One of the few variables that can be controlled by the surgeon and may influence prognosis in surgically treated colorectal liver metastases (CRLM) is the choice of surgical technique (anatomic vs. nonanatomic hepatectomy).1 Several studies have attempted to identify the optimal surgical technique to treat CRLM, but the results have varied.2,3,4 One possible explanation for the varied results may be the effect of anatomic versus nonanatomic hepatectomy relative to patient characteristics. For example, Margonis et al. reported that tumor biology, defined by specific biomarkers such as RAS mutation status, may impact outcomes following surgery relative to surgical technique.5 Specifically, patients with RAS-mutated CRLM benefited the most from an anatomic hepatectomy, which was contrary to the dogma of pursuing more limited resection for all CRLM tumors. In a separate study based on data from the University of Texas M.D. Anderson Cancer Center, no association between anatomic resections (AR) and survival was noted among patients with RAS-mutated or wild-type CRLM; in contrast, a multi-institutional study from Japan reported an association between AR and survival only among patients with RAS wild-type CRLM.6,7

A common limitation of previous studies that may explain the disparate findings is the limited sample size and statistical power. Therefore, a meta-analysis may help overcome this limitation by synthesizing the results of each study and increasing the statistical power. The aim of the current study is to perform a systematic review and meta-analysis of observational, cohort studies to determine whether tumor biology (i.e., RAS mutation status) can be used to select the surgical technique for patients with CRLM. In turn, such information may help shift the landscape from “one size fits all” to a more precise surgery based on tumor biology. Furthermore, we investigated whether a lower utilization of R1 resections may be the mechanism through which surgical technique benefits patients with a specific tumor biology.

Methods

Literature Search

A systematic literature search of the Medline, Embase, CENTRAL, CINAHL, Web of Science, and ClinicalTrials.gov databases was undertaken using the search terms “colorectal neoplasms,” “liver,” “hepatic,” “metastases,” “metastatic,” “resection,” “surgery,” “hepatectomy,” “kras,” and “nras,” combined with the Boolean operators AND/OR as appropriate for each database. Inclusion criteria were: (1) studies reporting on patients with known RAS status, (2) studies reporting on the type of resection performed (anatomic or nonanatomic), and (3) studies reporting on resection margin status.

After removing duplicate studies, the generated title and abstract list were independently screened by two authors with training in clinical investigation (D.P. and N.P.). The predetermined exclusion criteria were: (1) case reports and nonclinical studies, (2) studies not reporting patients’ RAS status, (3) studies published in a non-English language, (4) studies with overlapping patient populations, (5) clinical studies not reporting patients’ long-term survival outcomes, and (6) studies with patients managed primarily with modalities other than liver resection. Potentially eligible studies were identified and reviewed in full text, with a third author (D.D.) acting as a referee in cases of disagreement. Reference lists were also manually checked for other potentially relevant studies.

The present systematic review and meta-analysis were conducted according to Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) and Meta-Analysis of Observational Studies in Epidemiology (MOOSE) reporting guidelines (Supplementary file “MOOSE checklist”) and were registered in the “International Prospective Register of Systematic Reviews” (PROSPERO ID: CRD42022375583).8,9

Data Extraction and Outcomes of Interest

Data extraction was performed by two authors (D.P. and A.P.) to ensure data accuracy and completeness. The primary outcomes of interest were the reported hazard ratios regarding overall disease-free survival (DFS) and liver-specific disease-free survival (liver-specific DFS), stratified according to patient RAS status, type of liver resection, and resection margin status. Disease-free survival and liver-specific disease-free survival were chosen as the primary outcomes of interest because overall survival alone can obscure a causal relationship between a specific surgical technique and the prevention of recurrence. For example, even if a specific surgical technique prevents liver recurrence in patients with a distinct biology, overall survival of these patients may be similar to patients who undergo surgery without that specific technique and, consequently, have recurrence but benefit from a second hepatectomy. Furthermore, liver-specific relapse-free survival (RFS) was the primary endpoint in the “Anatomical Resection of Liver Metastases in Patients with RAS-Mutated Colorectal Cancer (ARMANI)” randomized trial (https://www.clinicaltrials.gov/study/NCT04678583). The ARMANI trial is the first of its kind, testing the hypothesis that anatomical resections exclusively benefit patients with RAS-mutated colorectal liver metastases. Overall survival (OS) was chosen as a secondary variable of interest.

Other data of interest were patient demographics, tumor-related characteristics (size of the largest tumor, number of tumors, distribution, and timing of disease), year of publication, and country of origin. Extracted data were entered into standardized Excel spreadsheets (Microsoft, Redmond, WA) for further tabulation.

Definitions

Patients with RAS mutations (mutRAS), irrespective of the type of mutation, were evaluated separately from their wild-type counterparts (wtRAS). Anatomic resections were defined as removal of at least one Couinaud segment containing the tumor, along with its respective portal triad. Nonanatomic resections were defined as parenchymal resections with no regard to segmental or sectional anatomy. Margin status was defined as R0 resection in cases with ≥ 1 mm distance between the tumor and specimen margin and as R1 resection in cases of < 1 mm margin.

Risk of Bias Assessment

The risk of bias was evaluated independently by two authors (D.P. and G.M.) using the ROBINS-I tool, which assesses studies across seven domains: bias due to confounding, bias due to selection of participants, bias in classification of interventions, bias due to deviations of intended interventions, bias due to missing data, bias in measurement of outcomes, and bias in selection of the reported results. For each domain, the risk of bias can be low, moderate, or serious.

Statistical Analyses

Statistical analyses were performed using Stata v. 17 (StataCorp. 2021. Stata Statistical Software: Release 17. College Station, TX: StataCorp LLC). Hazard ratios (HR) and the corresponding 95% confidence intervals (CI) were calculated for all survival-related outcomes (DFS, liver-specific DFS, and OS), while risk ratios (RR) were utilized to compute the risk for margin positivity. Hazard ratios were extracted from the published Kaplan–Meier (KM) survival curves using the “WebPlotDigitizer” software (https://automeris.io/WebPlotDigitizer) and the method described by Guyot et al.10 A random effects model (DerSimonianLaird) was selected due to anticipated high clinical heterogeneity in terms of patient and tumor baseline parameters. Statistical heterogeneity was quantified with the Higgin’s I2 statistic; values below 30% represent low heterogeneity, values between 30 and 60% represent moderate heterogeneity, and values above 60% represent substantial heterogeneity. For all statistical analyses, a p value less than 0.05 was considered statistically significant. Testing for publication bias could not be performed due to the small number of included studies.

Results

A total of 630 unique studies were screened during the study selection process. After applying the exclusion criteria, a total of ten studies (five studies assessing AR vs. NAR and five studies assessing R1 vs. R0) published between 2016 and 2023 were deemed eligible for inclusion in the quantitative analysis (Fig. 1).5,6,7,11,12,13,14,15,16,17 The study and patient characteristics are reported in Tables 1 and 2, respectively.

Fig. 1
figure 1

Prisma flowchart of study selection

Full size image
Table 1 Study and patient characteristics for patients undergoing anatomic versus non-anatomic resections
Full size table
Table 2 Study characteristics and baseline patient demographics for patients undergoing R1 versus R0 resections
Full size table

Outcomes of Anatomic versus Nonanatomic Resections in Patients with Known RAS Mutation Status

The association of AR versus nonanatomic resections (NAR) with DFS and liver-specific DFS in patients with known RAS status was evaluated in five studies with a total of 2018 patients (831 mutRAS and 1187 wtRAS).5,6,7,11,17 AR was associated with improved liver-specific DFS (HR = 0.60, 95% CI 0.44–0.81, p = 0.01; Fig. 2A) and overall DFS (HR = 0.72, 95% CI 0.54–0.95, p = 0.02; Fig. 3A) in mutRAS patients but not in wtRAS patients (Figs. 2B, 3B). Statistical heterogeneity was moderate to substantial across evaluated outcomes (I2 = 48.2–82.1%).

Fig. 2
figure 2

Forest plots of liver-specific disease-free survival in A mutRAS and B wtRAS patients

Full size image
Fig. 3
figure 3

Forest plots of overall disease-free survival in A mutRAS and B wtRAS patients

Full size image

R1 Resection Rates in Patients with Known RAS Mutation Status Who Underwent Anatomic versus Nonanatomic Resections

Among four studies incorporating 1289 patients (488 mutRAS and 801 wtRAS), AR was associated with a decreased incidence in R1 resection among mutRAS patients (HR = 0.6, 95% CI 0.4–0.91, p = 0.02; Fig. 4A) but not in wtRAS patients (HR = 0.93, 95% CI 0.69–1.25, p = 0.62; Fig. 4B). No statistical heterogeneity was detected.

Fig. 4
figure 4

Forest plots of risk for R1 resection after anatomic versus nonanatomic resections in A mutRAS and B wtRAS patients

Full size image

Association of R1 Margins with OS and Liver-Specific DFS

Five studies including 2580 patients (969 mutRAS and 1611 wtRAS) reported the association of R1 resection margin with OS for mutRAS and wtRAS patients.12,13,14,15,16 The R1 resection margin was significantly associated with poor OS in both groups (mutRAS: HR = 1.34, 95% CI 1.07–1.66, p = 0.01; wtRAS: HR = 1.8, 95% CI 1.44–2.25, p < 0.001), with low statistical heterogeneity (Fig. 5).

Fig. 5
figure 5

Forest plots of overall survival based on resection margin status in A mutRAS and B wtRAS patients

Full size image

Liver-specific DFS was evaluated in only two studies, for a total of 804 patients (318 mutRAS and 486 wtRAS patients).14,15 R1 resection margin was associated with worse outcomes among both mutRAS (HR = 1.46, 95% CI 1.04–2.05, p = 0.03) and wtRAS (HR = 2.64, 95% CI 1.99–3.5, p < 0.001) patients, with no statistical heterogeneity present (Supplementary Fig. 1).

Critical Appraisal and Risk of Bias Assessment

The ROBINS-I tool was used to assess risk of bias given the retrospective nature of the included studies. Across evaluated domains, bias due to confounding, deviation from intended interventions, and missing data were encountered. No studies were at serious risk of confounding bias, but moderate risk was noted in eight studies that did not utilize propensity score matching. A deviation from intended interventions was present in three studies in the AR versus NAR comparison, as tumors in the former group were larger, suggesting a predilection toward more aggressive surgery in such patients. In turn, it is likely that the true effects of AR may be even more favorable, as patients with mutRAS tumors who underwent AR had unfavorable tumor characteristics compared with their counterparts who underwent NAR. Moderate risk of bias due to missing data was encountered in a single study in which patients lost to follow-up were excluded from the final analysis. Overall, the risk of bias was judged to be moderate in eight studies, and low in one (Supplementary Table 1).

Discussion

Five studies have previously examined the association of AR vs NAR with overall and liver-specific DFS among patients with known RAS mutation status. A significant association between AR and superior overall and liver-specific DFS in mutRAS tumors was noted only in the largest two studies.5,6,7,11,17 Notably, the other three studies also reported an association of AR with improved overall and liver-specific DFS in mutRAS tumors but failed to reach statistical significance. Interpreting the lack of statistical significance as conflicting with the largest two studies may be misleading, since the estimated treatment benefit of AR was similar (Fig. 2A).18 As Goodman noted, “A nonsignificant difference does not make the null effect the most likely. The effect best supported by the data from a given experiment is always the observed effect, regardless of its significance.”18 Indeed, in examining the KM curves from the MD Anderson publication, AR was associated with better liver-specific DFS in both unselected (see Fig. 3C of that publication) and matched mutRAS cases (Fig. 3F) versus NAR.6 In contrast, there was no difference in the outcomes of AR versus NAR in both unselected (Fig. 3B) and matched wtRAS cases (Fig. 3E). A meta-analysis is an ideal methodological tool to use in situations such as this, in which smaller studies noted a trend toward efficacy as pooling individual studies allows for an increase in overall sample size, enhances the statistical power of the analysis, and reduces the confidence interval for the point estimate of the effect.19 This approach may unveil a true and significant benefit of therapy that would have been otherwise missed.20 In the current meta-analysis, AR was significantly associated with a 40% and a 28% improvement in liver-specific and overall DFS, respectively, versus NAR in patients with mutRAS tumors.

The benefit of AR in patients with mutRAS tumors may be related to more favorable characteristics of patients who underwent AR. However, this was not the case in any of the five studies, as AR was associated with unfavorable and not favorable characteristics. For example, in the study from M.D. Anderson, the only relevant differences in patient characteristics were in favor of NAR.6 Specifically, patients with mutRAS tumors who underwent AR were more likely to have larger and more CRLMs than their counterparts who underwent NAR. Unsurprisingly, when patients were matched for these differences, the benefit of AR over NAR became even more visually apparent (see Fig. 3F, C of that publication). Similarly, the median size of the largest CRLM was greater in patients with mutRAS tumors who underwent AR compared with NAR in the study from Johns Hopkins and the study from Fudan University.5,6,17 In addition, in the study from Fudan University, patients who underwent AR were more likely to have high (over 200 ng/mL) carbohydrate antigen (CA) 19-9 levels.17 Two additional significant differences include a lower rate of bilateral disease and less use of radiofrequency ablation (RFA) (only in the Johns Hopkins study) among individuals with mutRAS disease who underwent AR. Importantly, when all factors were adjusted in a multivariable analysis, the AR-associated improvement in overall and liver-specific DFS persisted in both studies [HR for overall DFS: 0.45 (p = 0.002) and 0.5 (p < 0.001) in the Johns Hopkins and Fudan University studies, respectively; HR for liver-specific DFS: 0.42 (p = 0.006) and 0.43 (p < 0.001) in the Johns Hopkins and Fudan University studies, respectively]. The remaining two studies identified a greater tumor size among patients who underwent AR as the only significant difference between patients with mutRAS tumors who underwent AR versus NAR.7,11 Thus, it is unlikely that the favorable effects of AR were confounded by other factors, as patients with mutRAS tumors who underwent AR had unfavorable tumor characteristics.

We also investigated the mechanism through which AR confers an oncologic benefit in patients with mutRAS but not wtRAS tumors. One possibility was the lower rate of R1 resections among patients with mutRAS who undergo AR, as a more extensive tumor growth pattern of mutRAS tumors can only be eradicated by AR. In contrast, a NAR may be sufficient to eliminate tumor cells of wtRAS tumors, and thus AR would not decrease the rate of R1 resections in this patient group. Indeed, among patients who underwent AR, only patients with mutRAS CRLM had a markedly lower rate of R1 resections. This novel finding again highlights the ability of meta-analyses to maximize statistical power and uncover relationships that are masked by small sample sizes of individual studies. For example, although not statistically significant, a trend toward lower rates of R1 resection after AR for mutRAS CRLM was reported in the study by Choi et al. Specifically, R1 resection was 0% versus 4.5% among patients with mutRAS tumors who underwent AR versus NAR, respectively (p = 0.550), and 3.9% versus 1.9% in those with wtRAS tumors who underwent AR versus NAR, respectively.11 In the study by Margonis et al., although AR was associated with lower rates of R1 in both mutRAS and wtRAS groups, the absolute reduction in R1 resection with AR was more pronounced in the mutRAS group at 12.4% versus 4.6%.5 In addition, the study by Joechle et al. reported a rate of R1 resection that was significantly lower in the mutRAS group who underwent AR versus NAR; similar to the previous studies, there was no difference in the wtRAS group.6

Subsequently, we evaluated whether R1 resection was associated with worse oncologic outcomes in patients with mutRAS tumors, which is important as the association between R1 and outcomes has generally been controversial.21,22 Of note, R1 resection was associated with worse liver-specific DFS in patients with mutRAS tumors. Given that only two studies were included in this analysis, we performed another analysis to examine the association of R1 with OS in patients with mutRAS CRLM, as this yielded more eligible studies. Similarly, we noted that R1 resection was also associated with worse OS in those with mutRAS tumors.

While the results presented in this analysis carry significant implications, it is important to acknowledge several limitations. The included studies, although generally sound in their overall methodology, predominantly utilized retrospective designs and often failed to adequately account for potential confounding factors. Notably, this was particularly evident in studies that compared anatomical resections and nonanatomical resections, as a tendency toward AR in larger tumors was observed in some studies. In turn, there was the potential for selection bias. Substantial statistical heterogeneity among the included studies was observed for many of the evaluated outcomes. While we employed a random-effects analysis model to mitigate some of the impact of this heterogeneity on the overall robustness of the analysis, a more comprehensive exploration of the causes of this heterogeneity through meta-regression analysis was not feasible due to the limited number of included studies. It is worth noting that meta-regression is generally not advisable when there are fewer than ten studies published on the topic.23 Lastly, we did not report on the association of surgical technique (AR vs. NAR) and OS, as only two studies reported relevant data, and therefore a meta-analysis was not possible.

Although the results obtained offer valuable insights into the impact of RAS status on the choice of the optimal surgical procedure and subsequent expected survival, it is important to highlight that a direct causal relationship between AR and DFS cannot be definitively established using observational data. For example, it is possible that wider margins accompanying AR, not AR itself, might be the reason for the observed association with improved DFS. If this was the case, one would anticipate that extended margins (e.g., > 1 cm) and AR would exhibit a similar impact on outcomes in CRLM. However, according to the most recent meta-analysis on margin width (> 1 cm vs. < 1 cm) and all meta-analyses comparing AR versus NAR for CRLM, this conclusion is not supported by the data.24,25,26,27 Specifically, the evidence suggests that while a margin width greater than 1 cm is associated with improvements in OS and DFS compared with a margin width less than 1 cm, AR has not been associated with improvements in OS or DFS compared with NAR. Future studies that independently evaluate margin width (as opposed to R0 vs. R1 status) and surgical technique (AR vs. NAR) among patients with RAS-mutated and wild-type tumors will be essential. Currently, only one study has provided data on both these factors, underscoring the need for additional research in this area.

In conclusion, the current systematic review and meta-analysis of surgical technique (AR vs. NAR) relative to tumor biology revealed a 40% decrease in the risk of liver recurrence and a 28% decrease in the risk of any recurrence for AR compared with NAR. The effect of AR versus NAR was only noted, however, among patients with mutRAS tumors. This finding serves to reopen the discussion on AR in RAS-mutated CRLM, which conflicts with existing dogma of pursuing more limited resections for all tumors. The current meta-analysis may be the best source of evidence until 2025 when the ongoing “Anatomical Resection of Liver Metastases in Patients with RAS-mutated Colorectal Cancer” (ARMANI) randomized trial will report prospective data on this topic.