The learning curve of a high-fidelity cataract surgical simulator Authors: Mingxue Ren, Hangdong Dai, Ziyi Dong, Lingxia Ye, An-Peng Pan, Kaijing Zhou, Yimin Yuan, Xu Xu Published: 2026 ABSTRACT AbstractObjectiveTo identify the number of iterations required to reach a performance plateau for each module on the HelpMeSee MSICS high‑fidelity cataract surgical simulator, and to evaluate the inter-rater reliability of the assessment rubric and the consistency of the simulator’s visuo-haptic feedback.MethodsTwenty medical students (Group A) and 12 residents (Group B) were recruited. All participants performed 10 iterations on five modules of the HelpMeSee MSICS high-fidelity surgical simulator. All operation videos were recorded and scored by two observers using the simulator assessment rubric. Intraclass correlation coefficient, Friedman test, and Mann-Whitney U test were used to evaluate inter-rater reliability, intra-module learning curves, and inter-group differences, respectively.ResultsLearning curves were demonstrated in both groups across all five modules and total score (all P < 0.01). Residents achieved higher initial performance and reached performance plateaus earlier than students. Inter-rater reliability was excellent for all modules and total score (ICC > 0.80, all P < 0.05).ConclusionMedical students and residents without independent cataract surgery experience improved and reached performance plateaus on the simulator through repeated practice, suggesting that the simulator provided consistent visuo-haptic feedback conducive to initial skill acquisition. However, these simulator-based improvements did not directly translate to improved real-life surgical performance and required separate validation through clinical outcome studies.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12909-026-09640-7.Keywords: Simulator, Cataract surgery, Manual small incision cataract surgery, Learning curve Introduction Manual small incision cataract surgery (MSICS) remains in high demand in developing countries [1, 2]. Although cataract surgery has a long learning curve and novice surgeons have higher complication rates [3–5], residency training often fails to provide sufficient supervised procedures [1, 2, 6] to meet recommended targets (e.g., 86 cases by ACGME [7] or 300 in some programs [8]. Surgical simulators have demonstrated construct validity and significant correlation with live surgery [9–13], and have been suggested to reduce complications [5, 14–25] and shorten the surgical learning curve [26–28]. HelpMeSee high-fidelity surgical simulator incorporates MSICS training modules with visuo-haptic feedback, which may provide a more realistic training experience than visual-only platforms [29]. Learning curve analysis is important in verifying simulator validity and optimizing training curricula [30, 31]. To our knowledge, this is the first module-specific learning curve analysis for an MSICS simulator incorporating visuo-haptic feedback. This study aimed to determine the number of iterations required for trainees with different prior experience levels to reach a performance plateau on each module of the HelpMeSee MSICS high-fidelity simulator, and to evaluate the inter-rater reliability of the assessment rubric and the consistency of the simulator’s visuo-haptic feedback. We hypothesized that (1) both groups would demonstrate measurable learning curves within 10 iterations; (2) residents with prior step-specific experience would reach the plateau earlier than medical students; and (3) the assessment rubric would demonstrate high inter-rater reliability. These findings will help optimize simulation-based training curricula by providing evidence-based, module-specific minimum training thresholds, ensuring trainees have achieved baseline simulator proficiency before transitioning to supervised live surgery. Methods and materials Participants Twenty medical students and twelve residents from Eye Hospital of Wenzhou Medical University participated in this study. None of the participants had been exposed to the high-fidelity cataract surgical simulator. Age, gender, dominant hand, number of prior cataract surgeries, and the single-step cataract surgery experience were recorded. Medical students (Group A, n = 20) had no prior experience in pterygium surgery or any step of cataract surgery. Residents (Group B, n = 12) had prior microsurgical experience, having performed pterygium surgeries under supervision (see Table 2 for details). They had no independent cataract surgery experience. During their residency, they had been exposed to simple cataract surgery steps under supervision, specifically creating paracentesis, performing viscoelastic exchange, and keratome entry into the AC. They had not performed more complex steps such as creating a scleral groove or tunnel dissection. None of the residents had performed several hundred MSICS operations; their operative exposure was limited to the supervised steps described above and a median of 9 prior pterygium surgeries (IQR 7–15). The exact number of repetitions for these cataract steps was not recorded. Table 2Demographics of participantsGroup AGroup BNumber2012Male: female6:145:7Age median [IQR]25.5 [25, 26]32.5 [30, 36]Dominant hand ratio100%100%Number of pterygium surgery09 [7, 14.75]Experience in performing single-step cataract surgeryNoYesExperience in performing cataract surgeryNoNo Demographics of participants Simulator HelpMeSee MSICS high-fidelity surgical simulator (Fig. 1) was used in this study (Version: 2.0.15.0). This simulator was developed for manual small incision cataract surgery training in conjunction with Moog Inc (New York, New York), InSimo (Strasbourg, France), and SenseGraphics (Kista, Sweden). The performance parameters in the simulator were peer-reviewed [32]. At the same time, the simulator was equipped with a set of specialized training courses, which include basic MSICS skills to help trainees fully understand and master the skills. Five modules used in this study (Fig. 2-1-5) were: Creating a scleral groove (Module 1, Fig. 2-1), Performing tunnel dissection (Module 2, Fig. 2), Creating paracentesis (Module 3, Fig. 2-3), Performing viscoelastic exchange (Module 4, Fig. 2-4), and Performing keratome entry into AC (Module 5, Fig. 2-5) (Table 1). The HelpMeSee MSICS simulator (New York, New York, USA) 1 Module 1 Creating a scleral groove. 2 Module 2 Performing tunnel dissection. 3 Module 3 Creating paracentesis. 4 Module 4 Performing viscoelastic exchange. 5 Module 5 Performing keratome entry into AC Table 1The modules on HelpMeSee MSICS high-fidelity surgical simulatorModuleDescriptionScore1Creating a scleral grooveUsing the Colibri forceps to stabilize the globe for cutting. Using visual and tactile cues during the scleral groove dissection to create a groove with proper dimensions and depth.272Performing tunnel dissectionUsing the crescent blade to make the scleral dissection, limbus dissection and corneal dissection for establishing the inner tunnel limit and proper depth of the tunnel.303Creating paracentesisTo make a paracentesis with 15°stab blade at the limbus.184Performing viscoelastic exchangeInjecting viscoelastic through the opening using the 25-gauge cannula with syringe.215Performing keratome entry into ACUsing the keratome blade to complete the inner tunnel opening across the entire path of the tunnel in the cornea.27 The modules on HelpMeSee MSICS high-fidelity surgical simulator Protocol Each participant in Group A and Group B received a brief introduction about the simulator before operation. Before the operation on the simulator, the theoretical training for each module was carried out. The theoretical training for each module was about 60 min. After the theoretical training, participants were arranged to perform 10 iterations of operation for each module on the simulator. The operation time of each module was about 30–60 min. This study collected the first 10 operation videos for each module. All participants completed the theoretical training courses and 10 iteration operations of 5 modules within 2 weeks. All operation videos were recorded for subsequent scoring. Two certified simulator trainers rated operations according to the HelpMeSee MSICS high-fidelity surgical simulator scoring scale. Statistical analysis Portions of the scoring data did not conform to the normal distribution; therefore, nonparametric tests were used for data analysis. The distribution of data was analyzed using box-plot analysis. Intraclass correlation coefficient (ICC) was calculated using a two‑way random effects model for absolute agreement, based on single measurements (ICC(2,1)). The Friedman test was used to evaluate performance changes across 10 iterations per module, and post-hoc comparisons were used to delineate the learning curve. Post-hoc pairwise comparisons following the Friedman test were performed using Bonferroni correction to control for multiple comparisons. P < 0.05 was considered statistically significant. A performance plateau was defined as the first iteration at which the score no longer showed a statistically significant difference from the final (10th) iteration score, as determined by post‑hoc multiple comparisons following the Friedman test. The Mann-Whitney U test was used to compare the differences between two groups in each iteration in the same module. P < 0.05 was considered statistically significant. SPSS (software version 26.0) was used for data analysis in this study. Results None of the participants had performed independent cataract surgery; Group A had no prior surgical exposure of any kind, whereas Group B had assisted in specific steps under supervision. The number of prior pterygium surgeries was recorded as a proxy for microsurgical experience, reflecting basic instrument handling and microscope familiarity (Table 2). Module-specific and total scores for both groups are presented in Table 3 and 4. Table 3Performance of five modules and total score for group A at selected iterations (1, 5, and 10)IterationModule 1Creating a scleral grooveModule 2Performing tunnel dissectionModule 3Creating paracentesisModule 4Performing viscoelastic exchangeModule 5Performing keratome entry into ACTotal score113.0 (11.625,14.375)15.25 (14.0,17.375)11.0 (8.625,12.0)12.5 (11.0, 13.0)14.5 (11.625,15.875)66.5 (64.0,68.125)520.25 (18.625,21.375)21.25 (16.875,23.375)14.25 (12.75,15.5)17.0 (16.0,18.5)19.25 (18.0,22.0)93.0 (87.625,95.875)1024.0 (23.125,25.375)28.0 (24.875,28.5)17.0 (16.125,17.0)19.0 (18.125, 19.0)24.5 (23.625,24.5)111.3 (107.625,113.875)Median (lower quartile, upper quartile)Complete iteration-by-iteration data are provided in Supplementary Table 1 Performance of five modules and total score for group A at selected iterations (1, 5, and 10) Median (lower quartile, upper quartile) Complete iteration-by-iteration data are provided in Supplementary Table 1 Table 4Performance of five modules and total score for group B at selected iterations (1, 5, and 10)IterationModule 1Creating a scleral grooveModule 2Performing tunnel dissectionModule 3Creating paracentesisModule 4Performing viscoelastic exchangeModule 5Performing keratome entry into ACTotal score114.25 (13.0,16.25)17.25 (14.125,22.0)13.0 (11.25,15.25)16.0 (14.125, 17.375)18.25 (14.75,21.0)76.0 (70.125,89.625)522.5 (19.625,24.25)27.5 (23.125,28.875)16.25 (14.0,17.0)18.0 (18.0,18.875)23.5 (22.25,24.5)104.0 (101.75,112.125)1024.5 (23.5,25.875)29.0 (27.25,29.875)17.0 (16.25,17.5)19.0 (18.625, 19.0)24.5 (24.0,24.875)114.0 (111.5,116.625)Median (lower quartile, upper quartile)Complete iteration-by-iteration data are provided in Supplementary Table 2 Performance of five modules and total score for group B at selected iterations (1, 5, and 10) Median (lower quartile, upper quartile) Complete iteration-by-iteration data are provided in Supplementary Table 2 Inter-rater consistency ICC values greater than 0.80 are generally considered to indicate excellent inter-rater reliability. In this study, the intraclass correlation coefficient was used to evaluate the inter-rater consistency of the HelpMeSee MSICS high-fidelity surgical simulator assessment rubric. Two HelpMeSee-certified simulator trainers independently rated all 1,600 videos (20 students and 12 residents, 10 iterations × 5 modules each). Table 5 summarizes the ICC of each module and the total score. Inter-rater reliability was excellent across all five modules and total score (all ICC > 0.80, P < 0.05), with total score showing the highest agreement (ICC = 0.955). Table 5Inter-rater reliability for HelpMeSee MSICS high-fidelity surgical simulator assessment rubricModulesICC95% CILower boundUpper bound1 Creating a scleral groove0.864*0.8330.8892 Performing tunnel dissection0.884*0.8690.9143 Creating paracentesis0.895*0.8710.9154 Performing viscoelastic exchange0.919*0.9010.9355 Performing keratome entry into AC0.888*0.8620.909Total score0.955*0.9450.964ICC values were calculated using a two‑way random effects model for absolute agreement, based on single measurements (ICC(2,1))*p < 0.05 Inter-rater reliability for HelpMeSee MSICS high-fidelity surgical simulator assessment rubric ICC values were calculated using a two‑way random effects model for absolute agreement, based on single measurements (ICC(2,1)) *p < 0.05 Box-plots Box-plots are often used to examine the distribution of data. In this study, the box plots for Group A and Group B across all modules and total score are shown in Fig. 3-1 and 2. In Group A, no outliers or extreme values were observed in any of the 5 modules and the total score. Except for viscoelastic exchange in Group B (Module 4), no outliers or extreme values were observed in any of the other modules. The data used for box plots in this study were the average values of two raters. Because averaged data reduce quartile spacing in box plots, some values appeared as outliers. Therefore, we retained all outliers in subsequent analyses. 1 The score distribution between 10 iterations of 5 modules and total score for Group A. 2 The score distribution between 10 iterations of 5 modules and total score for Group B The learning curve and the plateau The Friedman test indicated significant improvements across iterations for all modules and total score in both groups (all P < 0.0001). Technically demanding modules required more iterations to plateau than procedural steps of lower complexity. Residents reached plateaus 2–4 iterations earlier than students across all modules, although both groups achieved comparable performance by iteration 10. These differential learning trajectories are illustrated in Fig. 4-1 and 2; detailed iteration-by-iteration scores were provided in Table 3 and 4 and Supplementary Tables 1 and 2. 1 The learning curves for 5 modules and total score of group A. All data presented as median (interquartile range). Improvement between 10 iterations was significant for all modules and total score (all P < 0.0001). (a) Creating a scleral groove reaches the plateau of the learning curve at the eighth iteration, P = 0.509. (b) Performing tunnel dissection reaches the plateau at the seventh iteration, P > 0.99. (c) Creating paracentesis reaches the plateau at the sixth iteration, P > 0.99. (d) Performing viscoelastic exchange reaches the plateau at the sixth iteration, P = 0.682. (e) Performing keratome entry into AC reaches the plateau at the seventh iteration, P = 0.509. (f) The total score reaches the plateau at the seventh iteration, P = 0.155. (ns: not significant). 2 The learning curves for 5 modules and total score of group B. All data presented as median (interquartile range). Improvement between 10 iterations was significant for all modules and total score (all P < 0.0001). (a) Creating a scleral groove reaches the plateau of the learning curve at the fourth iteration, P = 0.187. (b) Performing tunnel dissection reaches the plateau at the fifth iteration, P > 0.99. (c) Creating paracentesis reaches the plateau at the fourth iteration, P = 0.077. (d) Performing viscoelastic exchange reaches the plateau at the third iteration, P = 0.385. (e) Performing keratome entry into AC reaches the plateau at the fourth iteration, P = 0.187. (f) The total score reaches the plateau at the fifth iteration, P = 0.168. (ns: not significant) The influence of single-step experience In this study, the Mann-Whitney U test was used to compare the two groups at each iteration within each module. As shown in Fig. 5, residents maintained significantly higher scores than students during early practice iterations, but this gap closed progressively and disappeared by iteration 10 across all modules. Significant inter-group differences were concentrated in early iterations (iterations 1–6 for most modules), reflecting the initial performance advantage conferred by residents’ prior step-specific experience. These differences progressively diminished in middle iterations as students rapidly improved. By iteration 10, no significant inter-group differences remained on any module or total score, indicating that sufficient repetition equalized performance between novices and trainees with limited prior exposure. These findings suggest that the simulator possesses construct validity in distinguishing baseline experience levels early in training, while simultaneously demonstrating its capacity to close the initial experience gap through structured repetition. Comparison of performance between group A (blue circles) and group B (red triangles) for all iterations across all modules. (a) Creating a scleral groove, (b) Performing tunnel dissection, (c) Creating paracentesis, (d) Performing viscoelastic exchange, (e) Performing keratome entry into AC, and (f) Total score. All data presented as median (interquartile range). *means P < 0.05 Discussion A key finding was that residents with prior step-specific experience achieved earlier performance plateaus than novices across all modules, yet both groups reached comparable simulator proficiency by iteration 10. This suggests that while prior exposure accelerates initial psychomotor acquisition, structured repetition on the simulator can effectively close the baseline experience gap. It is important to emphasize that the observed learning curves reflect initial psychomotor skill acquisition in a relatively novice population, not expertise refinement in experienced surgeons. Therefore, our findings describe the simulator’s capacity to support early skill development rather than the learning trajectory of the actual surgical technique in expert practitioners. Bartlett et al. noted that visuo-haptic feedback consistency was an important indicator of the simulator validity [30]. While the observed learning curves suggest that the simulator provides consistent visuo-haptic feedback conducive to initial skill acquisition, we caution that this does not directly prove haptic fidelity equivalent to human tissue, nor does it establish transfer validity to live surgery. Importantly, learning curve behavior alone should not be interpreted as direct evidence validating haptic fidelity or operative transfer; rather, it indicates only that the feedback was sufficiently consistent to support skill acquisition within the simulated environment. Demonstrating transfer validity to live surgery would require a separate study design incorporating Objective Structured Assessment of Cataract Surgical Skills (OSCAR) or similar validated operative assessments. Puri et al. demonstrated that supervised simulator or wet-lab training allows residents to be better prepared for live surgery [12]. The pattern of group differences across modules can be explained by the specific prior experience of residents. Residents started with significantly higher scores on Module 3 (paracentesis), Module 4 (viscoelastic exchange), and Module 5 (keratome entry) – the three steps they had actually performed before. For these modules, their early performance advantage was clear, and they reached plateaus 2–4 iterations earlier than students. In contrast, for Module 1 (scleral groove) and Module 2 (tunnel dissection), which residents had never performed, the initial performance gap between groups was smaller and, in the case of Module 1, did not become significant until later iterations. Module 2 showed a more persistent difference, suggesting that tunnel dissection requires substantial step-specific prior exposure. The lack of significant initial superiority in these two modules indicates that residents’ prior experience was limited and did not confer proficiency in steps they had never performed, confirming that both groups were effectively novice for these specific tasks. This suggests that prior exposure to specific steps accelerates learning on those exact tasks, but does not automatically transfer to unfamiliar steps that require different instrument handling (e.g., crescent blade for tunnel dissection). General microsurgical background from pterygium surgery (see Table 2) may have provided basic instrument handling and microscope familiarity, but not step‑specific procedural knowledge. Importantly, students without any prior experience eventually achieved comparable plateau performance on all modules after sufficient repetitions, indicating that the simulator can effectively teach even complex steps to novices. A limitation is that the exact number of prior repetitions for the experienced cataract steps was not quantified; future studies should record detailed logs to examine dose‑response relationships. The number of iterations required to reach a plateau varied across modules, with technically demanding steps (scleral groove, tunnel dissection) requiring more practice than procedural steps of lower complexity. Selvander et al. demonstrated that as module difficulty increased, the plateau would be delayed [33]. The performance of medical students in cataract navigation training module reached the plateau in the third iteration, while the overall score of capsulorhexis module did not reach the plateau within 10 iterations. Our module-specific findings extend prior aggregate-score analyses. Selvander et al. reported delayed plateaus for more difficult modules on a virtual-reality cataract simulator, and we similarly observed that Modules 1 and 2 required more iterations to plateau than Modules 3–5. However, our study differs in its focus on visuo-haptic MSICS simulation and its stratification of trainees by prior step-specific experience, which allowed us to distinguish transferable from non-transferable procedural skills. We emphasize that the plateau observed within 10 iterations represents initial stabilization of psychomotor skills in the simulated environment, not mastery of live cataract surgery. We acknowledge that this operational definition of the plateau—namely, the first iteration at which the score no longer showed a statistically significant difference from the 10th iteration score, as determined by post-hoc Bonferroni-corrected comparisons following the Friedman test—is inherently contingent upon the selected statistical approach and the limited 10-iteration design. Real-world cataract surgery learning curves extend over years and thousands of cases. The 10-iteration design was a methodological limitation; the restricted number of iterations may have limited detection of late plateaus, particularly in the student group where curves appeared to still be ascending at iteration 10. Future studies should employ extended iteration schedules to capture the complete learning trajectory. Educational implications The plateau thresholds identified in this study have direct implications for designing simulation‑based training curricula. The iteration numbers at which the plateau was reached may serve as a minimum evidence-based threshold for module-specific simulator training before supervised live surgery. However, as in live surgery, repetition beyond the initial plateau likely yields further refinement; therefore, ‘the more the better’ remains a prudent principle in curriculum design. For medical students (Group A), achieving proficiency on Module 1 (scleral groove) required 8 iterations, whereas Module 3 (paracentesis) required only 6 iterations. This suggests that training programs should allocate more practice time to technically demanding modules such as scleral groove and tunnel dissection. For residents with prior single‑step experience (Group B), plateaus were reached 2–4 iterations earlier, indicating that structured step‑by‑step exposure to cataract surgery before simulation can accelerate learning. Therefore, a stratified training schedule – with more repetitions for novices and shorter, targeted practice for those with partial experience – can optimize simulator resource utilization. Furthermore, the fact that students eventually reached the same plateau performance as residents suggests that simulator training can effectively close the initial experience gap, supporting its role as a standard component of ophthalmology residency preparation. The sample size is relatively small (n = 32). Although Bonferroni correction was applied to post-hoc Friedman comparisons, multiple Mann-Whitney U tests across iterations increased the risk of Type I error. Although facilitation-based instructor guidance and structured debriefing were provided as standard components of the training protocol, these pedagogical processes were not systematically quantified or analyzed in the present study. Future work should examine how educator feedback, debriefing quality, and learner self-assessment influence simulator learning trajectories. Furthermore, our outcomes are simulator-based only and do not directly reflect live surgical performance. We acknowledge that improvements in simulator performance do not automatically imply equivalent improvements in real‑life cataract surgery. Our conclusions regarding learning curves and plateaus are limited to the simulated environment. Further research is necessary to establish the transfer validity of the HelpMeSee MSICS simulator. Conclusion In conclusion, the HelpMeSee MSICS simulator demonstrates measurable module-specific learning curves with excellent inter-rater reliability. Trainees with limited prior experience can reach performance plateaus within 4–8 iterations per module, providing evidence-based minimum training benchmarks. However, simulator proficiency should be viewed as a prerequisite for, rather than a substitute for, supervised live surgical training. Integration of this simulator into standardized ophthalmology curricula may help address the global burden of cataract blindness by accelerating early psychomotor skill acquisition in a risk-free environment. Supplementary Information Supplementary Material 1: Supplementary Table 1. The performance of 5 modules and total score for Group A. Supplementary Table 2. The performance of 5 modules and total score for Group B. Acknowledgements Not applicable. Authors' contributions Mingxue Ren and Hangdong Dai contributed equally to the study conception and experimental design. Ziyi Dong and Lingxia Ye conducted the experiments and data collection. An-Peng Pan and Kaijing Zhou developed the methodology and provided critical reagents. Yimin Yuan performed formal analysis and data interpretation. Mingxue Ren drafted the original manuscript. Xu Xu reviewed and edited the manuscript with substantial intellectual input. All authors critically revised the manuscript and approved the final version for publication. Funding This work was supported by the General Research Project of Zhejiang Provincial Department of Education (Y202147087) and the Standardized Residency Training Education Reform Project of Wenzhou Medical University (2023XZP003). Data availability The datasets analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethical approval and consent to participate The anonymized student evaluation data used in this study were collected as part of routine curriculum assessment at Wenzhou Medical University and did not require separate ethical approval under institutional guidelines. Individual identifiers were removed prior to analysis. This study used fully anonymized surgical simulator performance data. According to Article 32 of China’s Ethical Review Measures for Life Sciences and Medical Research Involving Humans (2023), it qualifies for exemption from ethical review. Prior to the commencement of the study, written informed consent was obtained from all participants. All methods were performed in accordance with the Declaration of Helsinki. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Mingxue Ren and Hangdong Dai contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1: Supplementary Table 1. The performance of 5 modules and total score for Group A. Supplementary Table 2. The performance of 5 modules and total score for Group B. Data Availability Statement The datasets analysed during the current study are available from the corresponding author on reasonable request.