Simplified Staging System for Predicting the Prognosis of Patients With Resectable Liver Metastasis: Development and Validation

Hypothesis  Although several staging systems for colorectal liver metastasis have been proposed, simple and generally accepted staging systems are not available for this disease. We hypothesized that more detailed analysis of primary colorectal cancer may make it possible to develop a simple staging system and that its stratification ability may be demonstrated by validation against data from unrelated patients.

Design  Retrospective analysis of prospectively documented data, development of a stage, and validation against an unrelated cohort.

Setting  Four tertiary referral centers.

Patients  Twenty-two clinicopathologic factors were examined in 369 consecutive patients who underwent curative resection for liver metastasis from colorectal cancer (original cohort). Using the independent prognostic factors, a simplified staging system was developed and was validated by data from 229 unrelated patients (validation cohort).

Main Outcome Measures  Kaplan-Meier survival curve analyses between different prognostic groups in the cohorts.

Results  Multivariate analysis revealed several independent prognostic variables, including hepatic lymph node metastasis (relative risk 4.39), 4 or more colorectal lymph node metastases (RR 1.50), carcinoembryonic antigen level of 50 ng/mL or higher (RR 1.29), and multiple hepatic metastases (RR 1.27). Patients with hepatic lymph node metastasis were assigned to stage 4, and the remaining patients were divided according to number of factors: none, stage 1; 1, stage 2; 2 or 3, stage 3. In the original cohort, median survival in stages 1, 2, 3, and 4 was 7.2, 3.5, 2.0, and 1.3 years, respectively. In the validation cohort, these values were 9.6, 4.1, 2.8, and 1.6 years, respectively.

Conclusions  The proposed simplified staging system was easy to use, was highly predictive of patient outcome, and permitted categorization of patients into treatment groups. Although we validated this staging system, further validation and improvements are needed.

Liver metastases from colorectal cancer are classified by Union Internationale Contre le Cancer (UICC) staging criteria as stage IV, although the prognosis of patients with this disease varies widely.¹ Hepatic resection for colorectal liver metastasis remains the only treatment that has curative potential.² Many controversies exist about the treatment of liver metastasis, such as the effectiveness of adjuvant chemotherapy, the timing of resection for synchronous metastasis, and the operative indications for multiple metastasis or extrahepatic metastasis. As a result, there is an increasing need for a simple staging system that can reflect the prognosis and permit the stratification of patients for clinical trials.

Several staging systems for colorectal liver metastasis have been proposed. Gennari,^3-5 Fortner,⁶ and Gayowski⁷ and their colleagues proposed staging systems based on the size, number, and intrahepatic and extrahepatic extent of metastatic nodules. Cady and Stone⁸ developed a prognostic scoring system that weighs individual factors. Nordlinger,⁹ Fong,¹⁰ Iwatsuki,¹¹ and Schindl¹² and their colleagues developed staging systems by analyzing prognostic factors, but 5 to 7 factors had to be explored to determine the stage.

What are the requirements of a good staging system? First, it should be simple and easy to use. Second, it must provide reliable information on the prognosis of the disease. Third, it should permit the categorization of patients into various treatment groups. Based on these criteria, well-defined and generally accepted staging systems are not available for this disease. The primary goals of this study were to develop a staging system that will fulfill these requirements and to validate its prognostic reliability in an unrelated group of patients.

Between January 1, 1980, and December 31, 2002, 388 patients with hepatic metastasis from colorectal cancer underwent liver resection at the Department of Surgery, National Cancer Center (1980-1990), the First Department of Surgery, Shinshu University (1990-1994), and the Department of Hepato-Biliary-Pancreatic Surgery, University of Tokyo (1994-2002). The last author (M.M.) participated in all of the operations. Nineteen of these resections were not radical because of gross residual disease within or outside the liver, and the remaining 369 patients were included in the original cohort.

Selection criteria for surgery were the possibility of complete removal of all hepatic and extrahepatic lesions and the possibility of preserving at least 40% of the normal hepatic parenchyma. The total number of hepatic metastases, their unilateral or bilateral presentation, and the existence of extrahepatic metastasis were not considered exclusion criteria. No ablative strategies were used along with resection in any of these patients. The treatment policy for synchronous metastasis was simultaneous resection regardless of the number and extent of liver metastasis and the location of the primary cancer.

In all cases, the preoperative diagnostic workup included ultrasonography and plain and contrast-enhanced computed tomography to stage liver involvement and chest radiography, chest computed tomography, barium enema, and colonoscopy to assess the presence or absence of extrahepatic disease. Patients with advanced disease underwent bone scintigraphy or positron emission tomography. Intraoperative bimanual liver palpation and intraoperative ultrasonography (IOUS) were also performed in all patients, and all of the resections were IOUS-guided procedures. The mean duration of follow-up in the original cohort was 4.11 years (range, 1.1 months to 18.8 years).

The validation cohort consisted of 229 patients with colorectal liver metastases who underwent curative hepatic resections by colleagues of the last author (M.M.): 77 at the National Cancer Center between January 1, 1991, and December 31, 1997 (M.M. moved to Shinshu University in 1990), and 152 at Cancer Institute Hospital between January 1, 1997, and December 31, 2003. The selection criteria for hepatectomy and the preoperative and intraoperative diagnostic workup in these groups were comparable with those of the original cohort. The mean duration of follow-up in the validation cohort was 3.95 years (range, 2.5 months to 13.5 years). This retrospective study was approved by the institutional review boards in the respective institutions.

Survival time was calculated from the date of hepatic resection to death or censored date. Patients who died of colorectal cancer were treated as event observations, and patients who died of unrelated causes and were alive at the last follow-up were treated as censored observations. Survival curves were constructed using the Kaplan-Meier product-limit method and compared using the log-rank test. Significant prognostic factors in a univariate analysis were entered into a Cox proportional hazards model using stepwise selection to identify independent predictors of death. Statistical significance was defined as P<.05. A software program (SAS version 8; SAS Institute Inc, Cary, NC) was used for the statistical analyses.

The 3-, 5-, and 10-year survival of the original cohort were 52%, 38%, and 26%, respectively. There was no in-hospital death. We analyzed the effects of 15 clinicopathologic factors at hepatic resection (Table 1) and 7 at primary colorectal resection (Table 2) on survival after curative hepatic resection. Multiple liver metastases (P<.001), diameter of 5 cm or greater (P = .02), interval between primary cancer and liver resection less than 6 months (P = .04), carcinoembryonic antigen (CEA) level of 50 ng/mL or greater (P<.001), a resection margin less than 5 mm (P = .006), hepatic lymph node metastasis (P<.001), extrahepatic metastasis (P = .03), and extrahepatic invasion (P = .03) showed significant prognostic value for survival in a univariate analysis. Unilateral distribution of metastases was a favorable factor (P<.001), and 148 of 222 patients with unilateral metastasis had a solitary metastasis. Excluding patients with a single metastasis, distribution was not significant in patients with multiple metastases (P = .64) (Table 1). Survival curves stratified by the number of liver metastases are shown in Figure 1A. The prognosis according to the serum CEA level at hepatic resection is shown in Figure 1B. In this article, patients were divided into 2 groups according to the serum CEA level at hepatic resection (≥50 and <50 ng/mL) because the χ² statistic by the log-rank test reached a maximum (χ² = 21.8) when the boundary was set at 50 ng/mL.

Invasion to the serosa or another organ of primary colorectal cancer (pT4 by UICC classification) (P = .02), number of colorectal lymph node metastases of 4 or more (pN2 by UICC classification) (P<.001), and lymphatic duct involvement by the primary cancer (P = .03) also predicted an adverse outcome (Table 2). Nodal status of the primary cancer and long-term survival are shown in Figure 1C.

The univariate prognostic factors were entered into a multivariate model to identify independent predictors of long-term survival. Hepatic lymph node metastases had the greatest impact on survival (relative risk, 4.39), followed by 4 or more colorectal lymph node metastases (pN2) (relative risk, 1.50), CEA level of 50 ng/mL or greater (relative risk, 1.29), and multiple metastases (relative risk, 1.27) (Table 3).

Regional lymph node metastasis of the liver was clearly the most influential factor and was associated with a 4.39-fold increase in the likelihood of death if it was positive. Thus, these patients were assigned to stage 4. The other 3 independent prognostic factors (number of lymph node metastases around the primary cancer ≥4, CEA level ≥50 ng/mL, and multiple liver metastases) cannot be considered complete contraindications to resection because each alone was still associated with a sufficiently favorable outcome to justify an aggressive surgical procedure, and the increase in the likelihood of death ranged from 1.27 to 1.50. Therefore, these criteria were used to determine whether some combination could be used to dictate the choice of clinical options. Patients who had none of these 3 factors were assigned to stage 1, those with 1 factor to stage 2, and those with 2 or 3 factors to stage 3 (Figure 2). Survival curves for the original cohort, classified according to this simplified staging system, are shown in Figure 3. This simple staging was found to be highly predictive of the long-term outcome (P<.001) (Figure 3), and the differences in survival between the stages were significant (Table 4). Next, the original cohort was divided into 2 groups—synchronous vs metachronous metastasis—and the prognostic value of this simplified staging system was evaluated in each group. In the 187 patients with synchronous metastasis, 5-year survival for stages 1, 2, 3, and 4 were 65%, 38%, 18%, and 0%, respectively (P<.001). In the 182 patients with metachronous metastasis, 5-year survival for stages 1, 2, 3, and 4 were 54%, 48%, 30%, and 0%, respectively (P<.001).

The 3-, 5-, and 10-year survival of the validation cohort were 61%, 44%, and 35%, respectively. Of the 229 patients, 64 were assigned to stage 1, 93 to stage 2, 67 to stage 3, and 5 to stage 4. Median survival time and 5- and 10-year survival rates for each stage are summarized in Table 4. The assigned stage was highly predictive of patient outcome (P<.001) (Figure 4).

In the 1980s, 2 staging systems were developed by Gennari et al^3-5 and Fortner et al.⁶ These systems were based on the degree and extent of metastatic tumors and not on factors regarding primary colorectal cancer.^3-6 Cady and Stone⁸ proposed a scoring index based on 4 risk factors: surgical margin, CEA level, disease-free interval, and number of liver nodules. This staging also did not include factors regarding colorectal cancer, although the researchers pointed out that patients with poor differentiation and greater than 5 lymph node metastases in the primary cancer should have a poor prognosis, which would be governed by biological factors.¹³ Gayowski et al⁷ proposed a modified TNM staging system based on several factors: unilateral or bilateral, single or multiple, 2 cm or smaller or larger than 2 cm, and vascular or ductal invasion to a major branch. In this system, all metastases with bilateral distribution are considered modified T4. Generally, most patients with a single tumor have a unilateral distribution, and those with multiple nodules have a bilateral distribution. The worse outcome associated with multiple nodules affects the outcome with a bilateral distribution. As we have shown previously,¹⁴ the prognosis of patients with multiple tumors did not differ according to the distribution in the liver. In a multicenter study by Nordlinger et al,⁹ 1568 patients who had metastases confined to the liver and who received curative resection were analyzed, and 7 factors were found to be significant in a multivariate analysis: age 60 years or older, size 5 cm or larger, pT4 by UICC classification, pN1 or greater by UICC classification, disease-free interval of less than 2 years, 4 or more nodules, and margins less than 1 cm. Three stages were established based on the number of factors present: 0 to 2, 3 to 4, and 5 to 7. A similar method was used by Fong et al¹⁰ and Iwatsuki et al¹¹ in 1999, but cases with a positive margin, extrahepatic disease, or hepatic lymph node metastasis were either excluded or assigned to the highest stage. Patients without these factors were divided according to the number of the following factors: node-positive primary cancer, disease-free interval of less than 12 months, more than 1 hepatic tumor, largest hepatic tumor greater than 5 cm, and CEA level greater than 200 ng/mL by Fong et al¹⁰ and 3 or more tumors, tumor size greater than 8 cm, disease-free interval of 30 months or less, and bilateral tumor by Iwatsuki et al.¹¹ These staging systems were based on a multivariate survival analysis and reflected the prognosis but used 7 factors. Thus, all of the factors must be explored to determine the stage, which may make it difficult to use these staging systems.

It is essential for a good staging system to provide reliable information on the prognosis of the disease. To show that a staging system actually reflects the prognosis, it must be verified by validation against data from unrelated patients. The staging system proposed by Fong et al¹⁰ was validated by Mann et al¹⁵ in Australia. Schindl et al¹² developed a prognostic scoring system using Dukes stage, number of metastases, CEA level, alkaline phosphatase level, and albumin level and validated its prognostic reliability in an unrelated group of patients. The robustness of the present staging system was tested by validation against data from patients who were not included in the original cohort. The survival rates of each stage in the validation cohort approximate those in the original cohort, and the P values for stage 1 vs 2 and stage 3 vs 4 are significant. Regarding stage 2 vs 3, it seems reasonable to predict that it will be significant with increasing numbers of patients because the median survival time of each stage is monotonically decreasing with advancing the stage. Consequently, the present staging system may provide reliable information on the prognosis of patients with colorectal liver metastasis.

Extrahepatic extension, such as extrahepatic metastasis, extrahepatic invasion, local recurrence at the primary cancer, and hepatic node metastases, has been analyzed as a whole in most previous studies. Patients with these factors have long been considered to be contraindicated for hepatectomy because of their dismal outcome. However, lung metastases, intraperitoneal dissemination, and local recurrence have gradually gained acceptance for resection in some institutions because a favorable prognosis can be anticipated if the tumors are removed completely.^16-19

The incidence of macroscopic involvement of hepatic lymph nodes in patients who underwent hepatic resection reported in the literature is 3% to 6%, and 4 of 7 studies^{2,7,9,11,18,20,21} reported 5-year survival of 0%. In contrast, Elias et al¹⁸ showed 5-year survival of 27% in such patients after hepatectomy and lymph node dissection. The rate of microscopic involvement of hepatic lymph nodes has been reported to be 11% to 28%.^22-28 Although hepatectomy and lymph node dissection were performed in these patients, 5-year survival was reportedly 0% to 5%.^24,25,28 Rodgers and McCall²⁹ reviewed 15 studies that gave survival data on node-positive patients: 145 patients received hepatic resection, and only 5 (3.4%) survived 5 years. Based on these findings together with the present results, patients with hepatic lymph node metastasis were assigned to stage 4 in the simplified staging system. We should not operate on patients with hepatic lymph node metastasis.

Although many researchers^2,9,14,30-34 have noted that primary colorectal cancer affects the prognosis of patients who received hepatectomy for liver metastases, some^{7,13,20,24,35-40} have reported contrary results. This discrepancy may be due to rates of synchronous and metachronous metastasis in each study. As our group⁴¹ previously noted, the significant prognostic factors in patients with synchronous metastasis are different from those in patients with metachronous metastasis. In patients with synchronous metastasis, independent prognostic factors were 4 or more lymph node metastases around the colorectal cancer (P<.001) and multiple liver metastases (P = .003), whereas in patients with metachronous metastasis, CEA level (P = .002), 4 or more lymph node metastases around the colorectal cancer (P = .03), and hepatic lymph node metastasis (P = .03) were independently significant.⁴¹ Factors associated with colorectal cancer play a more important role in synchronous metastasis. In a study in which most patients have metachronous metastasis, the stage of the primary cancer may not play an important role in the prognosis.

In most studies, the factors of colorectal cancer were represented in terms of Dukes stage. We analyzed it more precisely: patients without mesenteric lymph node metastasis and those with 1 to 3 lymph node metastases had a similar prognosis, and those with 4 or more metastases showed a significantly worse outcome (Figure 1C). Therefore, it is more reasonable to separate patients according to the number of lymph node metastases (≥4 vs 0-3) than Dukes stage (A-B vs C). Moreover, the depth of the wall invasion by colorectal cancer is known to affect the prognosis. A tumor without regional lymph node invasion is classified as Dukes stage A if it invades the muscularis propria or less and as Dukes stage B if it infiltrates the subserosa or more. According to the present analysis on the depth of invasion and prognosis, tumors that perforated the visceral peritoneum or directly invaded other organs or structures (T4 by UICC classification) had a significantly poor outcome after hepatic resection, and no difference in survival was observed between tumors that invaded the submucosa (T1) or muscularis propria (T2) and tumors that invaded through the muscularis propria into the subserosa or into nonperitonealized pericolic or perirectal tissues (T3). A similar result was reported by Kato et al.⁴² Therefore, it may be more reasonable to separate patients with liver metastasis into T1 to T3 and T4 than Dukes stages A and B-C.

Many studies have demonstrated that the preoperative CEA level has prognostic value. However, little is known about the biological function of CEA, which might act as an adhesion molecule when expressed on the cell surface or as a secreted immune modulator.^43-47 It has also been noted that the tumor burden may not correlate with CEA levels,^48,49 that the prognostic value of a high serum CEA level was comparable with that of the presence of intraperitoneal tumor cells,⁵⁰ that CEA enhances liver metastasis by functioning as an attachment factor,⁵¹ and that an increased posthepatectomy CEA level was independently associated with extrahepatic recurrence.⁵² Based on these results, the precise function of CEA is not clear: a high serum CEA level may reflect a highly malignant nature of cancer cells, which induces peritoneal dissemination, liver metastasis, and extrahepatic recurrence. In the present series, a CEA level of 50 ng/mL or more was an independent prognostic factor that contributed to the construction of the staging in association with the number of mesenteric lymph node metastases and multiple liver metastases.

Solitary metastasis was a favorable prognostic factor in a multivariate analysis. The prognosis of 97 patients with 4 or more nodules was similar to that of 116 patients with 2 to 3 deposits (Figure 1A). This result may be a consequence of the complete removal of hepatic and extrahepatic metastases and treatment of postresectional recurrence. In the present series, all of the patients underwent careful examination by means of IOUS and IOUS-guided hepatectomy. Makuuchi et al⁵³ first introduced IOUS in 1979. Twenty-five years later, modern diagnostic instruments still cannot replace IOUS regarding its sensitivity in depicting liver nodules.⁵⁴ Choti et al⁴⁰ demonstrated that the patient's prognosis after hepatic resection was significantly improved with the use of IOUS. In our experience, approximately 1.5-fold as many nodules are visualized by means of IOUS in patients with 4 or more metastases, and, thus, one third of the nodules cannot be detected even with extracorporeal diagnostic modalities. If these nodules are left in place, the prognosis of patients with 4 or more metastases will be dramatically worsened. These occult nodules in 4 or more metastases may have played an important role in the poor prognosis. Characteristically, liver metastasis, especially 4 or more metastases, can easily lead to recurrent nodules in the remnant liver. The treatment of such recurrences can strongly affect the prognosis. Our choice of treatment for recurrent metastasis is repeated resection, performed immediately and without neoadjuvant chemotherapy. With this treatment, the prognosis of patients with multiple metastases has been remarkably improved.^55,56

This simplified staging system is easy to use, is highly predictive of patient outcome and survival, and permits the categorization of patients into various treatment groups. Patients with hepatic lymph node metastasis, who are categorized to stage 4 using the simplified staging system, should be excluded from hepatic resection. Patients in stage 1, 2, or 3 should receive hepatic resection, but it may be appropriate to apply adjuvant chemotherapy to patients with stage 3 disease. Our simplified staging system was validated by data from unrelated patients. However, further verification and refinement by other medical centers are necessary.

Correspondence: Masami Minagawa, MD, PhD, Department of Hepato-Biliary-Pancreatic Surgery, Department of Artificial Organ and Transplantation, Graduate School of Medicine, University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8655, Japan (minagawa-tky@umin.ac.jp).

Accepted for Publication: January 28, 2006.

Author Contributions:Study concept and design: Minagawa and Makuuchi. Acquisition of data: Minagawa, Yamamoto, Kosuge, and Miyagawa. Analysis and interpretation of data: Minagawa, Matsuyama, and Makuuchi. Drafting of the manuscript: Minagawa. Critical revision of the manuscript for important intellectual content: Minagawa, Yamamoto, Kosuge, Matsuyama, Miyagawa, and Makuuchi. Statistical analysis: Minagawa and Matsuyama. Administrative, technical, and material support: Minagawa and Kosuge. Study supervision: Yamamoto, Kosuge, Miyagawa, and Makuuchi.

Financial Disclosure: None reported.