INTRODUCTION
Patients with colorectal liver metastases (CLM) are classified as stage IV in the TNM staging system, although the prognosis of these patients varies widely [1]. Adoption of hepatic resection as standard treatment for colorectal cancer suddenly converted a stage IV cancer that was uniformly fatal within months of diagnosis into a heterogeneous population of patients with highly variable outcomes. Classification of patients based on their likelihood of survival would have several advantages including: accurate comparisons of results from various institutions, stratification of patients in clinical trials, assess the likely need for adjuvant therapy, and provide proper prognostic information to the patient [2-3].
Several authors have attempted to define clinical staging systems for this wide group of potentially curable patients [2,4-11]. In the 1999, Fong, et al. proposed a clinical risk score (CRS) system derived from data collected on 1001 patients resected for metastatic colorectal cancer, with the aim to select preoperatively patients for liver resection [8]. Since its publication, this scoring system has been validated by independent series [12-14] and used in several studies for risk stratification after liver resection of CLM [15-22]. In addition, it has also been shown that this scoring system predicts resectability and therefore yield of staging methods such as laparoscopy or PET scan [23-25].
Although Fong et al. found seven parameters independent predictors of outcome, two major findings such as positive margin and presence of extrahepatic disease were not included in the CRS because considered contraindications to liver resection, and often not available in the preoperative setting, therefore useless in the preoperative patient selection [8].
However, the criteria for resectability have been expanded over the last years leading to a more aggressive policy that extends the indications for surgery of colorectal liver metastases despite the presence of extrahepatic disease, if resectable, or in presence of minimal or positive margins [26-27- 28].
Therefore, we have hypothesized that for identifying patients at high risk for recurrence and poor prognosis after liver resection, the CRS could be improved by including the presence of totally resectable (R0) extrahepatic disease and positive surgical margins (R1).
METHODS
Using a multicenter database, we identified 413 consecutive patients who underwent liver resection for CLM between January 2000 and December 2004 at 3 hepatobiliary centers: Department of Surgery of Scientific Institute San Raffaele, Vita-Salute San Raffaele University (Milan, Italy), Unit of Hepato-Biliary and Digestive Surgery, Catholic University of the Sacred Heart School of Medicine (Rome, Italy), Liver Surgery Unit, Third Department of Surgery, Istituto Clinico Humanitas, University of Milan (Rozzano - Milan, Italy).
In all patients, a complete clinical, laboratory, and radiologic work-up was undertaken to detect evidence of local and disseminated disease. After hepatic resection, all patients were regularly followed for recurrence by serum CEA levels and a computed tomography scan every 2 to 3 months up to 2 years and then every 6 months thereafter. The following data were collected for each patient: pathologic features of the primary colorectal cancer; features of liver metastases, including size, number of lesions, margin status, presence of extrahepatic disease, preoperative CEA level, disease-free interval, and outcome.
The following conditions were considered absolute contraindications to liver resection: unresectable extrahepatic disease (such as diffuse peritoneal carcinomatosis, hepatic artery/celiac-axis lymph node, multifocal bilateral lung metastases --, bone, or brain metastases),extensive liver involvement (both portobiliary primary pedicles and/or all three hepatic veins involved or less than 20% of residual liver volume), and Child B or Child C cirrhosis. If suspect hepatic pedicle lymph nodes were evidenced on preoperative imaging (CT and PET) or during liver surgery, lymphadenectomy was performed. All patients with extrahepatic diseases, detected before surgery received preoperative chemotherapy and, in case of objective response, received it again after hepatic resection.
The CRS was calculated for each patient and then analyzed with respect to the postoperative survival. The CRS is comprised of five factors, (1) node-positive primary lesion; (2) disease free interval from resection of the primary tumor to time of metastasis of less than 12 months; (3) number of hepatic tumors greater than 1; (4) largest hepatic tumor greater than 5 cm; and (5) preoperative carcinoembryonic antigen level greater than 200 ng/mL. Each criteria is assigned one point and the sum represents the CRS, creating a minimum CRS of 0 and a maximum of 5 [8].
Then a modified clinical risk score (MCRS) was calculated adding two additional variables to the CRS: presence of resectable extrahepatic disease and surgical margin status (R0 vs R1). R1 was defined as the presence of tumor cells at the surface of inked resection margin or a free margin less than 1 mm thickness. Each criteria was assigned one point, creating a minimum score of 0 and a maximum of 7. Patients for whom all criteria were not known were excluded from the study. The performance of the original and modified CRS were evaluated and compared. Patients with incomplete information on the five prognostic factors included in the Fong et al model were excluded from the study.
STATISTICS
The five variables included in the CRS as well as surgical margin status and resectable extraepatic diseases were each individually assessed for their effect on survival through univariate analysis (Kaplan-Meier method with log rank test) and multivariate analysis (Cox regression model). The prognostic performance of each scoring system was statistically assessed, evaluating homogeneity within classification groups, and discriminatory ability of the score in the association between stages and survival rates. Firstly we evaluated, at univariate analysis, the capacity of each score to distinguish categories of patient with significantly different survival (homogeneity of the score). For each score, this performance was evaluated by comparing by log rank test the survival curves of the single categories, calculated using the Kaplan-Meier method. Finally, to evaluate the discriminatory ability for the prediction of survival, we evaluated the accuracy of prediction of death at one, three, and five years for each scoring system. This point was evaluated calculating the area under the receiver operating characteristic (ROC) curve for each score [29]. To perform this test, patients censored before one, three, and five years were excluded from the analysis. The discriminatory ability for the prediction of recurrence was also evaluated at one, two and three years. The area under the ROC represents the probability that a patient who died had a higher predicted probability of dying than a patient who survived. A ROC of 0.5 indicates that the model does not predict better than chance. The discrimination of a prognostic model is considered perfect if ROC =1, good if ROC >0.8, moderate if ROC is 0.6 to 0.8, and poor if ROC <0.6 [16].
Statistics were performed using SPSS version 14.0 (SPSS Inc, Chicago, Ill). All data are expressed as median values and ranges. P values 0.05 were considered significant.
RESULTS
Clinicopathologic Characteristics
Data of 413 patients were included in the analysis. There were 290 men and 123 women, and the mean age was 62± 9 years. Hepatic metastases were identified synchronously in 121 (29.5%) patients and 176 (42.6%) patients had a solitary metastasis. One-hundred and seventy patients (41.1%) had tumors larger than 5 cm and the majority of patients (n = 228; 55.2%) had positive lymph nodes on pathologic analysis of the primary colorectal tumor specimen. A CEA value higher than 200 ng/mL was found in 42 (10.1%) patients.
The type of operative procedure performed are included in Table 1. Although each patient had complete resection of the metastases macroscopically, 43 (10.4%) patients had microscopically positive pathologic surgical margins. Resection of extrahepatic disease was performed in 78 (18.8%) patients at time of liver surgery or later in case of lung metastases.
Extrahepatic disease included: lung metastases (n =26), hilar lymph node (n =17), peritoneal disease (n=11), pelvic lesions (n=10). Hepatic resection was performed in the setting of involvement of other organs by direct extension in 14 cases: diaphragm (n= 9), perinephric fascia (n=4), extrahepatic portal vein (n=3), extrahepatic biliary tree (n =3), and abdominal wall (n= 2).
Factors Influencing Probability of Survival
At a median follow-up of 34 months the median survival was 71.3 months. At the time of analysis 280 patients had been observed for at least 5 years. The 1-, 3-, and 5-year overall survival rates were 97%, 73%, and 58%, respectively. Recurrence at any site was diagnosed in 156 (37.7%) patients. The 1-, 3-, and 5-year disease-free were 94%, 72%, and 56%, respectively.
At univariate analysis, six of the seven risk factors were significantly predictive of survival : node status of the primary tumor (P=.0021), preoperative CEA level greater than 200 ng/mL (P<.0001), tumor size greater than 5 cm (P=.0035), number of tumors greater than 1 (P=.0048), surgical margin status (P=.0018), and extrahepatic disease (P<.0001) (Table 2). Disease-free interval less than 12 months (P=.73) was not found to be predictive of survival.
Multivariable analysis was performed on those factors correlating significantly to survival by univariate analysis. On multivariate analysis, node status of the primary tumor (HR =1.72, 95% CI = 0.81 -2.48, P = 0.021), preoperative CEA level greater than 200 ng/mL (HR =1.81, 95% CI = 1.10 -2.71, P = 0.001), tumor size greater than 5 cm (HR =1.64, 95% CI = 1.11 -2.31, P = 0.021), surgical margin status (HR =1.48, 95% CI = 1.05 -2.10, P = 0.045), and extrahepatic disease (HR =1.63, 95% CI = 1.12 -2.56, P = 0.0033) remained independent predictors of poor survival (Table 3).
CRS and MCRS as predictors of survival
The capacity of each score to distinguish categories of patient with significantly different survival was evaluated by comparing by log rank test the survival curves of the single categories, calculated using the Kaplan-Meier method. Table 4 and 5 show the results of retrospective staging of the 413 patients using the CRS and MCRS, with median survival times and death rates. The CRS was found to be predictive of long-term outcome (P= <0.001) (Fig. 1). Five-year survival rates for CRS of 0, 1, 2, 3 and 4 were found to be 61% 53%, 38%, 27.8%, and 21% respectively and was 0% for CRS>4 (Table 4). The MCRS was also found to be highly predictive of long-term outcome (P < 0.001) (Fig. 2). Five-year survival rates for MCRS of 0, 1, 2, 3, and 4 were found to be 72%, 69%, 45%, 28%, and 7% respectively and was 0% for MCRS >4 (Table 5).
Indeed, it appears possible to further stratify MCRS into 0 and 1, 2 and 3, 4 and 5, and 6 and 7 (P<.0001). Kaplan-Meier curves of survival after liver resection for colorectal liver metastases stratified across 4 groups of increasing postoperative scores are shown in Figure 3.
Discriminatory ability for death evaluated by ROC curve area analysis, were higher for MCRS compared with CRS at one (0.810 vs 0.690), three (0.781 vs 0.650), and five years (0.760 vs 0.640) (Table 6). Discriminatory ability for tumor recurrence at one (0.792 vs 0.670), three (0.771 vs 0.640), and five years (0.710 vs 0.610), were also higher for MCRS compared with CRS.
DISCUSSION
Surgical resection of CLM is associated to 5-year survival rates between 20% and 58% [2, 14-18]. This wide range of outcomes is mainly related to the heterogeneity of patients with hepatic colorectal metastases. Patients with a solitary hepatic lesion found years after the primary node-negative cancer clearly have a much better outcome that the patient with multiple lesions found synchronous to a node-positive primary, despite both being classified similarly as stage IV in current TNM staging. On this issue it has been recently underlined the urgent need of a new staging system of patients with advanced colorectal cancer [30].
Prognostic scoring for patients with CLM was introduced in 1991 by Cady and Stone [6]. They included surgical margin, the time to hepatic recurrence, the number of metastases, and serum CEA levels. However, this scoring system was not based on statistical analyses. Since then, several scoring systems have been developed to determine patients' prognosis [5-11].
Based on large experience of Memorial Sloan-Kettering, Fong et al. developed a simple clinical scoring system for preoperative selection of patients and predicting survival after hepatic resection [8]. The strength of the CRS is that it is predictive not only of survival but also of resectability and therefore yield of diagnostic modalities. Fong et al. analyzed prognostic factors in 1001 consecutive patients who underwent liver resection for CLM and found seven parameters independent predictors of outcome: positive surgical margin, presence of extrahepatic disease, number of tumors, preoperative CEA level, size of the largest tumor, nodal status of primary, disease-free interval from the primary to discovery of the liver metastases. The last five criteria were used to build the CRS, while positive surgical margin and presence of extrahepatic disease were not included in the CRS because they could not be obtained during the preoperative workup and therefore not useful in the preoperative patient selection, which was the main aim of this score. Furthermore Fong et al. argued to consider these last two factors as contraindications to liver resection [8]. However, since the publication of CRS in the 1999, the criteria for resectability of colorectal liver metastases have been significantly revised and expanded [3, 26-27].
Extrahepatic disease has been considered a contraindication to liver resection because subgroup analyses of large series published before 1990 showed a poor prognosis [31-32]. However, over the last years, some series have reported long term survival rates after resection of multiple lung metastases, extrahepatic disease synchronous to liver metastases, and of peritoneal carcinomatosis [33-36]. These studies have shown that resection of extrahepatic disease can achieve long-term survival, and even cure if the tumors are removed completely [34]. Consequently, the presence of disease outside the liver should no longer be considered a contraindication for liver resection provided that the disease outside the liver is resectable.
One of the classic contraindications to liver surgery was the presence of hilar lymph node metastases, considered by many authors as extrahepatic disease that cannot be successfully treated with hepatic resection. The incidence of macroscopic involvement of hepatic lymph nodes in patients who underwent hepatic resection reported in the literature is between 3% and 6% [5-6, 33]. However, data concerning involvement of hilar lymph nodes status were not evaluated in the original CRS [8]. Although early studies [37-38] reported few, if any, long-term survivors with hilar lymph node metastases, more recent studies [31, 39-40] have reported long-term survival in some patients with hilar nodal metastases and have concluded that this patient population may still benefit from hepatic resection. Jaeck et al. have reported that selected patients with limited regional nodal disease limited to the hepatoduodenal ligament have a significant benefit from liver resection. A recent review of the literature concluded that hepatectomy with en bloc lymphadenectomy resulted in a survival rate of 12-27 per cent in such patients, and that hilar node involvement does not absolutely contraindicate resection [41]. Furthermore, as improvements in systemic chemotherapy continue to evolve, particularly for metastatic colorectal cancer, the presence of positive hilar lymph nodes should not be used to determine resectability.
Extrahepatic disease, such as lung metastasis, peritoneal carcinomatosis, local recurrence at the primary cancer, and hilar lymph node metastases has been analyzed as a whole in the present study. Similarly, the results of a recent retrospective multivariate analysis suggest that the location of the extrahepatic disease if resectable, has not significant impact on outcome [35].
The role of the width of resection margin remains object of debate. Several earlier series concerning liver resection for CLM have reported that one should attain at least a 1-cm margin [37,42], and if not possible, this should be a relative contraindication to surgery [43]. Other authors reported that only resection margin less than or equal to 5 mm is associated with a greater risk of recurrence on the surgical margin, with a lesser rate of overall and disease-free survival [44]. Finally, it has been reported that the actual width of the surgical margin has no effect on survival as long as the margin is microscopically negative [45-46]. Experts now consider that resections should be performed whatever the margin, rather than no resection [27]. Indeed, a recent paper from Adam et al. reaches the conclusion that despite a higher recurrence rate, even the contraindication of R1 resection should be revisited in the current era of effective chemotherapy because survival is similar to that of R0 resection [28].
Taken together, these studies have demonstrated that the criteria for resectability of colorectal liver metastases have been significantly revised and expanded over the course of the last decade. Although clinicopathologic factors such as the presence of extrahepatic disease, or positive surgical margin are important prognostic factors, they should no longer be used to categorically exclude patients from surgical resection [3,26-27].
The present analysis confirms the value of the CRS point score system in preoperatively predicting the outcome following liver resection of CLM. Furthermore, the results of this study clearly demonstrated that the accuracy of CRS in predicting patients' survival is significantly improved by including resectable extrahepatic disease and surgical margin status as postoperative variables. In addition the MCRS was found more predictive not only of survival but also of tumor-recurrence. Although the relative risk for cancer-related death varied somewhat for the criteria included in the MCRS, we decided to maintain the original scoring system used for the CRS, assigning one point for every variable. Nevertheless, it is necessary to stress that the difference in relative risk for all the variables included in the MCRS were minimal.
Analyzed individually, 4 of the 5 variables considered in the original CRS were predictive of survival. Disease free interval from resection of the primary tumor was not found to be predictive of survival. However, previous studies have demonstrated the importance of this variables [5-6,8,47] and the prognostic value of the CRS has been previously validated [12-14]; consequently we have maintained the five original criteria in the MCRS.
Even though several series have demonstrated the usefulness of CRS, a recent study has disputed the value of prognostic scoring systems, including the CRS [2]. However, this study included patients who underwent liver resection over a long time period (1960-1995), during which many changes have being occurred in imaging, selection, and surgical technique. For these reasons in the present study we have considered necessary to limit our data analysis to a recent time period (2000-2004), which reflects recent progresses in liver surgery and chemotherapy for CLM.
Previous studies that specifically proposed prognostic scoring for patients with CLM included data from single institutions over a long time period. As such, these series are limited by their heterogeneous sample and potentially lack generalizability. The current study is important because it defines a score system in a large, modern, multi-institutional cohort of patients.
We found that both the CRS and MCRS are strong predictive of outcome. Moreover, in spite of the need of postoperative data, MCRS proved to be a more accurate predictor of survival and recurrence with a better discriminating power at one, three, and five years, than CRS. The MCRS, including two significant variables such as the presence of resectable extrahepatic disease and surgical margin status, offers a better stratification of patients based on their likelihood of survival. This remains true even if the patients are stratified into four main subgroups.
Because surgical margin status is not available preoperatively and extrahepatic disease may be detected only at the time of liver resection, the MCRS cannot be used for the selection of patients who may benefit from liver surgery, as the CRS.
Despite the growing knowledge about metastatic colorectal cancer, the issue of patient selection remains controversial. The recent trend of liver surgery for CLM is to include any patient in whom all known intra or extrahepatic disease can be removed and who has adequate hepatic volume/reserve [3,26-27]. Therefore, a simple risk scoring system such as that proposed here may be useful not in eliminating candidates for resection, but in selecting patients for different new protocols of adjuvant therapy, or for interinstitutional comparison of patients populations. Moreover, it is known that operative risk of re-resection is higher than first resection, both for technical reasons and for increased hepatic injury related to more prolonged chemotherapy [14,27]. In this specific group of patient a more accurate selection of good candidates to re-resection could be extremely useful .
CONCLUSIONS
This study validates and extends the prognostic model proposed by Fong et al for predicting patients survival following liver resection for CLM. Although the CRS remains the most widely used risk scoring system for preoperative selection of patients and prediction of patients survival, the recent expansion of criteria for resectability of CLM probably requires a revision of this score. The current study presents a Modified CRS that incorporates the same variables included in the CRS plus extrahepatic disease and surgical margin status. Differently from the CRS, the MCRS cannot be used for preoperative selection of patients, but as the CRS has the advantage of being simple, practical, and applicable to everyday practice because all the measures which adversely affect survival are easily available. However, compared to the CRS, the MCRS, provides a more accurate prediction of survival and recurrence following liver resection for CLM.
