Safety and learning curve of robotic pancreatoduodenectomy with limited robotic platform access: a propensity score-matched comparison with open surgery in a high-volume center

Abstract

Background

Pancreatoduodenectomy (PD) is a major surgical procedure that is still mostly performed via open approach [open PD (OPD)] since the laparoscopic approach (LA) is limited outside specialised centers. Robot-assisted PD (RPD) appears to decrease the difficulties associated with LA. However, previously published studies are partly biased because they stem from institutions with extensive access to robotic platforms (ARPs). The present study aims to analyse the safety of RPD compared with OPD and the specific learning curve when performed with limited ARP.

Methods

This prospective single-center study was conducted in a high-volume pancreatic surgery center but with limited ARP between May 2018 and October 2023. Safety (primary endpoint: 90-day mortality) was assessed using propensity score matching (2:1) to compare RPD and OPD. The learning curve was evaluated using cumulative sum (CUSUM) control chart analysis (split into three consecutive periods of 20 patients), focusing on operative time and complications.

Results

A total of 354 PD were analysed, of which 60 (17%) were RPD. A matched cohort of 120 OPD was used for comparison. The conversion rate was 15% in the RPD group. The 90-day mortality rate was 5% and 8.3% for RPD and OPD, respectively (P=0.55). The RPD learning curve showed a reduction in hospital stay (19.9 vs. 25.3 vs. 14.5 days, P=0.17) and a decrease in post-operative complications with increasing experience.

Conclusions

Even with limited ARP, setting up a robotic program for PD is feasible and safe but it requires a learning curve of 40 patients. During this initial phase, careful patient selection is mandatory.

Highlight box.

Key findings

• Robot-assisted pancreatoduodenectomy (RPD) has proven to be a safe and viable alternative to open pancreatoduodenectomy (OPD), even in centers with restricted access to robotic platforms (ARPs). The learning curve is achieved after 40 cases, demonstrating that RPD can be effectively integrated into surgical practice despite limited resources.

What is known and what is new?

• OPD is the standard approach, while laparoscopic pancreatoduodenectomy remains challenging outside specialised centers. RPD has shown promise but has mainly been studied in high-volume settings.

• This study confirms that RPD is feasible and safe even in centers with limited ARP, with a learning curve of 40 cases. Despite early higher complication rates, outcomes improved over time, hospital stays shortened, and oncological results remained comparable to OPD.

What is the implication, and what should change now?

• RPD is a safe and feasible alternative to OPD, even in centers with limited robotic access. The learning curve of 40 cases suggests that a wider implementation is possible, although robotic experience is required beforehand.

• More centers should adopt RPD with structured training, optimize robotic platform access, and refine patient selection to ensure safety during the learning phase. Further research on long-term outcomes is needed to support widespread adoption.

Introduction

Pancreatoduodenectomy (PD) remains one of the most challenging and pivotal surgical procedures and is the cornerstone treatment for pancreatic head carcinoma as well as other benign and malignant diseases of the periampullary region (1). Despite significant advances in surgical methods, pre-operative and post-operative care, the rate of pancreatic surgery complications is still high, affecting nearly 40% of patients undergoing PD (2,3). Morbidity is primarily driven by post-operative pancreatic fistula (POPF), post-pancreatectomy haemorrhage (PPH), and delayed gastric emptying (DGE) (4,5). Enhanced recovery after surgery (ERAS) programs have been developed with the aim of reducing morbidity associated with surgery, particularly in minimally invasive approaches. The application of ERAS protocols can play a crucial role in mitigating and preventing postoperative complications related to pancreatic surgery.

Since its first description in 1994 (6), the feasibility of laparoscopic PD has been demonstrated in many retrospective studies (7,8) but its reproducibility is still challenging and post-operative outcomes remain controversial (9,10). Several prospective randomised studies have been published (11-13). Notably, two significant prospective randomised studies have reported its feasibility: a Chinese study conducted by Zhang et al. reported a minimal reduction in hospital stay for highly experienced surgeons, whereas the LEOPARD 2 trial conducted by the Dutch study group was prematurely stopped owing to an increased mortality rate (14,15). Consequently, routine recommendation of laparoscopic PD is not advised outside of highly specialised centers (16).

In light of these challenges, robot-assisted PD (RPD) has emerged as a promising alternative to facilitate the implementation of minimally invasive pancreatic surgery, particularly during the reconstruction phase owing to the three-dimensional (3D) dexterity of the instruments and enhanced 3D visualization. Since the first RPD was reported in 2003 by Giulianotti et al. (17), several retrospective studies have demonstrated its safety and reproducibility in high-volume centers (18,19). RPD can offer further advantages such as reduced blood loss and shorter hospital stays without an increase in specific morbidity (20,21). However, due to the technical complexity of this procedure and the requirement for access to a robotic platform (ARP), its widespread adoption is limited outside expert centers (22) as reported in a recent review by Levi Sandri et al. (23) and defined by the Brescia Internationally Validated European Guidelines on Minimally Invasive Pancreatic Surgery (EGUMIPS) (24). Importantly, two recent prospective randomised trials have validated the safety of RPD (25,26). However, these studies were conducted in high-volume centers with extensive ARPs which could limit the reproducibility of the results and the widespread adoption of this approach since few centers worldwide have both elements.

The primary endpoint of the present study was to evaluate the safety and learning curve of RPD compared with the open PD (OPD) in a real-life context where ARP is limited. We present this article in accordance with the STROBE reporting checklist (available at https://hbsn.amegroups.com/article/view/10.21037/hbsn-2024-723/rc).

Methods

Study design

A retrospective analysis of a prospectively maintained database was conducted on all PDs performed in a high-volume pancreatic referral center between April 2018 and October 2023. A single surgeon, trained in robotic surgery (approximately 25 hiatal hernia repairs) and with over 500 PD procedures, conducted all RPD and more than 90% of the OPD. The operative theatre team, including nurses and anaesthesiologists already had experience, especially in robotic colorectal surgery, so they were already familiar with how to prepare and dock the system. Since the introduction of the Xi DaVinci System (Intuitive Surgical Inc., Sunnyvale, CA, USA) in April 2018, data have been prospectively collected, focusing on baseline characteristics [age, sex, body mass index (BMI)], surgical variables, and post-operative outcomes. All data were anonymised and securely stored. This study adhered to the strict ethical standards of the Declaration of Helsinki and its subsequent amendments, ensuring both integrity and compliance with ethical norms. The study was approved by the Rennes University Hospital Ethics Committee (No. 00045607A). All patients provided their consent.

Surgical technique

All patients underwent PD without pylorus preservation and with child’s reconstruction. Drainage of the abdominal cavity was systematically performed to provide amylase values on postoperative day 3 (POD3) and 5 (POD5), allowing for early detection of POPF. The robotic approach followed the 17 operative steps for RPD described by Giulianotti (27). After the resection phase, the first jejunal loop was passed behind the superior mesenteric vessels to be brought up to the supra-mesocolic compartment followed by the successive creation of the pancreatojejunostomy, hepaticojejunostomy, and finally the gastroenterostomy. The first two anastomoses were systematically performed with interrupted sutures at the beginning of the learning curve, gradually transitioning to the increased use of continuous running sutures. The gastroenterostomy was performed mechanically, laterally, side-to-side, and was initially supra-mesocolic but not antecolic. After the 40^th case, the systematic identification of the future gastroenterostomy site was done at the start of surgery to facilitate its creation in a antecolic position at the end of the procedure.

Definitions

Post-operative complications were classified according to the Clavien-Dindo classification (28). DGE and POPFs were defined according to the guidelines established by the International Study Group of Pancreatic Surgery (29,30). Pathological staging followed tumour-node-metastasis (TNM) classifications (31). Surgical duration included all surgical steps from incision to skin closure, including time for robot docking. Total hospitalisation costs were extracted from Program for the Medicalisation of Information Systems (PMSI) data, as assessed by the University Hospital of Rennes, based on patient severity, care, and medications administered during the hospital stay.

Patient selection for robotic approach

Patients aged 18 years or older with a surgical indication for PD due to benign or malignant conditions were initially considered for the surgeon’s first robotic procedures. As the surgeon’s expertise and the operating room team’s proficiency increased, patient selection criteria progressively expanded to include more complex cases. In the early phase, patients with obesity or those requiring vascular resections were excluded due to the heightened technical and logistical challenges posed by minimally invasive surgery.

Access to the robotics platform

During the study period, access to the robotic platform was available 1 to 2 times per month for the first 3 years, then increased to 2 to 3 times per month thereafter.

Statistical analysis

Categorical variables were presented as the number of patients in each category and the corresponding percentages. Group comparisons were performed using either χ² tests or Fisher’s exact tests. Continuous variables were presented as mean ± standard deviation (SD). Comparisons between two groups were carried out using Student’s t-tests or Mann-Whitney Wilcoxon tests, and between the three groups using analysis of variance (ANOVA) or Kruskal-Wallis tests.

To assess the safety of RPD compared with OPD, 2:1 propensity score matching was applied considering patient age, sex, Charlson Comorbidity Index score, and BMI to ensure a balanced comparison between the two approaches. The primary outcome was 90-day post-operative mortality while secondary outcomes included Clavien-Dindo grade ≥ III complications, length of hospital stay, surgical reoperation, hospital readmission rate, and the incidence of specific complications such as DGE, B/C POPF, and severe biliary fistulas (grade B/C) (27). A P value of <0.05 was considered statistically significant.

In order to evaluate the learning period, we applied cumulative sum (CUSUM) analysis based on operative duration. The difference between each procedure’s duration and the mean surgical duration was calculated and the CUSUM of these differences was plotted, identifying three distinct phases of 20 patients each. When the curve falls below zero, the learning curve is reached. We then compared these three groups in terms of mortality, major post-operative complications, and tumour types to analyse the learning curve of RPD.

All analyses were performed using SAS software, version 9.4 (SAS Institute, Cary, NC, USA).

Results

Demographic and operative data

During the study period, 354 patients underwent PD (193 men and 161 women). Among them, 60 patients (17%) underwent the RPD. The mean age at the time of surgery was 66 years. After applying (2:1) propensity matching, 120 OPD cases were compared with 60 RPD cases. The demographic data, tumour characteristics, and pre-operative treatments are reported in Table 1 and are detailed in Table 2 for the three consecutive groups of the robotic approach. No significant differences were found between the two groups regarding demographic data. However, tumours in the RPD group were significantly smaller (23.5±13.6 vs. 28.6±14.8 mm, P=0.009). Additionally, the RDP group presented fewer borderline tumours (8.3% vs. 25.8%, P=0.006) and therefore less need for neoadjuvant treatment (11.7% vs. 33.3%, P=0.002) compared with the OPD group.

Table 1. Baseline characteristics before and after matching.

Variables	Before matching					After matching
	Population (n=354)	OPD (n=294)	RPD (n=60)	P value		Population (n=180)	OPD (n=120)	RPD (n=60)	P value
Age (years)	66.0±10.7	66.1±10.6	65.8±11.6	0.84		66.5±10.3	66.9±9.6	65.8±11.6	0.49
Male	193 (54.5)	170 (57.8)	23 (38.3)	0.006		69 (38.3)	46 (38.3)	23 (38.3)	>0.99
BMI (kg/m2)	24.3±4.3	24.3±4.4	24.0±3.7	0.61		24.1±4.6	24.1±5.0	24.0±3.7	0.95
Charlson score	3.0±1.7	3.1±1.7	2.8±1.6	0.33		2.9±1.6	2.9±1.5	2.8±1.6	0.82
ASA score				0.055					0.11
1	81 (22.9)	73 (24.8)	8 (13.3)			38 (21.1)	30 (25.0)	8 (13.3)
2	236 (66.7)	188 (63.9)	48 (80.0)			126 (70.0)	78 (65.0)	48 (80.0)
3	37 (10.5)	33 (11.2)	4 (6.7)			16 (8.9)	12 (10.0)	4 (6.7)
Previous abdominal surgery	225 (63.6)	189 (64.3)	36 (60.0)	0.53		118 (65.6)	82 (68.3)	36 (60.0)	0.27
RHA	58 (16.4)	52 (17.7)	6 (10.0)	0.14		31 (17.2)	25 (20.8)	6 (10.0)	0.07
Tumour size (mm)	27.7±17.5	28.6±18.1	23.5±13.6	0.007*		26.9±14.6	28.6±14.8	23.5±13.6	0.009*
Immediately operable	269 (76.0)	215 (73.1)	54 (90.0)	0.005*		134 (74.4)	80 (66.7)	54 (90.0)	<0.001*
Borderline	68 (19.2)	63 (21.4)	5 (8.3)	0.02*		36 (20.0)	31 (25.8)	5 (8.3)	0.006*
Locally advanced	12 (3.4)	12 (4.1)	0	0.23		6 (3.3)	6 (5.0)	0	0.18
Pre-operative biliary drainage	180 (50.8)	146 (49.7)	34 (56.7)	0.07		95 (52.8)	61 (50.8)	34 (56.7)	0.07
Neoadjuvant chemotherapy	88 (24.9)	81 (27.6)	7 (11.7)	0.01*		47 (26.1)	40 (33.3)	7 (11.7)	0.002*

Data are presented as mean ± SD or n (%). *, P<0.05. ASA, American Society of Anaesthesiologists; BMI, body mass index; OPD, open pancreatoduodenectomy; RHA, right hepatic artery from superior mesenteric artery; RPD, robot-assisted pancreatoduodenectomy; SD, standard deviation.

Table 2. Baseline characteristics of RPD.

Variables	RPD (n=60)	RPD 1 to 20 (n=20)	RPD 21 to 40 (n=20)	RPD 41 to 60 (n=20)	P value
Age (years)	65.8±11.6	66.2±12.6	69.0±8.9	62.1±12.5	0.17
Male	23 (38.3)	5 (25.0)	10 (50.0)	8 (40.0)	0.26
BMI (kg/m2)	24.0±3.7	24.9±4.3	23.0±2.9	24.2±3.7	0.29
Charlson score	2.8±1.6	2.7±1.5	3.3±1.7	2.5±1.6	0.16
ASA score					0.09
1	8 (13.3)	4 (20.0)	2 (10.0)	2 (10.0)
2	48 (80.0)	16 (80.0)	14 (70.0)	18 (90.0)
3	4 (6.7)	0	4 (20.0)	0
Previous abdominal surgery	36 (60.0)	12 (60.0)	11 (55.0)	13 (65.0)	0.81
RHA	6 (10.0)	2 (10.0)	2 (10.0)	2 (10.0)	>0.99
Tumour size (mm)	23.5±13.6	22.6±12.2	21.4±10.8	26.3±17.0	0.72
Immediately operable	54 (90.0)	19 (95.0)	17 (85.0)	18 (90.0)	0.86
Borderline	5 (8.3)	1 (5.0)	3 (15.0)	1 (5.0)	0.60
Preoperative biliary drainage	34 (56.7)	10 (50.0)	12 (60.0)	12 (60.0)	0.76
Neoadjuvant chemotherapy	7 (11.7)	1 (5.0)	3 (15.0)	3 (15.0)	0.68

Data are presented as mean ± SD or n (%). ASA, American Society of Anaesthesiologists; BMI, body mass index; RHA, right hepatic artery from superior mesenteric artery; RPD, robot-assisted pancreatoduodenectomy; SD, standard deviation.

Operative data are summarised before and after matching in Table 3. The mean operative duration for RPD was significantly longer than that for OPD (567.2±46.3 vs. 361.2±63.9 minutes, P<0.001), reflecting the increased complexity of robotic procedures. The conversion rate for RPD, indicating cases where robotic surgery was converted to open surgery, was 15%. Transfusions were more common in the OPD group (7.5% vs. 1.7%, P=0.17). Similarly, vascular resections were more frequently required in the OPD group (35% vs. 6.7%, P<0.001).

Table 3. Operative data and outcomes before and after matching.

Variables	Before matching					After matching
	Population (n=354)	OPD (n=294)	RPD (n=60)	P value		Population (n=180)	OPD (n=120)	RPD (n=60)	P value
Operative time (min)	398.1±97.6	363.6±63.3	567.2±46.3	<0.001*		429.9±113.6	361.2±63.9	567.2±46.3	<0.001*
Conversion	9 (2.5)	0	9 (15.0)	0.16		9 (5.0)	0	9 (15.0)	0.16
Intra-operative transfusion	22 (6.2)	21 (7.1)	1 (1.7)	0.14		10 (5.6)	9 (7.5)	1 (1.7)	0.17
Venous resection	100 (28.2)	96 (32.7)	4 (6.7)	<0.001*		46 (25.6)	42 (35.0)	4 (6.7)	<0.001*
Gastroparesis	104 (29.4)	80 (27.2)	24 (40.0)	0.01*		56 (31.1)	32 (26.7)	24 (40.0)	0.02*
A	40 (11.3)	35 (11.9)	5 (8.3)			20 (11.1)	15 (12.5)	5 (8.3)
B	31 (8.8)	24 (8.2)	7 (11.7)			17 (9.4)	10 (8.3)	7 (11.7)
C	33 (9.3)	21 (7.1)	12 (20.0)			19 (10.6)	7 (5.8)	12 (20.0)
POPF	125 (35.3)	99 (33.7)	26 (43.3)	0.15		64 (35.6)	38 (31.7)	26 (43.3)	0.12
Grade B or C	34 (9.6)	26 (8.8)	8 (13.3)	0.28		16 (8.9)	8 (6.7)	8 (13.3)	0.14
Bile leakage	30 (8.5)	15 (5.1)	14 (23.3)	<0.001*		20 (11.1)	6 (5.0)	14 (23.3)	<0.001*
Grade B or C	16 (4.5)	9 (3.1)	8 (13.3)	0.28		16 (8.9)	8 (6.7)	8 (13.3)	0.14
Haemorrhage	44 (12.4)	36 (12.2)	8 (13.3)	0.82		23 (12.8)	15 (12.5)	8 (13.3)	0.88
Chyle leak	50 (14.1)	48 (16.3)	2 (3.3)	0.008*		22 (12.2)	20 (16.7)	2 (3.3)	0.01*
Sepsis	77 (21.8)	63 (21.4)	14 (23.3)	0.75		38 (21.1)	24 (20.0)	14 (23.3)	0.61
Post-op transfusion	64 (18.1)	53 (18.0)	11 (18.3)	0.96		34 (18.9)	23 (19.2)	11 (18.3)	0.89
Re-operation	44 (12.4)	37 (12.6)	7 (11.7)	0.84		24 (13.3)	17 (14.2)	7 (11.7)	0.64
Clavien-Dindo III–V	72 (20.3)	55 (18.7)	17 (28.3)	0.09		36 (20.0)	19 (15.8)	17 (28.3)	0.048*
Hospital stay (days)	16.3±12.2	15.7±11.2	19.9±16.1	0.47		17.0±12.1	15.5±9.2	19.9±16.1	0.26
Readmission rate	68 (19.2)	55 (18.7)	13 (21.7)	0.60		39 (21.7)	26 (21.7)	13 (21.7)	1
90-day mortality	35 (9.9)	32 (10.9)	3 (5.0)	0.16		13 (7.2)	10 (8.3)	3 (5.0)	0.55
Total value of hospital stay (€)	19.5±8.6	19.1±8.4	21.1±9.7	0.16		19.2±7.9	18.3±6.6	21.1±9.7	0.07

Data are presented as mean ± SD or n (%). *, P<0.05. OPD, open pancreatoduodenectomy; POPF, post-operative pancreatic fistula; RPD, robot-assisted pancreatoduodenectomy; SD, standard deviation.

Safety of RPD: results of propensity score matching

The primary outcome for evaluating the safety of the robotic approach was 90-day mortality which was lower in the RPD group (5% vs. 8.3%, P=0.55), although this difference did not reach significance. However, major complications (Clavien-Dindo grade ≥ III) as well as DGE were higher in the RPD group (28.3% vs. 15.8%, P=0.048) and (40% vs. 26.7%, P=0.02), respectively.

Other specific complications following PD such as POPF (B/C), severe biliary fistulas (B/C) and PPH were comparable between the two groups. In the same line, the rates of transfusion, reoperation, and readmissions were similar.

The rate of chylous ascites occurred less frequently with the robotic approach (3.3% vs. 16.7%, P=0.01).

Analysis of the learning curve: results of the CUSUM analysis

The CUSUM analysis, based on operative duration, enabled the division of patients into three consecutive groups of 20 patients (Groups A, B, and C) for the learning curve analysis. Operative time was longer for the first 20 patients compared with the subsequent groups (583.0±47.5 vs. 549.5±44.6 and 569.0±42.7 minutes, P=0.07) (Figure 1). This analysis revealed that the learning curve was reached at the 40^th procedure.

Figure 1.

Learning curve using CUSUM analysis. Operative time (blue) and CUSUM of the operative time (red). CUSUM, cumulative sum.

The demographic and pathological characteristics of patients undergoing RPDs over the three periods are presented in Table 2. Operative data and outcomes are shown in Table 4. The rate of DGE decreased significantly over time (60% in Group A vs. 35% in Group B vs. 25% in Group C, P=0.049). The conversion rate decreased without significance during the three periods (20% in Group A vs. 10% in Group B and 15% in Group C, P=0.90). Hospital stays were shorter for the final group of patients (19.9±14.2 vs. 25.3±21.6 vs. 14.5±8.7 days, P=0.17).

Table 4. Operative data and outcomes of RPD.

Variables	RPD (n=60)	RPD 1 to 20 (n=20)	RPD 21 to 40 (n=20)	RPD 41 to 60 (n=20)	P value
Operative time (min)	567.2±46.3	583.0±47.5	549.5±44.6	569.0±42.7	0.07
Conversion	9 (15.0)	4 (20.0)	2 (10.0)	3 (15.0)	0.90
Intra-operative transfusion	1 (1.7)	0	1 (5.0)	0	>0.99
Venous resection	4 (6.7)	1 (5.0)	2 (10.0)	1 (5.0)	>0.99
Gastroparesis	24 (40.0)	12 (60.0)	7 (35.0)	5 (25.0)	0.049*
A	5 (8.3)	4 (20.0)	0	1 (5.0)
B	7 (11.7)	5 (25.0)	1 (5.0)	1 (5.0)
C	12 (20.0)	3 (15.0)	6 (30.0)	3 (15.0)
POPF	26 (43.3)	8 (40.0)	11 (55.0)	7 (35.0)	0.41
Grade B or C	8 (13.3)	1 (5.0)	5 (25.0)	2 (10.0)	0.25
Bile leakage	14 (23.3)	6 (30.0)	5 (25.0)	3 (15.0)	0.64
Grade B or C	8 (13.3)	1 (5.0)	5 (25.0)	2 (10.0)	0.25
Haemorrhage	8 (13.3)	0	5 (25.0)	3 (15.0)	0.08
Chyle leak	2 (3.3)	1 (5.0)	0	1 (5.0)	>0.99
Sepsis	14 (23.3)	2 (10.0)	7 (35.0)	5 (25.0)	0.21
Post-op transfusion	11 (18.3)	2 (10.0)	6 (30.0)	3 (15.0)	0.34
Re-operation	7 (11.7)	0	5 (25.0)	2 (10.0)	0.06
Clavien-Dindo III–V	17 (28.3)	4 (20.0)	9 (45.0)	4 (20.0)	0.13
Hospital stay (days)	19.9±16.1	19.9±14.2	25.3±21.6	14.5±8.7	0.17
Re-admission rate	13 (21.7)	1 (5.0)	6 (30.0)	6 (30.0)	0.11
90-day mortality	3 (5.0)	0	1 (5.0)	2 (10.0)	0.77

Data are presented as mean ± SD or n (%). *, P<0.05. POPF, post-operative pancreatic fistula; RPD, robot-assisted pancreatoduodenectomy; SD, standard deviation.

In general, the various complications tended to decrease over the three periods except for 90-day mortality, but without ever reaching significance.

Pathological results and early oncologic outcomes

Tumour characteristics varied over time with a tendency towards more complex cases as shown in Table 5. In fact, ampullary tumours were more common at the beginning of the study (15% in both Groups A and B vs. 0% in Group C, P=0.23) whereas pancreatic cancers were more frequent in the later cases (30% in Group A vs. 25% in Group B vs. 55% in Group C, P=0.11). Tumour size (27.3±18.2 vs. 18.8±7.0 vs. 28.3±21.0 mm, P=0.16) and the number of resected lymph nodes (15.5±11.3 vs. 13.9±7.4 vs. 17.9±8.1, P=0.14) were higher in the last group of patients.

Table 5. Pathological data and early oncological outcomes of RPD divided in three periods.

Variables	Population (n=180)	OPD (n=120)	RPD (n=60)	P value†	RPD 1 to 20 (n=20)	RPD 21 to 40 (n=20)	RPD 41 to 60 (n=20)	P value‡
Malignant disease	138 (76.7)	92 (76.7)	46 (76.7)	0.92	16 (80.0)	13 (65.0)	17 (85.0)	0.40
Pancreatic ADC	82 (45.6)	60 (50.0)	22 (36.7)	0.08	6 (30.0)	5 (25.0)	11 (55.0)	0.11
Duodenal ADC	12 (6.7)	8 (6.7)	4 (6.7)	>0.99	3 (15.0)	0	1 (5.0)	0.31
Cholangiocarcinoma	19 (10.6)	14 (11.7)	5 (8.3)	0.48	1 (5.0)	3 (15.0)	1 (5.0)	0.60
Ampulloma	12 (6.7)	6 (5.0)	6 (10.0)	0.22	3 (15.0)	3 (15.0)	0	0.23
IPMN	9 (5.0)	4 (3.3)	5 (8.3)	0.17	1 (5.0)	2 (10.0)	2 (10.0)	>0.99
NET	13 (7.2)	9 (7.5)	4 (6.7)	>0.99	1 (5.0)	1 (5.0)	2 (10.0)	>0.99
Tumour size (mm)	26.6±16.1	27.2±15.6	25.3±17.2	0.30	27.3±18.2	18.8±7.0	28.3±21.0	0.16
Vascular embolism	39 (21.7)	22 (18.3)	17 (28.3)	0.15	5 (25.0)	6 (30.0)	6 (30.0)	0.92
Perineural invasion	83 (46.1)	59 (49.2)	24 (40.0)	0.19	8 (40.0)	9 (45.0)	7 (35.0)	0.81
Lymph nodes harvested	20.6±12.6	23.1±13.4	15.8±9.1	<0.001*	15.5±11.3	13.9±7.4	17.9±8.1	0.14
Lymph nodes				0.15				0.72
N1	41 (22.8)	32 (26.7)	9 (15.0)		3 (15.0)	3 (15.0)	3 (15.0)
N2	18 (10.0)	10 (8.3)	8 (13.3)		4 (20.0)	3 (15.0)	1 (5.0)
Margins R1	21 (11.7)	15 (12.5)	6 (10.0)	0.58	2 (10.0)	2 (10.0)	2 (10.0)	>0.99

Data are presented as mean ± SD or n (%). ^†, OPD vs. RPD. ^‡, RPD 1 to 20 vs. RPD 21 to 40 vs. RPD 41 to 60. *, P<0.05. ADC, adenocarcinoma; IPMN, intraductal papillary mucinous neoplasm; NET, neuroendocrine tumour; OPD, open pancreatoduodenectomy; RPD, robot-assisted pancreatoduodenectomy; SD, standard deviation.

Discussion

The robotic approach in pancreatic surgery is a significant advancement in surgical innovation, particularly in complex procedures such as PD due to its technical advantages and promising results. However, these results came from studies carried out in centers combining high-volume pancreatic surgery and unlimited ARPs. The present study, conducted in a tertiary referral center for pancreatic surgery with limited ARP, demonstrates that RPD is a safe and effective alternative to OPD. Indeed, using propensity matching analysis, this study showed a 90-day mortality rate similar in the RPD group compared with OPD. The absence of mortality during the first 20 RPD procedures highlights the safety of this approach. It is important to note that during the early learning phase, patient selection for PD was more restrictive. This criterion involved cases with lower BMI, a higher occurrence of ampullary and neuroendocrine tumours (NETs) and fewer borderline cases of PDAC.

Inequality of ARPs remains a significant issue in the widespread adoption of robotic surgery. At institutions where robotic platforms are limited, as in cases where multiple specialties must share the same few systems, competition for access can create bottlenecks. This shared use often limits the frequency of robotic procedures in each specialty, potentially slowing the learning curve and reducing the ability to maximise the benefits of this technology. This limitation underscores the need for programs that enhance the feasibility of robotic procedures by ensuring adequate access and training opportunities (32). In fact, as noted in the Paris Jury-Based Consensus Conference, limited ARP remains a significant challenge in the learning curve for RPD (33). In addition, in a difficult economic climate, hospitals sometimes choose to provide greater access to surgical specialties with shorter and/or more profitable surgical procedures. This can make it even more difficult to develop procedures such as RPD. In the present work, this limited and shared access explains the limited number of RPD cases performed annually with an average of 12 procedures.

Despite this limited access, the results in the present work revealed a learning curve reached at the 40^th procedure. These results are in line with previously published studies performed in centers with more extensive (or full) ARPs (34,35). These two studies highlight the importance of experience in optimizing RPD: the learning curve identified (requiring 40 and 30 patients respectively), suggests that as surgeon gain proficiency, both the operating time and length of stay can be significantly reduced. This finding represents the most significant contribution of the present work. In fact, even under less-than-ideal conditions, the robotic approach appears to enable the establishment of a minimally invasive surgical program for PD with no loss of opportunity for the patient.

The role of the learning curve was evident over 60 interventions which were divided into three periods of 20 patients each. As the surgeon’s experience increased, patient selection criteria became progressively and naturally less restrictive. This progressive adaptation reflects the mastery learning curve, wherein surgeons effectively manage increasingly complex cases as they gain experience (32). This gradual expansion allowed for the safe introduction of RPD with no 90-day mortality observed during the first period. The conversion rate for all RPDs was 15% which is comparable to or lower than the rates reported in the literature (23). Although the indications broadened in later periods, the 90-day mortality rate remained comparable to that OPD, demonstrating the safety of RPD as the surgeon’s experience increased. Moreover, the rate of Clavien-Dindo ≥ III major complications was consistent with the benchmarks in pancreatic surgery (36).

The early increase in POPF (B/C) in the RPD group gradually decreased as experience was gained. This trend can be influenced by several factors including tumour type with more ampullomas than pancreatic adenocarcinomas in the RPD group. Indeed, pancreatic texture and the calibre of the Wirsung duct are recognized risk factors for pancreatic fistula (37,38). Recently, Bruna et al. reported no difference in POPF rates between minimally invasive (including robotic and laparoscopic) PD and open procedures, but with more severe consequences when minimally invasive-POPF were present (39). We did not find these results in our work.

The impact of learning was also reflected in the major complications rate which was almost comparable between the final RPD group and OPD. The shorter hospital stays observed in the final RPD cases are consistent with other studies on the advantages of minimally invasive surgery (22). In contrast, hospital stays were significantly longer for the first RPD patients which was probably due to the higher incidence of DGE in the early periods. In fact, we observed a high rate of DGE, reaching 60% and 35% during the first and second phases, respectively. This rate decreased to 25% in the third phase, aligning with the rate reported by the OPD group. The primary hypothesis links this improvement to the modified gastroenterostomy site after the 40^th procedure. Specifically, as described in the “Methods” section, from the 40^th procedure onward, the future gastroenterostomy site was identified in a pre-colic position at the beginning of the surgery.

Overall, the present results confirmed that RPD offers comparable outcomes to OPD (18). This is comparable with other studies, both retrospective and prospective, that have shown the potential of RPD as a viable alternative to traditional open surgery (22,23,40,41).

Conversely, laparoscopic PD has not achieved similar results and has become less popular since a prospective randomised study was halted owing to increased mortality (14). The technical challenges associated with laparoscopy, including limited instrument mobility, lack of 3D vision, and ergonomic difficulties explain the reason that it has not been widely adopted in a complex procedure such as PD. In comparison, robotic surgery overcomes these limitations, offers greater precision, improved ergonomics, and a more controlled surgical environment.

Concerning early oncological outcomes, the RPD group had the same R0 resection rate as the OPD group, indicating that RPD did not compromise oncological outcomes. These results were in line with a previous study (42). However, the number of harvested lymph nodes was higher in the OPD group which was probably due to the higher proportion of pancreatic cancers in that cohort which necessitated more extensive lymphadenectomy. RPD is therefore a viable option for treating malignant tumours without compromising the quality and standard of oncological surgery.

The results presented in this limited access condition which are consistent with those published with wider ARPs, are likely to be related to the expertise of the entire OR team, including nurses and anaesthetists who have been involved in other robotic procedures.

From an economic point of view, RPD is initially more expensive than OPD due to the costs associated with advanced technology and longer operative time. However, as the surgeon becomes more experienced, the cost differential decreases, primarily owing to shorter hospital stays. A more detailed medico-economic analysis would be beneficial to fully explore the long-term cost benefits of RPD (43,44).

In perspective, the present study revealed similar and perhaps even slightly better results in terms of shorter hospital stay, as previously described (22). An interesting avenue for future research would be to assess patient-reported outcomes. Evaluating post-operative quality of life, functional recovery, and patient satisfaction could provide valuable insights to complement clinical outcomes. These factors, which are often underexplored in comparisons between robotic and open approaches, could reveal additional advantages of robotic surgery, particularly in terms of post-operative comfort, return to daily activities, and overall well-being. Such evaluations could further strengthen the case for broader adoption of robotic technology in hospitals.

Despite its strengths, the present study has several limitations. First, it is a non-randomised study, although we attempted to make the two groups as comparable as possible using 2:1 propensity score matching analysis. Secondly, we did not collect data on intra-operative blood loss which has been lower in RPD in some studies (45). In addition, the fact that all RPD were performed by a single surgeon is a strength of this study, but it could be considered as a bias. Indeed, this fact could underestimate the reproducibility of this major procedure.

Conclusions

Limited ARPs should not exclude RPD since the results of this study show its effectiveness even in resource-constrained settings. Despite sharing a small number of systems across multiple specialties, the present study demonstrated that RPD can be performed safely with outcomes comparable to OPD, especially during the learning curve. Analysis of the learning curve suggests that 40 procedures are required for achieving better outcomes with greater ARPs. During this initial phase, careful patient selection is essential. These results should be reassuring and should encourage centers with limited ARPs to start an RPD program.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The Study protocol complied with the ethical guidelines of the Declaration of Helsinki and its subsequent amendments and was approved by the Rennes University Hospital Ethics Committee (N°00045607A). All patients provided their consent.

Data Sharing Statement

Available at https://hbsn.amegroups.com/article/view/10.21037/hbsn-2024-723/dss

hbsn-14-06-937-dss.pdf ^{(59.7KB, pdf)}

DOI: 10.21037/hbsn-2024-723