500 Minimally Invasive Robotic Pancreatoduodenectomies

Abstract

Objectives:

This study aims to present the outcomes of our decade-long experience of robotic pancreatoduodenectomy and provide insights into successful program implementation.

Background:

Despite significant improvement inmortality over the past 30 years, morbidity following open pancreatoduodenectomy remains high. Weimplemented a minimally invasive pancreatic surgery program based on the robotic platform as one potential method of improving outcomes for this operation.

Methods:

A retrospective review of a prospectively maintained institutional database was performed to identify patients who underwent robotic pancreatoduodenectomy (RPD) between 2008 and 2017 at the University of Pittsburgh.

Results:

In total, 500 consecutive RPDs were included. Operative time, conversion to open, blood loss, and clinically relevant postoperative pancreatic fistula improved early in the experience and have remained low despite increasing complexity of case selection as reflected by increasing number of patients with pancreatic cancer, vascular resections, and higher Charlson Comorbidity scores (all P<0.05). Operating room time plateaued after 240 cases at a median time of 391 minutes (interquartile rang 340–477). Major complications (Clavien >2) occurred in less than 24%, clinically relevant postoperative pancreatic fistula in 7.8%, 30- and 90-day mortality were 1.4% and 3.1% respectively, and median length of stay was 8 days. Outcomes were not impacted by integration of trainees or expansion of selection criteria.

Conclusions:

Structure dimple mentation of robotic pancreatoduodenectomy can be associated with excellent outcomes. In the largest series of RPD, we establish benchmarks for the surgical community to consider when adopting this approach.

Pancreatoduodenectomy (PD) remains one of the most challenging abdominal operations. Despite its low mortality at high volume centers, morbidity remains a significant clinical and economic burden.^1–6 Since the description of the original resection by Kausch in 1912 and the 1 step procedure by Whipple in 1935, PD has undergone several minor iterations in technique.¹ One of the transformative modifications was the description of laparoscopic PD in 1994.² Early results demonstrated the safety and feasibility of laparoscopic pancreatoduodenectomy (LPD).^3,4 Although select surgeons have shown a benefit to this approach, the technical complexity of LPD coupled with the limited range of motion and poor ergonomics afforded by laparoscopy resulted in a demanding learning curve that restricted wider dissemination.⁵

With the approval of the da Vinci robot in 2000, interest in applying a different minimally invasive platform to PD renewed.⁶ Improvements in 3-dimentional imaging, near 360-degree range of articulating instruments, and enhanced precision facilitated a surge in reports of robotic pancreatoduodenectomy (RPD) since 2003.^7–9 Our first RPD was performed in 2008, and while early reports focused on safety and feasibility, continued growth in our robotic program enabled us to refine our technique and identify an RPD learning curve after which several comparative effectiveness studies were performed.^10,11 Some of those studies have shown a benefit to the robotic approach when applied to select patient cohorts.^12–14

Since the inception of our RPD program however, expansion of selection criteria, modifications in intraoperative technique, and the advent of enhanced postoperative care pathways have resulted in improvements in operative performance and patient outcomes.^11,15,16 Coupled to the implementation of a robotic pancreas training program for trainees, continued improvement in outcomes has motivated us to disseminate our experience to early adopters. In this study, we review the evolution and outcomes of RPD at our institution over the last decade, providing insights into successful implementation, expansion of selection criteria, and incorporation of trainee participation.

METHODS

Study Design, Definitions, and Outcomes

A retrospective review of a prospectively maintained database was performed of consecutive patients who underwent RPD at the University of Pittsburgh Medical Center between October 2008 and September 2017. All cases were performed by 1 of 5 attending HPB surgeons (HJZ, DLB, AJM, MEH, and AHZ). This study was approved by the University of Pittsburgh Institutional Review Board (PRO14120203).

Patient demographics, clinicopathologic variables, and operative and perioperative outcomes were collected prospectively. Operative time was defined as length of time from incision to skin closure and therefore included the time required to dock and instrument the robot. Transfusions were recorded both in the operating room and perioperatively. The pancreato-jejunostomy anastomosis was constructed in a duct to mucosa fashion in 2 layers in the majority of cases according to the modified Blumgart technique.¹⁷ Postoperative pancreatic fistula (POPF) and delayed gastric emptying were classified using the International Study Group for Pancreatic Fistula definitions.^18,19 Complications were graded according to the Clavien–Dindo classification system.²⁰ Cancer staging was determined according to American Joint Committee on Cancer 8th edition.²¹ A negative margin on final pathology was defined as 0 mm (no tumor at margin). All perioperative outcomes were followed to 90 days after surgery.

Implementation of the Robotic Training Curriculum

After 2012, fellows were required to maintain a record of the robotic cases they performed at the console and the specific steps of the procedure they were responsible for. The following year, a structured curriculum was introduced that was continually expanded until 2015. This curriculum included virtual reality simulation, inanimate biotissue drills, video review, adaptive intraoperative coaching based on case logs and experience, and prospective assessment of operative performance and outcomes.^22–26 Data on fellow participation and which steps of each RPD they performed was recorded as part of the institutional prospective clinical database which facilitated assessment of trainee integration into the RPD program.²⁵

Statistical Analysis

For descriptive statistics, continuous data were summarized by reporting mean and standard deviation when normally distributed data or by median and interquartile rang (IQR) when not normally distributed. Categorical data were summarized using frequency and percentage. Spearman Correlation and linear regression were used to test trends in continuous data. Measures of operative performance and postoperative outcomes were analyzed consecutively to detect improvements across time in patients who underwent RPD. All testing was of 2-tailed nature with P value of < 0.05 considered statistically significant. Statistical analyses were performed using the Intercooled Stata 13.0 statistical software package (Stata Corporation, College Station, TX).

RESULTS

Patient Selection

In just under a decade, 500 RPDs and 521 open PDs were performed at our institution. Over time, the use of RPD increased (Rho = 0.95, P < 0.01) whereas the use of open PD significantly decreased (Rho = −0.89, P = 0.02, Fig. 1A). A total of 9 surgeons performed RPD during the study period. Three surgeons (AHZ, MEH, and HJZ) were the primary surgeon in 90% (n = 450) of RPD operations and acted as an assistant in the remaining 10% (n = 50). Those same 3 surgeons performed 34% (n = 175) of open PD operations during the same time period. The adoption of the robotic platform amongst the 3 highest volume surgeons occurred more swiftly over time and is shown in Figure 1B.

FIGURE 1.

A, Approach to pancreatoduodenectomy over time. From 2010 the proportion of patients undergoing RPD at our institution has continued to increase over time and open PD has decreased. B, The adoption of the robotic platform amongst the 3 highest volume surgeons occurred more swiftly over time.

Patient demographics are shown in Table 1. Mean age was 66.5 years old, 45.6% were women and mean age adjusted Charlson Comorbidity Index was 4.6. As the experience matured case selection expanded to include patients with a greater number of comorbidities, evidenced by a significant increase in the age-adjusted Charlson Comorbidity Index over the study period (Rho 0.12, P < 0.05). Of the 229 (45.9%) patients with pancreatic ductal adenocarcinoma (PDAC), 37% received neoadjuvant chemotherapy. Both the number of patients with PDAC (Rho = 0.097, P < 0.031) and those receiving neoadjuvant therapy (Rho = 0.269, P < 0.01) increased over time (Fig. 2). Sixty-five patients (13%) required a vascular resection at the time of RPD. Although vascular resections increased over time, the most dramatic increase was observed after the first 200 RPDs (Rho = 0.156, P < 0.01).

TABLE 1.

Patient Demographics and Clinical Characteristics of Patients Undergoing Robotic Pancreatoduodenectomy.

	500 RPD			Last 100 RPD
Patient Demographics		n, %		n, %
Age			66.48 ± 11.43	66.17 ± 11.08
Sex (F)			228 (45.6)	45 (45)
ASA
	1		1 (0.2)	0(0)
	2		87 (17.4)	17 (17)
	3		387 (77.4)	78 (78)
	4		25 (5)	5(5)
CCI
	Age adjusted		5 (4–6)	5 (4–6)
	Age unadjusted		2 (2–3)	2 (2–3)
BMI			27.63 ± 5.66	27.74 ± 5.71
Preoperative albumin, g/dL			3.67 ± 0.57	3.83 ± 0.51
Prior abdominal surgery			269 (54.1)	52 (52)
*Neoadjuvant therapy			134 (59)	40 (77)
Pancreas characteristics
	Soft texture		277 (59.6)	45 (52.7)
	Dilated duct		224 (47.6)	46 (52.9)
Pathologic variables		n, %
PDA			229 (45.9)	52 (52)
Malignancy			413 (82.8)	84 (84)
Tumor stage
	0		5 (1.25)	1 (1.19)
	1		40 (10)	10 (11.90)
	2		47 (11.75)	7 (8.33)
	3		285 (71.25)	62 (73.81)
	4		23(5.75)	4 (4.76)
Tumor size (cm)			2.5 (1.9–3.5)	2.5 (2.0–3.5)
Lymph node stage
	0		149 (36.7)	34 (37.78)
	1		247 (60.84)	53 (58.89)
	2		10 (2.46)	3 (3.33)
LN positive			261 (52.6)	56 (56)
# Positive LNs			1 (0–4)	1 (0–5)
Lymphovascular invasion			296 (66.4)	62 (68.9)
Perineural invasion			302 (66.5)	66 (70.2)

All values presented as n (%), mean ± SD, or median (IQR).

^*Chemotherapy ± radiation therapy.

FIGURE 2.

Expansion of selection criteria for patients undergoing robotic pancreatoduodenectomy. The number of patients with a cancer diagnosis, those with pancreatic ductal adenocarcinoma, those who received neoadjuvant therapy, and those who required a vascular resection increased as the experience matured.

Intraoperative Outcomes

Intraoperative metrics are shown in Table 2. Mean operative time for the entire cohort was 415 minutes. There was a sharp and statistically significant reduction in operative time early in the experience (Fig. 3A). Using linear regression modeling, the greatest reduction in operative time occurred during the first 240 cases (Rho = 0.558, P < 0.01). After 240 cases, there was no further statistically significant reduction in operative time. The mean estimated blood loss was 363 mL and—similar to operative time—continued to decrease as the experience matured (Rho = −0.012, P < 0.01). A blood transfusion was given in 74 patients (14.8%). Only 26 patients (5.2%) underwent conversion to an open approach; nearly half were within the first 100 cases. With regards to short-term oncologic outcomes, the mean number of harvested lymph nodes was 28, with a statistically significant increase throughout the entire experience (Rho = 0.433, P < 0.001). A negative margin was achieved in 439 (87.8%) patients and this remained steady throughout despite expanding selection criteria.

TABLE 2.

Outcomes of 500 Robotic Pancreatoduodenectomies

	500 RPD	Last 100 RPD
Intraoperative Outcomes	n,%	n, %
Operative time, min	415 ± 107	373 ± 76
Conversion to open	26 (5.2)	5 (5)
EBL, mL	250 (150–400)	200 (100–355)
EBL ≥ 500 mL	111 (22.3)	21 (21)
Blood transfusion	74 (14.8)	4 (4)
Lymph node harvest	28 (20–37)	34 (25–40)
Margin negative resection	439 (87.8)	85 (85)
Postoperative outcomes
POPF	101 (20.2)	19 (19)
CR-POPF	39 (7.8)	3 (3.0)
DGE	117 (23.5)	29 (29.3)
DGE grade
A	43 (8.6)	9 (9)
B	37 (7.4)	6 (6)
C	37 (7.4)	14 (14)
Wound infection	63 (12.6)	5 (5.1)
Pseudoaneurysm	21 (4.2)	5 (5.0)
Reoperation	25 (5.01)	6 (6.1)
Length of ICU stay (d)	0 (0–1)	2 (2–5)
Length of hospital stay (d)	8 (6–11)	7 (5–8)
Readmission (90 d)	177 (35.5)	37 (37)
Overall morbidity	344 (68.8)	58 (58)
Serious morbidity	124 (24.8)	17 (17)
Max. Clavien grade
1	59 (11.8)	15 (15)
2	161 (32.3)	26 (26)
3	55 (11.0)	7 (7)
4	54 (10.8)	7 (7)
5	15 (3)	3 (3)
30-d mortality	9 (1.8)	3 (3%)
90-d mortality	15 (3)	3 (3%)
Adjuvant chemotherapy*	157 (75)	31 (80)
Follow-up, mo	36.7 (31.8–41.3)	NA

All values presented as n (%), mean ± SD, or median (IQR).

^*In patients with pancreatic cancer.

FIGURE 3.

Optimization of select outcomes in patients undergoing robotic pancreatoduodenectomy. Operative time improved significantly during the first half of the experience (A). The red line indicates the predicted operative time as the series matured with accompanying 95% confidence interval (gray outline). The orange line denotes the mean operative time of the entire series (415 min). The greatest reduction in operative time occurred during the first 240 cases (arrow). After 240 cases, there was no further statistically significant reduction in operative time. The incidence of CR-POPF (B), number of patients who experienced a serious morbidity (Clavien > 2) (C) and postoperative length of stay (D) were reduced as the experience matured.

Postoperative Outcomes

Postoperative outcomes are displayed in Table 2. POPF occurred in 101 patients (20.2%), of which, 12.4% were biochemical leaks and 7.8% were CR-POPF; 26 (5.2%) were Grade B, and 13 (2.6%) were Grade C. A steady and significant reduction in the number of patients who developed CR-POPF was noted across the series (Grade B/C, Rho = −0.31, P = 0.002, Fig. 3B). Delayed gastric emptying occurred in 117 patients (23.5%).

Twenty-four patients (4.8%) required an unanticipated return to the operating room. Of these, 12 (2.4%) were for placement of enteral access (PEG/PEJ tube) for nutritional supplementation. Indications for the remaining 12 patients included the following: wound complications = 5, gastrojejunostomy leak at anastomatic staple line = 1, marginal ulcer perforation = 1, bleeding/evacuation of hematoma = 3, removal of an infected port = 1, and video assisted thoracoscopy for evacuation of a parapneumonic empyema = 1.

Major complications (Clavien grade > 2) occurred in 124 patients (24.8%) with a steady reduction in the number of major complications across the series (Rho = −0.075 P = 0.096, Fig. 3C). Mortality within 30 days of surgery occurred in 9 patients (1.8%) while 90-day mortality occurred in 15 (3%). Thirty-day mortalities included: myocardial infarction (n = 3), postoperative pseudoaneurysm (n = 1), aspiration pneumonia (n = 1), pulmonary embolism (n = 1), acute mesenteric ischemia (n = 1), and bacteremia with septic shock (n = 2).

Median length of hospital stay for the entire cohort was 8 days (IQR 6–11) and steadily improved over the first 400 cases after which it reached a plateau of a median of 6 days (Rho = −0.42, P < 0.001, Fig. 3D). Readmission was required in 177 (36.1%) patients. In patients with a diagnosis of PDAC (n = 229), 134 (59%) received neoadjuvant chemotherapy. Of patients with PDAC (n = 229), 157 (75%) went on to receive adjuvant therapy. Excluding patients who died within 90 days, 199 patients with PDAC (89%) received neoadjuvant and/or adjuvant chemotherapy.

Postoperative outcomes of the last 100 RPDs are shown in Table 2. Mean operative time was 373 ± 76 minutes and median estimated blood loss was 200 mL (IQR 100, 355). Sixteen patients (16%) underwent vascular resection, only 4% needed a blood transfusion, and 5% required conversion to open. The median number of lymph nodes harvested was 34 (IQR: 25,40) and 85 patients (85%) had an R0 resection. Three patients (3%) developed a CR-POPF (Grade B/C), while 29% developed delayed gastric emptying. The median length of stay was 7 days (IQR 5.8).

Utilization of the Robotic Platform and Integration of Fellows Over Time

After the first 80 cases, new staff members, surgical oncology (SSO accredited), and hepatopancreatobiliary (HPB accredited) trainees were integrated into the RPD program (Supplemental figure 1, http://links.lww.com/SLA/B747). Trainee involvement and the degree of participation steadily increased over the experience, such that by the latter portion of the experience, fellow participation increased to over 80% of cases. Additionally, fellows were routinely performing over 50% of the case including the dissection/resection in over 50% of cases (Fig. 4).

FIGURE 4.

Progressive increase in trainee participation in 500 robotic pancreatoduodenectomies.

Finally, most RPDs were performed using the da Vinci Si system. In 2015, our group trialed the Xi system in 9 cases (2%). This platform was quickly abandoned in favor of the older Si model due to a higher conversion rate (56%) experienced within the short trial period (P < 0.0001). This higher conversion rate using the Xi platform was borne out despite no change patient and disease-related characteristics and was confirmed on logistic regression analysis (Supplemental tables 1–3, http://links.lww.com/SLA/B747).

DISCUSSION

To our knowledge this experience represents the largest single institutional series of RPD to date. We implemented a novel approach to PD, optimized performance over the course of 500 RPDs, and incorporated a novel, mastery-based robotics curriculum for the instruction of RPD in under a decade. Optimization of RPD occurred early in the experience, allowing for expansion of selection criteria and increased resident and fellow training without adversely affecting patient outcomes. This experience serves as a model for the safe implementation and adoption of advanced technology in complex gastrointestinal operations.

In our previous report of our first 200 RPDs significant reductions in estimated blood loss and conversions to open surgery following 20 cases, pancreatic fistula after 40 cases, and operative time following 80 cases were demonsrated.²⁷ In the current analysis of 500 RPDs, we demonstrate continued improvements in estimated blood loss and operative time within the first 240 cases. Importantly, other outcomes continued to improve despite expansion of selection criteria to include borderline resectable tumors and use of neoadjuvant therapy. Although the plateau in operative time reduction coincided with the introduction of trainees into the operative team, operative times did not increase after integration of trainees. This was mainly due to implementation of a structured training program for RPD in 2014. This novel, mastery-based training curriculum, allowed trainees (with little to no previous robotic experience) to perform an increasing number of resections and anastomoses under direct supervision from one of 3 attendings (HJZ, MEH, and AHZ).^{22,24–26,28,29} Implications of this training program for improving patient safety, reducing the learning curve, and credentialing are an active area of investigation.^28,30 Based on our experience, we strongly advocate early adopters to undergo similarly structured training that involves simulation, inanimate drills with feedback, operative video review, and case observations prior to embarking on RPD.

Conversion during a minimally invasive PD can be associated with adverse outcomes as shown by 2 national studies.^31,32 National data suggest that rates of conversion to open are lower with the robotic platform compared to pure laparoscopy.^31,32 In the current analysis, the number of conversions remained at 5% beyond our initial learning curve, despite an increasing proportion of pancreatic cancer cases and use of neoadjuvant therapy. A statistically significant increase in the number of conversions was noted in 2015 while trialing the da Vinci Xi system. It is the opinion of the senior authors that while the da Vinci Xi may have certain advantages (ease of docking, table movement, multiquadrant surgery), it has important limitations when performing this highly complex procedure including poor resolution and distorted color perception with the 8 mm Xi camera (as opposed to the high-resolution 12 mm SI system) and diminished responsiveness (smoothness of movement) at the consol. These limitations are not as critical when performing purely ablative procedures such as the distal pancreatectomy. This observation has significant implications given the manufacturer’s (Intuitive Surgical Inc, Sunnyvale, CA) stated goal of phasing out this highly functional model by 2024.

Throughout the experience, mortality remained low with rates commensurate with other high-volume OPD centers.^33–36 Morbidity rates were also low and compare favorably to recently published national data.^22,37 For the last 100 RPDs, the incidence of CR-POPF was 3.3% which is lower than that observed in patients treated with open PD according to 1 large national study.³⁷ Similarly, a significant reduction in major morbidity (down to 15%) and median LOS (down to 6 d) was observed for the last 100 RPDs. It is uncertain if these recent improvements in the last 100 cases were a result of further refinement in technique (tail end of the learning curve), or due to other factors such as incorporation of an enhanced recovery pathway (which was implemented in 2016) into our management.¹⁶ It should be noted that these outcomes are superior to proposed benchmarks of open PD in the United States and of open PD and MI-PD from a recent international study of 23 high-volume centers.^38,39 Taken together these findings further support the safety and efficacy of RPD.

Concerns over the oncologic adequacy of minimally invasive pancreatic resections have recently been raised in the context of a phase III randomized controlled trial of minimally invasive surgery in patients with cervical cancer.⁴⁰ In the present study, the rates of negative margin resection (defined as a 0 mm distance) remained low throughout the experience at 87.8%, despite expanding selection criteria to include patients with borderline resectable disease. On average, 28 lymph nodes were harvested with each specimen and the number of lymph nodes harvested during RPD increased over the experience to an average of 34 lymph nodes for the last 100 cases. This may reflect better oncologic dissection along the SMA plane but could also be an artifact of pathologic sampling. Median survival in patients with PDAC was 22.6 months (95% confidence interval 18.6–28.1), which is consistent with more recent reports of OPD.⁴¹ Given the high rate of distant failure in patients with pancreatic cancer, it is unlikely that a prospective trial could be powered to compare the impact of operative approach on local control in this disease. As such, the outcomes from this study support the oncologic safety of RPD when performed by high-volume surgeons within a dedicated program.

Several studies have recently emerged in support of a minimally invasive approach to PD.^31,42–49 In a large North American multi-institutional retrospective study comparing RPD to OPD performed at high-volume centers by surgeons beyond the learning curve, RPD was associated with longer operative times but reductions in major morbidity.¹² A randomized controlled trial from Spain (PADULAP trial) comparing LPD and OPD demonstrated LPD to be associated with a longer median operative time, but reduced length of stay and serious morbidity (Clavien-Dindo grade > 2) compared with open PD.⁴³ Similarly, Palanivelu et al demonstrated an improved LOS for LPD compared with OPD in another randomized trial from India. Conversely, however, a randomized controlled trial of LPD versus OPD in the Netherlands was suspended prematurely because of concerns of higher mortality in the laparoscopic arm (10% vs 2%). It is notable that this trial was suspended despite lack of statistical significance between the treatment arms (RR of operative mortality = 4.90, 95% CI 0.59–40.4, P = 0.20). The authors of this trial are now preparing to examine the robotic platform in a multicenter, nationwide initiative. Three of the senior authors (AHZ, MEH, HJZ) are serving as proctors for this group to ensure technical proficiency using a similar training program as previously described that incorporates simulation and virtual reality, biotissue drills, video review, proctorship, and prospective monitoring and reporting of operative performance and postoperative outcomes.⁵⁰ Although the impact of the Dutch trial on rates of minimally invasive PD may be too early to analyze, recent data suggest that surgeons are increasingly adopting the robotic platform over pure laparoscopy for pancreatoduodenectomy.^5,8

There are several limitations to the present study. While the data used for this study were gathered prospectively, the retrospective nature of the study introduces selection bias. In addition, the optimized outcomes presented are a collaborative effort amongst several surgeons who often worked in tandem during the initial learning curve to ensure patient safety. As such, the results may not be generalizable to smaller centers that are unable to allocate the required resources and infrastructure. This has important implications with regards to the learning curve for an individual surgeon and the cost of implementing this platform at a national level where the majority of pancreatoduodenectomies in the United States are currently being performed at low-volume centers (less than 10 cases a year).⁴⁶

CONCLUSION

In summary, this series represents the largest analysis of RPD to date. Our experience demonstrates the importance of a systematic approach to establishing the safety and efficacy of this complex procedure and its subsequent dissemination. Our data suggest continued improvements beyond an initial steep learning curve, with no compromise in postoperative or oncologic outcomes after expansion of selection criteria and meaningful participation of trainees as part of an advanced robotics curriculum. This work establishes a framework and benchmark for the surgical community to consider when implementing RPD.