the learning curve of robotic lobectomy

The learning curve of robotic lobectomy

Mark Meyer2*Farid Gharagozloo1

Barbara Tempesta1

Marc Margolis2

Eric Strother1

Douglas Christenson2

1University of Arizona College ofMedicine, Tucson, AZ, USA2Washington Institute of Thoracic andCardiovascular Surgery, GeorgeWashington University Medical Center,Washington, DC, USA

*Correspondence to: M. Meyer,Washington Institute of Thoracic andCardiovascular Surgery, GeorgeWashington University MedicalCenter, 2175 K Street NW,Washington, DC 20037, USA.E-mail: [email protected]

Abstract

Background Robotic lobectomy has been shown to be feasible, safe andoncologically efficacious. The actual learning curve of robotic lobectomy has yet tobe defined. This studywas designed to define the learning curve of robotic lobectomy.

Methods Weperformed a retrospective review of prospectively accrued patientsat our institution who underwent robotic lobectomy from January 2004 untilDecember 2011. Six scatter graphs were constructed, comparing operative time,conversion rate, morbidity, mortality, length of stay and surgeon comfort withthe number of consecutive cases. In each graph, a regression trendline was drawnand the change in the slope of the curve corresponding to the beginning of theplateau defined the learning curve. The overall learning curve was defined asmean� SD of the sum of the individual learning curves.

Results Based on operative times, mortality and surgeon comfort, the overalllearning curve was 18� 3 cases. The learning curve based on operative times,mortality and surgeon comfort was 15, 20 and 19 cases, respectively. There was noassociation between the need for conversion and number of consecutive cases. Therewas a trend towards lower morbidity and decreased length of stay with greaterexperience. However, these parameters did not define a specific learning curve.

Conclusions Operative time, mortality and surgeon comfort were found tobe key parameters for the learning curve of robotic lobectomy when performedby surgeons who are experienced with video-assisted thoracic surgery (VATS).The learning curve was 18� 3 cases. Copyright © 2012 John Wiley & Sons, Ltd.

Keywords learning curve; robotic lobectomy; lung cancer

Introduction

Robotics provides three-dimensional (3D) visualization and greater instru-ment manoeuvrability in a confined space and has the potential of enhancingminimally invasive endoscopic lobectomy. Robotic lobectomy has beenreported to be feasible, oncologically efficacious and safe (1–3).The use ofthe robot for lobectomy is associated with a learning curve. Althoughpreviously published papers have suggested the presence of a learning curve,the actual learning curve has not been defined. This study was designed todefine the learning curve of robotic lobectomy.

Patients and methods

Patients

A retrospective review was conducted of patients who underwent roboticlobectomy from January 2004 to December 2011. These patients were selected

ORIGINAL ARTICLE

Accepted: 18 July 2012

Copyright © 2012 John Wiley & Sons, Ltd.

THE INTERNATIONAL JOURNAL OF MEDICAL ROBOTICS AND COMPUTER ASSISTED SURGERYInt J Med Robotics Comput Assist Surg (2012)Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/rcs.1455

from a larger cohort of patients who underwent lobec-tomy by other techniques at our institution. In patientswith lung cancer, inclusion criteria were clinical stage Iand II lung cancer, predicted ability to achieve resectionby lobectomy and the physiological state of the patient.Exclusion criteria were clinical stage greater than I andII, endobronchial tumours visible at bronchoscopy, acentral tumour and preoperative induction therapy.Preoperative evaluation included a comprehensive historyand physical examination, computed tomographic (CT)scans, positron emission tomography (PET), cardiac eval-uation, pulmonary function testing, peripheral venousultrasound examination and assessment of functionalstatus by Eastern Cooperative Oncology Group (ECOG)scoring. The procedures were performed by two surgeonsat a single institution who had a combined experience of400 video-assisted thoracic surgery (VATS) lobectomies.

Operative technique

Robotic lobectomy was performed using the same stan-dard technique which has been described elsewhere (4).At the outset of the procedure, a camera was inserted intothe pleural space and a final decision was made whetherto proceed with robotic lobectomy. Contra-indications torobotic lobectomy included extensive adhesions and anobliterated fissure. After each procedure the surgeon wasasked to score his overall comfort level as: 0, apprehensive,procedure did not feel routine; or 1, routine procedure. Thesame surgical assistant was present in all procedures.

The data were gathered prospectively and reviewed ret-rospectively. The data points were age, sex, operativetime, histology, resected lobe, resected nodal stations,upstaging, length of stay, complications, mortality andsurgeon comfort. Data points are reported as mean� SDand median.

Six scatter plots (graphs) were constructed. Figure 1was designed to evaluate the relationship of operativetime to the extent of experience, defined as the numberof consecutive cases. Figure 2 was designed to evaluatethe relationship of conversion to thoracotomy to the ex-tent of experience, as defined by the number of

consecutive cases. The criteria for conversion to thoracot-omy were a calcified bronchus, inability to find the pulmo-nary artery in an incomplete fissure after 15min of dissec-tion, and bleeding from the proximal pulmonary arterywhich could not be controlled with a specially designedhilar vascular clamp. Figure 3 was designed to evaluatethe relationship between the number of complicationsand the extent of experience, as defined by the numberof cases. Figure 4 was designed to evaluate the relation-ship of mortality to the extent of experience, as definedby the number of cases. Figure 5 was designed to evaluatethe relationship between hospital stay and the extent ofexperience, as defined by the number of cases. Figure 6

Figure 1. Operative time vs consecutive cases. The regressiontrendline is shown with the beginning of the plateau at case 15

Figure 2. Number of conversions vs consecutive cases. Theregression trendline is shown and is flat

Figure 3. Operative morbidity vs consecutive cases. The regres-sion trendline is shown to be decreasing with greater experience

Figure 4. Operative mortality vs consecutive cases. The regres-sion trendline is shown to have a sharp decline after 20 cases

M. Meyer et al.

Copyright © 2012 John Wiley & Sons, Ltd. Int J Med Robotics Comput Assist Surg (2012)DOI: 10.1002/rcs

was designed to evaluate the relationship betweensurgeon comfort and the extent of experience, as definedby the number of cases.

In each graph, a regression trendline was drawn andthe change in the slope of the curve corresponding tothe beginning of the plateau defined the learning curve.The overall learning curve was defined as mean� SD ofthe sum of the individual learning curves.

This study was reviewed and determined to be exemptfrom institutional review board approval under 45 CFR46.101 (b) (4).

Results

During the period of this study, robotic lobectomy wasperformed in 185 consecutive patients. There were 72men and 113 women, with a mean age of 65� 9 years.

Indication for lobectomy is summarized in Table 1. Themost common indication for lobectomy was lung cancer(91%). Table 2 summarizes the distribution of lobectomies.The majority of lobectomies were performed in the upperlobes. The patients with lung cancer who underwentrobotic lobectomy were in clinical stages I (145 patients)and II (23 patients).

Operative times were in the range 102–454min, asdepicted in Figure 1, with a mean of 211�60min. There

were three (1.6%) emergent conversions to thoracotomyfor bleeding from the proximal pulmonary artery.

A total of 47 complications were seen in 31 patients(31/185, 16.8%) (Table 3). The majority of complications(31/47, 66%) were minor, defined as atrial fibrillation,atelectasis, hydropneumothorax, lymphatic leak, acuterenal failure, pleural effusion requiring drainage andwound infection. The remaining complications (16/47,34%) were classified as major, defined as pneumonia, pro-longed air leak, pulmonary embolism, incisional bleedingrequiring exploration, respiratory failure, cardiopulmo-nary arrest, heparin-immune thrombocytopenia (HIT)and liver failure.

There were three deaths, for a mortality rate of 1.6%.All deaths were in the postoperative period. There wereno deaths that were attributable to the robotic technique.Two patients died of respiratory insufficiency and onepatient died of unexplained cardiac arrest. All deathsoccurred in the first phase of the experience. There wereno deaths in the last 165 patients.

Figure 5. Hospital stay vs consecutive cases. The regressiontrendline is shown to be decreasing with greater experience

Figure 6. Surgeon comfort vs consecutive cases. The regressiontrendline is shown to have an abrupt change after 19 cases

Table 2. Distribution of robotic lobectomies

Lobe (n)

Right upper lobe 52Right middle lobe 16Right lower lobe 33Left upper lobe 53Left lower lobe 29Lingula 2Total 185

Table 3. Postoperative complications after robotic lobectomy

Complication (n) (%)

Atrial fibrillation 16 8.6Prolonged air leak 6 3.2Atelectasis 6 3.2Pleural effusion requiring drainage 5 2.7Pulmonary embolism 3 1.6Respiratory failure 2 1.1Acute renal failure (ARF) 1 0.5Incisional bleeding requiring exploration 1 0.5Hydropneumothorax 1 0.5Liver failure 1 0.5Pneumonia 1 0.5Cardiopulmonary arrest 1 0.5Heparin-induced thrombocytopenia (HIT) 1 0.5Lymphatic leak 1 0.5Wound infection 1 0.5Total 185

Table 1. Indications for robotic lobectomy

Indication (n)

Lung cancer 168Central metastatic disease 6Benign central lesions 4Inflammatory disease involving entire lobe 3Pulmonary sequestration 2Central inflammatory mass 1Aspergilloma 1Total 185

Learning curve of robotic lobectomy


Hospitalization ranged from 2 to 21 days with a medianof 4 days.

Figure 1 depicts operative time vs number of consecu-tive cases. There is a sharp change in the slope of theregression trendline, which corresponds to the beginningof the plateau of operative times and correlates with case15. Based on operative time alone, the learning curve was15 patients.

Figure 2 depicts conversions to thoracotomy vs thenumber of consecutive cases. There was no associationbetween the requirement for conversion and the overallexperience.

Figure 3 depicts morbidity vs the number of consecutivecases. There is a slight trend towards lower morbiditywith greater experience, without a specific deflectionpoint.

Figure 4 depicts mortality vs the number of consecutivecases. There were no deaths following case 20 in the last165 cases. Based on mortality, the learning curve was 20cases.

Figure 5 depicts length of stay vs number of consecutivecases. There is a slight trend toward lower length of staywith greater experience, without a specific deflectionpoint.

Figure 6 depicts surgeon comfort vs number of consec-utive cases. The surgical procedure was scored as routineafter the 19th case.

A learning curve was seen in Figures 1, 4 and 6. Basedon these graphs, the mean cumulative learning curve foroperative times, mortality and surgeon comfort was18� 3 cases.

Discussion

The surgical robot provides magnified high-definition 3Dvisualization and increased instrument manoeuvrabilityin a confined space. It has been proposed that the surgicalrobot can enhance the minimally invasive approach tolobectomy. However, robotic lobectomy is associated witha number of controversies. These controversies include com-parability to lobectomy by thoracotomy or VATS, feasibilityof the operation, oncologic efficacy, cost, morbidity, mortal-ity, postoperative pain, hospital stay, duration of chest tubedrainage and the learning curve.

In a retrospective analysis comparing lobectomy bythoracotomy to robotic lobectomy, Veronesi and collea-gues (1) concluded that robotic lobectomy is feasible,safe, oncologically efficacious and is associated withshorter hospital stay. A number of other retrospective caseseries have reported similar safety and feasibility forrobotic lobectomy (3,5,6).

It has been proposed that robotic lobectomy is associatedwith a steep learning curve. In this study we evaluated thelearning curve in 185 consecutive robotic lobectomiesperformed in a standardized fashion in the same institution.There is agreement that a learning curve provides an objec-tive assessment of the skills that are necessary for basic

mastery of a surgical procedure (7,8). However, the defini-tion of a learning curve is difficult. It has been proposed thatthe learning curve should include a combination of opera-tive times, morbidity, mortality and surgeon comfort (7).Although surgeon comfort is important, it remains a subjec-tive parameter which is difficult to quantify objectively. Interms of robotic lobectomy, Veronesi et al. (1) reportedthe learning curve to be 18 cases. However, these authorsdid not expand on the factors which were used in definingthe learning curve. In this study, we looked at operativetime, morbidity, mortality, conversion rate, length of stayand surgeon comfort as independent variables and createda regression trendline comparing these parameters tothe consecutive number of cases. For each parameter, thelearning curve was defined as the change in the slope ofthe curve corresponding to the beginning of the plateau.

When operative times are plotted against consecutivecases, the slope of the regression trendline shows anabrupt change at 15 cases, which corresponds to thebeginning point of the plateau in operative times.Although operative times will vary with the complexityof individual cases, it appears that the abrupt decrease(learning curve) corresponds to 15 cases. Operative timesare an indirect measure of the complexity of the caseas well as the basic mastery of the surgical task andsurgeon comfort.

At the outset of the robotic experience, the decision wasmade to commit to a robotic lobectomy following initialexamination of the pleural space with videoendoscopy.Using this strategy, patients with extensive adhesions oran obliterated fissure were excluded. It is likely that thisstrategy played a significant role in the overall lower rateof conversion in this series. There were 3/185 (1.6%)emergent conversions to thoracotomy. A number ofpatients in this series had intraoperative bleeding. In themajority of patients, bleeding was controlled using acombination of robotic and endoscopic techniques. It isour view that minor bleeding should be controlledusing videoendoscopic techniques if at all possible andthoracotomy should be reserved for uncontrolled bleeding.In three patients, bleeding occurred at the proximal pulmo-nary artery. In these patients, initial control was obtainedby compression of the bleeding point. A thoracotomy wasperformed. Proximal control was obtained and the pulmo-nary artery was repaired. As depicted on Figure 2, therewas no correlation between the need for conversion andthe number of cases. Conversion to a thoracotomy shouldnot be seen as a rescue manoeuvre in patients whoundergo robotic lobectomy. Conversion to thoracotomymay be determined by surgeon experience and comfortand is difficult to objectify. Based on the findings in thisseries, we can conclude that conversion to thoracotomydoes not play a role in the assessment of the learningcurve for robotic lobectomy.

Figure 3 depicts complications vs the number of consecu-tive cases. The overall complication rate was 31/185(16.8%). There was a trend towards lower morbidity withgreater experience. The rate and nature of the postoperativecomplications are consistent with previously reported series.

M. Meyer et al.


Figure 4 depicts the mortality rate vs consecutive cases.The three deaths in this series occurred among the first 20patients. Although the deaths occurred in the postopera-tive period and were not the result of the robotictechnique, they most likely represent poor patient selec-tion for the early part of the robotic experience. We haveobserved that in patients with poor FEV1 and diffusioncapacity of the lung for carbon monoxide, the lung doesnot collapse and the robotic manoeuvres are hindered bythe confined pleural space. As a result, for these patients,the operative times were longer and may have contributedto the postoperative complications and the poor outcome.These patients should be excluded from the early part of arobotic experience. Based on this case series, usingmortality as an independent parameter, the learningcurve for robotic lobectomy was 20 cases.

Assessment of length of stay as a parameter in determin-ing the learning curve for robotic lobectomy is complicated.Length of stay is a function of intraoperative and postopera-tive complications, co-morbidities, institutional approach topostoperative care and social issues. In this series, althoughlength of stay varied from 2 to 21 days, themedian length ofstay was 4 days. There was a trend towards a decrease inlength of stay with greater experience.

In this series, based on the assessment of the indepen-dent parameters vs the number of consecutive cases,operative times, mortality and surgeon comfort were theprimary determinants of the learning curve. Based onoperative times, the learning curve was 15 cases. Basedon mortality, the learning curve was 20 cases. Based onsurgeon comfort, the learning curve was 19 cases.Undoubtedly, surgeon comfort is a subjective criterionwhich can be affected by the operating room personnel,presence of pleural adhesions, incomplete fissures andoverall experience with minimally invasive surgery. In thisseries, preoperative planning for bleeding complicationsand other complications necessitating conversion to athoracotomy was implemented in order to factor out theseissues from the assessment of surgeon comfort in terms ofthe use of the robot.

The mean overall learning curve was 18� 3 cases.These results are consistent with the experience reportedby Veronesi and colleagues (1). In both studies thesurgeons performing robotic lobectomy had extensiveexperience with lobectomy by VATS. The learning curvemay be steeper for surgeons who are transitioning fromlobectomy by thoracotomy to robotic technique.

Conclusions

Robotic lobectomy is safe and feasible. Based on thisstudy, operative time, mortality, and surgeon comfortwere found to be key parameters in determining the learn-ing curve for robotic lobectomy for surgeons who areexperienced with VATS lobectomy. For these surgeons, thelearning curve is 18� 3 cases.

References

1. Veronesi G, Galetta D, Maisonneuve P, et al.. Four-arm roboticlobectomy for the treatment of early-stage lung cancer. J ThoracCardiovasc Surg 2010; 140(1): 19–25.

2. Ninan M, Dylewski MR. Total port-access robot-assisted pulmo-nary lobectomy without utility thoracotomy. Eur J CardiothoracSurg 2010; 38(2): 231–232.

3. Gharagozloo F, Margolis M, Tempesta B, et al.. Robot-assistedlobectomy for early-stage lung cancer: report of 100 consecutivecases. Ann Thorac Surg 2009; 88: 380–384.

4. Gharagozloo F, Margolis M, Tempesta B. Robot-assisted thoraco-scopic lobectomy for early-stage lung cancer. Ann Thorac Surg2008; 85: 1880–1886.

5. Giulianotti PC, Buchs NC, Caravaglios G, et al.. Robot-assistedlung resection: outcomes and technical details. Interact CardiovascThorac Surg 2010; 11: 388–392.

6. Park BJ, Flores RM, Rusch VW. Robotic assistance for video-assisted thoracic surgical lobectomy: technique and initial results.J Thorac Cardiovasc Surg 2006; 131: 54–59.

7. Kaul S, Shah NL, Menon M. Learning curve using robotic surgery.Curr Urol Rep 2006; 7: 125–129.

8. Maniar H, Council M, Prasad S, et al. Comparison of skill trainingwith robotic systems and traditional endoscopy: implications ontraining and adoption. J Surg Res 2005; 125: 23–29.

Learning curve of robotic lobectomy


the learning curve of robotic lobectomy

Documents