Kaplan-Meier survival analysis in all patients of the original cohort with good (score, 0-10), moderate (score, 11-25), and poor (score, >25) prognoses; good vs moderate, P = .003; moderate vs poor, P<.001; and good vs poor, P<.001; log-rank test.
Kaplan-Meier survival analysis in patients of the original cohort who underwent hepatic resection with good (score, 0-10), moderate (score, 11-25), and poor (score, >25) prognoses; good vs moderate, P = .008; moderate vs poor, P = .02; and good vs poor, P = .001; log-rank test.
Kaplan-Meier survival analysis in all patients of the validation cohort with good (score, 0-10), moderate (score, 11-25), and poor (score, >25) prognoses; good vs moderate, P = .01; moderate vs poor, P<.001; and good vs poor, P<.001; log-rank test.
Schindl M, Wigmore SJ, Currie EJ, Laengle F, Garden OJ. Prognostic Scoring in Colorectal Cancer Liver MetastasesDevelopment and Validation. Arch Surg. 2005;140(2):183-189. doi:10.1001/archsurg.140.2.183
A prognostic scoring system for colorectal cancer liver metastases that is derived from unselected patients referred for hepatic resection would improve the applicability and increase the accuracy of prognostication.
Retrospective analysis of prospectively documented data; validation against an unrelated cohort from another institution. The median follow-up was 16.4 months (95% confidence interval, 15.0-17.8 months) (original cohort).
Two tertiary referral centers at unrelated university hospitals.
Independent prognosticators of survival were derived from 337 patients with colorectal cancer liver metastases referred for consideration of liver resection, and prognostic scores were calculated in 269 patients (79.8%) (original cohort). Calculation of prognostic scores was also applied to 193 patients referred and treated in an unrelated institution (validation cohort).
Main Outcome Measures
Kaplan-Meier survival curve analysis (log-rank test) between different prognostic groups in the original and the validation cohorts.
Independent prognosticators of survival were Dukes stage, number of metastases, and serum concentrations of carcinoembryonic antigen, alkaline phosphatase, and albumin. Significant differences were found in cumulative overall survival between patients assigned to good, moderate, and poor prognoses in the original and validation cohorts (P<.05). Liver resection improved survival in all prognostic groups. However, no patient with poor prognosis and only 19.7% (13 of 66) of patients with moderate prognosis survived 5 years, compared with 62.5% (10 of 16) of patients with good prognosis (P<.001).
This prognostic scoring system is derived from and can be applied to patients with colorectal cancer liver metastases at the time of referral for consideration of surgery. Patients with poor prognosis have no long-term benefit from curative liver resection and should therefore be considered for combined multimodal treatment.
For patients with colorectal cancer liver metastases (CRCLM), radical resection offers the only chance of long-term survival, with 30% to 40% of patients surviving 5 years.1- 6 Recent advances in surgical techniques and intensive care, together with a relaxation of traditional exclusion criteria for surgery, have increased the number of patients suitable for hepatic resection and the safety of the procedure.7- 11 However, despite routine adjuvant chemotherapy for advanced colorectal cancer and promising results of novel cytotoxic agents in adjuvant and neoadjuvant trials,12- 15 many patients with CRCLM still undergo major liver resection, despite a low likelihood of long-term survival. Several risk factors have been identified1,3,6,8,16- 20 and have been used within prognostic scoring systems to assess the individual risk for disease recurrence and death2,4,5,21,22 (Table 1). Ideally, a prognostic scoring system should be applicable to patients referred for liver surgery during their preoperative evaluation. However, all of the scoring systems published so far have used variables that were derived from patients who had already undergone liver resection; therefore, their accuracy when applied to all patients referred can be questioned. The aims of the present study were to develop a preoperative prognostic scoring system that can be applied accurately to all patients with CRCLM referred for liver surgery and to validate its prognostic reliability in an unrelated group of patients.
The original cohort was derived from 337 patients who were evaluated for surgical treatment of CRCLM at the Division of Hepatobiliary Surgery, Department of Surgery, The Royal Infirmary of Edinburgh, between October 1, 1988, and January 31, 2002. Preoperative investigations included imaging of the chest, abdomen, and pelvis by contrast-enhanced triple-phase computed tomographic scan with a slice thickness of 7 mm, determination of serum concentration of carcinoembryonic antigen (CEA), and assessment of liver function from routine blood test results. A detailed description of patients’ characteristics, primary tumor stage and metastatic behavior, liver function variables, and CEA serum concentration is provided in Table 2 and Table 3. Computed tomography arterial portography was performed in 184 patients (54.6%) and diagnostic laparoscopy with laparoscopic ultrasonography in 71 patients (21.1%). Intraoperative ultrasonography was performed routinely in all patients who underwent laparotomy. Patients were considered to have resectable disease if there was local control of the primary cancer, no extrahepatic disease existed, and complete removal of all hepatic lesions was expected, leaving enough hepatic parenchyma to prevent liver failure. One hundred fifty (44.5%) of 337 patients were treated by hepatic resection. Thirty patients (8.9%) underwent open exploration without resection, and 157 patients (46.6%) were deemed not suitable for resection at initial assessment and were referred for palliative chemotherapy. Patients normally received chemotherapy after resection of a Dukes C primary tumor, but chemotherapy was not administered routinely before or after hepatic resection. In 5 symptomatic patients (1.5%), palliative chemoembolization was performed. The median follow-up in patients comprising the original cohort was 16.4 months (95% confidence interval [CI], 15.0-17.8 months).
The validation cohort consisted of 193 patients with CRCLM who underwent surgery for an intended liver resection at the Division of General Surgery, Department of Surgery, University of Vienna, between January 1, 1998, and December 31, 2002 (Table 2 and Table 3). The initial staging and selection criteria for surgery were comparable to those of the original cohort, except that computed tomography arterial portography and diagnostic laparoscopy were rarely used in this group. One hundred fifty-seven patients (81.3%) underwent hepatic resection, 20 patients (10.4%) received intra-arterial hepatic chemotherapy alone through a vascular catheter implanted during surgery, and 10 patients (5.2%) received open local tumor ablative therapy. Six patients (3.1%) underwent exploration by laparotomy without further intervention. The median follow-up in this group was 16.5 months (95% CI, 14.5-18.4 months).
Single variables were explored regarding their prognostic value by Kaplan-Meier survival curve analysis with factors compared by log-rank test statistic and by Cox proportional hazards technique. Variables that showed significant values in all patients of the original cohort and in a subgroup of patients who underwent hepatic resection were entered into a Cox proportional hazards multiple regression model, and stepwise selection of independent prognostic variables was performed manually by significant changes in likelihood ratio.23
A mathematical equation was built that included all significant variables and coefficients from the final regression model as follows:
where Dukecode indicates Dukes stage A/B (score, 0) or C (score, 1); Metcode3, 1 to 3 metastases (score, 0) or more than 3 metastases (score, 1); ln Alkphos, natural logarithmic function of the serum concentration of alkaline phosphatase (normal range, 40-125 U/L); ln CEA, natural logarithmic function of the serum concentration of CEA (normal range, <5 μg/L); and Albumin, the serum concentration of albumin (normal range, 3.6-4.7 g/dL).
The coefficients were multiplied by 10 and rounded to achieve whole number multipliers, and a constant term was added to avoid negative scores. To assign scores to prognosis, mean cumulative 5-year survival was calculated in patients of the original cohort who underwent liver resection. Scores with estimated cumulative 5-year survival greater than 50% were assigned to a good prognosis, between 30% and 50% to a moderate prognosis, and less than 30% to a poor prognosis. Calculation of prognosis was applied to the original cohort and the validation cohort, and differences between prognosis groups were evaluated by Kaplan-Meier survival curve analysis with log-rank test statistic. The statistical software package SPSS 11.0 (SPSS Inc, Chicago, Ill) was used for all steps of data analysis. The assumption of normality and linearity was confirmed for all selected variables.24 However, some of the variables required logarithmic transformation.
Median survival in patients of the original cohort was 17.7 months (95% CI, 14.6-20.9 months), and the mean ± SE estimated 1-, 3-, and 5-year cumulative survival was 69.9% ± 2.6%, 26.4% ± 2.6%, and 14.9% ± 2.2%, respectively. Median survival in patients who underwent hepatic resection was 37.7 months (95% CI, 28.4-47.0 months), and the mean ± SE estimated 1-, 3-, and 5-year cumulative survival was 84.4% ± 3.0%, 52.2% ± 4.5%, and 36.0% ± 4.6%, respectively, compared with 13.3 months (95% CI, 11.6-15.0 months) and 56.6% ± 3.7%, 7.8% ± 2.1%, and 0%, respectively, in patients without hepatic resection (P<.001, log-rank test).
Dukes stage, site of primary tumor, diameter of the largest liver lesion, and serum concentration of CEA, alkaline phosphatase, and albumin showed significant prognostic value for crude survival in all patients of the original cohort and in a subgroup that underwent hepatic resection. Number of lesions, bilobar disease, age, and serum concentrations of bilirubin and γ-glutamyltransferase were significant only in the entire group of patients (Table 4 and Table 5). Among all significant variables, Dukes stage, number of metastases, and serum concentration of CEA, alkaline phosphatase, and albumin were identified as independent prognosticators and were used to calculate prognostic scores (Table 6). Variables encoding for hepatic resection and extrahepatic metastases also showed significant independent prognostic value for survival. However, because the study aimed to develop a scoring system primarily to be applied during preoperative assessment, these variables indicative of consecutive treatment were not considered for further analysis.
Prognostic scores were calculated in 270 (80.1%) of 337 patients of the original cohort. Scores ranging from 1 to 57 were derived, and patients were allocated based on their individual scores into good (score, 0-10), moderate (score, 11-25), and poor prognoses (score, >25). The overall median survival was 35.7 months (95% CI, 0.3-96.4 months), 23.5 months (95% CI, 19.3-27.6 months), and 10.6 months (95% CI, 7.6-13.5 months) in patients assigned to good, moderate, and poor prognoses, respectively (Figure 1) (good vs moderate, P = .003; and moderate vs poor and good vs poor, P<.001). Prognostic scoring was possible in 131 (87.3%) of 150 patients who underwent hepatic resection, and median survival was 59.8 months (95% CI, 32.3-68.4 months), 32.0 months (95% CI, 20.0-44.1 months), and 21.9 months (95% CI, 12.3-31.6 months) for good, moderate, and poor prognoses, respectively (Figure 2) (good vs moderate, P = .008; moderate vs poor, P = .02; and good vs poor, P = .001). A significant survival benefit for patients who underwent hepatic resection compared with those without resection was seen in each prognostic group. However, of all patients of the original cohort who underwent hepatic resection, only 13 (19.7%) of 66 patients assigned to moderate prognosis and none of 12 patients assigned to poor prognosis were alive at 5-year follow-up. In contrast, 10 (62.5%) of 16 patients with good prognosis were alive at this time (P<.001). The median recurrence-free survival after hepatic resection was 63.2 months (95% CI, 37.4-73.7 months), 23.8 months (95% CI, 16.2-31.5 months), and 6.8 months (95% CI, 0.7-17.5 months) for good, moderate, and poor prognoses, respectively (good vs moderate, P = .16; good vs poor, P<.001; and moderate vs poor, P<.001).
The overall median survival in patients from the validation group was 26.3 months (95% CI, 21.5-31.0 months), and the mean ± SE estimated 1-, 3-, and 5-year cumulative survival was 75.2% ± 3.1%, 40.6% ± 3.3%, and 24.4% ± 5.2%, respectively. Patients who underwent hepatic resection had an overall median survival of 26.7 months (95% CI, 16.4-37.1 months) and a mean ± SE estimated 1-, 3-, and 5-year cumulative survival of 79.3% ± 3.3%, 46.2% ± 5.3%, and 28.2% ± 6.0%, respectively, compared with 13.3 months (95% CI, 8.3-18.3 months) and 57.0% ± 8.5%, 19.2% ± 7.8%, and 0%, respectively, in patients without hepatic resection (P<.001, log-rank test). Median survival in patients assigned to good, moderate, and poor prognoses was 41.8 months (95% CI, 25.4-59.1 months), 27.5 months (95% CI, 17.4-37.6 months), and 15.8 months (95% CI, 10.6-20.9 months), respectively (Figure 3) (good vs moderate, P = .01; moderate vs poor, P<.001; and good vs poor, P<.001). In patients who underwent hepatic resection, median survival was 52.3 months (95% CI, 29.1-67.4 months), 35.3 months (95% CI, 23.8-46.7 months), and 16.3 months (95% CI, 13.7-18.9 months) for good, moderate, and poor prognoses, respectively (good vs moderate, P = .02; moderate vs poor, P<.001; and good vs poor, P<.001). Individual prognostic scores showed significant prognostic value for crude survival by Cox regression analysis (P<.001) in the entire validation group and in patients who underwent liver resection.
At present, radical liver resection is still the only curative treatment for patients with CRCLM. Despite this, fewer than 40% of patients undergoing resection are alive at 5 years.1,8,21,25- 27 It is therefore evident that surgery alone is appropriate in few patients and that attempts should be made to identify those at high risk of developing early recurrence and death, despite having had a major liver resection.
Prognostic scoring for patients with CRCLM was introduced years ago by Nordlinger and colleagues2 from a French multicenter study group. They identified 3 groups of patients with low, intermediate, and high risk for early death based on the evidence of 7 risk factors. Since then, at least 4 other scoring systems have been developed to calculate patients’ prognosis, but scores were always derived from patients who had undergone liverresection.4,5,21,22 However, a prognostic score should ideally become available during preoperative assessment and would therefore involve all patients referred rather than being limited to patients undergoing liver resection. We have chosen a novel approach to develop a prognostic scoring system based on variables that were derived from significant independent prognosticators in “all comers” for consideration of liver surgery.
Prognostic scores were calculated from a mathematical equation based on variables and their correlation coefficients and related to good, moderate, and poor prognoses. When applied in daily clinical practice, mathematical equations may not be as practicable as scores derived by giving 1 point for each risk factor evident.5 However, the use of correlation coefficients as risk multipliers is likely to provide more reliable prediction of survival than singular addition of risk factors present.
Some studies28- 34 have addressed the prognostic value of serum concentrations of alkaline phosphatase and albumin in patients with advanced colorectal cancer. Furthermore, there is evidence that analysis of serum-derived tumor-specific isoforms of alkaline phosphatase may increase the predictive accuracy of this marker.35,36 Apart from CEA, no other blood-derived variable has been used in prognostic scoring in CRCLM, to our knowledge. We found that determination of serum alkaline phosphatase and albumin provides significant independent information about prognosis in all patients referred for surgery and in patients who underwent liver resection. The combination of clinical and biochemical variables may further improve prognostication in patients with malignant neoplasms by considering not only the behavior of the tumor but also the host response.
Three groups with different prognoses were identified among all comers for liver surgery. Although liver resection improved survival prospects in all patients regardless of their prognosis, none of those patients assigned to a poor prognosis were alive at 5-year follow-up. Prognostic scoring systems such as the one presented herein have the potential to improve the accuracy of patients’ selection for surgery and to allocate patients to combined treatment modalities. From our results, it is obvious that patients with high prognostic scores should not undergo liver resection alone rather than being considered for surgery together with an adjuvant or neoadjuvant treatment protocol.12- 14,29 Moreover, extensive preoperative staging involving positron emission tomography and magnetic resonance imaging might be performed in patients who have high scores but are otherwise suitable for major liver resection.37- 41 Similarly, intensities of follow-up may be organized according to individual prognosis.
The robustness of this scoring system has been tested by cross-validation against data from an unrelated international center, which demonstrated similar survival distribution between prognosis groups. It is evident that patients in the validation cohort have not been “unselected” rather than being already considered for liver resection. However, comparable results from prognostication in the original and validation cohorts demonstrate that the presented scoring system can be widely applied to patients with CRCLM referred for liver surgery. In modern medicine, patients are much more involved in the decision-making process about their treatment. Prognostic scoring such as this offers a greater insight into the risk-benefit balance of undergoing liver resection and may inform treatment selection.
Correspondence: Stephen J. Wigmore, MD, FRCSE, Division of Hepatobiliary Surgery, Department of Surgery, The Royal Infirmary of Edinburgh, 51 Little France Crescent, Edinburgh EH16 4SA, Scotland (firstname.lastname@example.org).
Accepted for Publication: July 27, 2004.
Funding/Support: Dr Schindl was supported by the Austrian Science Fund (FWF), Wien, and by the Austrian Surgical Society.