Evaluation of the learning curve for totally robotic single-anastomosis duodenal–ileal bypass with sleeve gastrectomy by CUSUM analysis

Single-anastomosis duodenal–ileal bypass with sleeve gastrectomy (SADI-S) has similar ecacy and lower complication rate in the treatment of morbid obesity and obesity-associated metabolic diseases compared with the biliopancreatic diversion with a duodenal switch. The use of a robotic surgical system reduces surgical diculties and improves surgical outcomes. The learning curve reects the rate of skills or knowledge acquired in a certain period of time. However, the learning curve for robotic SADI-S has not been estimated.We used the cumulative sum analysis method to investigate the learning curve of totally robotic SADI-S. Textbook outcome analysis was performed to comprehensively dene surgical success or failure.Multivariate analysis was performed to predict independent risk factors for complications and operative time. The moving average method was used to reect the trends in operative time.This study showed that the learning curve for totally robotic SADI-S was 27 cases. Surgeon experience (case number and successful case number) was an independent predictor of the total operative time. A successful case number was the only independent predictor of surgical success or failure in this study (β = 0.084; P = 0.001). Except for the rst assistant level, scrubbed nurse level, operative time, and proportion of abdominal drainage tube, there was no signicant difference between the learning stage and mastery stage groups.The learning curve for totally robotic SADI-S was 27 cases. Surgeon experience including case number and successful case number were identied as independent predictors affecting the total operative time. A successful case number was the only independent predictor of surgical success or failure.


Introduction
Bariatric and metabolic surgery-evolved from laparotomy to the laparoscopic approach-is switching to a robot surgical system. Compared with conventional laparoscopy, the robotic surgical system offers several advantages [1][2][3][4][5][6][7] , including 7 degrees of freedom, tremor ltration, three-dimensional highde nition visualization, and superior ergonomics, which increase surgical safety and improve surgical outcomes. Since Cadiere reported the world's rst robotic bariatric surgery in 1999, the robotic surgical system is being used in various bariatric procedures such as sleeve gastrectomy (SG), Roux-en-Y gastric bypass (RYGB), and biliopancreatic diversion with duodenal switch (BPD/DS); several studies have reported the learning curve for SG, RYGB, and BPD/DS [8][9][10][11] . Single-anastomosis duodenal-ileal bypass with sleeve gastrectomy (SADI-S) is being used for the treatment of morbid obesity for 14 years since it was rst proposed by Torres et al. in 2007 [12] . SADI-S has been recommended by the International Federation for the Surgery of Obesity and Metabolic Disorders in the treatment of morbid obesity and obesity-related metabolic diseases [13] . However, only a few studies are present on robotic SADI-S [14][15][16][17] .
Moreover, the learning curve for robotic SADI-S has not been estimated. Therefore, this study aimed to estimate the learning curve for totally robotic SADI-S.

Patient and Clinical Data
Seventy-one consecutive patients undergoing totally robotic SADI-S between March 2020 and April 2021 were included in this study. All the surgeries were performed by the same surgeon. We recorded and analyzed the following factors: patient gender, age, preoperative body weight, body mass index, waistline, standard live volume, American Society of Anesthesiology Physical Status Classi cation, operative time, rst assistant level, scrubbed nurse level, number of staplers used, the proportion of abdominal drainage tube, length of postoperative stay, complications (30-day), conversion to laparotomy, mortality, reoperation, and readmission within 30 days. The Dindo-Clavien classi cation was used to classify the severity of complications [18] .This research including all experimental methods was approved by the ethics committee of the China-Japan Union Hospital of Jilin university,and informed consent was obtained from all participants. All procedures performed in studies involving human participants were carried out in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Operative Technique
The Da Vinci Xi® model was used for totally robotic SADI-S. The patient undergoing surgery was in reverse Trendelenburg position with open legs and arms. The rst trocar for the 30°camera (8 mm, robotic arm 3) was placed at the lower edge of the navel. Other trocars were inserted under visual inspection. The second trocar for the stapler insertion (12-mm, robotic arm 1) was placed at the junction of the right anterior axillary line and right end of the greater curvature of the stomach. The third trocar for liver retractor (8 mm, robotic arm 2) was placed at the junction of the right midclavicular line and costal margin. The fourth trocar (8-mm, robotic arm 4) was placed at the junction of the left midclavicular line and left end of the greater curvature of the stomach. All the trocars were spaced more than 8 cm, avoiding each other's interference of robotic arms. A 300-cm common channel was measured retrograde from the ileocecal valve and marked by sutures. SG was performed about 4 cm from the pylorus over a 34 Fr bougie tube. Complete duodenum transection was performed about 2 cm from the pylorus. Finally, duodenal-ileal anastomosis was performed continuously by using an absorbable 3 − 0 barbed suture.

De nitions
According to a review of existing "Textbook Outcome" metrics in the literature [19][20][21][22][23] , the de nition of surgical failure was subsequently expanded for totally robotic SADI-S-speci c outcomes selected based on clinician consensus among a team of bariatric surgeons at our institution. The nal de nition of surgical failure in totally robotic SADI-S included the following 5 key parameters: operative time > 210 min, length of postoperative stay > 7 days, postoperative morbid event ≥ Clavien grade II, conversion to laparotomy, and rehospitalization or death after totally robotic SADI-S. Surgical failure was recorded when any of the aforementioned parameters were observed. The level of the surgeon, rst assistant, and scrubbed nurse was represented as the number of totally robotic SADI-S they participated in.
Cumulative Sum (CUSUM) analysis method The CUSUM analysis method and best-t model were used to estimate the learning curve for totally robotic SADI-S. This method was used for all study cases considering operative time. First, all the patients were numbered from 1 to 71 on the basis of the earliest to the latest date of surgery and then, the CUSUM value of operative time for each case was calculated using the formula mentioned below [24] . The scatter plot on the learning curve of totally robotic SADI-S was drawn by case number as X-axis and CUSUM (operative time) as Y-axis. The IBM® SPSS® Statistics software version 22.0 is used to t the learning curve. The tting model test is judged by the P value, and P 0.05 means that the tting of learning curve is successful. The goodness-of-t is judged by the coe cient of determination (R2). The closer to the 1 for R2 value, the higher the goodness-of-t. The model with the largest R2 value will be selected as the best tting model. The peak of the CUSUM curve was considered as the end of the learning process. Based on the peak of the CUSUM curve, each patient was classi ed into either the learning stage group or the mastery stage group.
(Xi: the operative time for the corresponding patient; µ: the mean operative time for all cases; n: the casenumber)

Moving Average Method
The moving average method is a simple, smoothing, forecasting technique. According to time series, sequential time averages containing a certain number of items are sequentially calculated by item by item to re ect the long-term trend [25]. We employed the moving average method to calculate the operative time of totally robotic SADI-S using the formula mentioned below: SMA(n) = Xn+ Xn − 1 + Xn − 2 + + Xn − 13 14 Statistical Analysis SPSS 22.0 was used for statistical analysis. Measurement data were expressed as mean ± standard deviation and were analyzed by the independent-sample Student's t-test (Normality data) or Mann-Whitney U test (skewed data), as appropriate. The calculated data were analyzed by the Chi-square test. A P value of < 0.05 was considered statistically signi cant.

Results
Seventy-one consecutive patients undergoing totally robotic SADI-S were included in this study and the overall follow-up rate was 100%. Among the 71 patients, 42 were women and 29 were men with a mean age of 33.65 years (range, 18-59 years). The patient demographic data are summarized in Table 1. The best tting equation of learning curve for totally robotic SADI-S was as follows: CUSUM (operative time) = 96.21 38.75X-1.21X 2 9.25E-3X 3 (X is surgical case number). The P value of tting model test was 0.000, which meant that the tting of learning curve was successful. The coe cient of determination (R2) was 0.933, which represented the goodness-of-t was excellent. The CUSUM (operative time) of learning curve reached its peak when the number of surgical cases accumulated to the 27th case.This suggests that the learning curve for totally robotic SADI-S in a CUSUM manner was 27 cases (Fig. 1). Subsequently, all the patients were classi ed into the learning stage group (the rst 27 patients) and the mastery stage group (the last 44 patients). No signi cant difference was observed between the two groups in terms of patient demographic data ( Table 2). The operation-related parameters such as the number of staplers in SG, complications, reoperations, readmission, and postoperative hospital stay were similar in both groups (Table 3). However, the rst assistant level and the scrubbed nurse level were signi cantly less in the learning stage group compared with the mastery stage group (Table 3). Moreover, operative time and proportion of abdominal drainage tubes were also signi cantly higher in the learning stage group compared with the mastery stage group ( Table 3). The operative time on the moving average curve decreased throughout the study period. However, the slope was more pronounced in the learning stage group (until 27th case) than the mastery stage group (Fig. 2).  classi cation (one with seroperitoneum, one with postoperative abdominal bleeding, one with delayed gastric emptying, and one with duodenal-ileal anastomotic leakage), and all of them were cured successfully by the conservative treatment. Two patients suffered from gastric leakage (grade IIIb) and required reoperation. One patient was transferred to the intensive care unit because of postoperative acute respiratory failure (grade IV) and was eventually cured.
Multivariate analysis identi ed surgeon experience (case number and successful case number) as the independent predictor of operative time (Table 4). However, successful case number was found to be the only independent predictor of surgical success or failure (β = 0.084; P = 0.001). The comparison of patient demographic data between the successful and unsuccessful surgery groups is shown in Table 5. The patient demographic data between the successful surgery group and the unsuccessful surgery group was not signi cantly different (Table 5). Therefore, the patient demographic data is not correlated with the success of the surgery. SG: Sleeve gastrectomy. The de nition of successful case in totally robotic SADI-S included freedom from the following factors: operative time > 210 min, the length of postoperative stay > 7 days, postoperative morbid event ≥ Clavien grade II, conversion to laparotomy, and rehospitalization or death after totally robotic SADI-S. The de nition of successful case in totally robotic SADI-S included freedom from the following factors: operative time > 210 min, the length of postoperative stay > 7 days, postoperative morbid event ≥ Clavien grade II, conversion to laparotomy, and rehospitalization or death after totally robotic SADI-S.

Discussion
SADI-S is being used as a surgical procedure in the treatment of morbid obesity for 14 years since it was rst proposed in 2007 by Torres et al. as a simpli ed procedure of BPD/DS [12] . SADI-S is preferred over BPD/DS because of the reduced operative risk by eliminating one anastomosis and similar weight loss and remission of metabolic diseases. Nonetheless, SADI-S has some technical challenges; large waistline, large liver, thick abdominal walls, and substantial visceral fat make exposure, dissection, and reconstruction di cult. Among laparoscopic SADI-S, duodenoileostomy is the most challenging, which requires advanced laparoscopic skills, including suturing and intracorporeal knot tying. These skills are di cult to master because of laparoscopic limitations such as a two-dimensional view and a limited range for instrument movement. Compared with conventional laparoscopy, the robotic surgical system offers several advantages [1][2][3][4][5][6][7] including 7 degrees of freedom, tremor ltration, three-dimensional highde nition visualization, and superior ergonomics, which increase surgical safety and improve surgical outcomes. Using a robotic system, surgeons can easily and precisely perform a 2-layer hand-sewn duodenoileostomy, avoid circular stapler use, eliminate the di culties and risks in anvil introduction, and decrease the stricture rates as gastrojejunostomy during gastric bypass [26] .
To our knowledge, this study is the rst to estimate the learning curve for robotic SADI-S. According to our results, the learning curve of totally robotic SADI-S is 27 cases and can be divided into Phase 1 and Phase 2. Phase 1 includes the rst 27 patients and represents the initial learning stage. Phase 2 represents the mastery stage, with a signi cant reduction in the operative time and proportion of abdominal drainage tube, and a signi cant increase in the rst assistant level and scrubbed nurse level.
The learning curve for other bariatric procedures has been reported previously. Previous studies [9,27] have shown that the learning curve for robotic SG is 20-25 cases. The learning curve for robotic RYGB ranged from 14 to 84 cases [10,[28][29][30][31][32] . Sudan et al. [11] reported that the learning curve for robot-assisted BPD/DS was 50 cases. The learning curve depends on various parameters including the surgeon's skill level, assistant surgeon, scrubbed nurse, ancillary staff, procedure complexity, and patient-related risk factors. In this study, by multifactorial analysis, we analyzed independent risk factors that may affect the learning curve for totally robotic SADI-S. The results indicated that the independent predictor affecting the total operative time was the surgeon's experience (including case number and successful case number). Our results are consistent with the results reported previously [11,29] . Therefore, surgeons assisting in robotic SADI-S need to receive professional robotic training to successfully pass phase 1 of the learning curve.
The mean total operative time in this study was 183.19 min, which is within the range of those reported previously for robotic SADI-S (145-204 min) [14][15][16][17] . An increase in surgeon's pro ciency with the increase in surgical experience was re ected in the operative time required; of note, the slope was more pronounced in the learning phase.
This study showed that totally robotic SADI-S is a feasible and safe surgical approach for morbid obesity. A total of 7 patients developed complications (9.9%), including seroperitoneum (grade II), delayed gastric emptying (grade II), abdominal bleeding (grade II), duodenal-ileal anastomotic leakage (grade II), gastric leakage (2-grade IIIb), and postoperative acute respiratory failure (grade IV). In addition to two patients with gastric leakage (grade IIIb) requiring reoperation, the other 5 patients with the complications were cured by the respective treatment. In general, surgeons are concerned about a higher complication rate during the initial learning phase of the learning curve, wherein they plan to develop a new procedure. In this study, no statistical difference was observed in terms of morbidity between the rst 27 patients (learning stage) and the last 44 patients (mastery stage) (9.7% vs. 9.1%; P = 1.000). Thus, totally robotic SADI-S for patients with obesity is completely safe even during the initial phase of the learning curve.
The main limitation of robotic surgery is the perceived higher cost compared with that of laparoscopy.
Most previous studies reported that the use of a robotic surgical system increases the cost of the procedure [32][33][34] . However, Hagen et al. [35] reported that the overall cost of robotic RYGB is less compared with laparoscopy. We did not analyze the cost in our study; however, more safety may equalize higher cost. Moreover, the cost of robotic surgery is not necessarily higher than conventional laparoscopy [35] . This study is the rst to report the learning curve for totally robotic SADI-S. However, the study has some limitations. On the one hand, it was a retrospective study. One the other hand, the cost of robotic SADI-S was not analyzed. Randomized controlled trials using large sample sizes are required for further study.

Conclusion
The learning curve for totally robotic SADI-S was 27 cases. Surgeon experience (case number and successful case number) affected the total operative time more than other factors. Concerning a composite event de ning surgical success or failure, successful case number was the only independent predictor identi ed in this study.

Declarations
Author contributions L.W. and T.J. participated in the design of the study and helped to draf the manuscript. ZY.W. performed the statistical analysis and helped to draf the manuscript. All authors reviewed the manuscript.

Funding
No funding source. Figure 1 The CUSUM curve showed that CUSUM value for operative time reached a peak at 27th case, followed by a rapid decrease.This suggests that the learning curve for totally robotic SADI-S in a CUSUM manner was 27 cases.