SUB-CLASSIFICATION OF LAPAROSCOPIC LEFT HEPATECTOMY BASED ON HIERARCHIC INTERACTION OF TUMOR LOCATION AND SIZE WITH PERIOPERATIVE OUTCOMES

Abstract

Introduction:

The aim of this multicentric study was to investigate the impact of tumor location and size on the difficulty of Laparoscopic-Left Hepatectomy(L-LH).

Methods:

Patients who underwent L-LH performed across 46 centers from 2004 to 2020 were analyzed. Of 1236 L-LH, 770 patients met the study criteria. Baseline clinical and surgical characteristics with a potential impact on LLR were included in a multi-label conditional interference tree. Tumor size cut-off was algorithmically determined.

Results:

Patients were stratified into 3 groups based on tumor location and dimension: 457 in antero-lateral location(Group 1), 144 in postero-superior segment (4a) with tumor size ≤40 mm(Group 2), and 169 in postero-superior segment (4a) with tumor size >40 mm(Group 3). Patients in the Group 3 had higher conversion rate (7.0% vs. 7.6% vs. 13.0%, p-value 0.048), longer operating time (median, 240 min vs. 285 min vs. 286min, p-value <0.001), greater blood loss (median, 150ml vs. 200ml vs. 250 ml, p-value <0.001) and higher intraoperative blood transfusion rate (5.7% vs. 5.6% vs. 11.3%, p-value 0.039). Pringle’s maneuver was also utilized more frequently in Group 3 (66.7%), compared to Group 1 (53.2%) and Group 2 (51.8%) (p=0.006). There were no significant differences in postoperative stay, major morbidity, and mortality between the 3 groups.

Conclusion:

L-LH for tumors that are >40mm in diameter and located in PS Segment 4a are associated with the highest degree of technical difficulty. However, post-operative outcomes were not different from L-LH of smaller tumours located in PS segments, or tumors located in the antero-lateral segments.

INTRODUCTION

Laparoscopic liver surgery has taken off globally in light of its widely demonstrated short-term benefits such as reduced blood loss, lower post-operative morbidity, and shorter hospital stay compared to traditional open resection, with no compromise in oncologic outcomes^1,2. Despite the many advantages, it remains a technically challenging procedure, hence consensus guidelines stress the importance of stepwise progression along the learning curve in order to minimize adverse patient outcomes^3,4. It is thus imperative to categorize the technical difficulty of a laparoscopic liver resection (LLR) preoperatively in order to risk stratify and counsel patients regarding the peri-operative risks, as well as to select cases appropriately for surgeons depending on their individual skillset and expertise level⁵. This is also critical when evaluating the implementation of novel surgical procedures and auditing outcomes to ensure a fair comparison⁶.

The degree of difficulty of LLR may be affected by various factors including patient and tumour characteristics, liver anatomy and type of resection. Tumor size and location are two variables which may significantly impact the technical complexity of LLR. Large liver lesions are more difficult to manipulate laparoscopically with a higher risk of rupture^7,8. Furthermore, if these are situated deep within the liver parenchyma, close to major hepatic vessels and require more extensive resections, resulting in higher blood loss, bile leak and greater post-operative morbidity⁹. Tumour size has been incorporated into widely-validated Difficulty Scoring Systems (DSS) such as the Iwate score⁵ and the Southampton score¹⁰, although there is no clearly defined optimal size cut-off. An important limitation of these previous studies was that the impact of tumor size had not been correlated with the extent of LLR Intuitively, the impact of tumor size on the difficulty of LLR would differ according to the type and extent of LR.

Tumour location as a predictor of surgical difficulty was included in the Ban DSS¹¹, which was later modified to the Iwate criteria based on expert opinion⁵. In both these scoring systems, lesions in superficial, anterolateral segments are allocated fewer difficulty points than lesions in more challenging postero-superior locations. Similarly, in the Hasegawa score, lesions in anterolateral segments 2/3/4 are accorded lower scores (0 points) versus lesions in S5/6 (1 point) and S7/8 (2 points)¹². The recent Tong score which aims to predict the probably of open conversion and post-operative morbidity following LLR also identified tumour location as an important risk factor¹³.

Hence, we performed this study in order to evaluate the impact of tumour size and location on the difficulty of laparoscopic left hepatectomies (L-LH), as well as to identify an optimal tumour size cut-off/s to distinguish between ‘easy’ and ‘difficult’ L-LH.

METHODS

This was an international multicentric retrospective analysis of 17,602 patients who underwent pure LLR at 46 international centers from January 2004 to December 2020. Amongst these cases, 1236 were pure L-LH. We excluded patients who underwent concomitant major surgical procedures (colectomies, gastrectomies, hilar lymphadenectomies, bile duct resections), repeat liver resections as well as surgical procedures for gallbladder cancer, cysts/ cystic tumors or abscesses. Finally, 770 patients were included in this study. All centers obtained their respective approvals according to local requirements. This study was approved by the Singapore General Hospital Institution Review Board (CIRB 2020/2802) and the need for patient consent was waived. The anonymized data was collected in the individual centers, and subsequently collated and analyzed centrally at the Singapore General Hospital.

Definitions

Left hepatectomy was defined according to the 2000 Brisbane classification¹⁴ as the anatomical resection of segments 2, 3 and 4. Anterolateral segments were defined as Couinaud segments 2, 3, 4b whereas segments 4a was defined as posterosuperior segments^5,15,16. For cases with multiple tumours, the diameter of the largest lesion was recorded. The difficulty of resections was graded according to the Iwate score⁵. Post-operative complications were classified according to the Clavien-Dindo classification, and recorded for up to 30 days or during the same hospitalization¹⁷.

Statistical analysis

Continuous and categorical variables were summarized using medians (interquartile range, IQR) and proportions, respectively. Tests of inequality across different categories were performed using Kruskal-Wallis and Fisher’s exact tests for continuous and categorical variables.

The objective of this study was to examine the interactions between predictors of laparoscopic difficulty and to propose a simple sub-classification system to discriminate between ‘easy’ and ‘difficult’ left hepatectomy. Preoperative factors (shown in Table 1) with potential impact on laparoscopic difficulty were included for consideration in a multivariate conditional interference tree, as shown in Fig. 1[12]. For continuous predictors such as tumor size, splits were implemented about a cutoff which was algorithmically determined as described below. We considered intraoperative outcomes to be the best surrogate for laparoscopic difficulty; hence, the response variables used in this multivariate conditional inference tree included rates of open conversion, operative time, estimated blood loss, application of Pringle’s maneuver, and requirement for intraoperative blood transfusion. Postoperative outcomes were also analyzed although it must be acknowledged that it is poorer surrogate of difficulty as postoperative outcomes may be confounded by other important factors such as patients’ age/ comorbidities and the extent of liver resection.

Table 1:

Comparison of baseline clinical and surgical characteristics of patients who underwent L-LH, stratified by the three categories of left hepatectomy.

	All N = 770	Group 1 (Anterolateral location) N = 457	Group 2 (Posterosuperior location & tumor size ≤ 40mm) N = 144	Group 3 (Posterosuperior location & tumor size > 40mm) N = 169	P-value (Inequality between groups)
Median age (IQR), yrs	64 (53–71)	63 (53–71)	66 (58–73)	63 (50–72)	0.032
Male sex, n (%)	474/769 (61.6%)	281/456 (61.6%)	99/144 (68.8%)	94/169 (55.6%)	0.058
Year of surgery  2004–2012  2013–2021	102 (13.2%) 668 (87.8%)	61 (13.3%) 396 (86.7%)	15 (10.4%) 129 (89.6%)	26 (15.4%) 143 (84.6%)	0.437
Previous abdominal surgery, n/total (%)	258/762 (33.9%)	152/450 (33.8%)	50/144 (34.7%)	56/168 (33.3%)	0.965
Concomitant minor surgery, n (%)	53 (6.9%)	25 (5.5%)	17 (11.8%)	11 (6.5%)	0.039
Multiple liver resections, n (%)	70 (9.1%)	15 (3.3%)	31 (21.5%)	24 (14.2%)	<0.001
ASA score, n/total (%)  1  2  3  4	113 (14.7%) 462 (60.0%) 192 (24.9%) 3 (0.4%)	75 (16.4%) 274 (60.0^) 106 (23.2%) 2 (0.4%)	19 (13.2%) 82 (56.9%) 42 (29.2%) 1 (0.7%)	19 (11.2%) 106 (62.7%) 44 (26.0%) 0 (0.0%)	0.423
Malignant neoplasm, n (%)	679 (88.2%)	403 (88.2%)	136 (94.4%)	140 (82.8%)	0.005
Cirrhosis, n/total (%)	196 (25.5%)	121 (26.5%)	35 (24.3%)	40 (23.7%)	0.747
Portal hypertension, n/total (%)	37/766 (4.8%)	23/455 (5.1%)	5/144 (3.5%)	9/167 (5.4%)	0.732
Median tumor size, mm (IQR)	43 (28–70)	40 (29–65)	25 (18–33)	70 (50–100)	<0.001
Multiple tumors, n (%)	199 (25.8%)	88 (19.3%)	58 (40.3%)	53 (31.4%)	<0.001
Posterosuperior segments (IVa), n (%)	313 (40.7%)	0 (0.0%)	144 (100.0%)	169 (100.0%)	<0.001
Median Iwate difficulty score, (IQR)	9 (8–9)	8 (7–9)	9 (8–10)	10 (9–10)	<0.001
Median Iwate difficulty score excluding tumor size, (IQR)	8 (7–9)	7 (7–8)	9 (8–9)	9 (8–9)	<0.001
Iwate difficulty, n (%)  Low  Intermediate  High  Expert	0 (0.0%) 52 (6.8%) 529 (67.7%) 189 (24.5%)	0 (0.0%) 52 (11.4%) 376 (82.3%) 29 (6.3%)	0 (0.0%) 0 (0.0%) 99 (68.8%) 45 (31.2%)	0 (0.0%) 0 (0.0%) 54 (32.0%) 115 (68.0%)	<0.001

^†Kruskal-Wallis test for continuous variables and Fisher’s exact test for categorical variables.

Figure 1.

Multi-label Conditional Inference Tree Depicting Hierarchic Interaction of Tumor Size and Location in Determining Surgical Outcomes

We required that each terminal node had a minimum of n=100 patients to avoid generating an unnecessarily complex classification system with little clinical utility. A stringent multiplicity-adjusted Monte-Carlo P-value < 0.05 threshold was used to reduce the risk of overfitting and to avoid the need for pruning, and recursive partitioning analysis was performed through maximization of 1 – multiplicity-adjusted P-values. This approach ensures that the right-sized tree is grown and no form of internal validation (eg, cross-validation) or pruning is required. We illustrated the effect of tumor size on perioperative outcomes in the overall cohort by iteratively dichotomizing the tumor size at each integer value, and computing effect sizes using the ‘cutoff-finder’ package¹⁸ Additionally, area under the curve (AUC)/F1 score and root mean squared error (RMSE) were computed for binary and continuous outcomes respectively. Coefficient of variation of RMSE (CVRMSE)% was calculated using the RMSE/mean value x 100%.

Statistical analyses were completed with R v4.0.2, and nominal p < 0.05 was considered to indicate statistical significance.

RESULTS

Baseline clinical and surgical characteristics of patients who underwent L-LH

A total of 770 patients who underwent pure L-LH and met the study criteria were included [Table 1]. Median age was 64 years (IQR 53–71), 61.6% of them were male and 33.9% of them had previous abdominal surgery.

The multi-label interference tree models for tumor dimension are illustrated in Figure 1. According to this model, L-LH was first sub-divided by tumor location into L-LH for tumors located in antero-lateral, “easy” segments, and L-LH for tumors located in postero-superior (PS) segments. Subsequently, a cut-off value of tumor diameter 40 mm was found to increase difficulty, applicable only for postero-superiorly located lesions. Tumor size had limited impact on perioperative outcomes for L-LH of antero-laterally located tumours [Figure 2].

Figure 2:

Cutoff analysis for anterolateral tumors. The impact of tumor size cutoffs on surgical outcomes was systematically investigated by iteratively dichotomizing the tumor size at each 10mm-interval and computing treatment effect sizes local to that cutoff.

Based on the resulting model, the study population was divided into 3 groups of increasing difficulty: Group 1: tumor located in anterolateral segments (2–3-4b) of any size (457 patients, 59%), Group 2: tumor located in postero-superior segment (4a) with tumor size ≤40 mm (144 patients, 19%), Group 3: tumor located in postero-superior segment (4a) with tumor size >40 mm (169 patients, 22%). The baseline characteristics of each group are summarized in Table 1. There were differences in the proportion of patients with concomitant minor surgery (Group 1: 5.5% (n=25); Group 2: 11.8% (n=17); Group 3: 6.5% (n=11), p-value 0.039), multiple liver resections (Group 1: 3.3% (n=15); Group 2: 21.5% (n=31); Group 3: 14.2% (n=24), p-value <0.001), malignant neoplasms (Group 1: 88.2% (n=403); Group 2: 94.4% (n=136); Group 3: 82.8% (n=140), p-value 0.005) and multiple tumors (Group 1: 19.3% (n=88), Group 2: 40.3% (n=58); Group 3: 31.4% (n=53), p-value <0.001) among the groups. Apart from this, the groups had similar proportions of patients with underlying liver cirrhosis and portal hypertension.

Median Iwate difficulty score was 8 (IQR 7–9) for Group 1, 9 (IQR 8–10) for Group 2 and 10 (IQR 9–10) for Group 3 (p-value <0.001). After excluding the variable tumor size, median Iwate difficulty score was 7 (IQR 8–9) for Group 1, 9 (IQR 8–9) for Group 2 and 9 (IQR 8–9) for Group 3 (p-value <0.001). In all three groups, there were no resections which were classified as low difficulty, while groups 2 and 3 did not have any resections which were of intermediate difficulty. In group 1, 52 (11.4%) L-LH were classified as intermediate, 376 (82.3%) as high and 29 (6.3%) as expert (p-value <0.001) in difficulty. In group 2, 99 (68.8%) resections were classified as high and 45 (31.2%) as expert level (p-value <0.001). In group 3, 54 (32%) L-LH were high level and 115 (68%) were expert level (p-value <0.001).

The AUC/F1 score, RMSE, CV RMSE of the new model and the Iwate model are summarized in Supplementary 1 and 2. These were similar for both models.

Perioperative outcomes of patients who underwent L-LH

Perioperative outcomes of patients who underwent L-LH are detailed in Table 2. Among the 65 patients (8.4%) who underwent open conversion, 32 patients (7.0%) were in Group 1, 11 (7.6%) in Group 2 and 22 (13.0%) in Group 3 (p-value 0.048). Median operating time (OT) was longer for patients in Group 3 (286 min, IQR 215–355), compared to Group 2 (285 min, IQR 218–379) and Group 1 (240 min, IQR 180–310) (p-value <0.001). Median blood loss was also greater in Group 3 (250 ml, IQR 100–450), compared to Group 2 (200 ml, IQR 100–350) and Group 1 (150 ml, IQR 50–300) (p-value <0.001). Intraoperative blood transfusion rate was 5.7% (n=26) in Group 1, 5.6% (n=8) in Group 2 and 11.3% (n=19) in Group 3 (p-value 0.039). Finally, Pringle’s maneuver was more utilized more frequently in Group 3 (66.7%), compared to Group 1 (53.2%) and Group 2 (51.8%) (p=0.006).

Table 2:

Comparison of perioperative outcomes of patients who underwent L-LH, stratified by the three categories of laparoscopic left hepatectomy.

	All N = 770	Group 1 (Anterolateral location) N = 457	Group 2 (Posterosuperior location & tumor size ≤ 40mm) N = 144	Group 3 (Posterosuperior location & tumor size > 40mm) N = 169	P-value (inequality between groups) †
Open conversion, n (%)	65 (8.4%)	32 (7.0%)	11 (7.6%)	22 (13.0%)	0.048
Median operating time (IQR), min	260 (195–330)	240 (180–310)	285 (218–379)	286 (215–355)	<0.001
Median blood loss (IQR), ml	200 (100–350)	150 (50–300)	200 (100–350)	250 (100–450)	<0.001
Intraoperative blood transfusion, n (%)	53 (6.9%)	26/457 (5.7%)	8/143 (5.6%)	19/168 (11.3%)	0.039
Pringle maneuver applied, n (%)	423 (55.9%)	240/451 (53.2%)	73/141 (51.8%)	110/165 (66.7%)	0.006
Median postoperative stay (IQR), days	6 (4–8)	6 (5–7)	6 (4–8)	6 (4–8)	0.987
Overall morbidity, n (%)	140/769 (18.2%)	82/457 (17.9%)	23/143 (23.1%)	25/169 (14.8%)	0.165
Major morbidity (Clavien-Dindo grade>2)	43/769 (5.6%)	25/457 (5.5%)	12/143 (8.4%)	6/169 (3.6%)	0.189
30-day readmission, n (%)	26 (3.4%)	15/453 (3.3%)	5/142 (3.5%)	6/168 (3.6%)	0.962
30-day mortality, n (%)	6 (0.8%)	5 (1.1%)	0 (0.0%)	1 (0.6%)	0.636
90-day mortality, n (%)	8 (1.0%)	4 (0.9%)	1 (0.7%)	3 (1.8%)	0.602
Close margins, n (%)	100/766 (13.1%)	47/454 (13.4%)	29/143 (20.3%)	24/169 (14.2%)	0.009

^†Kruskal-Wallis test for continuous variables and Fisher’s exact test for categorical variables.

Postoperative outcomes were as follows: overall median postoperative stay was 6 days (IQR 4–8), overall morbidity rate was 18.2% (140/679), major morbidity (Clavien Dindo grade >2) rate was 5.6% (43/769), 30-day readmission rate was 3.4%, 30-day mortality rate was 0.8% and 90 day-mortality was 1.0%. These outcomes were comparable across all three groups. Of note, the frequency of close margins was significantly higher for resection of postero-superiorly located segments (13.4% (n=47) for Group 1, 20.3% (n=29) for Group 2 and 14.2% (n=24) for Group 3, p-value <0.009).

DISCUSSION

With the increasing adoption of LLR worldwide even by liver surgeons in low-medium volume centers, there is a need to develop DSS which are able to precisely estimate surgical risks associated with different laparoscopic resections¹⁹. Several scoring systems for LLR have been created to date^{5,10–13,16,20}, however there is no consensus as to which is the most effective⁶. Studies which sought to validate and compare the performance of different DSS against each other found varying levels of success in predicting specific intra- and post-operative outcomes by each DSS, as well as differing degrees of correlation between the scoring systems^21–24. Cipriani et al recently described a novel risk score for the prediction of conversion for pure laparoscopic right hepatectomy²⁵, however there are no existing scoring systems specific to L-LH. L-LH is often the first major LLR attempted by surgeons on their learning curve, as it is perceived to be ‘easier’ due to the relative ease of access of the left lobe of the liver compared to the right side, as well as the less extensive mobilization required. In the IMM score, left hepatectomy and anterolateral segmentectomy are accorded 2 difficulty points, versus 3 points for right hepatectomy, right posterior sectionectomy, central hepatectomy, posterosuperior segmentectomy and extended left/right hepatectomy¹⁶. However, not all L-LHs are of the same difficulty level, hence it would be useful to have a DSS which is able to accurately distinguish the degree of complexity pre-operatively for this procedure. We therefore performed this study in order to create a simple, easily reproducible system for the sub-classification of difficulty of L-LH based on the hierarchic interactions of tumour location and size.

In this study, we first used a multi-label conditional interference tree as described by Hothorn et al²⁶ to stratify patients into a machine-learning derived model of three groups accordingly to perioperative outcomes: L-LH for tumors in anterolateral location (S2–3,4b), L-LH for tumors in PS location (S4a) with tumor size ≤ 40mm and L-LH in PS with tumor size > 40 mm. Notably, this model identified that tumor location had the greatest influence on difficulty of L-LH, while tumor size mattered only if the lesion was located in PS segments, with a cut-off value of 40 mm. Patients with tumours > 40mm located in PS segments had higher rates of conversion, longer operation times, greater blood loss, higher rates of blood transfusion and Pringle maneuver utilization, as well as greater frequency of close margins. However, postoperative outcomes were largely similar across the three groups.

In L-LH, only segment 4a is considered to be a PS segment, and lesions located here are allocated 4 difficulty points according to the Iwate score, versus 1, 2 and 3 points for lesions in anterolateral segments 3, 2 and 4b respectively⁵. Similarly, the Tong score accords 2 points to lesions located in PS segments 1, 4a, 7 and 8¹³. Interestingly, in the Hasegawa score, lesions in 2 3 4 are all considered ‘easy’ and given 0 points only¹². Laparoscopic resection of liver tumours located in PS segments is widely considered to be technically challenging due to the limited access when adopting the caudal-to-cranial approach, as well as the proximity to major tributaries of hepatic veins and the inferior vena cava²⁷.

Increasing tumor size increases the difficulty of laparoscopic resection, for reasons mentioned earlier. Several groups have reported that LLR for large tumours may be associated with higher conversion rates, longer operative time, higher blood loss and transfusion rates, greater morbidity and longer hospital stays^7,28–30. There is, however, no consensus on the optimal size cut-off for safe LLR. The Iwate score⁵ uses 3cm whereas the Southampton score¹⁰ uses 5cm. A recent study sought to further refine the Iwate score by using a trichotomy (<30mm, 30–69mm, ≥70mm) to provide additional granularity³¹, however the authors did not consider how tumour size may have differing effects on varying extents of liver resections (i.e., tumor size cut-off 3 cm may impact a mono-segmentectomy differently from a major hepatectomy). Similarly, Ivanecz et al reported that a new threshold for tumour size index of 38mm may further improve the ability of the Iwate score to predict intra-operative and post-operative complications³². Notably, these studies included all type of LLR including major and minor resections. Hence, a major limitation of these studies was that the proposed size cut-offs was not tailored to the extent of the LLR performed. Moreover, many of the cut-off values were arbitrarily chosen and not based on robust statistical analysis. Intuitively, one would expect that a tumor size >3 cm would have a greater impact on the complexity of LLR in the case of a monosegmentectomy but to a lesser extent in the case of right hepatectomy.³³ Hence, the present study focused purely on patients who underwent L-LH only, and precise statistical methods were employed to establish the cut-off of 40mm

In the sub-classification system proposed here, the three patient groups correspond well to the Iwate score, which is the most widely validated score to date⁶. Although the patients in Group 2 appeared to be more complex at baseline (greatest proportion of concomitant minor surgery, multiple liver resection and malignant neoplasms), the median Iwate score of this Group was 9 (Iwate ‘advanced’ level) versus 10 (Iwate ‘expert’ level) for Group 3. Accordingly, patients in Group 3 had the worst peri-operative outcomes. This reinforces the concept that in L-LH, the complex interplay between tumour size and location must be considered in combination when assessing the difficulty of surgery. Large tumours in PS S4a may be situated close to the root of the middle and/or left hepatic veins, and require more challenging, extensive resections with greater blood loss. A greater parenchymal transection surface area may also be required for these lesions, which has been shown to correlate with intraoperative difficulty and post-operative complications³⁴. However, post-operative morbidity and length of stay were similar across the three groups. This may be partly attributed to the fact that all participating centers in this study were high-volume centers where patients were managed according to standardized protocols following surgery, resulting in fairly uniform post-operative outcomes. Ruzzenente et al analyzed four different DSS using a machine learning algorithm, and reported that while current scoring systems may predict the risk of intra-operative complications, they are limited in their ability to predict post-operative morbidity³⁵. Nevertheless, the novel classification system we propose here is the first to sub-classify the difficulty of L-LH using only two variables which can be readily obtained from preoperative imaging. Using this tool, surgeons may be able to discern which cases would be more technically challenging, and select the most suitable surgical approach as well as the most experienced teams for these. Patients who are counseled appropriately may also adjust their expectations accordingly, and would be more accepting of adverse peri-operative events should they occur. The current novel simple 2-variable hierarchic model demonstrated a similar performance to the more complicated multi-parameter Iwate classification (Supplementary 1 and 2).

There are several strengths to this study. One common critique of existing DSS is that they were formulated using small patient cohorts from either Eastern or Western centers, where the indications for surgery (hepatocellular carcinoma in the East versus colorectal liver metastasis in the West) as well as patient profiles differed. To the best of our knowledge, this study analyzed the largest cohort of purely laparoscopic left hepatectomies from multiple specialized units across the globe, hence there was adequate representation of different groups of patients. Traditional DSS were created based on the probability of predicting the occurrence of one or two surrogates of surgical difficulty only (such as open conversion, operation time, blood loss and post-operative complications), and may not be able to discriminate for the development of other endpoints. The classification system we have proposed here has been demonstrated to correlate with 5 intra-operative outcomes, as well as one additional post-operative outcome. Although this classification system analyzed the impact of technical factors (tumour location and size) only, it is worth noting that 33.9% of patients in this series had previous abdominal surgery, 88.2% of resections were for malignant neoplasms, and 25.5% of patients had underlying cirrhosis, all of which are regarded as patient and organ-specific variables which increase surgical difficulty^5,10. For these reasons, we theorize that the classification system proposed here may be widely applicable to most patient populations.

This study has several limitations. Firstly, its retrospective nature renders it susceptible to bias, although most of the centers collected data in a prospectively maintained database. Secondly, while focusing on a single type of LLR allowed us to obtain a precise value for cut-off size, this model may be applied for L-LH cases only. Thirdly, although the system correlated significantly with intraoperative outcomes which are important surrogates of surgical difficulty, the postoperative outcomes did not differ. However, this observation is not surprising as postoperative outcomes of LLR such as morbidity are heavily influenced by other confounding factors such as patient age and fitness. Fourthly, being a multi-center study, it is important to consider differences in caseloads of the participating centers and experience levels of the individual surgeons involved. Finally, another limitation is that the proposed risk stratification was not externally validated as the sample size was limited and we required a minimum of 100 patients in each terminal node, however, an internal 10-fold cross-validation assessment was performed. Nonetheless, it could also be argued that this ‘limitation’ is irrelevant to the aim of this study, as the aim of this study was never to propose a new model for wider application or to replace existing difficulty scoring systems (eg, Iwate system), but to merely demonstrate that tumor location interacts with tumor size to differentially affect post-operative outcomes (ie, to simply demonstrate that the effect of tumor size on laparoscopic difficulty varies depending on the location of the tumor). We found that tumor size only affects outcomes when the tumor is located in posterosuperior segments, but not when the tumor is located in an anterolateral segment with respect to left hepatectomies. Hence, the present findings suggest that existing difficulty models such as the Iwate score can be further optimized. Despite these shortcomings, this is the first international multi-center study to assess the influence of tumor size on difficulty of L-LH, using a novel machine learning technique to establish an evidence-based cut-off value. Stringent inclusion criteria was used to ensure that patients were fairly comparable across the groups at baseline.

Conclusion

This large, multi-center cohort study demonstrated that L-LH for tumours that are >40mm in diameter and located in PS Segment 4a is associated with the highest degree of technical difficulty, and resulted in inferior intraoperative outcomes. However, post-operative outcomes were not different from L-LH of smaller tumours located in PS segments, or tumour located in anterolateral segments. This simple 2-variable hierarchic model was as robust as the more complicated Iwate classification Further, prospective studies are required to corroborate these findings.

LIST OF ABBREVIATION:

L-LH

laparoscopic left hepatectomy

LLR

laparoscopic liver resection

LLS

laparoscopic liver surgery

IMM

Institute Mutualiste Montsouris

DSS

difficulty scoring systems

PS

postero-superior

International robotic and laparoscopic liver resection study group investigators

1. Mikel Gastaca (Hepatobiliary Surgery and Liver Transplantation Unit, Biocruces Bizkaia Health Research Institute, Cruces University Hospital, University of the Basque Country, Bilbao, Spain)

2. Henri Schotte (Department of Digestive and Hepatobiliary/Pancreatic Surgery, Groeninge Hospital, Kortrijk, Belgium)

3 Celine De Meyere (Department of Digestive and Hepatobiliary/Pancreatic Surgery, Groninge Hospital, Kortrijk, Belgium)

4. Felix Krenzien (Department of Surgery, Campus Charité Mitte and Campus Virchow-Klinikum, Charité-Universitätsmedizin, Corporate Member of Freie Universität Berlin, and Berlin Institute of Health, Berlin, Germany)

5. Moritz Schmelzle (Department of Surgery, Campus Charité Mitte and Campus Virchow-Klinikum, Charité-Universitätsmedizin, Corporate Member of Freie Universität Berlin, and Berlin Institute of Health, Berlin, Germany)

6. Kit-Fai Lee (Division of Hepatobiliary and Pancreatic Surgery, Department of Surgery, Prince of Wales Hospital, The Chinese University of Hong Kong, New Territories, Hong Kong SAR, China

7. Diana Salimgereeva (Department of Hepato-Pancreato-Biliary Surgery, Moscow Clinical Scientific Center, Moscow, Russia)

8. Ruslan Alikhanov (Department of Hepato-Pancreato-Biliary Surgery, Moscow Clinical Scientific Center, Moscow, Russia)

9. Lip-Seng Lee (Hepatopancreatobiliary Unit, Department of Surgery, Changi General Hospital, Singapore)

10. Jae Young Jang (Department of General Surgery, CHA Bundang Medical Center, CHA University School of Medicine, Seongnam, Korea)

11. Yutaro Kato (Department of Surgery, Fujita Health University School of Medicine, Aichi, Japan)

12. Masayuki Kojima (Department of Surgery, Fujita Health University School of Medicine, Aichi, Japan)

13. Asmund Avdem Fretland (Interventional Centre and Department of HPB Surgery, Oslo University Hospital, Oslo, Norway)

14. Jacob Ghotbi (Interventional Centre and Department of HPB Surgery, Oslo University Hospital, Oslo, Norway)

15. Fabricio Ferreira Coelho (Liver Surgery Unit, Department of Gastroenterology, University of Sao Paulo School of Medicine, Sao Paulo, Brazil)

16. Jaime Arthur Pirola Kruger (Liver Surgery Unit, Department of Gastroenterology, University of Sao Paulo School of Medicine, Sao Paulo, Brazil)

17. Victor Lopez-Lopez (Department of Surgery, Virgen de la Arrixaca University Hospital, Murcia, Spain)

18. Paolo Magistri (HPB Surgery and Liver Transplant Unit, University of Modena and Reggio Emilia, Modena, Italy)

19. Margarida Casellas I Robert (Hepatobiliary and Pancreatic Surgery Unit, Department of Surgery, Dr. Josep Trueta Hospital, IdIBGi, Girona, Spain)

20. Roberto Montalti (Department of Clinical Medicine and Surgery, Division of HPB, Minimally Invasive and Robotic Surgery, Federico II University Hospital Naples, Naples, Italy)

21. Mariano Giglio (Department of Clinical Medicine and Surgery, Division of HPB, Minimally Invasive and Robotic Surgery, Federico II University Hospital Naples, Naples, Italy)

22. Alessandro Mazzotta Department of Digestive, Oncologic and Metabolic Surgery, Institute Mutualiste Montsouris, Universite Paris Descartes, Paris, France

23. Boram Lee (Department of Surgery, Seoul National University Bundang Hospital, Seoul National University College of Medicine, Seoul, Korea)

24. Mizelle D’Silva, Department of Surgery, Seoul National University Bundang Hospital, Seoul National University College of Medicine, Seoul, Korea

25. Hao-Ping Wang Department of Surgery, Chang Gung Memorial Hospital, Kaohsiung

26. Mansour Saleh (Department of Hepatobiliary Surgery, Assistance Publique Hopitaux de Paris, Centre Hepato-Biliaire, Paul-Brousse Hospital, Villejuif, France)

27. Franco Pascual (Department of Hepatobiliary Surgery, Assistance Publique Hopitaux de Paris, Centre Hepato-Biliaire, Paul-Brousse Hospital, Villejuif, France)

28. Phan Phuoc Nghia (Department of Surgery, University Medical Center, Ho Chi Minh City, Vietnam)

29. Chetana Lim, (Department of Digestive, HBP and Liver Transplantation, Hopital Pitie-Salpetriere, Sourbonne Universite, Paris, France)

30. Qiu Liu (Faculty of Hepatopancreatobiliary Surgery, The First Medical Center of Chinese People’s Liberation Army (PLA) General Hospital, Beijing, China)

31. Prashant Kadam (Department of Hepatopancreatobiliary and Liver Transplant Surgery, University Hospitals Birmingham NHS Foundation Trust, Birmingham, United Kingdom)

32. Chung-Ngai Tang (Department of Surgery, Pamela Youde Nethersole Eastern Hospital, Hong Kong SAR, China)

33. Zewei Chen (Department of Hepatobiliary Surgery, Affiliated Jinhua Hospital, Zhejiang University School of Medicine, Jinhua, China)

34. Shian Yu (Department of Hepatobiliary Surgery, Affiliated Jinhua Hospital, Zhejiang University School of Medicine, Jinhua, China)

35. Ugo Giustizieri (HPB Surgery, Hepatology and Liver Transplantation, Fondazione IRCCS Istituto Nazionale Tumori di Milano, Milan)

36. Davide Citterio (HPB Surgery, Hepatology and Liver Transplantation, Fondazione IRCCS Istituto Nazionale Tumori di Milano, Milan)

37. Francesco Ardito (Hepatobiliary Surgery Unit, Fondazione Policlinico Universitario A. Gemelli, IRCCS, Catholic University of the Sacred Heart, Rome, Italy)

38. Simone Vani (Hepatobiliary Surgery Unit, Fondazione Policlinico Universitario A. Gemelli, IRCCS, Catholic University of the Sacred Heart, Rome, Italy)

39. Tiing Foong Siow (Division of General Surgery, Far Eastern Memorial Hospital, Taipei)

40. Federico Mocchegianni (HPB Surgery and Transplantation Unit, United Hospital of Ancona, Department of Experimental and Clinical Medicine Polytechnic University of Marche)

41. Giuseppe Maria Ettorre (Division of General Surgery and Liver Transplantation, S. Camillo Hospital, Rome, Italy)

42. Marco Colasanti (Division of General Surgery and Liver Transplantation, S. Camillo Hospital, Rome, Italy)