Defining the learning curve of robotic portal segmentectomy in small pulmonary lesions: a prospective observational study

Although robotic segmentectomy has been applied for the treatment of small pulmonary lesions for many years, studies on the learning curve of robotic segmentectomy are quite limited. Thus, we aim to investigate the learning curve of robotic portal segmentectomy with 4 arms (RPS-4) using prospectively collected data in patients with small pulmonary lesions. One hundred consecutive patients with small pulmonary lesions who underwent RPS-4 between June 2018 and April 2021 were included in the study. Da Vinci Si/Xi systems were used to perform RPS-4. The mean operative time, console time, and docking time for the entire cohort were 119.2 ± 41.6, 85.0 ± 39.6, and 6.6 ± 2.8 min, respectively. The learning curve of RPS-4 can be divided into three different phases: 1–37 cases (learning phase), 38–78 cases (plateau phase), and > 78 cases (mastery phase). Moreover, 64 cases were required to ensure acceptable surgical outcomes. The total operative time (P < 0.001), console time (P < 0.001), blood loss (P < 0.001), and chest tube duration (P = 0.014) were reduced as experience increased. In conclusion, the learning curve of RPS-4 could be divided into three phases. 37 cases were required to pass the learning phase, and 78 cases were needed to truly master this technique.


Introduction
In recent years, with the widespread use of computed tomography (CT) in clinical practice and lung cancer screening, a growing number of small pulmonary lesions have been detected [1]. Recently, several studies demonstrated that segmentectomy was an alternative to lobectomy for small pulmonary cancer lesions, especially for those shaped with ground-glass opacities (GGOs) [2][3][4]. Early in this century, robotic-assisted thoracoscopic surgery (RATS) was first used for lung resection [5,6]. Since then, RATS has developed rapidly, and many thoracic surgeons have tried to apply RATS for segmentectomy [7][8][9][10][11]. It is therefore important to investigate the learning curve of robotic segmentectomy to shed light on education and technique with the purpose of better serving patients. Unfortunately, studies on the learning curve of robotic segmentectomy are quite limited [12,13], and prospective studies on this issue are still lacking.
In our previous prospective study, we defined the learning curve of robotic portal lobectomy with 4 arms (RPL-4) [14]. Herein, we set a prospective study to further define the learning curve of robotic portal segmentectomy with 1 3 4 arms (RPS-4) [15,16] using the first 100 RPS-4 s performed by a single surgical team in a single center.

Patient selection
In June 2018, we began to build a database including all consecutive cases receiving RATS performed by Dr. Hao-Xian Yang's surgical team using the Da Vinci Si/ Xi system (Intuitive Surgical Inc., USA) [14]. In April 2021, we completed the 100th RPS-4 and then used these 100 consecutive RPS-4 s to assess the learning curve of this technique. The authenticity of this article has been validated by uploading the key raw data into the Research Data Deposit public platform (www. resea rchda ta. org. cn), with the approval RDD number RDDA2021338634 on 16 September 2021.
The inclusion criteria for this study were as follows: (1) age from 18 to 80 years; (2) pulmonary nodule located in the outer one-third of the lung field; (3) nodule size ≤ 2 cm; and (4) the consolidated part of the nodule being ≤ 50% if the tumor size was over 1 cm. The exclusion criteria for this study were as follows: (1) surgeries other than segmentectomy: wedge resection, lobectomy, bilobectomy, and pneumonectomy; (2) history of thoracotomy at the same side; and (3) nodule located in the right middle lobe.

Data collection
A comprehensive case report form (CRF) was previously designed for each included case as we described previously [14]. The general clinical characteristics, such as sex, age, tumor location, and clinical stage, were recorded before the operation. The cost of time for each step of the operation, such as the console time, docking time, and time for incision closure, were recorded simultaneously by an assistant during the operation. The postoperative variables were also collected and recorded in a timely manner in the CRF, once observed. The surgical complexity was recorded for each operation, which was defined by Oizumi and colleagues in 2011 [17]. During the study period, the thoracic drainage tube could be removed if the following two criteria were met simultaneously: (1) no air leakage and (2) thoracic drainage for 24 h < 250 ml. The 8th edition of lung cancer stage classification by the American Joint Committee on Cancer [18] was applied in this study. Intra/postoperative complications were recorded and classified according to the Clavien-Dindo classification system [19].

Operative technique
One surgical team led by Dr. Hao-Xian Yang as the attending surgeon performed all 100 RPS-4s. The team had an experience of over 2000 cases of video-assisted thoracic surgery (VATS) procedures, in which more than 400 cases were segmentectomies; the other procedures included lobectomies, minimally invasive thymectomies, and minimally invasive esophagectomies, etc. All of the circulating nurses and scrub nurses were from the same robotic surgery team. RPS-4 procedures were used in this cohort [15,16]. The port setting for RPS-4 was similar to that for RPL-4, which was described previously [14]. Figure 1 shows the RPS-4 port location strategy based on the specific lobe. However, for right-side S2 segmentectomy, the assistant port was set at the point of lower lobe segmentectomy to enable a suitable direction to handle the anatomical structures with staplers. We used various methods to deal with blood vessels depending on the diameter and specific anatomy of the vessel branches. For general principal, we used ultrasonic scalpel or silk ligation to deal with very small vessels and used vessel clips or staplers to deal with big vessel branches. After the vessel branches and the segmental bronchus were taken, the intersegmental borderlines could be identified and we continued an anatomical dissection by lifting the residue of the segmental bronchus along the intersegmental veins to make a better display of the intersegmental interfaces. The fluorescence were used in only three cases whose intersegmental interfaces were not well displayed, and the modified inflation-deflation method was not used in the study period because of the use of CO 2 insufflation. At last, staplers were used to resect the targeted segments.

Cumulative sum analysis
In this study, the learning curve was analyzed by cumulative sum (CUSUM) analysis, which was similar to our previous study on the learning curve of RPL-4 [14]. The CUSUM statistic of the first case was equal to the total operative time of the first case minus the average total operative time of all 100 cases. The CUSUM statistics of the second and subsequent cases were equal to the difference between the total operative time and the average added CUSUM statistic of the previous case [12,[20][21][22][23]. CUSUM analysis based on console time was also performed to reduce the bias caused by different surgical assistants [14].

Risk-adjusted cumulative sum analysis
Furthermore, risk-adjusted CUSUM (RA-CUSUM) was also used to investigate clinical indicators that may impact the learning curve [24]. In this study, surgical failure was defined as any of the following events that occurred: conversion to open, perioperative complications, and postoperative complications [12,14]. Variables with P < 0.15 in univariate logistic models were included in the multivariate analysis. The probability of surgical failure was calculated using a multivariate logistic regression model. The RA-CUSUM was defined as RA- indicates surgical success, represents the observed event rate, and P i represents the expected rate of surgical failure of each case [24].

General statistical analysis
IBM SPSS Statistics (version 25.0, IBM Corp.) was used to perform all statistical analyses. A two-sided P < 0.05 was considered statistically significant. The Mann-Whitney U test or Kruskal-Wallis H test was used to compare continuous variables between groups, and the Pearson χ 2 test or Fisher's exact test was used for categorical variables. Scatter plots of CUSUM and RA-CUSUM against consecutive cases were drawn and fit with curves based on nonlinear regression (least-squares fit method) in GraphPad Prism 6.0 (GraphPad, La Jolla, CA, USA).

Clinicopathological characteristics of patients
From June 2018 to April 2021, a total of 340 robotic thoracic surgeries were performed. A flowchart of the inclusion process is shown in Figure S1. The general clinicopathological characteristics of the entire cohort are summarized in Table 1. A total of 68.0% of the patients were female, and the mean age of the entire cohort was 48.8 ± 11.4 years (median: 49.0 years, range 26-80 years). The mean tumor diameter was 1.0 ± 0.3 cm (median: 0.9 cm, range 0.5-2.0 cm). The average total operative time was 119.2 ± 41.6 min (median: 110 min, range 60-259 min). Adenocarcinoma (69/100, 69.0%) was the predominant histological type, and pT1aN0M0 accounted for most of these adenocarcinoma cases (40/69, 58.0%). No conversion to open surgery or perioperative death occurred in this series.

The learning curves
Based on CUSUM analysis, the scatter plot of the CUSUM for total operative time (skin to skin) against consecutive cases was fit to a fifth-order polynomial, representing the learning curve of total operative time (R 2 = 0.8641; Fig. 2). To reduce bias caused by different surgical assistants, the scatter plot of CUSUM for console time (time of operating console) against consecutive cases was further fit to a fifth-order polynomial, representing the learning curve of console time (R 2 = 0.8871; Fig. 3). Whenever total operation time or console time was used to reflect the learning process, the learning curve could be divided into three phases: the learning phase (cases 1-37, phase 1), the plateau phase (cases 38-78, phase 2), and the mastery phase (cases > 78, phase 3).
Furthermore, the univariate logistic regression model indicated that FEV1% and blood loss were risk factors associated with surgical failure, and the multivariate logistic regression model further confirmed that FEV1% and blood loss were independent risk factors associated with surgical failure (Table S1, online only). Based on this multivariate logistic model, RA-CUSUM analysis was performed, and a scatter plot fit with a fifth-order polynomial was depicted (Fig. 4). As Fig. 4 shows, a continuing downward trend was found after 64 cases, indicating that at least 64 RPS-4 cases were needed to ensure a satisfactory surgical outcome.

Comparison between phases
The comparison of clinicopathologic characteristics is shown in Table S2 (online only). All the clinicopathologic characteristics were well balanced among the three phases     Table S2, online only) and surgical system (P < 0.001; Table S2, online only). The comparison of perioperative outcomes between different phases is shown in Table 2. As predicted, reductions in total operative time (P < 0.001) and console time (P < 0.001) were observed over the three phases, but the docking time showed a decreasing trend over the three phases without statistical significance (P = 0.071). Interestingly, blood loss (P < 0.001) and chest tube duration (P = 0.014) showed significant downward trends over the three phases. Comparison between every two adjacent phases showed that the total operative time and console time were different between phases 1 and 2 (P = 0.003, P = 0.007, respectively), and this difference was also observed between phases 2 and 3 (P = 0.007, P = 0.001, respectively). However, the rates of intraoperative complications in phases 1 and 2 were up to 8.1% and 7.3%, respectively, but this rate decreased to 0 in phase 3. Similarly, the rates of postoperative complications in phases 1 and 2 were up to 13.5% and 19.5%, respectively, but this rate decreased to 9.1% in phase 3.

Discussion
Since the first use of RATS for lung resection early in this century [5,6], RATS has developed rapidly and has been widely applied in all types of lung resections including segmentectomy [7]. Previous studies demonstrated that robotic segmentectomy was safe and feasible and even had better short-term outcomes than video-assisted thoracoscopic segmentectomies [7][8][9][10][11]. Learning to master this technique easily and quickly is crucial for any surgeon who performs these types of procedures and utilizes these technologies. Unfortunately, limited studies have analyzed the learning curve of robotic segmentectomy in pulmonary lesions [12,13], and it is therefore important to share the experience of overcoming the learning curve from different thoracic surgeons. Herein, we aim to embody and visualize the process of learning the RPS-4 and share our experience with other surgeons. This study is unique not only because it is specific to a single type of operation (RPS-4) by a single surgical team but also because, more importantly, it is a prospective study. The results of this study suggested that the learning curve could be divided into three phases: the learning phase (phase 1, cases 1-37), the plateau phase (phase 2, cases 38-78), and the mastery phase (phase 3, > 78 cases). As predicted, reduced operation time and console time were observed as operation experience accumulated. The blood loss and chest tube duration showed a downward trend with the increase in consecutive surgeries. RA-CUSUM analysis further suggested that at least 64 cases were required to guarantee satisfactory surgical outcomes of the RPS-4 procedure, indicating the mastery of RPS-4.
Pardolesi et al. previously reported a retrospective study including 17 patients with robot-assisted segmentectomies [7]. A 3-or 4-incision strategy with a 3-cm utility incision in the anterior fourth or fifth intercostal space was performed during the surgical procedure [7]. A similar procedure was also reported by Toker et al. for 21 patients who underwent robotic segmentectomies [8]. Both studies suggested that robotic segmentectomy was a feasible and safe procedure, but the sample sizes of these studies were not large enough to elucidate the learning curve of robotic segmentectomy. Cerfolio et al. reported 100 consecutive robotic segmentectomies using a complete portal procedure that was similar to this study, which confirmed the feasibility and safety of this technique [10]. Zhang et al. reported the learning curve of robotic segmentectomy from 104 consecutive cases [12]. Their data suggested that 21 cases were needed to cross the learning curve, and 40 cases were required to assure feasible perioperative outcomes [12]. However, Zhang et al. only analyzed the total operative time, but the console time was not available due to the respective nature. Console time excluded the potential bias caused by different surgical assistants, and we therefore believe that it is important to elucidate the learning curve. Le Gac et al. also reported that 30 cases of experience were required to pass the learning curve of robotic segmentectomy in a retrospective study [13]. However, they did not further elucidate how many cases were needed to truly master this technique. It is worth noting that they used a 3-arm, not a 4-arm technique, which is different from our method. Compared to previous studies, we believe that the prospective nature of this study makes the results more reliable.
In this study, we did not only consider operative times, but reductions in blood loss and chest tube duration were also observed among the three phases. Our results demonstrated that the occurrence of intra/postoperative complications in phase 3 had a decreasing trend compared with that in phase 1 or phase 2. This outcome indicated that passing the plateau phase (phase 2: cases 38-78) was essential to minimize the incidence of surgical complications and to truly master this technique. RA-CUSUM analysis further confirmed that at least 64 cases were required to guarantee satisfactory perioperative outcomes. Based on these findings, we propose that the mastery of RPS-4 for satisfactory surgical outcomes requires at least 78 cases of experience.
During the study period, the surgical team not only performed RPS-4 but also other robotic thoracic surgeries, including robotic lobectomies and mediastinal tumor resections. This must be considered when interpreting our data. However, it is worth noting that segmentectomy is considered more technically demanding than lobectomy because of its complexity [12]. This study suggested that at least 37 cases of experience were required to overcome the learning curve of robotic segmentectomy, but our previous study suggested that only ten cases of experience were required to overcome the learning curve of robotic lobectomy [14].
The surgical system used was not controlled between the three phases. The Da Vinci Xi system was more likely to be utilized in phase 3, due to timing. In our center, the Da Vinci Si system was applied in January 2016, and the Da Vinci Xi system was applied in April 2020. We therefore could only use the Si system during the initial period of this study. There may be some advantages in docking for the Da Vinci Xi system, but the port location, surgical technique, instrument choice, and surgical team were all the same during the study period. Moreover, because the vascular stapler of Da Vinci Xi system has not yet been authorized in mainland China, pulmonary vessels and bronchus were transacted by assistant using staplers during the whole study period. The results of this study also demonstrated that the surgical system was not associated with surgical failure (P = 0.714,  Table S1, online only). However, using two robot systems must also be considered when interpreting our data. The limitations of this study need to be considered. First, there were 100 cases in this study, but the learning curve might be different if more cases were included. In addition, the learning curve of RPS-4 for individual segments might be different. We are building the prospective database continuously and will update the data of the learning curve in the future. Second, the surgical team in this study was led by a senior thoracic surgeon with rich experience in VATS and had experience with over 300 robotic thoracic surgeries during the study period. The experience of the surgeon must be considered when interpreting our results. Nevertheless, a unique surgical team in a single center may guarantee the homogeneity of the data by excluding other surgeons' experience and making the outcomes reproducible.
In conclusion, our data indicate that the learning curve of RPS-4 can be divided into three phases. Thirty-seven cases were required to overcome the learning curve, and at least 78 cases were needed to pass the plateau phase for mastery of RPS-4.