Introduction

In surgery, the learning curve refers to the period during which a surgeon attains proficiency in a specific surgical procedure or technique. Initially, surgeons experience longer operation times and higher rates of complications. As experience and repetition increase, these metrics improve over time, followed by a plateau where limited additional improvements are observed1. As a surrogate for surgeons’ performance in low-rate complication procedures, it is generally accepted that operative time change over a series of procedures should be adopted2. Investigating learning curves of surgical procedures is of utmost importance for several reasons such as patient safety, quality of care, efficiency, surgeon training, outcome optimization, benchmarking and research and innovation3,4. Key reasons why learning curves in robotic surgery are of interest to surgeons include the need to optimize surgical technique by improving outcomes and reducing morbidity and mortality; the importance of objectively assessing surgical proficiency, with operative time serving as a practical and measurable indicator; and the broader aim of improving cost-effectiveness by achieving health recovery efficiently. While the comprehensive evaluation of these aspects often requires complex analyses, operative time—particularly when adjusted for case complexity—represents a valid and widely accepted surrogate for surgical performance and treatment efficiency in this context.

The widespread adoption of robotic systems in recent years has posed new challenges to surgeons and surgical trainees. Surgeons worldwide have had to learn new techniques and instrumentation, requiring significant time and the development of new surgical skills5. To date, several articles have been published regarding learning curves with new robotic systems2,5,6. Some of these studies have investigated robotic-assisted hiatal hernia repair and have reported heterogeneous results7,8,9. This heterogeneity likely reflects differences in surgical experience, case mix, and the lack of standardized metrics to assess robotic performance. Unlike conventional laparoscopy, robotic surgery introduces unique technical challenges and learning requirements, including system docking, instrument control, and three-dimensional visualization. These factors underscore the need for adjusted, objective measures—such as operative time corrected for case complexity—to better define the learning process in robotic surgery and to guide training and clinical implementation2,3.

The aim of this study was to assess the learning curve of robotic-assisted hiatal hernia repair.

Materials and methods

Patient selection, data collection and outcomes

We retrospectively searched on our prospectively maintained database (https://www.herniamed.de) all patients who underwent robotic-assisted hiatal hernia repair at our institution from Mai 2018 to April 2024. We included all types of primary hiatal hernia according to the Allison classification10 and recurrent hiatal hernia operated with the DaVinci (Intuitive) robotic system. No exclusion criteria were applied.

Collected data included demographic and clinical characteristics (age, sex, body mass index (BMI), American Society of Anesthesiology (ASA) score, symptoms), indication for surgery, type of hiatal hernia, operative time, type of surgery (including mesh placement, if any), intraoperative complications, postoperative complications according to the Clavien-Dindo classification11 and the Cumulative Complication Index12, length of postoperative stay, 30-days follow-up visits and long-term follow-up including symptom improvement by screening patients’ medical records. The operating surgeons already had substantial experience in laparoscopic hiatal hernia repair and basic robotic-assisted operations. All surgeons attended specific robotic surgery training programs and performed exercises on the robotic surgery virtual simulator available at our institution (Mimic dV-Trainer – Mimic Technologies, Inc., Seattle, WA).

The primary outcome was the assessment of the learning curve for robotic-assisted hiatal hernia repair, defined as the number of operations required to achieve a significant reduction in operative time. Operative time was defined as the duration from the first skin incision to the final skin suture. Secondary outcomes included intraoperative and postoperative complications, as well as the length of hospital stay.

Surgical technique

Patients were placed in a 20° anti-Trendelenburg supine position and received antibiotic prophylaxis. Pneumoperitoneum was established with a Veress needle. Four 8-mm robotic trocars, along with one 5-mm trocar, were inserted, and the robot was docked (Fig. 1). For type 3 hiatal hernias, we used an anterior approach to gradually free the hernia sac in the lower mediastinum until sufficient distal esophageal length was achieved within the abdominal cavity. For type 1 hiatal hernias, we started by dividing the lesser omentum up to the right diaphragmatic pillar. On the greater curvature, the short gastric vessels were divided, and the left pillar was mobilized to reposition the esophagogastric junction into the abdomen. The diaphragmatic pillars were sutured posteriorly with non-absorbable 0 polyester sutures. The hiatus was reinforced with an absorbable mesh (Cook Biodesign Hiatal Hernia Graft or Phasix ST Mesh) in case of hiatal hernia type 2, 3, or 4, in case of recurrence or BMI ≥ 30 kg/m2. In most cases, a loose, 2 cm long wrap (short-floppy Nissen) was constructed. For post-esophagectomy paraesophageal hernias, the surgical technique involved reducing the intestinal content into the abdominal cavity, followed by suturing the anterior diaphragmatic pillar with non-absorbable 0 polyester sutures and reinforcing it with the absorbable mesh.

Fig. 1
figure 1

Robotic trocar placement.

Full size image

Ethical statement

All procedures were performed in compliance with relevant laws and institutional guidelines and have been approved by local ethic committee (Comitato Etico Cantonale Ticino, 2019 − 01132 CE 3495). The privacy rights of human subjects have been observed and informed consent was obtained from all patients. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement was followed13.

Statistical analysis

Descriptive statistics were presented as frequencies for categorical variables and as means with standard deviations (SD) for continuous variables. The chi-square test was used to compare dichotomous variables, while the Student t-test was employed for continuous variables. According to previously published studies3, an analysis of covariance (ANCOVA) was performed to estimate adjusted operative times and residuals, taking into account age, sex, BMI, ASA score, and hiatal hernia grade. A cumulative sum control chart (CUSUM analysis) was utilized to monitor changes over time and assess the learning phases for each surgeon based on the adjusted operative time. We reported both primary components of the CUSUM analysis graph: the positive and negative curves. These curves together provide a visual representation of operative time behavior as the caseload increases. The significance level was set at 0.05. All analyses were conducted using MedCalc® Statistical Software version 22.021 (MedCalc Software Ltd, Ostend, Belgium; https://www.medcalc.org; 2024).

Results

At our institution, during the study period, three surgeons performed 101 consecutive robotic-assisted hiatal hernia repairs. The mean age was 65.7 ± 14.3 years, 69 (68.3%) patients were female, the mean BMI was 26.7 ± 4.3 kg/m2, and 87 (86.1%) patients had at least one comorbidity. Patients’ characteristics were reported in Table 1.

Table 1 Patients’ characteristics.
Full size table

Regarding the primary outcome, we first performed an ANCOVA, which identified hernia grade as a factor strongly impacting operative time (p < 0.001), while age, sex, BMI, and ASA score did not. After calculating the operative time corrected for covariates for each patient, we proceeded with the CUSUM analysis. All curves clearly visualized an initial learning phase characterized by longer operative times and a subsequent proficiency phase with shorter operative times. To visually identify the point where the learning curve was overcome, we chose the point where the CUSUM analysis no longer showed an increase. The estimated number of cases was 15 for the first surgeon (Fig. 2), 21 for the second surgeon (Fig. 3), and 16 for the third surgeon (Fig. 4). The overall CUSUM analysis indicated that the learning curve should be surpassed after 16–21 cases (Fig. 5).

Fig. 2
figure 2

CUSUM analysis for the first surgeon. The learning phase included the first 15 cases.

Full size image
Fig. 3
figure 3

CUSUM analysis for the second surgeon. The learning phase included the first 21 cases.

Full size image
Fig. 4
figure 4

CUSUM analysis for the third surgeon. The learning phase included the first 16 cases.

Full size image
Fig. 5
figure 5

CUSUM analysis for all surgeon. The learning phase included the first 16–21 cases.

Full size image

We carried out an analysis of the learning and post-learning phases. Patients’ characteristics were similar in both groups, as shown in the tables. Intraoperatively, a Nissen fundoplication was performed on 88 (87.1%) patients, a Toupet in 5 (5.0%) patients, a Dor in 1 (1.0%) patient and a fundophrenicopexy in 7 (6.9%) patients, with no case of Collis gastroplasty. A mesh was implanted in 63 (62.4%) patients. One minor complication occurred during the learning phase. During the repair of a type 3 hiatal hernia, the hernia sac was adherent to the left pleura and was inadvertently opened, resulting in pneumothorax and subcutaneous emphysema. No specific treatment was deemed necessary, and the patient experienced an uneventful postoperative course. No conversion to open surgery or laparoscopy was required. Table 2.

Table 2 Intra- and postoperative results.
Full size table

Postoperatively, 19 (18.8%) complications occurred. Most were low-grade, including 3 cases of hypotension, 2 cases of hypertension, 2 urinary retentions, 1 case of desaturation, and 1 case of postoperative nausea, classified as grade 1 according to the Clavien-Dindo classification. Grade 2 complications included 3 cases of pneumonia, 1 conservatively managed postoperative bleeding, 1 pulmonary embolism, 1 central catheter infection, and 1 decubitus ulcer. Grade 3 complications consisted of 1 case of early recurrence causing dysphagia necessitating reoperation, 1 case of fundoplication perforation requiring robotic-assisted treatment, and 1 case of pleural effusion necessitating chest tube placement. All patients recovered well without long-term complications, and no mortality was recorded. Complications were evenly distributed between the learning and post-learning phases.

After a mean follow-up of 26.5 ± 18.0 months, 13 (12.9%) cases of recurrence were recorded. These were all small axial recurrences following repair of type 3–4 hiatal hernias, and none required surgical revision. During follow-up, data regarding symptoms were retrieved for 98 patients (97.0%). Among patients with reflux, 28 (58.3%) experienced an improvement in symptoms without requiring medications, 15 (31.2%) improved but required daily medications, and 5 (10.4%) had a severe recurrence of symptoms. All patients with dysphagia experienced an improvement in symptoms. Only 6 (5.9%) out of 101 patients reported transient postoperative dysphagia, with only one case — as already mentioned among the complications — requiring surgical revision. Although slightly higher in the proficiency phase, symptom improvements were evenly distributed between the learning and post-learning phases.

Discussion

Our study demonstrated that performing robotic-assisted hiatal hernia repair required between 15 and 21 cases to achieve a significant reduction in operative time. Other outcomes, including intra- and postoperative complications, as well as recurrence during follow-up, were equally distributed between the learning and post-learning phases.

The concept of the learning curve was originally introduced by Theodore Paul Wright in 1936 within aircraft manufacturing. In the 1980s, this term was adopted within the healthcare sector, particularly with the advent of minimally invasive surgery. The surgical learning curve refers to the number of procedures necessary for a surgeon to achieve proficiency and consistently achieve favorable outcomes independently4. Several factors influence the learning curve, including the surgeon’s understanding of anatomy, manual dexterity, the structure of the training program, and various other factors14. Training using surgical models and animal tissue has been demonstrated to facilitate and enhance the learning process14. This preparatory training is instrumental in equipping surgeons with the skills and confidence needed to navigate the complexities of real surgical scenarios effectively.

The introduction of a new surgical technique requires thorough evaluations, including safety, feasibility, effectiveness compared to existing methods, financial considerations, and the time needed for surgeons to become proficient. Soliman et al.15 reported that robot-assisted hiatal hernia repair improved short-term outcomes compared to laparoscopic hiatal hernia repair, resulting in shorter hospital stays and lower complication rates. Conversely, Benedix et al.16 found no clear advantage of robotic surgery over conventional laparoscopy in terms of short-term outcomes and postoperative morbidity. The systematic adoption of robotics in hiatal hernia repair, as with other procedures, remains a topic of debate. In our experience, we transitioned to a fully robotic program in 2017, which included hiatal hernia repair. Although we did not include a control group, the rate of intra- and postoperative complications was relatively low, with only 3 major complications among 101 patients. In terms of functional outcomes, all patients operated on for dysphagia experienced a notable improvement, while 89.6% of patients operated on for reflux reported symptom improvement, with a slightly higher recurrence rate during the learning phase. Moreover, in our experience, robotic assistance allowed for sufficient distal esophageal mobilization in all patients, and no Collis gastroplasty was required. This suggests that robotic surgery may reduce the need for esophageal lengthening procedures by enabling safer and more extensive mediastinal dissection.

While we did not collect financial data due to the absence of a control group, it is already established that robotic surgery tends to be more expensive than conventional laparoscopy for simple operations17. However, it may also be cost-effective for more complex procedures18, which might include hiatal hernia repair. Panse et al. in their economic evaluation of laparoscopic and robotic paraesophageal hernia repair concluded that laparoscopy was likely to be more cost-effective for most institutions, however, robotic surgery may improve outcomes enough to be cost-effective after surpassing the learning curve.

Defining and measuring a learning curve is challenging due to multiple confounders related to the surgeons, patients, specific procedures, and the institution where the surgery takes place6. Identifying proper outcomes is essential. In their systematic review of robot-assisted surgery, Soomro et al.5 found that time-based metrics were used as assessment tools in 42 of the 49 studies. Similarly, Kassite et al.6 reviewed 166 studies and identified 46 endpoints, with total operating time and total robotic time being commonly used measures. We deemed it appropriate to measure the learning curve using operative time as the primary outcome, while noting that the conversion rate to open surgery and intra- and postoperative complications were relatively low. This posed a challenge as hiatal hernia surgeries can range from simple to highly complex. To account for this variability, we conducted an ANCOVA to estimate adjusted operative time and residuals from a regression analysis, factoring in the level of difficulty and patient characteristics. A similar statistical approach was employed by Bernardi et al.3, who assessed the learning curve of robotic-assisted hepatic resections by adjusting for the level of surgical difficulty.

In our experience, the learning curve for robotic-assisted hiatal hernia repair was estimated to be between 15 and 21 cases. The robustness of our model was supported by several factors. First, the CUSUM analysis clearly visualized an improvement over time on the curves, with a similar caseload for all included surgeons. Significant improvement was also demonstrated by shorter operative times in the post-learning phase compared to the learning phase. The difficulty of cases was homogeneously distributed over time, minimizing the likelihood of significant selection bias. Our results are consistent with the available literature. Straatman et al.9 conducted a retrospective study with 109 patients and 4 surgeons, finding that each surgeon performed between 22 and 32 cases, with a significant CUSUM analysis inflection point after 7–15 cases within a structured learning pathway with proctoring. Sarkaria et al.7 retrospectively studied 24 patients undergoing symptomatic giant paraesophageal hernia repair and found that the initial learning curve is relatively short for experienced minimally invasive surgeons. Conversely, Lin et al.8 examined a series of robotic hiatal hernia repairs performed by a single surgeon, reporting that the training phase was achieved after 40 cases and a high level of mastery after approximately 85 cases. In our study, we included only surgeons with prior experience in laparoscopic hiatal hernia surgery, which likely accounts for the shorter learning curve observed. Among the articles investigating the learning curve in robotic-assisted hiatal hernia surgery, ours was unique in employing ANCOVA analysis to adjust for surgical difficulty and estimate corrected operative times, potentially leading to more reliable results.

This study has several limitations. We conducted a retrospective analysis using a prospectively collected database with a relatively small number of patients. Given that some studies have shown significant improvements even after hundreds of cases, one might argue that performing 101 hiatal hernia repairs across three surgeons may not be sufficient. However, our findings indicate a significant reduction in operative time after only 15–21 cases, with further marginal gains expected as caseloads increase substantially. Another limitation concerns the varying difficulty levels of the hiatal hernias included in our analysis. Procedures involving paraesophageal hernias with mesh implantation are typically more complex and time-consuming compared to simpler type 1 hiatal hernias. To address this, we adjusted for the difficulty level and patient characteristics by calculating adjusted operative times. Unfortunately, standardized pre- and postoperative questionnaires were not administered, limiting the ability to objectively assess symptom improvements during the follow-up. All three surgeons were already proficient in laparoscopic hiatal hernia surgery, thus the learning curve we assessed applies specifically to experienced surgeons and may not generalize to novices. Additionally, the relatively low annual caseload could be considered a limitation; nonetheless, we maintained consistency by having the same three surgeons perform all hiatal hernia operations. Furthermore, potential biases may arise from variations in the competence levels of table assistants and operating room nurses, which can influence aspects of operative times such as docking and instrument exchange times.

Despite these limitations, our study provides valuable insights, particularly noting the short learning phase observed for expert surgeons adopting robotic-assisted hiatal hernia repair. These findings are relevant and applicable to surgeons and centers integrating robotic surgery into their approach for hiatal hernia repairs.

Conclusions

Our findings showed that experienced surgeons performing robotic-assisted hiatal hernia repair have a rapid learning curve. Specifically, we observed a significant reduction in operative time after completing an estimated 15 to 21 cases. Intraoperative and postoperative complication rates were low and evenly distributed during and after the learning phase, underscoring the procedure’s safety and effectiveness in experienced surgical teams.