Learning Curves of Robotic Spine Surgery in a System Lack of Active Perception: Potential Roles of Teamwork and Unmet Needs

Yu-feng Su Kaohsiung Medical University Chung-Ho Memorial Hospital Tai-Hsin Tsai Kaohsiung Medical University Chung-Ho Memorial Hospital Keng-Liang Kuo Kaohsiung Medical University Chung-Ho Memorial Hospital Chieh-Hsin Wu Kaohsiung Medical University Chung-Ho Memorial Hospital Cheng-Yu Tsai Kaohsiung Medical University Chung-Ho Memorial Hospital Yen-Mou Lu Kaohsiung Medical University Chung-Ho Memorial Hospital Shiuh-Lin Hwang Chi-Hsien Spine Hospital, Kaohsiung, Taiwan Pei-Chen Lin Kaohsiung Medical University Ann-Shung Lieu Kaohsiung Medical University Chung-Ho Memorial Hospital Chih-Lung Lin Kaohsiung Medical University Chung-Ho Memorial Hospital Chih-Hui Chang (  chchang20@gmail.com ) Kaohsiung Medical University Chung-Ho Memorial Hospital

The teamwork, including human-robot team interaction, workplace culture, and experiences of surgeons can affect the processes and outcomes of robotic endoscopic surgery. (12) Interestingly, the relationship between teamwork and the learning curve has rarely been mentioned in the eld of robotic spine surgery. Furthermore, active perception, one of key components in robotics, has never been mentioned in any study or research protocol of spinal robotic surgery.
Kaohsiung Medical University Hospital (KMUH) set up the robotic spine surgery team in May 2013. Up until July 2017, the members of the team had performed 688 cases and implanted 3,896 transpedicular screws. The orthopedics department joined the robotic spine surgery team in August 2014. At that time, team, mainly cooperating with the department of neurosurgery ( Figure 1). The perioperative teamwork of this team has been emphasized especially by active communication and the "triple-check process".
This study aimed to investigate the learning curve of the robotic spine surgery via analyzing the accuracy and the surgical time of transpedicular screw placement. Case series with speci c time sequences were collected for analysis to clarify the possible correlation among team, teamwork, individual surgeons, and learning curves.

Ethics statement
This clinical study was approved by the Institutional Review Board of Kaohsiung Medical University Hospital (No: KMUHIRB-E(I)-20150167). All methods were performed in accordance with the relevant guidelines and regulations. Written informed consent was obtained from all participants. Prior to analysis, patient data were de-identi ed and anonymously analyzed.

Patient selection
Between May 2013 and June 2017, there were 688 patients consecutive who received thoracolumbar surgery using the Renaissance robotic system (Renaissance TM ; Mazor Robotics Ltd., Caesarea, Israel) at KMUH. A total of 3,896 screws were implanted successfully. We retrospectively analyzed all the patients. Cases of spinal malignancy or spinal infection were excluded from the study. Three groups of patients were enrolled and analyzed: (1) Neurosurgeon (NS) early 50: The rst 50 cases performed by the department of neurosurgery and the robotic spine surgery team between May 2013 and July 2013.
(2) Orthopedic early 50: The rst 50 cases performed by the department of orthopedics and the robotic spine surgery team between August 2014 and September 2015.
(3) NS later 50: 50 consecutive cases performed by the department of neurosurgery and the robotic spine surgery team between August 2014 and November 2014 (just after the department of orthopedics joined the team).
The characteristics of patients in the 3 groups were reviewed and analyzed (Table1). The time per screw, assessed as the learning curve of the team, was calculated and plotted respectively ( Figure 2).
Separately, the individual learning curves of surgeons were investigated with the short-segment surgery (2 segments, 4 pedicle screws). All the surgeons in this study are experienced spine surgeons with high surgical volume (up to 300 cases/year). The rst 10 cases with a 100% accuracy for 5 individual surgeons (50 cases total, 40 screws total per surgeon) were enrolled for analysis. The time per screw was measured and plotted (Table 2 and Figure 3).

Robotic spine surgery techniques
The Renaissance robot-guided system has been described in previous studies (13,14). Five main steps were carried out to place transpedicular screws with the Renaissance robot-guided system: 1. Preoperative planning: spiral computed tomography (CT) scanning (1-mm intervals) of the spine was performed preoperatively to reconstruct the 3D plane of the spine and to select the optimal screw placement strategy.
2. Mounting: the mounting system was attached to an appropriate bony structure of the spine to maintain stability during registration. The robotic arm of the guided system was also xed to the bony structure to provide intraoperative guidance.
3. Registration: the Renaissance robot-guided system automatically registered the intraoperative images with the preoperative CT images by comparing anteroposterior and oblique views of radiographic images. 4. Drilling: the robot was attached to a mounting frame and moved to the position selected during preoperative planning. After directing the guiding tube to the pedicle, the surgeon performed the drilling. The K-wire was then placed with robotic guidance. During the process of drilling, there is no passive nor active perception processs/mechanims provided by the robotic system.

5.
Intraoperative accuracy evaluation: after the K-wire implantation was completed, anteroposterior and lateral uoroscopic images were obtained. The accuracy was evaluated and repositioning of the Kwire was achieved manually in cases where the accuracy was not acceptable. Secondary registration, developed by the team at KMUH, with anteroposterior and oblique uoroscopic images and the Renaissance robot-guided system, was carried out in cases after January 2015 (13,14). Reposition of the K-wire was done again with the robotic system. Manual reposition was only applied when the reposition was still unacceptable.

Team building and teamwork
Team building The robotic spine surgery team at KMUH is made up of quali ed surgeons (7 neurosurgeons and 3 orthopedic surgeons), coordinators, radiological technicians, specialist nurses, surgical technicians, and surgical nursing staff. Surgeons and surgical technicians are all quali ed in the Renaissance robotics system training programs, including cadaver workshop. The coordinators and specialist nurses manage the preoperative preparation and perioperative care. The parameters and image quality of CT scans are examined carefully by radiological technicians and surgeons.

Human-robot interaction
Page 6/15 The members meet the day prior to surgery to discuss the surgical plan. The process and possible factors in uencing the placement of K-wires (14) are previewed and examined. Five main steps mentioned above are carried out to place transpedicular screws with the Renaissance robot-guided system. The "triplecheck process" is applied during the surgery, i.e. pointing, reading, and con rming all parameters of the settings and instruments with team members involved in every step. Pointing means the members should point with ngers the parameters shown on the screen of robot. Reading means the members should read together the parameters they point. Con rming means the member agree the parameters they point and read. The process and results are recorded immediately following every procedure.
Workplace culture The Joint Commission International's (JCI) accreditation standards for hospitals (Joint Commission International, Oak Brook, Illinois, USA) are followed by the team perioperatively, as KMUH has partnered with the JCI. Patient-safety Reporting system has been setup and staffs are encouraged to communicate actively and report possible adverse events with name or anonymously.

Statistical analysis
The Chi-squared test and Wilcoxon sign rank test were applied for categorical and continuous variables. Mean values are presented as mean±standard deviation (SD). The learning curves (team and individual) were t and plotted on a nonlinear platform. The following equation was used: Where a=growth rate, b=in ection point, and c=asymptote.
The parallelism of learning curves was measured and tested with the parallelism F test. All statistical analyses were performed using the JMP 12 software (SAS Institute Inc., Buckinghamshire, UK). The level of signi cance was set at 0.05.

Results
The learning curves of the team and departments Table 1 shows the characteristics of the 3 groups. There were no signi cant differences among the basic characters of patients. The orthopedic early group performed more long segments cases ( ≧3 levels and ≧ 6 screws). The total 150 cases and 841 screws in the 3 groups had a time per screw of 9.56±4.19, 7.29±3.64, and 8.74±5.77 minutes for the NS early, orthopedic early, and NS later groups, respectively. The NS early group took signi cantly more time per screw (p=0.0017), but this was not signi cant after the nonlinear parallelism test (growth rate estimate was -0.01±0.01, -0.05±0.17, and -0.01±1.25, p=0.85, Figure 1). The accuracy for each group was 99.6% (253/254) for NS early, 99.5% (361/363) for orthopedic early, and 99.1% (222/224) for NS later. There was no signi cant difference in the accuracy between the 3 groups.

The individual learning curves of surgeons
Five surgeons (4 neurosurgeons and 1 orthopedic surgeon) and their rst 10 cases of short segment surgery (2 levels and 4 screws) were enrolled for the analysis of individual learning curves ( Table 2). The time per screw of each surgeon was 12.28±5.21, 6.38±1.54, 8.68±3.10, 6.33±1.90, and 6.73±1.81 minutes. The rst surgeon who initiated the robotic spine surgery in the team took signi cantly more time per screw (p=0.001), and the nonlinear parallelism test (Figure 2) also showed that only the rst surgeon had a steeper learning curve (growth rate estimate was -0.13±0.17, 0.08±0.92, -1.41±0.0, -0.57±1.6, 0.75±0.0, p<0.0001).

The learning curves between team and individual surgeons
This study demonstrates the learning curve of both the team as a whole as well as individual surgeons.
With an established-team and standardized teamwork, the learning curve of a newly joined but quali ed member of the surgical team may be parallel with the other experienced members in the team. To the best of our knowledge, this could be the rst study to show the learning curves of robotic spine surgery with quanti cational and nonlinear analysis.
The power law of practice is well described quantitatively for human learning curve study in psychology (15). This law states that the logarithm of the reaction time for a particular task decreases linearly with the logarithm of the number of practices. Therefore However, the characteristics of the smooth power law may potentially mask more complex dynamics underpinning individual learning curves. Therefore, using a single power law to predict or analyze individual performance may obscure more complex learning dynamics (16).
The accuracy of robotic spine surgery The accuracy of spinal instrumentation is another important issue in robotic spine surgery. Previous studies have shown that the rate of successful robotic-assisted pedicle screw placement became consistent after 20 or 30 cases (6, 10). In our study, the rate of accuracy did not change signi cantly between the 3 groups of 50 cases during 3 speci c time intervals. In a meta-analysis of robotic spine surgery by Joseph et al. (17), including 22 retrospective case series and prospective randomized trials, the consistency and high accuracy rates of robotic spine surgery were also recognized. Ringel et al. (3) also stated that accuracy did not improve through the course in their study. Obesity, osteoporosis, and congenital scoliosis have been recognized as risk factors for screw malposition and surgeons in the initial stage of using a robot are suggested to avoid performing surgery on patients with these risk factors (1).
One of our previous studies developed a secondary registration protocol that increased the success rate and intraoperative accuracy by the same robotic system (13). In the studies we published in 2016 and 2017 (13,14) showed that the K-wire needed to be repositioned manually is 1.26% (4/317 K-wires, with secondary registration) and 0.15% (1/662 K-wires, with third registration). Factors in uencing accuracy can be errors in preoperative planning, mounting, registration, drilling, or robot assembly (14). All of these factors could be eliminated or minimized by a well-established team and teamwork, according to the results of this study.

Potential roles of teamwork
Effective teamwork can be measured by examining the quality of output, the process and the members' performance (18). Team dynamics are important for e cient teamwork. Team dynamics include open communication to avoid con icts, effective coordination to avoid confusion, e cient cooperation to perform the tasks in a timely manner and produce the required results, and high levels of interdependence to maintain high levels of trust, risk-taking, and performance (19). The smooth and parallel learning curves of the team and individuals imply the potential roles of teamwork in robotic spine surgery.
Communication and workplace culture are thought to be important factors in the human-robot team interaction (12), but the evaluation and measurements are usually di cult in complex operation rooms. In the surgery with the Renaissance robot-guided system, a potential error could take place during registration due to inconsistencies between the numbers of station number set on the computer and on the mounting platform. The "triple-check process" applied in robotic spine surgery, i.e. pointing, reading, and con rming all parameters of settings and instruments with team members in every step, prevents our team from errors. There is little literature related with human-robot interaction in the eld of robotic spine surgery. Though we did not propose new experimental study design about the human-robot interaction in this study, the high accuracy of transpedicular screw implantation in this study demonstrates not only the e ciency of surgical robot but also the e ciency of the good human-robot interaction.
KMUH has been an academic medical center in southern Taiwan since the 1970s. It has received accreditation every three to ve years by the o cial accreditation organizations in Taiwan. Furthermore, the partnership with the international accreditation organization also encourages and regulates the team to follow rules and guidelines more strictly.
We believed this is the rst article emphasizing the importance of team building, human-robot interaction and workplace culture in robotic spine surgery.
Robotic spine Surgery -a system lack of active perception and unmet needs Modeling and control strategies for perception, de ning active perception (20), are missing in the robotic system we used for this study. According to our previous study, skiving over a steep slope of bony surface is the main factor affecting accuracy (14). However, during the process of drilling, there is no passive nor active perception process or mechanism provided. The members of surgical team use their naked eyes and ngers to detect possible skiving of guiding tube before, during, and after the drilling procedure. Contrary to intraoperative image-guided spinal navigation , there is no guiding image or visual feedback provided by the robotic system during the drilling procedure and screw placement (21). These defects are also barriers to surgeons to use or trust this robotic system. Intelligent control strategies according to the data from detecting possible deviation of guiding tube, are apparently the unmet needs for robotic spine surgery.

Conclusion
With teamwork, the learning curve of robotic spine surgery for a newly employed surgeon can be smooth and parallel with other experienced surgeons. The accuracy is also high and consistent. Communication and workplace culture are important for teamwork meanwhile triple-check process is advocated during robotic spine surgery. We propose active perception is the unmet need for current robotic spine surgery.

Consent for publication
No images, personal or clinical details of participants are presented that compromise anonymity

Con ict of Interest
The authors declare that the research was conducted in the absence of any commercial or nancial relationships that could be construed as a potential con ict of interest.   Groups learning curves The learning curve of the 3 groups and the test of parallelism were plotted. The upper part (A) is the curves of the 3 groups put combined. The lower part of the gure (B) is the curve for each group. The parallelism of these curves was tested using the Parallelism F test and showed no signi cance.