Long-Term and Oncologic Outcomes of Robotic Versus Laparoscopic Liver Resection for Metastatic Colorectal Cancer: A Multicenter, Propensity Score Matching Analysis

Abstract

Background

To assess long-term oncologic outcomes of robotic-assisted liver resection (RLR) for colorectal cancer (CRC) metastases as compared to a propensity-matched cohort of laparoscopic liver resections (LLR). Although safety and short-term outcomes of RLR have been described and previously compared to LLR, long-term and oncologic data are lacking.

Methods

A retrospective study was performed of all patients who underwent RLR and LLR for CRC metastases at six high-volume centers in the USA and Europe between 2002 and 2017. Propensity matching was used to match baseline characteristics between the two groups. Data were analyzed with a focus on postoperative and oncologic outcomes, as well as long-term recurrence and survival.

Results

RLR was performed in 115 patients, and 514 patients underwent LLR. Following propensity matching 115 patients in each cohort were compared. Perioperative outcomes including mortality, morbidity, reoperation, readmission, intensive care requirement, length-of-stay and margin status were not statistically different. Both prematching and postmatching analyses demonstrated similar overall survival (OS) and disease-free survival (DFS) between RLR and LLR at 5 years (61 vs. 60% OS, p = 0.87, and 38 vs. 31% DFS, p = 0.25, prematching; 61 vs. 60% OS, p = 0.78, and 38 vs. 44% DFS, p = 0.62, postmatching).

Conclusions

Propensity score matching with a large, multicenter database demonstrates that RLR for colorectal metastases is feasible and safe, with perioperative and long-term oncologic outcomes and survival that are largely comparable to LLR.

Introduction

Laparoscopic liver resection (LLR) has been compared to open liver resection (OLR), and the non-inferiority of a laparoscopic approach is well-established in terms of perioperative outcomes [1–4]. Studies comparing laparoscopic and open hepatectomy for metastatic colorectal cancer (CRC) have shown analogous oncologic outcomes including margin positivity, recurrence and survival [1–3, 5–7]. Robotic-assisted surgery has been increasingly described as an alternative to laparoscopy for minimally invasive liver resection. Between 2010 and 2019 there were 21 series, each including at least nine patients, which described 753 robotic liver resections (RLR) [8–28]. Advantages offered by RLR include a greater range of motion with articulating instruments, ease of fine movements, improved three-dimensional field of vision and minimization of physiologic tremor [29]. Disadvantages are increased cost and the loss of tactile feedback [8, 29].

To the best of our knowledge, there are fourteen series that compare laparoscopic and robotic liver surgery [9, 11, 12, 14, 15, 17, 19, 21–26, 28]. Five studies demonstrated longer operative times, four studies showed higher rates of Pringle maneuver, and three showed higher blood loss with robotic-assisted surgery [9, 11, 12, 14–17, 19, 24, 26, 28]. Other intraoperative and postoperative outcomes were equivalent. One recent meta-analysis comparing clinical outcomes of RLR and LLR concluded identified RLR as having longer operative times, more intraoperative blood loss and higher cost [30].

Though the perioperative and short-term outcomes of RLR have been well described and shown to be comparable to LLR, oncologic and long-term data are largely lacking, though one recent study did compare survival and oncologic outcomes between 111 LLR and 61 RLR for a heterogeneous group of malignancies [28]. This primary aim of this multicenter study was to provide long-term and oncologic outcomes of RLR performed for CRC metastases, with propensity score (PS) matching to an overlapping laparoscopic cohort, with a secondary aim of comparing perioperative outcomes.

Materials and methods

Patients and indications

Participating centers included five from the USA and one from Europe: University of Pittsburgh Medical Center, Ghent University Hospital Medical School in Belgium, Memorial Sloan Kettering Cancer Center, Cleveland Clinic, Tulane University and Carolinas Medical Center. All centers and all surgeons had adequate LLR volume to have surpassed the 45–75 case range that is the suggested learning curve, and all centers had adopted RLR as well with at least 10 RLR having been done at each center [31–33]. All LLR and RLR cases performed for metastatic CRC at each center were included in the database.

Patients were diagnosed with liver metastasis from CRC cancer primaries using cross-sectional imaging and liver biopsy when deemed necessary, and diagnoses were confirmed by surgical pathology of the resected liver. All patients were evaluated in a multidisciplinary fashion prior to any surgical intervention to determine appropriate neoadjuvant and adjuvant therapies. Indications and candidacy for robotic-assisted liver surgery are similar to those previously described for laparoscopic liver surgery [5]. Tumor characteristics that precluded a minimally invasive approach included proximity to major vasculature and concern for adequate margin negative resection. All surgeons were hepatobiliary specialists and had extensive laparoscopic experience as well as training on the robotic platform.

Operative approach and management

Specific details of the surgical techniques used for robotic-assisted hepatic resections on included patients have been previously described [10, 18, 25, 34]. The da Vinci Surgical System (Intuitive, Inc., Sunnyvale, CA) robotic-assisted platform with four arms was used by all surgeons. In general, central venous pressures were kept low during parenchymal transection, intraoperative ultrasound (US) was used to guide resection, and a Pringle maneuver was used at the discretion of the surgeon. Intraabdominal placement of drains and the decision to observe patients in the intensive care unit postoperatively were at the discretion of each primary surgeon.

Data collection

Patient data were collected from each center in prospective databases and then compiled into a single database for retrospective analysis. Complications were classified according to the Clavien–Dindo system with those graded at three or higher were considered serious [35]. Tumor characteristics reported included number, location and size of tumors as well as histologic differentiation, margin status and distance of closest margin on surgical pathology. The study was carried out by all of all participating institutions with approval from the Institutional Review Board (IRB PR017020317).

Classification of hepatic resections

Resection types were described according to the Brisbane classification of liver anatomy. Hepatic resections were defined as major when patients underwent removal of four or more liver segments [36]. Minor resections were defined as any excision of less than four hepatic segments, and non-anatomic wedge resections were defined as the excision of less than one Couinaud’s segment [37].

Propensity matching

PS matching was used to minimize differences between the RLR and LLR cohorts. The PS was generated using a logistic regression model including the following covariates: age, sex, American Society of Anesthesiologists (ASA) classification, BMI, synchronous liver lesions, administration of neoadjuvant chemotherapy, classification of resection (major or minor), size of largest liver lesion (more or less than 5 cm), anatomic tumor location (anterolateral, posterosuperior, caudate or multiple) and tumor grade (well, moderately or poorly differentiated). Nearest-neighbor method was used to perform one-to-one matching without replacement.

Statistical analysis

The database was compiled using Microsoft Excel and analyzed using STATA/SE 15.1 software. After assessing the quality of the data provided by each center, data were assembled into a single dataset. Continuous variables were expressed as means with standard deviation or medians with range, and percentages were used for discrete variables. The comparison of discrete variables between groups was determined using Chi-square test. T test was performed for normal distributed variables, and rank-sum test for non-normal distributed variables. Five-year survival analysis was performed using Kaplan–Meier for overall and disease-free survivals. Survival comparisons were performed using log-rank test. Patients who had no events at the end of 5-year period or at the time of closure of this study were censored. Patients who were lost at follow-up were contacted by the primary surgical team of each participating institution by phone call or by letter to assess patient’s conditions. Patients who could not be reached by their primary surgical specialists were censored at the time of their last follow-up visit. All tests were two-tailed, and p values ≤ 0.05 were considered significant.

Results

Patient characteristics

The study population included a total of 115 patients who underwent RLR resection for CRC metastases from January 2008 to October 2016 at six high-volume tertiary referral centers, as well as a cohort of 514 patients undergoing LLR for metastatic CRC between July 2002 and October 2017 at the same institutions. Prior to PS matching, patient characteristics were compared (Table 1). Patients in the RLR group were slightly younger, had higher ASA scores and were more likely to have had concomitant procedures at the time of operation. Patients in the RLR group were also more likely to have had neoadjuvant chemotherapy prior to resection of their primary tumor. Resections types differed between the two groups, though the majority of procedures in both groups were minor resections. Conversion to an open procedure was required in 6 patients in the RLR group and 63 patients in the LLR group (5.2 vs. 12.2%, p = 0.03). Reasons for conversion in the RLR group included bleeding (n = 3), technical difficulty (n = 1), greater than anticipated disease burden (n = 1) and inadequate tumor visualization with intraoperative US (n = 1). Rates of intraoperative transfusion were higher in the LLR cohort. On surgical pathology, tumor grade differed between the groups, and in a number of patients who had undergone neoadjuvant chemotherapy, only tumor necrosis without any viable malignant cells was observed. PS matching was performed to account for the effect of covariates on overall and disease-specific survival. After PS matching, 230 patients in the balanced cohort were compared (115 RLR vs. 115 LLR). The differences in the prematching patient characteristics were alleviated after matching (Table 1).

Table 1.

Patient characteristics of the overall cohort and after propensity score (PS) matching

	Robotic resection (n = 115)	Laparoscopic resection (n = 514)	p (overall cohort)	Laparoscopic resection after PS matching (n = 115)	p (after PS matching)
Mean age [years ± SD]	61 ± 11	63 ± 12	0.06	61 ± 12	0.57
Sex [n (%)]			0.34		1.00
Female	39 (33.9)	199 (38.8)		40 (34.8)
Male	76 (66.1)	314 (61.2)		75 (65.2)
Site of primary colon cancer [n (%)]			0.05
Rectum	42 (37.2)	161 (32.1)
Sigmoid	23 (20.4)	159 (31.7)
Right colon	17 (15.0)	94 (18.3)
Left colon	11 (9.7)	43 (8.6)
Transverse colon	10 (8.8)	19 (3.8)
Ileocecal	8 (7.1)	21 (4.2)
Multifocal	2 (1.8)	7 (1.4)
Unknown	2 (1.8)	10 (1.9)
Patients presenting with synchronous liver lesions [n (%)]	49 (42.6)	237 (46.2)	0.53	44 (38.3)	0.59
Prior abdominal surgery [n (%)]	102 (88.7)	418 (82.0)	0.10
Neoadjuvant chemotherapy prior to primary resection	64 (55.7)	56 (14.7)	<0.001
Neoadjuvant radiation prior to primary resection	8 (9.0)	60 (14.6)	0.18
Adjuvant chemotherapy after primary resection	81 (71.1)	219 (64.4)	0.44
Preoperative chemotherapy prior to liver resection	63 (54.8)	283 (62.3)	0.16
Adjuvant chemotherapy after liver resection	83 (76.9)	307 (73.1)	0.52
Body mass index (BMI) [mean ± SD]	28 ± 6	28 ± 5	0.29	29 ± 6	0.74
American Society of Anesthesia classification [n (%)]			<0.001		0.64
Class 1	0 (0)	49 (9.7)		0 (0)
Class 2	21 (18.3)	192 (37.9)		16 (13.9)
Class 3	82 (71.3)	252 (49.7)		88 (76.5)
Class 4	12 (10.4)	14 (2.8)		11 (9.6)
Number of tumors [n (%)]			0.58
1 metastasis	78 (67.8)	320 (62.7)
2 metastases	24 (20.9)	117 (22.9)
3 or more metastases	13 (11.3)	73 (14.3)
Longest diameter of largest tumor in cm [median (range)]	2.5 (0.3–12.2)	2.4 (0.3–15.2)	0.06
Longest diameter of largest tumor					0.84
< 5 cm	97 (85.8)			84 (87.5)
≥ 5 cm	16 (14.2)			12 (12.5)
Anatomic tumor location [n (%)]			0.20		0.85
Anterolateral	57 (49.6)	245 (47.9)		58 (50.4)
Posterosuperior	22 (19.1)	136 (26.6)		24 (20.9)
Caudate	3 (2.6)	6 (1.2)		1 (0.9)
Multiple	33 (28.7)	124 (24.3)		32 (27.8)
Type of resection [n (%)]			<0.0001
Single segment or non-anatomic wedge hepatic resection	63 (54.8)	379 (74.2)
Left lateral sectionectomy	31 (27.0)	69 (13.5)
Right hepatectomy	9 (7.8)	33 (6.5)
Left hepatectomy	6 (5.2)	18 (3.5)
Caudate resection	4 (3.5)	4 (0.8)
Right trisectionectomy	2 (1.7)	1 (0.2)
Left trisectionectomy	0 (0)	1 (0.2)
Central hepatectomy	0	6 (1.2)
Type of resection [n (%)]					0.73
Minor	97 (84.3)			94 (81.7)
Major	18 (15.7)			21 (18.3)
Patients who underwent concomitant procedures at operation [n (%)]	72 (62.6)	180 (35.2)	<0.001
Operative time in minutes [mean ± SD]	272 ± 115	253 ± 118	0.12
Patients receiving intraoperative blood transfusions [n (%)]	11 (9.6)	166 (32.5)	<0.001
Required conversion to an open procedure [n (%)]	6 (5.2)	63 (12.1)	0.03
Tumor grade [n (%)]			0.005		0.70
Well differentiated	16 (13.9)	68 (13.2)		20 (17.4)
Moderately differentiated	63 (54.8)	218 (42.4)		62 (53.9)
Poorly differentiated	4 (3.5)	27 (5.3)		3 (2.6)
Not reported	25 (21.7)	189 (36.8)		27 (23.5)
Unable to classify as no viable tumor observed in surgical specimen	7 (6.1)	12 (2.3)		3 (2.6)
Closest margin for R0 resections in mm [mean ± SD]	0.8 ± 0.8	0.8 ± 0.8	0.72

BMI body mass index, ASA American Society of Anesthesiologists

Perioperative outcomes

There was one perioperative mortality in the LLR group and one in the RLR group from cardiac arrest. Other perioperative outcomes after PS matching are shown in Table 2, and no outcomes were statistically different between the two cohorts. The most common complications in the RLR cohort were biloma or intraabdominal abscess (n = 7), followed by cardiac arrhythmias (n = 5) and ileus (n = 4). Other complications included empyema, urinary retention, pleural effusion, persistent oxygen requirement, vocal cord paralysis, dehydration, fluid retention requiring diuretics, pneumonia, atelectasis, acute kidney injury and venous thromboembolic events.

Table 2.

Postoperative outcomes after propensity score (PS) matching

	Robotic resection (n = 115)	Laparoscopic resection (n = 115)	p
Perioperative death within 30 days [n (%)]	1 (0.9)	1 (0.9)	1.00
Complications [n (%)]	36 (31.3)	32 (27.8)	0.66
Serious complication (Clavien–Dindo ≥ 3) [n (%)]	12 (10.4)	17 (14.8)	0.43
Patients requiring reoperation within 30 days [n (%)]	1 (0.9)	4 (3.5)	0.37
Patients requiring readmission within 30 days [n (%)]	8 (7.0)	8 (7.0)	1.00
Intensive care unit admission [n (%)]	17 (14.8)	21 (18.3)	0.59
Length-of-stay in days [median (IQR)]	5 (3–6)	4 (2–6)	0.46
Margin status [n (%)]			0.18
R0	84 (73.7)	89 (77.4)
R1	19 (16.7)	23 (20.0)
R2	4 (3.5)	1 (0.9)
Unable to classify as no viable tumor observed in surgical specimen	7 (6.1)	2 (1.7)

Oncologic outcomes and survival

Margin status following resection was not significantly different between the laparoscopic and robotic cohorts (p = 0.18, Table 2). The median follow-up period for the overall cohort was 2.9 years, with a median follow-up of 3.1 years for the RLR group and of 2.8 years for the LLR group. Kaplan–Meier survival curves are shown in Figs. 1 and 2. There were no statistical differences in overall survival (OS) and disease-free survival (DFS) between the RLR and LLR groups prior to PS matching (61 vs. 60% OS, p = 0.87, and 38 vs. 31% DFS, p = 0.25) or after PS matching (61 vs. 60% OS, p = 0.78, and 38 vs. 44% DFS, p = 0.62).

Fig. 1.

Kaplan–Meier curves for overall (a) and disease-free (b) survival following laparoscopic (LLR) and robotic-assisted liver resection (RLR) in the prematching cohort. a Five-year survival rate 0.60 for LLR and 0.61 for RLR, p = 0.87 (log-rank test); b 5-year survival rate 0.31 for LLR and 0.38 for RLR, p = 0.25 (log-rank test)

Fig. 2.

Kaplan–Meier curves for overall (a) and disease-free (b) survival following laparoscopic (LLR) and robotic-assisted liver resection (RLR) in the after PS matching. a Five-year survival rate 0.60 for LLR and 0.61 for RLR, p = 0.78 (log-rank test); b 5-year survival rate 0.44 for LLR and 0.38 for RLR, p = 0.62 (log-rank test)

In the RLR cohort, 60 patients (52.2%) had recurrence of their metastatic colorectal cancer following liver resection. The majority of recurrences occurred in the liver (n = 24) followed by the lungs (n = 21), multiple organs (n = 11) and the site of primary CRC (n = 7). Management of recurrence was primarily by chemotherapy (n = 40), followed by repeat surgical resection (n = 12), liver directed therapies including ablation and Yttrium-90 (n = 5) and none or palliation (n = 4).

Discussion

The introduction of minimally invasive techniques for liver resections has been slow as compared to other fields of general surgery. Concerns have been raised that LLR might not be oncologically equivalent to open surgery in patients with large lesions or with metastatic burden in anatomically unfavorable segments. Recently, several series from the USA, Europe and Japan have described robust experiences with LLR performed for CRC metastases, often with comparisons to contemporary patients undergoing OLR and including long-term outcomes [38–41]. Results from the first trial to randomize patients with colorectal liver metastases to open or laparoscopic surgery were recently published and included 280 patients from a single center, 133 of whom underwent LLR. They primarily examined perioperative outcomes and found that LLR was associated with less postoperative complications than OLR [42].

Platforms for robotic-assisted surgery have improved visualization and the ability to perform complex surgical resections. If robotic-assisted surgery is to be deemed an acceptable option for management of CRC liver metastases, then results, including oncologic outcomes, should be comparable to those observed with LLR. Of the 21 major series reporting on robotic-assisted liver surgery, 13 reported R0 resection rates which ranged from 85 to 100%. Only eight studies provided data on duration of follow-up, which ranged from 9.6 to 75 months, and only those eight studies provided any data on short-term recurrence rates or survival [13–16, 20, 25, 27, 28]. The dearth of oncologic data is also emphasized by a by a recent systemic review of RLR which included 31 studies, only 4 of which included oncologic outcomes [43]. The recent series published by Lim et al. compares RLR and LLR with a focus on resection margins and long-term outcomes, in addition to perioperative results. The majority of their cohort was comprised of patients with hepatocellular carcinoma, though metastatic CRC patients were also included [28]. To our knowledge, this series is the largest multicenter study on RLR performed for CRC metastases, with the inclusion of a number of major hepatectomies, comparison to an overlapping laparoscopic cohort and a focus on long-term and oncologic outcomes.

This series reports results of a multicenter experiences in LLR and RLR performed for CRC metastases, both before and after applying PS matching, to analyze whether RLR is a safe, feasible and non-inferior approach to LLR with comparable short and long-term outcomes. A strength of the study is the multicenter nature with both LLR and RLR patients contributed from six geographically diverse, high-volume hepatobiliary centers, where both surgeons and centers have surpassed the 45–75 case range that has been reported as the learning curve for LLR [31–33]. Generalizability is therefore maximized, particularly to centers already routinely performing LLR and interested in transitioning to or adding RLR to their repertoire. It is important to emphasize that these operations should be performed by experienced, specialized hepatobiliary surgeons in institutions well equipped to care for such patients. As emphasized by a recent review of RLR performed for malignancy, experience with conventional laparoscopy is necessary for successful implementation of a robotic liver surgery program [44].

In order to maximize numbers and allow for PS matching between cohorts, all LLR and RLR performed for metastatic CRC at participating centers were included, resulting in a long observation period of 15 years. As laparoscopic technology and the adoption of LLR preceded that of RLR, LLR were performed between 2002 and 2017 as compared to RLR which were performed from 2008 to 2016, thereby making the experiences overlapping but not strictly contemporaneous. The main operating surgeons at these institutions remained fairly consistent, though there was some expected turnover. Likewise, techniques and technology changed over this time period, which does raise concern for disparity in the laparoscopic cohort but reflects accurately the actual experience of adopting new approaches. It is important to interpret the comparison between cohorts with such caveats in mind.

Both before and after PS matching, more than 10% of patients in each cohort underwent major resections, which is a greater proportion than previously reported in other studies, and perhaps representative of increasing experience levels and comfort with RLR [10]. It is also notable that 45.5% of patients in the overall cohort presented with synchronous liver lesions, which is higher than the 20–25% reported in the literature and may reflect some inherent bias in reviewing a cohort of metastatic CRC patients who all underwent liver resection for their disease [45]. Owing to the nature of metastatic CRC, the majority of patients (> 80%) in both pre-matched groups had prior abdominal surgery, primarily consisting of previous colon or rectal resections. Interestingly, though more patients in the RLR group underwent concomitant procedures at operation (62.6 vs. 35.2, p < 0.001), operative times were similar. The types of resections performed differed, with more patients in the LLR group undergoing a single segment or wedge resection (74.2 vs. 54.8%) and proportionally more patients in the RLR group undergoing caudate resections (3.5 vs. 0.8%), right hepatectomies (7.8 vs. 6.5%) and left hepatectomies (5.2 vs. 3.5%). Rates of conversion to laparotomy, however, were significantly higher in the LLR group (12.1 vs. 5.2%, p = 0.03).

Margin status was not significantly different between the two cohorts, and R0 rates were 73.7% for RLR and 77.4% for LLR. More recent series of LLR for CRC metastases have reported somewhat higher R0 resection rates between 78.5 and 93.7%, and this difference may be accounted for by the 15 year collection period in this study as compared to the more recent time periods of other series [38–42]. Prior series reporting on RLR, including procedures done for various malignancies and not solely CRC, have reported high R0 rates of 89–100% [9–11, 13, 14, 18, 20, 21, 24, 25, 46]. No differences were observed in OS or DFS between LLR and RLR prior to PS matching (61 vs. 60% OS, p = 0.87, and 38 vs. 31% DFS, p = 0.25) or after PS matching (61 vs. 60% OS, p = 0.78, and 38 vs. 34% DFS, p = 0.62) at 5 years. Recent laparoscopic series for CRC metastases have reported 5-year OS between 56.8 and 70.1%, and DFS between 39.7 and 53.4% [38–42].

An important point that is frequently raised regarding robotic-assisted surgery is the increased cost. Given the multi-institutional nature of the study and the inclusion of a European center, accurate data on cost that would allow for a comparison between the RLR and LLR groups are not available. Other study limitations include the retrospective nature and selection bias in choosing patients to undergo RLR. Additionally, though PS matching largely balances the cohorts and attempts to decrease selection bias, it may omit unforeseen confounders which impact outcomes.

This study represents, to the best of our knowledge, the largest series of RLR to date and the only one reporting solely on long-term oncologic outcomes of CRC metastases. After PS matching, both perioperative and long-term oncologic outcomes were compared to a contemporaneous LLR cohort and were largely equivalent. Though more studies comparing the two approaches are needed, the present results provide valuable information regarding the non-inferiority of RLR as an alternative to LLR.