Abstract
Background: We explored and validated the association of postoperative carcinoembryonic antigen (CEA) with disease-free survival (DFS) and overall survival (OS) in stage III colon cancer.
Methods: Patients with stage III colon cancer from the MOSAIC and PETACC-8 trials were enrolled. The relation between CEA and outcomes was continuously modeled with the restricted cubic splines (RCS) method. Association of CEA with outcomes was assessed by the Kaplan–Meier method, with two risk groups among patients with a CEA level ≤5 ng/mL. Multivariate Cox proportional hazard models were constructed.
Results: The CEA level was available in 1,292 (96%) and 2,477 (97%) patients in the discovery and validation cohorts. The RCS analysis confirmed that patients with a CEA level >5 ng/mL were at highest risk of recurrence or death and those with a CEA level ≤5 ng/mL presented a heterogeneous risk population. In the discovery cohort, the 3-year DFS rate was 75%, 65%, and 45% in a group of patients with CEA level of 0–1.30 ng/mL (n = 630), 1.30–5 ng/mL (n = 613), and >5 ng/mL (n = 49), respectively (P < 0.001). CEA was independently associated with endpoints. All findings were confirmed in the validation cohort.
Conclusions: Postoperative CEA level was highly and independently associated with DFS and OS, especially in patients with a CEA level of ≤5 ng/mL, suggesting that this cutoff is not optimal.
Impact: CEA levels should be applied more accurately in future trials and clinical practice.
Introduction
Colon cancer is the third most common cancer in men and women (1, 2). When diagnosed at a localized stage, curative management is proposed to patients including surgery and adjuvant chemotherapy if lymph node involvement is observed (stage III colon cancer). In stage III disease, fluoropyrimidine-based adjuvant treatment improved patient outcomes (3). The addition of oxaliplatin to a fluoropyrimidine chemotherapy backbone has produced additional benefit by increasing the 3-year disease-free survival (DFS) rate (4–6), and the oxaliplatin-containing combination became the worldwide standard adjuvant treatment in the setting of colon cancer. Recently, the findings of the International Duration Evaluation of Adjuvant Therapy collaboration have provided useful information in helping oncologists discuss the duration of adjuvant therapy (7–10). To optimize the value of this treatment, better tumor markers or better awareness of tumor markers to improve the recurrence risk detection rate are needed.
Carcinoembryonic antigen (CEA) is a well-known, low-cost biological tumor marker for colon cancer (11, 12), overexpressed in 90% of colorectal cancer cases (13). The European Society of Medical Oncology (ESMO) guidelines recommend CEA determination in a postoperative follow-up period (14). Previous large studies have shown that elevated preoperative CEA levels are associated with worse prognosis in all stage colon cancer patients (15) and in stage I and II colon cancer patients who did not receive adjuvant chemotherapy (16).
The results by Konishi and colleagues showed that the preoperative CEA level is not a relevant marker of recurrence in localized colon cancer if CEA is normalized after surgery (17), thus underling the importance of postoperative CEA. Another study by Lu and colleagues identified high postoperative CEA level of ≥5 ng/mL as an independent prognostic predictor of early relapse (18). The CEA upper limit of normal (ULN) levels vary between different institutions, ranging from 3.0 to 5.0 ng/mL, with a cutoff of 5.0 ng/mL used by most centers even though this value remains debated (15, 16, 19). Recently, Margalit and colleagues defined preoperative CEA of 2.35 ng/mL as an optimal cut-off point for predicting survival of stage I and II colon cancer (16). Likewise, Kim and colleagues suggested preoperative CEA of 3 ng/mL as the best cut-off value to identify stage III colon cancer patients with low and high risk of death or recurrence (19).
The aims of this post hoc analysis were (i) to assess and precisely validate the postoperative CEA prognostic value for DFS and OS in patients with stage III colon cancer receiving standard adjuvant treatments; (ii) to challenge the standard 5 ng/mL as threshold value to discriminate patients with low and high risk of recurrence or death; and (iii) to determine the additional value of postoperative CEA for the TNM classification to better stratify patients for risk of death and of recurrence.
Materials and Methods
Population
Stage III colon cancer patients for a discovery cohort were selected from the MOSAIC phase III trial (NCT00275210; refs. 4, 20). The MOSAIC study showed a survival benefit with the addition of oxaliplatin to adjuvant 5-fluorouracil (5-FU) and leucovorin (LV; the FOLFOX4 regimen) in patients with colon cancer. External validation cohort was obtained with the PETACC-8 phase III trial population (NCT00265811; ref. 21). The PETACC-8 study failed to show a survival benefit for adjuvant combination of FOLFOX and cetuximab in patients with stage III colon cancer. Both trials were performed in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines and approved by the appropriate Ethics Committees.
Follow-up
In both study cohorts, the radiologic evaluation was made at randomization and every 6 months during the first 5 years of follow-up.
The MOSAIC database was locked on April 25, 2014, and the PETACC-8 database on October 27, 2016.
Data extraction
Demographics, cancer history, pathologic, clinical, biological parameters, and efficacy outcomes (DFS and OS) were prospectively collected at the time of randomization.
Postoperative CEA serum measurements
For both study cohorts, a postoperative CEA value (MOSAIC: <10 ng/mL; PETACC-8: ≤1.5× ULN) was an inclusion criterion. The serum CEA measurements were not centralized. In the PETACC-8 study, the ULN for CEA was available for all patients; the median CEA value was 5 ng/mL [interquartile range (IQR), 3.8–5].
Statistical analysis
Median (IQR) values and proportions (percentage) were used for continuous and categorical variables, respectively. Median and proportions were compared using the Wilcoxon–Mann–Whitney test and the χ2-test (or the Fisher exact test, if appropriate), respectively.
DFS was defined as the time between randomization and local/distant relapse, second colorectal/rectal occurrence, or death, whichever occurred first. Alive patients without relapse and second colorectal cancer were censored at the date of their last follow-up. OS was defined as the time between randomization and death from any cause. Patients known to be alive were censored at the date of their last follow-up.
DFS and OS were estimated using Kaplan–Meier method and described using median or rate at specific time points with their 95% confidence intervals (95% CI). Follow-up was calculated using the reverse Kaplan–Meier method (22).
When used continuously, the association between CEA and survival was investigated using the fractional polynomials method (23–25) and was validated by the restricted cubic splines method with graphical evaluation. Different cutoffs of CEA level in the discovery cohort were explored to investigate an association between CEA and survival (categorical). These were based on the percentiles (median and tercile) and optimal cut-off point (log-rank maximization) methods (26).
The association of demographic, clinical, biological, and molecular factors with survival was assessed with univariate Cox proportional hazards model. Variables with P values of less than 0.10 and/or clinically relevant variables were entered into an intermediate multivariable Cox regression model. Final multivariable model was constructed with statistically significant variables of the intermediate model. The correlation between variables was verified before construction of the multivariate models, in order to deal with potential colinearity. All multivariate models were stratified on the basis of adjuvant treatment received. Hazard proportionality was assessed for each categorical variable using the Schoenfeld residual-based test and a graphical method.
The predictive value and the discrimination ability of the final model were assessed with the Harrell's concordance index (C-index; ref. 27). The robustness of the statistical models was assessed by multiple imputation procedure (assessment of potential bias arising from missing data for parameters involved in the multivariable model; ref. 28).
The differential treatment effects on DFS and OS among CEA risk groups were evaluated with an interaction term in the Cox proportional hazards model.
All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC) and R software version 2.15.2 (R Development Core Team; http://www.r-project.org). P values of less than 0.05 were considered statistically significant, and all tests were two-sided.
Results
Characteristics of patients
In the discovery cohort, 1,347 patients with stage III colon cancer from the MOSAIC study were enrolled. Postoperative CEA was available in 1,292 (95.9%) patients; 1,243 (96.2%) had a CEA level below 5 ng/mL. The median time between surgery and CEA measurement was 4.3 weeks and the median follow-up was 9.6 years (Table 1).
Characteristics of the discovery (n = 1,347) and external validation (n = 2,552) populations
In the external validation cohort, 2,552 patients with stage III colon cancer from the PETACC-8 study were enrolled. The median time between surgery and CEA measurement was 5.1 weeks and the median follow-up was 5.85 years (Table 1).
Association of postoperative CEA in its continuous form with survival
In the discovery cohort, the restricted cubic splines approach showed that patients with a CEA level of >5 ng/mL were at highest risk of recurrence or death. Patients with a CEA level of ≤5 ng/mL did not present a homogeneous risk population. A gradual risk suggested a square root relation between CEA and DFS (Fig. 1A). The square root transformation was the best to model the relation between CEA and survival endpoints with the fractional polynomials method. Similar observations were made for the external validation cohort (Fig. 1B) and for OS endpoint (Supplementary Fig. S1). These results allowed to identify different risk populations among patients with a level of CEA lower than 5 ng/mL.
Restricted cubic splines modeling of hazard ratio (HR) for DFS as a function of postoperative CEA level. HR for DFS in stage III colon cancer as a function of postoperative CEA level in the discovery cohort (A) and in the validation cohort (B). Association between CEA and DFS was modeled using restricted cubic splines, showing a square root relation between those variables. Estimates were derived from Cox model. The purple area around the blue line represents the 95% CI.
Survival according to postoperative categorical CEA
The 5 ng/mL cut-off value of CEA allowed to identified two populations with distinct DFS risk profiles (CEA ≤ 5 ng/mL population: the 3-year DFS rate of 70%; and CEA > 5 ng/mL population: the 3-year DFS rate of 45%; Fig. 2A).
Association between CEA and DFS in the discovery and validation cohorts. Kaplan–Meier survival curves for DFS of stage III colorectal cancer patients in the MOSAIC cohort according to the standard CEA levels (A; ≤ or >5 ng/mL), in the MOSAIC cohort according to the new CEA levels (B; ≤1.30 ng/mL, 1.30–5 ng/mL, >5 ng/mL), in the PETACC-8 cohort according to the standard CEA levels (C; ≤ or >5 ng/mL), and in the PETACC-8 cohort according to the new CEA levels (D; ≤1.30 ng/mL, 1.30–5 ng/mL, >5 ng/mL). P values were estimated using the log-rank test. The discrimination ability of the CEA level was evaluated with the C-index. HR and 95% confidence intervals were obtained with a Cox regression method.
Considering the important number of patients with CEA ≤ 5 ng/mL, the median and tertile approach were used to determine 2 and 3 populations in these patients (Table 1). Figure 2B shows a distinct DFS risk profile in the two median based populations identified in patients with CEA ≤ 5 ng/mL. The 3-year DFS rate was 75%, 65%, and 45% for CEA < 1.30 ng/mL, CEA 1.30 to 5 ng/mL, and CEA > 5 ng/mL groups, respectively (P < 0.0001). The optimal cut-off value determined by the Hothorn and Lausen method (Supplementary Fig. S2) was equal to 1.30 ng/mL for patients with a CEA level of ≤ 5 ng/mL, reinforcing the choice of CEA 1.30 to determine different risk populations in these patients. These results were fully confirmed in the external validation cohort (Fig. 2C and D) and were also validated for OS in both cohorts (Supplementary Fig. S3). Similar results were observed for DFS and OS with the tertile approach for both study cohorts (Supplementary Fig. S4).
Consistent observations were found in a subgroup analysis of treatments and of times between surgery and CEA measurement for both cohorts (Supplementary Figs. S5–S8).
Patients' characteristics according to CEA level groups are presented in Supplementary Tables S1 and S2.
Postoperative CEA level as independent prognostic factor for survival
In univariate and multivariate analyses, CEA level was found to be significantly associated with DFS. The independent prognostic value of CEA was also confirmed in multivariate analysis. Patients with CEA between 1.30 and 5 mg/mL had a HR of 1.67 (95% CI, 1.28–2.19) compared to those with CEA < 1.30 ng/mL (Table 2). Similar results were obtained for OS (Table 2).
Univariate and multivariate Cox proportional hazards model for DFS in the discovery cohort
Univariate and multivariate Cox proportional hazards model for OS in the discovery cohort
In the sensitivity analysis, multiple imputation procedure provided analogous results in the discovery cohort (Supplementary Tables S3 and S4). All results were replicated in the external validation cohort (Table 3; Supplementary Table S5).
Multivariate proportional hazards Cox model for DFS in the validation cohort
Additional value of postoperative CEA for DFS risk stratification within TNM risk subgroups
In the discovery cohort, 729 (54.3%) and 614 (45.7%) patients had low-risk (T3N1) and high-risk (T4 or N2) stage III colon cancer, respectively. As expected, DFS differed according to these subgroups (Supplementary Fig. S9). Among the low and high-risk subgroups, the postoperative CEA level classification allowed to identify populations with different DFS and OS risk profiles (Fig. 3; Supplementary Figs. S9 and S10). Similar results were obtained in the external validation cohort for both DFS and OS (Fig. 3; Supplementary Fig. S9).
Association between CEA and DFS within stage III colon cancer risk groups. Kaplan–Meier survival curves for DFS of stage III colorectal cancer patients in the MOSAIC low-risk (T3N1) group (A), in the MOSAIC high-risk (T4 or N2) group (B), in the PETACC-8 low-risk group (C), and in the PETACC-8 high-risk group according to the new CEA levels (C; ≤1.30 ng/mL, 1.30–5 ng/mL, >5 ng/mL). P values were estimated using the log-rank test.
Chemotherapy regimen benefit across CEA risk groups
In the discovery cohort, the survival benefit from the addition of oxaliplatin to 5-fluorouracil was not different across different CEA groups in term of DFS and OS. No statistically significant interaction between CEA groups and treatment the arm (5-FU/LV vs. FOLFOX) was found (P = 0.5371 and P = 0.6761, respectively; Supplementary Fig. S11).
In the validation cohort, no interaction between the CEA risk groups and the treatment arm (cetuximab-FOLFOX vs. FOLFOX) in terms of DFS and OS was found (P = 0.8629 and P = 0.2412, respectively).
Discussion
This study confirmed strong prognostic value of postoperative CEA level for DFS and OS in two large cohorts of patients with stage III colon cancer treated with adjuvant chemotherapy. Patient with a CEA level of >5 ng/mL had a worse risk profile for recurrence and death. Distinct risk profiles were identified in patients with CEA ≤ 5 ng/mL currently considered as a single population with normal CEA values.
One of the questions that arises from our results is the choice of the CEA cutoff for risk prediction. Kim and colleagues showed in their recent retrospective study of 965 patients with stage III colon cancer that the optimal cutoff for preoperative CEA was 3 ng/mL in the training and validation sets (19). Other studies evaluating preoperative CEA also showed that colon cancer patients with CEA levels inferior to 2.5 or 2.7 ng/mL had better OS and DFS (29–31). Recently, in a large study of 45,449 patients with stage I and II colon cancer not receiving adjuvant chemotherapy, preoperative CEA level of 2.35 ng/mL was defined as the best cutoff to predict survival (16). The results of these studies suggest that a CEA value of 5 ng/mL is not optimal. We identified considerable DFS and OS heterogeneity in the population with postoperative CEA ≤ 5 ng/mL, providing new data in favor of a lower CEA cutoff in the postoperative setting for stage III colon cancer. Defining a lower cutoff for postoperative CEA will allow to construct more accurate prediction models and will better stratify patients for adjuvant treatment.
Nicholson and colleagues showed that the sensitivity of a CEA level of 2.5 ng/mL is better than the standard 5 ng/mL value (82% vs. 71%), but with rather low specificity (80% vs. 88%; ref. 32) Nevertheless, in adjuvant colorectal cancer, higher sensitivity is more relevant for detecting disease recurrence. If the sensitivity is good for a lower CEA cutoff, the question about reproducibility of the measures below 5 ng/mL has to be addressed. Analytic sensitivity of most automated immunoassay systems for measuring CEA is under 1 ng/mL. For instance, analytic sensitivity of the Abbott immunoassay used in our center is 0.5 ng/mL, whereas for Beckmann immunoassays, which are also widely used, is less than 0.7 ng/mL. Thus, reliable management decisions can be made with CEA levels below 5 ng/mL.
Recently, molecular and immunologic prognostic factors in predicting colon cancer patients' prognosis have been identified (33–36). In our study, the postoperative CEA performance seems similar to these molecular/immunologic scores for DFS and OS risk stratification in patients with stage III colon cancer. These observations and the fact that CEA is a low-cost biological tumor marker reinforce its interest.
With the breakthrough of liquid biopsy, circulating tumor DNA (ctDNA) has been assessed in early-stage colon cancer (37–39). Very high prognostic value of ctDNA for OS and RFS has been showed in very small cohorts (37, 39). However, no study to date has assessed the prognostic value of ctDNA in stage III colon cancer. Currently, three studies are ongoing (NCT03416478, NCT02842203, NCT03312374). Our results suggest that prognostic value of ctDNA should absolutely be evaluated together with CEA as a covariate in multivariate analysis. Similarly, Lu and colleagues showed that combining postoperative CEA levels assessment with the presence of postoperative circulating tumor cells may be valuable tool in predicting early relapse and survival rate in patients with stage II and III colon cancer (18). Such combination was also found to be of interest for better disease prediction in the study of non–small cell lung cancer (40).
We identified the additional prognostic value of postoperative CEA for survival endpoints among conventional TNM risk classification. Recently, the results of the IDEA study suggested that shorter adjuvant chemotherapy could be administered in the subgroup of patients with T3/N1 disease (7–10). However, given that these results are still debated, postoperative CEA level might assist in clinical decision-making in this setting.
Although prediction tools already exist in localized colon cancer (41–44), these do not include postoperative CEA level. In the light of our results, a careful consideration of this measurement should be made in scoring tools to predict DFS and OS for localized colon cancer in the future.
Our study has several strengths. We evaluated postoperative CEA in adjuvant stage III colon cancer that was of more interest than preoperative CEA. This study was constructed in the framework of two large international multicenter phase III trials offering a broad spectrum of reliable parameters including molecular status at inclusion and the largest postoperative CEA assessment data in stage III colon cancer. Moreover, patients in these studies were treated with 5-fluorouracil and leucovorin in combination with oxaliplatin in majority of cases (50%). We built our analysis in a rigorous methodologic framework respecting the transparent reporting of the multivariate model TRIPOD statement (25). Showing the same results for both cohorts, we confirmed the robustness of our data.
The current study also has some limitations. Given that CEA level was an inclusion criterion in the MOSAIC and PETACC-8 trials, this analysis includes a low number of patients with high CEA. Importantly, CEA level may be affected by factors that were not available in our cohorts, mainly smoking status and other comorbidities (45). However, the robustness and the magnitude of the association observed support the notion of a cancer-derived effect on the outcomes. A second potential limitation is that in our study CEA assessment is established within different ranges, according to each manufacture, and most of the time low threshold values are considered as insufficiently sensitive. The large DFS and OS differences between subgroups of patient with CEA <5 ng/mL suggest that the findings apply across CEA values obtained from a broad range of laboratories. Despite the heterogeneity of the adjuvant regimens received and the multiples laboratories that performed the CEA measurement, postoperative CEA identified in the overall population was still relevant in all subgroups.
To conclude, in these two large cohorts, we showed that the postoperative CEA level is a highly independent prognostic factor for DFS and OS. We further confirmed that patients with CEA ≤5 ng/mL should not be considered as a single-risk population. These results support a routine postoperative CEA testing, suggesting its future interest for optimizing clinical trials design and for defining risk-adapted and personalized strategies for stage III colon cancer management.
Disclosure of Potential Conflicts of Interest
G. Folprecht reports receiving a commercial research grant from Merck-Serono (for a clinical study), has received speakers bureau honoraria from Merck-Serono, Roche/Genentech, Sanofi-Aventis, Lilly, Amgen, and is a consultant/advisory board member for Merck-Serono, Bayer, MSD, BMS, Shire, Roche/Genentech, Servier, Laboratoire HRA Pharma, Amgen, Mundipharma. J. Tabernero is a consultant/advisory board member for Array Biopharma, AstraZeneca, Bayer, BeiGene, Boehringer Ingelheim, Chugai, Genentech, Inc., Genmab A/S, Halozyme, Imugene Limited, Inflection Biosciences Limited, Ipsen, Kura Oncology, Lilly, MSD, Menarini, Merck Serono, Merrimack, Merus, Molecular Partners, Novartis, Peptomyc, Pfizer, Pharmacyclics, ProteoDesign SL, Rafael Pharmaceuticals, F. Hoffmann-La Roche Ltd, Sanofi, SeaGen, Seattle Genetics, Servier, Symphogen, Taiho, VCN Biosciences, Biocartis, Foundation Medicine, HalioDX SAS, and Roche Diagnostics. T. Hickish is a medical director at IQHealth Tech and is a consultant/advisory board member for Sobi International and Lilly. T. André has received speakers bureau honoraria from Sanofi and is a consultant/advisory board member for Roche. D. Vernerey is a consultant/advisory board member for OSE, HalioDX, Pfizer, and Jansen. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: E. Auclin, J. Taieb, C. Louvet, J. Tabernero, D. Vernerey
Development of methodology: E. Auclin, J. Taieb, D. Vernerey
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Lepage, T. Aparicio, R. Faroux, E. Mini, G. Folprecht, R. Salazar, M. Banzi, C. Louvet, J.-L. Van Laethem, J. Tabernero, T. Hickish, A. de Gramont, T. André
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Auclin, J. Taieb, E. Mini, R. Salazar, M. Benetkiewicz, J. Tabernero, T. André, D. Vernerey
Writing, review, and/or revision of the manuscript: E. Auclin, J. Taieb, C. Lepage, T. Aparicio, R. Faroux, E. Mini, G. Folprecht, R. Salazar, M. Benetkiewicz, C. Louvet, J.-L. Van Laethem, J. Tabernero, A. de Gramont, T. André, D. Vernerey
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Tabernero, T. André
Study supervision: J. Tabernero, D. Vernerey
Other (providing final approval of the version to be submitted): M. Benetkiewicz
Acknowledgments
We thank Pr. Jean-Paul Feugeas and Dr. Laurent Bermont for their help regarding the automated immunoassay systems.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Footnotes
Cancer Epidemiol Biomarkers Prev 2019;28:1153–61
- Received August 6, 2018.
- Revision received November 16, 2018.
- Accepted April 16, 2019.
- Published first July 1, 2019.
- ©2019 American Association for Cancer Research.
References
Volume 28, Issue 7
