Introduction

Liver resection is acknowledged as the only curative treatment for colorectal liver metastases (CRLM). There has been a steady movement away from mandatory anatomic resection (AR) and toward non-anatomic resection (NAR) in situations in which the latter can achieve an R0 resection. While a recent review of 12 studies examining short- and long-term outcomes of both techniques did not show any survival differences,1 NAR is associated with lower risk of postoperative liver failure and morbidity after surgery.1,2,3,4,5,6 Additionally, the appropriate use of NAR offers equivalent risk of intrahepatic recurrence, but allows a higher chance to perform a re-resection if recurrence occurs, which may lead to longer survival after recurrence.2

While most favor NAR over AR, assuming a margin-negative resection can be achieved, some authors have suggested that a more aggressive resection strategy may be required on the basis of the disease biology. In particular, RAS mutation has been identified as an important risk factor for worse long-term outcomes in patients with metastatic colorectal cancer,7 leading some to question the appropriateness of applying NAR in patients harboring RAS mutation, whether for lung or for liver metastases.8,9 In this context, the primary aim of this study was to compare the outcomes of patients undergoing AR versus NAR for CRLM, stratified by RAS mutation status.

Patients and Methods

Study Population

After obtaining approval from the Institutional Review Board of The University of Texas MD Anderson Cancer Center (protocol #PA19-0242), we retrospectively identified patients with known RAS mutation status who underwent AR or NAR for CRLM between 2006 and 2016 from a prospectively maintained departmental database. Patients who underwent simultaneous ablation at the time of resection, two-stage hepatectomy, R2 resection, or a combined AR and NAR were excluded.

AR was defined as segmentectomy, multi-segmentectomy, sectionectomy, or segmentectomy plus sectorectomy, as described by Couinaud10 and The Brisbane 2000 Terminology of Liver Anatomy and Resections.11 Thus, formal segmentectomy, sectorectomy, right hepatectomy, left hepatectomy, extended right hepatectomy, and extended left hepatectomy were considered an AR. A NAR was defined as the resection of at least one metastasis, which includes a margin of microscopically normal liver tissue, regardless of hepatic anatomy.

Perioperative Management

The decision to perform an AR or NAR was made by the surgeon and determined during preoperative evaluation. Intraoperative liver ultrasound was performed. A resection margin on final pathological examination of < 1 mm was defined as an R1 resection. Follow-up, including physical examination, laboratory testing, and cross-sectional imaging, was conducted every 3–4 months.

Data Collection

The following patient data were recorded: sex, age, American Society of Anesthesiologists (ASA) physical status classification, body mass index, CRLM characteristics (number of tumors, largest tumor diameter, unilobar vs bilobar location, and timing of metastasis), type of hepatectomy, pre-hepatectomy chemotherapy, and perioperative outcomes (blood loss, operative time, resection margin, postoperative complications, length of hospital stay, unplanned readmission 45 days after discharge,12 primary tumor location, and primary nodal status). Postoperative complications were classified according to Clavien-Dindo classification,13 with grade IIIa or higher being considered a major complication.

Recurrence Definitions

Patients were considered to have experienced recurrence when imaging showed an area of concern that was either biopsy proven to be recurrent disease or was treated as such by the treatment team. Intrahepatic recurrence was defined as evidence of metastatic disease within the liver at the time of first recurrence, whether or not synchronous extra-hepatic recurrence was also present. In patients with intrahepatic recurrence, we closely examined the site of recurrence. Local recurrence was defined as an intrahepatic recurrence at the resection margin or within the same segment of the liver, as determined by imaging review.

Propensity Score Matching

Propensity score (PS) matching without replacement in an optimal algorithm was used to adjust for differences in baseline characteristics between AR and NAR patients and was separately applied for RAS wild-type and RAS-mutated patients. The variables that were considered most clinically relevant and directly associated with undergoing AR or NAR were included in a logistic regression model to calculate the PS. These variables were age, sex, ASA score, body mass index, bilateral vs unilateral liver metastases, number of metastases, largest tumor diameter (as a continuous variable), neoadjuvant chemotherapy, extrahepatic disease, and repeated hepatectomy. After PS generation, AR and NAR patients underwent 1:1 nearest available matching of the logit of the PS, with a caliper width of 0.05 of the standard deviation of the score. If both AR and NAR patients did not meet the matching criteria, they were excluded.

Statistical Analysis

Statistical analysis was performed using JMP Pro 14 software (SAS Institute, Inc., Cary, NC). Continuous variables were analyzed using a Wilcoxon signed-rank test and expressed as medians and ranges. Categorical variables were compared using χ2 test or Fisher’s exact test, as appropriate, and expressed as absolute values and percentages. Recurrence-free survival (RFS), liver-specific recurrence-free survival (L-RFS), and overall survival (OS) rates were calculated from the day of liver surgery and estimated using the Kaplan-Meier method. The 1-, 3-, and 5-year RFS, L-RFS, and OS rates were compared using log-rank statistics. All tests were two-sided, and P < 0.05 was considered to be statistically significant.

Results

Patient Characteristics

Between 2006 and 2016, 1474 patients underwent hepatectomy for CRLM at our institution. The RAS mutation status was known in 1025. After exclusion criteria were applied, 622 patients were included in analysis (Fig. 1). Of these, 338 (54%) underwent AR and 284 (46%) underwent NAR. RAS mutation was found in 274 (40%) patients, with a similar distribution of AR and NAR (55% and 45%, respectively). The remaining 348 (60%) patients were RAS wild type, again with a similar distribution of AR and NAR (54% and 46%, respectively). As expected, NAR patients presented with fewer metastases (P = 0.002), a smaller median tumor diameter (P < 0.0001), and were more likely to undergo repeat hepatectomy (P = 0.002). These differences persisted when patients were evaluated according to RAS mutation status, except for repeat hepatectomy, which was similar between AR and NAR in the RAS mutation patients (Supplementary Table 1).

Fig. 1
figure 1

Flowchart of patient selection

Full size image

PS-Matched Patient Characteristics

After 1:1 PS matching, 360 patients (180 AR vs 180 NAR) were included in the analysis, 210 (58%) with RAS wild-type and 150 (42%) with RAS mutation; exactly half in each group underwent AR and NAR. The differences in patient characteristics between AR and NAR in the original cohort were well adjusted (Supplementary Table 1).

Perioperative Outcomes

Patients’ perioperative outcomes are shown in Table 1. In the original cohort, patients with AR had greater intraoperative blood loss (P < 0.001), were more likely to develop major complications (P = 0.001), and had a longer hospital stay (P < 0.001); this was true regardless of RAS status. In the RAS mutation group, patients who underwent NAR were more likely to have an R1 margin (P = 0.036). After PS matching, significant differences in intraoperative blood loss (P = 0.001) and length of hospital stay (P = 0.006) persisted. R1 resection rates were similar in patients who underwent AR and NAR.

Table 1 Perioperative outcomes after anatomic vs non-anatomic resection by RAS mutation status
Full size table

Survival Analysis Before Matching

The median follow-up period in the original cohort was 43.9 months, with 45 months in the AR group and 43 months in the NAR group. Patients who underwent NAR had a significantly longer OS duration than did those who underwent AR for the whole cohort (P = 0.021); the same trend was found for RAS mutation patients (P = 0.052). No difference was found for patients with RAS wild-type (Supplemental Fig. 1a, 1b, 1c). While the RFS rate was similar for AR and NAR in the entire cohort and for RAS mutation patients, it was higher for NAR in patients with RAS wild-type (P = 0.034) (Fig. 2a–c). The L-RFS rates were similar between NAR and AR for all groups (Fig. 3a–c).

Fig. 2
figure 2

Recurrence-free survival in patients after anatomic (AR) vs non-anatomic resection (NAR) according to RAS mutation status, before matching for a the whole cohort, bRAS wild-type, and cRAS mutation and after matching for d the whole cohort, eRAS wild-type, and fRAS mutation

Full size image
Fig. 3
figure 3

Liver-specific recurrence-free survival in patients with anatomic (AR) vs non-anatomic resection (NAR) according to RAS mutation status, before matching for a the whole cohort, bRAS wild-type, and cRAS mutation and after matching for d the whole cohort, eRAS wild-type, and fRAS mutation

Full size image

Survival Analysis After Matching

After PS matching, the median follow-up duration was 43.1 months, with 45.3 months for patients undergoing AR and 40.8 months for patients undergoing NAR. The OS was similar between NAR and AR for the whole cohort, RAS wild-type, and RAS mutation groups (Supplemental Fig. 1d, 1e, 1f). The RFS rate was similar between NAR and AR for all groups, (Fig. 2d–f), as was L-RFS for all groups (Fig. 3d–f).

Local Recurrence

Three hundred forty-two patients (54.9%) in the original cohort developed intrahepatic recurrence, with no difference in the RAS mutation group between AR and NAR (55.0% vs 57.7%, P = 0.647). In the matched cohort, 187 patients (51.9%) developed intrahepatic recurrence, again with no difference in the RAS mutation group between AR and NAR (49.0% vs 54.7%, P = 0.513). There were no differences in local recurrence after AR or NAR in either the RAS wild-type (14.3% vs 7.6%, P = 0.122) or RAS mutation (10.7% vs 5.3%, P = 0.229) groups (Table 2).

Table 2 R1 resection, and intrahepatic and local recurrence after matching
Full size table

Of 622 patients in the overall cohort, 96 (15.4%) had R1 resections. In the matched cohort, 35 patients (9.7%) had R1 resections. There was no difference in R1 resection rate between NAR and AR in the RAS wild-type group (10.5% vs 11.4%, P = 0.825) or the RAS mutation group (6.7% vs 8.0%, P = 0.754) (Table 2). Local recurrence was not more common after R1 resection than R0 resection for the RAS mutation group (9.1% vs 7.9%, P = 0.838).

Discussion

In this study, we demonstrated that both AR and NAR can be performed with similar oncologic outcomes, regardless of RAS mutation status. After PS matching, we found no difference in OS, RFS, or L-RFS rates and durations after AR or NAR in the overall cohort or in patients with wild-type or mutant RAS. Moreover, there was no difference in local recurrence between AR and NAR, before or after matching.

The superiority of NAR or AR as the optimal surgical approach for resection of CRLM has been evaluated in several retrospective studies. DeMatteo et al. showed a benefit from more extensive resection in a retrospective review published in 2000, but this study pre-dated the ubiquitous use of modern perioperative chemotherapy and thus is not likely applicable in the current era.14 While no prospective, randomized trials have compared these two techniques, more recent studies have shown no detriment in oncologic outcomes with NAR, but have shown improved salvageability and survival after recurrence for patients who have undergone NAR for small, solitary metastases.1,2,5,15,16 Meanwhile, multiple studies have shown increased morbidity and mortality with AR,5,17 just as we observed increased blood loss, complications, and length of stay in the current study. Considering the decrease in perioperative risk without sacrificing oncologic outcomes, we advocate for NAR for CRLM, assuming a negative margin can be obtained.

While NAR has become an acceptable approach in most patients with CRLM, some have argued that the decision for AR or NAR should be different for patients harboring RAS mutations, as these mutations are known to be a marker of more aggressive disease.18,19,20 In a recently published study that evaluated the impact of AR vs NAR for colorectal lung metastases,9 patients with RAS mutation who underwent NAR had a significantly higher resection-margin recurrence rate, shorter time to pulmonary recurrence, and lower OS compared to those undergoing AR. This finding in the setting of lung metastases calls into question the appropriateness of NAR for patients with RAS mutant CRLM. Indeed, a recently reported study of AR versus NAR in patients with known RAS mutation showed a significant improvement in disease-free survival and intrahepatic disease-free survival for RAS-mutated patients treated with AR, while there was no difference in RAS wild-type patients.8 The authors hypothesized that RAS mutation CRLM may spread through the portal system, and would thus benefit from en bloc removal of the segmental portal pedicle, mandating an AR. While there is evidence for such spread in hepatocellular carcinoma,21 this has not been shown in CRLM.22

In contrast, our study did not show improved OS or RFS with AR in patients with RAS mutation, but instead shorter OS in the overall cohort, and a trend toward a shorter OS in RAS-mutated patients. This finding is somewhat related to more aggressive disease at baseline in the AR group, but, at minimum, this shows that AR does not significantly improve OS or RFS. In an effort to account for these differences, we performed PS matching, but we still found no benefit from AR in terms of OS or RFS in RAS-mutated patients or any other group. This finding is consistent with previous data demonstrating that, while RAS mutation is a risk factor for more aggressive systemic disease and likely predicts poor response to chemotherapy, this risk is not specific to intrahepatic recurrence and is not prevented by AR.7 Thus, a more aggressive liver resection will not affect overall outcomes such as OS and RFS, and may actually decrease the ability to perform a re-resection when these patients experience recurrence in a remote site within the liver.2

Perhaps most importantly, there was no difference in L-RFS between groups, before or after matching. This contradictory result when compared to previous work may directly reflect our exclusion of patients undergoing intraoperative ablation, which likely accounted for at least some of the intrahepatic recurrences.23 Most importantly, we closely evaluated patients with intrahepatic recurrence and examined the site of recurrence. We found no decrease in local recurrence, which we defined to include any recurrence at the resection margin or within the same segment as the resected lesion, with AR. If RAS-mutated CRLM required an AR to clear the segmental portal pedicle, this outcome would be the most likely to show a difference. While previous studies have shown that 40% of CRLM present with microvascular invasion and 20% with micrometastases, mostly occurring in a 2- to 4-mm area around the tumor,24,25 there is no clear relation to the presence of RAS mutation.22 Moreover, there is no convincing evidence that colorectal liver metastases spread via the portal pedicle, which would require an AR. Therefore, an adequate surgical margin, rather than an AR, is required to fully remove these sites of metastatic disease, even with RAS mutation.

Our study has some important limitations that should be considered. First, we used a retrospective design, evaluating a selected cohort of patients at a single institution, which introduced selection bias. Patients who underwent AR had more aggressive disease with more tumors, larger tumors, and potentially tumors that were closer to major vascular structures, biasing surgeons against NAR. We did attempt to account for this bias using PS matching, but it still must be considered when interpreting our data. Second, the definitions of AR and NAR are not universal. For this study, we specifically addressed the concern of tumor spread via the segmental pedicle; therefore, we included any formal segmentectomy and sectionectomy in the AR group. We used a prospectively maintained database with this data point filled by the faculty surgeon in each case. When ambiguity existed, operative reports were read to determine whether or not an AR was performed. Despite these limitations, this study represents the largest to date addressing this important question of AR vs NAR for RAS-mutated patients.

Conclusion

In conclusion, both AR and NAR can be performed with similar long-term outcomes in patients with CRLM, regardless of RAS mutation status. While RAS mutation may be a risk factor for more aggressive systemic disease, it did not impact outcomes between AR or NAR. These data do not support a universal requirement for AR in RAS mutant CRLM when R0 resection can be achieved with NAR.