unconscious bias related to rater experience, workload, or cognitive factors (\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). In our study, regression analysis revealed that self-reported overall laparoscopic skill level was significantly associated with the magnitude of disagreement between AI and expert ratings, whereas postgraduate year, gender, handedness, and training status were not. This finding suggests that discrepancies were not driven by formal training level or demographic factors, but may reflect differences in how performance thresholds are implicitly applied across experience strata. Importantly, this pattern does not indicate systematic bias across the overall performance range, but rather highlights the sensitivity of qualitative assessment at higher skill levels. The relative stability of AI-generated scores across repeated evaluations suggests that AI-based assessment may apply more consistent evaluative criteria, which could contribute to increased standardization and fairness in formative surgical assessment, particularly when evaluating advanced or borderline performances. AI-based assessments should be viewed as decision-support tools rather than autonomous evaluators, complementing human supervision especially in summative or high-stakes contexts.\u003c/p\u003e \u003cp\u003eAccording to Connor et al., mentorship provides a safe space for reflection and support, enabling exploration of strengths and weaknesses, self-challenge, insight generation, and goal orientation (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). AI-supported mentorship programs may facilitate structured learning pathways, provide timely feedback, and contribute to continuous and fair evaluation (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e). This assumes that such tools can help trainees adhere to rigorous standards, acquire requisite skills, and be treated fairly and inclusively (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eSeveral limitations should be acknowledged. Although the LLM appeared largely impartial with respect to most participant characteristics, it may undervalue performance in individuals with greater experience or higher self-reported proficiency. This may reflect limitations in recognizing nuanced skill cues or stricter implicit thresholds in higher-skill subgroups, underscoring the need for further calibration to ensure fairness and validity. In addition, a carefully prepared prompt is required to generate qualitative results, and fully continuous human-like video analysis is not yet feasible. Current workflows rely on manual clip segmentation, which may introduce human influence, and the analysis depends on automated timestamped logging of actions.\u003c/p\u003e \u003cp\u003eFuture studies should evaluate these tools in larger populations, refine standardized metric extraction, define appropriate benchmarks, and test performance across diverse skill levels and settings. Advancing these automated procedures - and strengthening collaboration between surgeons and software engineers - will be essential for robust, fair, and scalable deployment in surgical education.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrates that a large language model can reliably assess laparoscopic surgical performance in a simulation setting with expert-level agreement and high test\u0026ndash;retest stability. By enabling standardized qualitative skill assessment without continuous expert presence, AI-based video analysis addresses key limitations of current evaluation approaches. While further optimization is required for selected skill domains and validation in larger cohorts, this approach has the potential to support scalable, reproducible, and resource-efficient integration of qualitative assessment into proficiency-based surgical training programs.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting interests:\u003c/h2\u003e \u003cp\u003eAll authors declare no financial or non-financial competing interests.\u003c/p\u003e \u003c/p\u003e\u003cp\u003e \u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e \u003cp\u003eAll methods were carried out in accordance with relevant guidelines and regulations. We obtained written informed consent from all study participants. This research followed the guidelines and regulations of the WMA Declaration of Helsinki and the Swiss Federal Act on \u0026ldquo;Research involving Human Beings 810.30\u0026rdquo; about scientific research in human beings. The Swiss Ethics Committee \"Swissethics.ch\" grants a general waiver for the use of purely anonymized data and an application with approval is not requested.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding:\u003c/h2\u003e \u003cp\u003eThis study was supported by the national \"Proficiency\" research project, funded by the Swiss Innovation Agency Innosuisse in 2021 as one of 15 flagship initiatives. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://surgicalproficiency.ch/\u003c/span\u003e\u003cspan address=\"https://surgicalproficiency.ch/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConception and design of the work: DC, MS, DH, SBData acquisition and analysis: DC, MS, PB, DK, AV, BS, DH, SBInterpretation of data: DC, MS, PB, DK, AV, BS, DH, SBDrafted the manuscript: DC, SBSubstantively revised the manuscript: DC, MS, PB, DK, BS, DH, SB\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors like to thank the congress organization of the \"Davos Course\" (davoscourse.ch) for the organizational support.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets generated and/or analyzed during the current study are included in this published article and its supplementary information files or are available from the corresponding author on request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGallagher AG, Ritter EM, Champion H, Higgins G, Fried MP, Moses G, et al. Virtual reality simulation for the operating room: proficiency-based training as a paradigm shift in surgical skills training. Ann Surg. 2005;241(2):364\u0026ndash;72.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShaker D. Cognitivism and psychomotor skills in surgical training: from theory to practice. Int J Med Educ. 2018;9:253\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShahrezaei A, Sohani M, Taherkhani S, Zarghami SY. The impact of surgical simulation and training technologies on general surgery education. BMC Med Educ. 2024;24(1):1297.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurri P, Chatziisaak D, Sparn M, Bischofberger S. [Learn playfully, operate seriously: The new era of surgical training]. Chirurgie (Heidelb). 2025;96(3):236\u0026ndash;42.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeymour NE, Gallagher AG, Roman SA, O'Brien MK, Bansal VK, Andersen DK, et al. Virtual reality training improves operating room performance: results of a randomized, double-blinded study. Ann Surg. 2002;236(4):458\u0026ndash;63; discussion 63\u0026thinsp;\u0026ndash;\u0026thinsp;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartin JA, Regehr G, Reznick R, MacRae H, Murnaghan J, Hutchison C, et al. Objective structured assessment of technical skill (OSATS) for surgical residents. Br J Surg. 1997;84(2):273\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVassiliou MC, Feldman LS, Andrew CG, Bergman S, Leffondr\u0026eacute; K, Stanbridge D, et al. A global assessment tool for evaluation of intraoperative laparoscopic skills. Am J Surg. 2005;190(1):107\u0026ndash;13.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFaherty A, Counihan T, Kropmans T, Finn Y. Inter-rater reliability in clinical assessments: do examiner pairings influence candidate ratings? BMC Med Educ. 2020;20(1):147.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTavares W, Eva KW. Exploring the impact of mental workload on rater-based assessments. Adv Health Sci Educ Theory Pract. 2013;18(2):291\u0026ndash;303.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHashimoto DA, Rosman G, Rus D, Meireles OR. Artificial Intelligence in Surgery: Promises and Perils. Ann Surg. 2018;268(1):70\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLillemoe HA, Geevarghese SK. Stopping the Progression of Moral Injury: A Priority During Surgical Training. Ann Surg. 2021;274(6):e643-e5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWalter SD, Eliasziw M, Donner A. Sample size and optimal designs for reliability studies. Stat Med. 1998;17(1):101\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15(2):155\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiller E. The signed-rank (Wilcoxon)test. Lancet. 1969;1(7590):371.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChikwe J, de Souza AC, Pepper JR. No time to train the surgeons. Bmj. 2004;328(7437):418\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSparn MB, Teixeira H, Chatziisaak D, Schmied B, Hahnloser D, Bischofberger S. Virtual reality simulation training in laparoscopic surgery - does it really matter, what simulator to use? Results of a cross-sectional study. BMC Med Educ. 2024;24(1):589.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChatziisaak D, Sparn M, Krstic D, Bauci G, Warschkow R, Brunner W, et al. Be prepared! Impact of structured video-assisted coaching on performance in a simulated bleeding exercise during laparoscopic surgery. Surg Endosc. 2024;38(10):6120\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChatziisaak D, Burri P, Sparn M, Hahnloser D, Steffen T, Bischofberger S. Concordance of ChatGPT artificial intelligence decision-making in colorectal cancer multidisciplinary meetings: retrospective study. BJS Open. 2025;9(3).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarcelin JR, Siraj DS, Victor R, Kotadia S, Maldonado YA. The Impact of Unconscious Bias in Healthcare: How to Recognize and Mitigate It. J Infect Dis. 2019;220(220 Suppl 2):S62-s73.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSherman MD, Ricco J, Nelson SC, Nezhad SJ, Prasad S. Implicit Bias Training in a Residency Program: Aiming for Enduring Effects. Fam Med. 2019;51(8):677\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConnor MP, Bynoe AG, Redfern N, Pokora J, Clarke J. Developing senior doctors as mentors: a form of continuing professional development. Report Of an initiative to develop a network of senior doctors as mentors: 1994-99. Med Educ. 2000;34(9):747\u0026ndash;53.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBhattacharya S, Bhattacharya K, Bhattacharya N. Surgical Mentorship in the Era of Artificial Intelligence. Indian Journal of Surgery. 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSatapathy P, Hermis AH, Rustagi S, Pradhan KB, Padhi BK, Sah R. Artificial intelligence in surgical education and training: opportunities, challenges, and ethical considerations - correspondence. Int J Surg. 2023;109(5):1543\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8801088/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8801088/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground:\u003c/h2\u003e \u003cp\u003eQualitative assessment of surgical skills is essential for proficiency-based training but remains resource-intensive and difficult to scale due to reliance on expert observation. Automated simulator metrics, while objective, are largely limited to quantitative parameters and fail to capture qualitative aspects of surgical performance. Artificial intelligence\u0026ndash;based video analysis may address this gap by enabling standardized and reproducible qualitative assessment.\u003c/p\u003e\u003ch2\u003eMethods:\u003c/h2\u003e \u003cp\u003eIn this prospective observational study conducted at the 41st Annual Davos Surgical Course (2024), 50 first- and second-year surgical residents performed laparoscopic cholecystectomy on porcine simulation models. Procedures were video recorded, anonymized, segmented, and independently assessed by expert raters and an AI-based model using the Global Operative Assessment of Laparoscopic Skills (GOALS). AI test\u0026ndash;retest reliability and agreement with expert ratings were evaluated.\u003c/p\u003e\u003ch2\u003eResults:\u003c/h2\u003e \u003cp\u003eAI demonstrated excellent test\u0026ndash;retest reliability for the total GOALS score (ICC 0.91) and good reliability across most domains. Agreement between AI and expert raters was excellent for the total score (ICC 0.92) and good to excellent for individual domains, with minimal bias on Bland\u0026ndash;Altman analysis.\u003c/p\u003e\u003ch2\u003eConclusion:\u003c/h2\u003e \u003cp\u003eAI-based video analysis enables reliable, reproducible qualitative assessment of laparoscopic surgical skills in a simulation setting and may support scalable integration of qualitative evaluation into surgical training programs and other video-based clinical skill assessment domains.\u003c/p\u003e","manuscriptTitle":"AI-Driven Qualitative Skill Assessment in Laparoscopic Training: A Prospective Observational Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-17 14:50:07","doi":"10.21203/rs.3.rs-8801088/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-09T23:52:50+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-08T22:13:56+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-03T15:04:56+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-02-23T08:58:55+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"237034468228796496647282025787784625776","date":"2026-02-13T17:49:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"75531954927022774791063478373028762524","date":"2026-02-12T08:11:55+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"119568461802146613367328638925423364159","date":"2026-02-12T07:04:59+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-12T03:25:32+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-12T01:20:03+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-11T16:55:07+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2026-02-05T22:23:02+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"fee20806-bb22-4852-9127-ef1da86de6d0","owner":[],"postedDate":"February 17th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":62995957,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":62995958,"name":"Health sciences/Gastroenterology"},{"id":62995959,"name":"Health sciences/Health care"},{"id":62995960,"name":"Physical sciences/Mathematics and computing"},{"id":62995961,"name":"Health sciences/Medical research"}],"tags":[],"updatedAt":"2026-05-08T16:54:07+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-17 14:50:07","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8801088","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8801088","identity":"rs-8801088","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}