Limited liver resections in the posterosuperior segments: international multicentre propensity score-matched and coarsened exact-matched analysis comparing the laparoscopic and robotic approaches Free

Abstract

Introduction

Minimally invasive liver resection (MILR) has undergone a paradigm shift since the first procedure was performed in 1991¹. After three consensus meetings and successful creation of the International Laparoscopic Liver Society, the adoption of MILR has been rapidly increasing worldwide^2–4. MILR in the anterolateral segments has been considered safe and effective ever since the first consensus meeting in 2008. However, it was only after nearly a decade that MILR for tumours in the posterosuperior segments⁵ and major resections⁶, including donor hepatectomy, gained approval as safe procedures⁷.

The posterosuperior segments are considered ‘difficult segments’ (Fig. 1) because their location high up in the abdominal cavity, beneath the ribs and towards the posterior abdominal wall, makes these segments difficult to access, especially via a minimally invasive approach.

Fig. 1

Illustration of posterosuperior liver segments

S, segment.

Open in new tabDownload slide

Hence, when bleeding occurs during MILR, it becomes more challenging to control owing to poor visualization and difficulty in accessing the bleeding site. These limitations also affect the ability of the surgeon to achieve oncological safety by ensuring adequate surgical margins⁸. There has been an increasing number of studies reporting on MILR for tumours in the posterosuperior segments^5,9,10. These studies from expert centres have demonstrated not only the feasibility of MILR for tumours in the posterosuperior segment but also superior perioperative outcomes compared with those of conventional open surgery^11–13.

The application of robotic surgery for liver resections is currently gaining traction worldwide¹⁴. The added advantages of three-dimensional visualization, better manipulation of instruments, stable movements, and improved dexterity are features that are contributing to increasing adoption of robotic surgery¹⁵. These features probably contribute to the reportedly shorter learning curve associated with robotic compared with conventional laparoscopic liver resection¹⁴. However, the main barrier to the widespread adoption of robotic surgery is the limited access to the technology resulting from its high cost¹⁶. Despite its many theoretical advantages, there is currently limited evidence in the literature demonstrating the clinical advantages of robotic surgery over conventional laparoscopy for liver resection^17,18. Furthermore, although it has been postulated that the increased dexterity of the robotic Endowrist^® (Intuitive Surgical, Sunnyvale, Ca) would be especially useful for limited resections of tumours located in the posterosuperior segments, there is scant evidence to date supporting this hypothesis¹⁹.

Taken the limited available evidence in the literature today, this large multicentre study was conducted with the primary objective of comparing the outcomes of robotic limited liver resections (R-LLRs) versus laparoscopic limited liver resections (L-LLR) for tumours located in the superior segments (VII, VIII, IVa) of the liver. To the best of the authors’ knowledge, this is the largest study to date to compare the outcomes of R-LLR versus L-LLR for tumours in the posterosuperior segments.

Methods

This was an international multicentre retrospective analysis of patients who underwent L-LLR or R-LLR of tumours in the posterosuperior segments (IVa, VII, VIII) (Fig. 1) at 24 centres between 2010 and 2019. The institutions obtained approvals according to their respective local centre’s requirements. This study was approved by the Singapore General Hospital Institution Review Board. As this was a retrospective study of pseudoanonymized data, the need for patient consent was waived. All data were collected centrally and analysed at the Singapore General Hospital.

Only patients who underwent pure laparoscopic or robotic surgery were included. Laparoscopically assisted (hybrid) and hand-assisted laparoscopic resections were excluded. Other study exclusion criteria were patients undergoing multiple concomitant liver resections in different segments, repeat liver resections, or had concomitant other major operations such as colectomy, hepaticojejunostomy, gastrectomy, stoma reversal, or hilar lymphadenectomy.

Patients who had concomitant minor operations such as cholecystectomy or hernia repair were included. The selection of patients is summarized in Fig. S1. All procedures were performed or supervised by specialized liver surgeons of consultant/attending level.

Definitions

The posterosuperior segments included segments VII, VIII and IVa, as defined according to the 2000 Brisbane classification²⁰. Notably, tumours in segment 1 were excluded from this study. LLRs were classified as wedge resection, segmentectomy or bisegmentectomy (segments VII/VIII or IVa/VIII) (Fig. 1). Of note, right posterior sectionectomies, right anterior sectionectomies, and left medial sectionectomies were not considered LLR in this study.

Postoperative complications were graded according to the Clavien–Dindo classification and recorded for up to 30 days or during the same hospital admission²¹. Difficulty of resections were graded according to the Iwate score²².

Statistical analysis

To ensure the robustness of the conclusions, analyses were carried out using two approaches in the causal inference toolbox for minimizing confounding and selection bias: propensity-score matching (PSM) and coarsened exact matching (CEM).

One-to-one CEM^23–26 was used to identify approximately-exact matches between patients who were assigned to R-LLR and L-LLR, and matched for all variables except sex as this factor was not considered to affect outcomes. Continuous variables were coarsened using an automatic binning algorithm based on Sturge’s rule.

Propensity scores^27–34 were developed using mixed-effects logistic regression modelling of all variables, with a random-effects term to account for between-centre heterogeneity. This model exhibited good discrimination (area under the curve 0.803, bias-corrected 95 per cent c.i. 0.758 to 0.847) and calibration (P = 0.800 from Hosmer–Lemeshow test with 10 deciles; slope 1.001; calibration in the large −0.000) (Figs S2 and S3). PSM in a 1 : 3 ratio was carried out without replacement within a caliper of 0.25 times the standard deviation of the linear predictor (log odds of the propensity score).

Comparisons of patient characteristics and perioperative outcomes between patients undergoing R-LLR or L-LLR in the unmatched cohort were accomplished using the Mann–Whitney U test and Pearson’s χ² test for continuous and categorical variables respectively. Comparisons in the PSM and CEM cohorts took into account the paired nature of the data; therefore, mixed-effects regression models (in which a random-intercept was used to denote the matched groups) and conditional logit models were used for continuous and binary variables.

Statistical analyses were done in Stata^® version 16.0 (StataCorp, College Station, TX, USA), and two-sided P < 0.050 was considered statistical significant.

Results

Some 983 of 1566 patients who underwent L-LLR or R-LLR of tumours in the posterosuperior segments met the study criteria, of whom 159 underwent R-LLR and 824 underwent L-LLR. There was no significant difference in the distribution of R-LLR and L-LLR over the study interval before and after matching (Fig. 2).

Fig. 2

Assessment of bias in historical cohorts

Distribution of procedures over year of surgery in the overall cohort a before matching, b after 1 : 3 propensity score matching (PSM), and c after 1 : 1 coarsened exact matching (CEM). L-LLR, laparoscopic limited liver resection; R-LLR, robotic limited liver resection. a  P = 0.280 (Pearson’s test), b  P = 0.671 (conditonal logit), c  P = 0.358 (McNemar’s test).

Open in new tabDownload slide

Overall clinicopathological features and perioperative outcomes are summarized in Tables 1 and 2. The median duration of operation was 206 (i.q.r. 145–290) min, and there were 57 conversions to open surgery (5.8 per cent). Some 309 patients (31.4 per cent) had cirrhosis and 80 of 981 patients (8.2 per cent) required intraoperative blood transfusion. The overall postoperative morbidity rate was 17.6 per cent (173 patients) and the major morbidity rate was 6.8 per cent (67 patients). The median duration of postoperative hospital stay was 5 (3–7) days.

Comparison between R-LLR and L-LLR in entire unmatched cohort

Table 1 shows the preoperative characteristics of patients in both groups, and perioperative outcomes are compared between R-LLR and L-LLR in Table 2. There were significant differences in the median operating time, median blood loss, blood loss of at least 500 ml, intraoperative blood transfusion, median duration of Pringle manoeuvre, conversion to open surgery, median duration of postoperative stay, and major morbidity in the unmatched cohort.

Comparison between R-LLR and L-LLR in matched cohorts

To ensure unbiased analysis, 1 : 3 PSM and further 1 : 1 CEM were performed. After matching, both groups were well balanced for all variables (Table 3 and Figs S4–S6). Standardized differences of less than 0.1 (or 10 per cent) may be interpreted to indicate a negligible difference in the mean or proportion of a continuous or categorical variable respectively (Fig. S6)³⁵. After PSM, 127 patients remained in the R-LLR group and 381 in the L-LLR group. The number in each group reduced further to 104 after CEM (Table 3).

After PSM, patients were matched adequately for preoperative characteristics including cirrhosis, Child–Pugh class, median tumour size, and Iwate grade (Table 3 and Fig. S6). Comparison of perioperative outcomes demonstrated that median blood loss (100 (i.q.r. 40–200) versus 200 (100–500) ml; P = 0.003), blood loss of at least 500 ml (9 (7.4 per cent) versus 94 (27.6 per cent); P < 0.001), intraoperative blood transfusion rate (4 (3.1 per cent) versus 38 (10.0 per cent); P = 0.025), open conversion rate (1 (0.8 per cent) versus 30 (7.9 per cent); P = 0.022), median duration of Pringle manoeuvre when applied (30 (20–46) versus 40 (25–58) min; P = 0.012), and median operating time (175 (130–255) versus 224 (155–300) min; P < 0.001) were lower for R-LLR than L-LLR. There was, however, no significant difference in other outcomes between the groups (Table 4).

The results after CEM showed a significant difference in the median duration of operation (P = 0.035), median blood loss (P < 0.001), and blood loss of at least 500 ml (P = 0.001), but no significant difference in intraoperative transfusion rate (P = 0.521) between the two groups. There was no difference in the median duration of Pringle manoeuvre (P = 1.000); however, there still remained a higher rate of conversion to open surgery in the L-LLR group (P = 0.034). Other key perioperative outcomes, including median postoperative stay (P = 0.931), 30-day readmission (P = 1.000), as well as morbidity (P = 1.000) and mortality (P = 1.000) rates, remained similar for the two approaches (Table 4).

Discussion

In this matched study of selected patients, both robotic and laparoscopic access for limited liver resections were associated with safe outcomes for tumours in the posterosuperior segments in well selected patients. Robotic resection was associated with less blood loss, a shorter operating time, and lower rate of conversion to open surgery than laparoscopic access.

Liver resection for tumours in the posterosuperior segments poses a challenge mainly because of its location behind the ribs. These resections present difficulty for surgeons even in open surgery, leading to larger incisions to obtain adequate exposure and greater blood loss. Liver resection by open surgery for small tumours in these difficult locations frequently leads to a large incision, resulting in greater physiological insult from the trauma of the surgical incision than from the liver resection itself. The pain resulting from this long incision is frequently the main determinant of the patient’s postoperative recovery³⁶. Although MILR for tumours in the posterosuperior segments has been proposed to overcome these limitations of open surgery, the difficulty of performing L-LLR for tumours located in the posterosuperior segment has been well established^11,37. Numerous studies have demonstrated inferior perioperative outcomes of LLR for posterosuperior segments, such as increased operating time, longer hospital stay, increased open conversion rate, and greater blood loss, compared with those after resections for tumours in the anterolateral segments⁹.

During the second international consensus meeting in 2015, LLR was considered a relative contraindication to liver resection for tumours in the posterosuperior segment³. However, with increasing experience, laparoscopy has now been shown to be associated with outcomes similar to and even superior to those of open surgery^12,13. Nonetheless, it must be acknowledged that numerous challenges exist when L-LLR is performed for tumours in the posterosuperior segments. The two-dimensional vision, along with the limited degree of freedom and difficulty in manipulating rigid instruments in a curvilinear plane of the posterior section, frequently lead to inadequate exposure as well as difficulty in arresting haemorrhage, especially from large blood vessels such as the hepatic veins³⁸. Nonetheless, recent technological advancements have made laparoscopy technically less challenging. These include introduction of the flexible three-dimensional laparoscope, which improves visualization of the surgical field, as well as the adoption of new techniques to approach this difficult-to-access region, such as the use of intercostal ports^39–41.

MILR for tumours in the posterosuperior segments is well established in many expert centres⁴². However, it is important to emphasize that MILR for such tumours should be undertaken only when surgeons have mounted their learning curve for simpler MILR. Surgeons should always adopt a stepwise approach, with proper selection of patients when attempting more complicated MILR, such as resection of tumours located in the posterosuperior segments⁴². Several retrospective comparative studies^11–13 have reported superior perioperative outcomes of LLR over open surgery for tumours in the posterosuperior segment without compromising oncological safety. A recent subset analysis of the OSLO-COMET RCT⁴³ showed comparable perioperative outcomes between laparoscopy and open surgery for tumours in the posterosuperior segments.

Robotic surgery has been applied for liver resections in order to overcome the limitations of conventional laparoscopy, particularly for tumours in the posterosuperior segments^44,45. The technical advantages of R-LLR over L-LLR include improved three-dimensional vision, along with elimination of tremor, and improved dexterity from the Endowrist may theoretically be particularly advantageous for tumours located in the posterosuperior segments^46,47. However, potential limitations of robotic surgery include lack of availability of instruments such as the Cavitron Ultrasonic Surgical Aspirator (CUSA), which is widely used for parenchymal transection in both open and laparoscopic liver surgery. Hence, results from studies^48–52 comparing the outcomes of R-LLR versus L-LLR are mixed, and the superiority of robotic liver resection remains unclear, despite its higher costs. A recent systematic review¹⁴ analysing the learning curves of MILR suggested that a potential advantage of robotic surgery was that it was associated with a shorter learning curve than laparoscopy. Furthermore, in a multicentre study by Nota et al.⁵³ comparing 51 robotic with 145 open minor liver resections in the posterosuperior segments, robotic surgery was reported to be safe and feasible, and associated with a shorter hospital stay. This was in contrast to an earlier study¹⁹ which reported that laparoscopic resection was associated with a higher complication rate than robotic resections.

There are limited data in the literature comparing the outcomes of patients undergoing R-LLR versus L-LLR specifically for tumours in the posterosuperior segments^54,55. In the present study, after matching, R-LLR was associated with a significantly shorter operating time, lower rate of conversion to open surgery, decreased blood loss, and lower blood transfusion rate than L-LLR. There are several possible reasons for these findings. First, the superior perioperative outcomes of R-LLR may be due to the technical advantages of the robotic platform; in particular, the wristed instruments result in superior dexterity and easier access to the posterosuperior segments compared with rigid laparoscopic instruments. This may enable surgeons to quickly control bleeding, particularly from hepatic vein tributaries, by suturing the bleeding site, which is more difficult to perform laparoscopically. Second, a potential major confounding factor is the experience of the surgeon. It is likely that many of the surgeons performing R-LLR would have had previous experience with L-LLR before embarking on R-LLR. Furthermore, it is important to highlight that, because of the high barrier of entry to robotic surgery owing to its limited accessibility, it is likely that mainly experienced surgeons with a keen interest in MILR would have committed the time to perform R-LLR. This contrasts with the situation for laparoscopic surgery: a larger number of surgeons with a broader range of experience would have performed L-LLR because of its easier accessibility. Finally, because of the higher costs associated with R-LLR and increased patient expectations, surgeons may have adopted more stringent selection criteria when performing R-LLR and this may have accounted for the extremely low open conversion rate of only 0.6 per cent.

Two statistical methods were used in this study to improve the robustness of the analyses. PSM is widely used to reduce selection bias by largely balancing the cohorts, but has the main limitation of overlooking unmeasured confounding variables that may have an influence on outcomes. Therefore, CEM was also performed which, unlike PSM, has been shown to yield estimates of causal effects that are lowest in variance and bias for any given sample size⁵⁶. With the use of two matching modalities, the robustness of the analysis was strengthened as the results from both methods were consistent.

The main limitation of this study is its retrospective nature which is inevitably associated with information and selection bias. Although, PSM and CEM were undertaken to mitigate these biases, residual biases were likely to remain which could only be completely eliminated by performing a prospective RCT. Data on certain factors that may have affected outcomes, for example the exact location of tumour in the superficial or deep segment of the liver, were not available and could not be matched. Furthermore, as this was a multicentre study, the participating surgeons used various techniques and instruments to perform R-LLR and L-LLR. There was also likely to be variation in the expertise and experience of the surgeons and centres. Another potential confounding factor was that it is conceivable that surgeons performing R-LLR in general had already been trained in laparoscopic surgery and then moved on to robotic surgery. This could mean that surgeons in the robotic group were more experienced than those in the laparoscopic group. As information on the cumulative experience of each participating centre and individual surgeons in MILR was not known, it was not possible to determine and control for biases resulting from surgeon or centre experience. A major strength of this study was the large number of procedures performed over a relatively short interval, allowing robust statistical analysis. The effect of the pioneering phase and learning curve was minimized by limiting the analysis to between 2010 and 2019. Furthermore, there was no significant difference in the proportion of R-LLR and L-LLR over the study interval, suggesting absence of historical bias.

Collaborators

International robotic and laparoscopic liver resection study group investigators: M. Gastaca (Biocruces Bizkaia Health Research Institute, Cruces University Hospital, University of the Basque Country, Bilbao, Spain), H. Schotte, C. De Meyere (Groeninge Hospital, Kortrijk, Belgium), E. C. Lai (Pamela Youde Nethersole Eastern Hospital, Hong Kong SAR, China), F. Krenzien, M. Schmelzle (Campus Charité Mitte and Campus Virchow-Klinikum, Charité-Universitätsmedizin, Corporate Member of Freie Universität Berlin, and Berlin Institute of Health, Berlin, Germany), P. Kadam (University Hospitals Birmingham NHS Foundation Trust, Birmingham, United Kingdom), M. Giglio, R. Montalti (Federico II University Hospital Naples, Naples, Italy), Q. Liu (The First Medical Center of Chinese People’s Liberation Army (PLA) General Hospital, Beijing, China), K. F. Lee (Prince of Wales Hospital, The Chinese University of Hong Kong, New Territories, Hong Kong SAR, China, D. Salimgereeva, R. Alikhanov (Moscow Clinical Scientific Center, Moscow, Russia), L. S. Lee (Changi General Hospital, Singapore), J. Y. Jang (CHA Bundang Medical Center, CHA University School of Medicine, Seongnam, Korea), C. Lim (Hopital Pitie-Salpetriere, APHP, Sorbonne Université, Paris, France), K. P. Labadie (University of Washington Medical Center. Seattle, USA).

Funding

Dr Kingham was partially supported by the US National Cancer Institute MSKCC Core [Grant number P30 CA008747].

Disclosure

The authors declare no conflict of interest. The following authors have disclosures outside of this work. B.K.P.G. has received travel grants and honoraria from Johnson and Johnson and Transmedic, the local distributor for the Da Vinci^® robot; M.V.M is a consultant for CAVA Robotics. J.P. reports a research grant from Intuitive Surgical, and personal fees or non-financial support from Johnson & Johnson, Medtronic, AFS Medical, Astellas, CHG Meridian, Chiesi, Falk Foundation, La Fource Group, Merck, Neovii, NOGGO, pharma-consult Peterson, and Promedicis. M. Schmelzle reports personal fees or other support outside of the submitted work from Merck, Bayer, ERBE, Amgen, Johnson & Johnson, Takeda, Olympus, Medtronic, and Intuitive. F. R. reports speaker fees and support outside the submitted work from Integra, Medtronic, Olympus, Corza, Sirtex, and Johnson & Johnson.

Supplementary material

Supplementary material is available at BJS online.

References

Author notes