Characteristics of carcinoembryonic antigen-producing colorectal cancers: A population based study

Objectives: Serum carcinoembryonic antigen is a tumor marker often found to be elevated in colorectal cancer patients. Elevated carcinoembryonic antigen has been strongly associated with poor prognosis. However, little is known about the patient and tumor characteristics between carcinoembryonic antigen-secreting and non-secreting tumors. Methods: We performed a retrospective analysis of all patients (N=164,187) in the Surveillance, Epidemiology and End Results database diagnosed with colorectal adenocarcinoma from 2010 to 2014. All patients were designated as having either positive/elevated (C1) or negative/normal (C0) pretreatment serum carcinoembryonic antigen level. Results: Of the 164,187 patients, 68,833 (57.0%) had available carcinoembryonic antigen information, and 33,412 (48.5%) had positive/elevated (C1) antigen levels. Median age was 65 years, and 36,464 (53.0%) were male. Patients with C1 cancers were more likely to be female (Odds ratio 1.06), black (Odds ratio 1.62), separated or never married (Odds ratios 1.50 and 1.49, respectively), higher grade (Odds ratios 1.35, 1.64, and 1.72 of moderately, poorly, and undifferentiated cancers, respectively), and of signet ring cell histology (Odds ratio 1.47) compared to males, whites, married participants, well differentiated grade, and adenocarcinoma histology respectively (P<0.001). Multivariate analysis showed that non-Caucasian race, female gender, unmarried status, distal to sigmoid colon location, increasing tumor invasion beyond muscular layer, increasing nodal involvement, and presence of metastases were independent factors associated with the C1 diagnosis. Conclusions: About half of all colorectal adenocarcinomas are associated with elevated pre-treatment serum carcinoembryonic antigen levels. Our study is the first nationwide population-based study quantifying the prevalence of serum carcinoembryonic antigen elevation in the colorectal cancer population, and identifying patient and tumor characteristics associated with elevated carcinoembryonic antigen.


INTRODUCTION
In 2018, an estimated 140,250 Americans will develop colorectal cancer (CRC) and 50,630 will die from the disease (1). Carcinoembryonic antigen (CEA) is a serum tumor marker often found in colorectal cancer patients, and elevated serum CEA has been strongly associated with poor oncological prognosis (2,3). However, not all colorectal cancer patients produce CEA, and little is known about the patient and tumor characteristics between CEA-secreting and non-CEA-secreting tumors. The TNM staging system of the American Joint Committee on Cancer (AJCC) utilizes the size and local extent of the primary tumor (T-stage), regional lymph node involvement (N-stage), and presence or absence of metastasis (M-stage) as core elements for staging, given the importance of these components in estimating oncological prognosis (4). With increasing evidence to show that the highest pretreatment serum CEA level is associated with poor prognosis, it has also been recommended to be included in the staging system for colorectal cancer (5)(6)(7). Incorporation of CEA into the TNM staging system has shown to significantly impact survival estimates (2,3). However, the proportion of CRC patients who have elevated CEA levels is unknown. In this study, we aim to determine the prevalence of elevated serum CEA amongst patients with colorectal adenocarcinoma and identify the factors associated with such elevation. www.onk.ns.ac.rs/Archive • Volume 25 • Issue 1 • January 2019 adenocarcinomas, 68,833 patients (57.0%) had CEA information available for analysis. Age in years, sex, and marital status when the patient was first diagnosed with cancer were obtained. Race information was available as whites; blacks; Asian or Pacific Islanders; American Indians or Alaska Natives; and unknown. Grade of primary tumor, when known, was available as well differentiated; moderately differentiated; poorly differentiated; undifferentiated; or unknown. Tumor histology was coded in the database as per the International Classification of Diseases for Oncology, Third Edition (ICD-O-3) and was accessed using the variable "Site and Morphology. ICD-O-3 Hist/behav, malignant."

Statistical Analysis
Categorical covariates were summarized with frequencies and percentages, and continuous covariates were summarized with means, medians, minimums, maximums, and standard deviations. We analyzed the data using univariate and multivariate logistic regression to model the odds of a patient having elevated serum CEA levels (i.e. being C1). Exponentiated maximum-likelihood estimates on model coefficients were reported along with p-values, based on the Wald test. From these models we obtained odds ratios (OR) and corresponding 95% confidence intervals (95% CI). As serum CEA elevation is strongly correlated with probability of metastatic disease at diagnosis, we performed further univariate and multivariate analyses on the subset of metastatic patients (defined as patients with stage IV CRC). Statistical analysis was performed using SAS software (9).

Patient and Tumor Characteristics
We performed a univariate analysis of CRC patients characterized by CEA status (C-stage) ( Table 1). Of the 68,833 (57.0%) patients with available CEA information, 33,412 (48.5%) patients had positive/elevated (C1) antigen levels. Median age was 65 years and 36,464 (53.0%) patients were male. Compared to C0 cancers, patients with C1 cancers were significantly more likely (P<0.001) to be female and non-white. On analysis of the marital status, patients who were single, divorced, widowed or separated at the time of diagnosis was associated with more than 50% chance of having elevated CEA compared to married individuals. However, individuals who were unmarried but lived with their significant other were significantly less likely to be C1 compared to married individuals. Analysis of tumor characteristics associated with CEA elevation showed that compared to C0 cancers, C1 cancers were significantly more likely (P<0.001) to be of higher grade and of more advanced stage, within each TNM category, as well as overall AJCC stage (Table 1). Signet ring cell pathology was the most common morphology associated with C1 cancers, while adenocarcinoma in preexisting adenomatous polyps and medullary histology were the most common morphologies associated with C0 cancers.

Risk Factors for CEA Secretion
We performed a multivariate analysis to identify independent factors associated with diagnosis C1 tumors ( Figure 2). African-American and Asian/Pacific Islander race were more likely to have elevated CEA. Female gender was also associated with slighted higher risk of C1 tumors. Unmarried marital status variables, as described in the univariate analysis, were also associated with C1 tumors. Other factors that emerged as independent factors associated with CEA elevation included left-side tumor location (sigmoid, rectosigmoid junction, and rectum); increasing tumor invasion beyond the muscularis propria (T stage); increasing extent of nodal disease (N stage); and presence of metastases (M stage). Increased grade was not associated with diagnosis of C1 tumors on multivariate analysis.

CEA Elevation in Metastatic Cancers
Presence of metastases was strongly associated with the highest risk of C1 tumors, but all patients with metastatic disease do not have elevated serum CEA levels. Hence, we performed an analysis of the subset of patients with metastatic disease only, to identify the factors independently associated with diagnosis of C1 cancers (Figure 3). We included site of metastatic disease as a factor in the analysis. C1 patients with metastases were significantly more likely to have metastasis to liver (OR 2.86) and lungs (OR 1.53) compared to brain metastasis.

DISCUSSION AND CONCLUSIONS
Colorectal cancer is the third leading cause of cancer and cancer-related mortality in the United States. There have been multiple staging systems in the last several decades, in an attempt to categorize patients into different prognostic groups (i.e. 'stages') and streamline management options. In 1978, the AJCC published their first edition of their TNM system and today, it is the most widely used method for staging colorectal cancers, as well as all other types of cancer (10). The T stage describes the size and local extent of the primary tumor; the N stage describes regional lymph node involvement; and the M stages describes the presence or absence of metastasis (4). The overall stage is determined by combining the scores from these individual categories. By using serum CEA elevation to represent poor prognosis, we can describe the subset of people who do not have this marker and begin to uncover the environmental and genetic influences that play a protective role. Having a better understanding of the patient and tumor characteristics associated with this aggressive subset of colon cancers will help to develop more personalized treatment plans and effective management strategies.
Using survival data analyses largely from the Surveillance, Epidemiology, and End Results (SEER) database, the AJCC regularly updates the text with changes to TNM categories, criteria, and prognostic stage groups, and in 2010 they published their seventh edition (11,12). The limitation with the TNM system is that it is a purely anatomical method of assessing prognosis, without taking into consideration other biological markers associated with oncological natural history and pattern of aggressive behavior. The AJCC is beginning to incorporate non-anatomical prognostic factors into their site-specific staging guidelines (4, 13). Examples include age in thyroid cancer, histologic grade in esophageal cancer and sarcoma, mitotic rate in gastrointestinal stromal tumors, prostate-specific antigen levels and Gleason score in prostate cancer, and serum tumor markers in testicular cancer (11). Several prominent organizations, including the Colorectal Working Group of the AJCC, the American Society of Clinical Oncology (ASCO), the European group on tumor markers recommended the addition of serum carcinoembryonic antigen (CEA) levels to colorectal staging guidelines, (5-7) but the AJCC has yet to incorporate it into colorectal staging guidelines. CEA is normally secreted on the apical side of colorectal epithelial cells and excreted with feces. However, in CRC, due to loss of polarity in cancer cells, CEA could get expressed on the whole tumor cell surface, allowing for secretion into blood vessels (14). High preoperative serum CEA levels have been well-correlated with increased recurrence and decreased survival in CRC patients (2,(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25). Furthermore, elevated serum CEA has shown to be a more effective predictor of worse oncological prognosis than N stage (3,26) and hence, an important factor that could potentially guide the course of treatment (27). CEA is a highly specific serum tumor marker for colorectal cancer, but it is not a sensitive tumor marker because not all colon cancer patients have elevated serum CEA levels (28,29). While it is generally associated with larger tumor size and more  extensive metastasis, (30) we have shown in this study that approximately 50% of CRC cases do not have elevated CEA levels. Another recent US study reviewing the National Cancer Database (NCDB) also reported that approximately 66% of the CRC patients had normal CEA levels (31). Their study emphasized the importance of evaluating preoperative serum CEA, as it was associated with increased hazards of mortality. However, the study did not associate these findings with any specific race or gender. In our study, African-American race was found to be an independent risk factor associated with diagnosis of colorectal cancer with elevated CEA. Supplemental data from another recent NCDB study on early onset CRC also showed that black patients were the most likely to have elevated CEA levels (32). These US population studies strongly support our findings related to elevated CEA as an important prognostic biomarker for improved disease outcome among black patients with CRC. These studies on serum CEA in the clinics can be complemented or validated by tissue CEA expression (33). Many of the previous CRC epidemiological studies including SEER data studies have not focused on gender associations with any biomarkers in CRC. Our findings related to female gender being an independent risk factor associated with elevated CEA level are unique and need further stratification to show any association with age, race or ethnicity. Analysis of SEER data has shown that marital status at diagnosis has been well-associated as an independent factor for poor prognosis in many types of cancers, (34)(35)(36)(37)(38) including colorectal cancer (39). Often, this is explained by social support mechanisms available for patients who are married at the time of diagnosis. Our study throws new light to a rather new explanation -the possibility that marital status is related to the diagnosis of biologically different cancers or other socioeconomic conditions associated with marital status may also contribute. But, such analysis or discussion is beyond the scope of this manuscript. There are several limitations in our study, and most of them are inherent to any large population-based database retrospective analysis. The SEER database represents only 28% of the United States population via 20 different registry sites. Although such extensive coverage is appropriate for analyzing incidence and survival on a large scale, uniformity, consistency and accuracy of data entry cannot be verified by authors. Several data points have unrecorded entries, especially for race, marital status, histological grade, site of metastasis, and CEA information. However, we decided to keep all these patients and categories data-unknown patients as a variable in analysis. A drawback of our unusually large sample size is that such large sample sizes may drive even small differences to positive statistical significance, while they may not be clinically significant in daily practice. Univariate analysis of tumor characteristics showed that C1 cancers were significantly more likely to be located in the sigmoid colon, rectosigmoid junction, or rectum and have a higher histological grade. The latter may be explained by the differences in venous drainage anatomy between colon and rectum. Colonic venous blood drains into either the inferior or superior mesenteric vein, which later drains into the portal venous system that enters the liver, where CEA is metabolized. On the other hand, rectal venous drainage enters the systemic venous system via the inferior vena cava bypassing hepatic metabolism, thereby leading to relatively higher levels of serum CEA for rectal tumors compared to colonic tumors. This is also reflected in our study results, which shows that diagnosis of C1 tumors is associated with more distal location of the primary tumor. In other words, tumors located in the ascending, transverse, and descending colon may need to produce quantitatively more CEA (i.e. to be larger in size and/or more invasive in terms of extent of the tumor) in order to demonstrate the rather equivalent levels of serum CEA as tumors located in the sigmoid colon or rectum. Further studies are needed to study this possibility in more detail.
Although CEA is a well-established marker of prognosis in CRC, about 15% of our initial patient cohort did not have the assay ordered prior to first course of treatment, and 27% were missing CEA information all together. With further accumulation of evidence to support prognostic value of CEA, we hope that the use of presence of elevation of serum CEA on diagnosis (i.e., C-stage) would become more prevalent. This will also drive more physicians to routinely order serum CEA levels on diagnosis.
In brief, serum CEA measurement on diagnosis should be considered standard of care in the initial workup of colorectal malignancies. Future studies studying the factors associated with the non-availability of serum CEA level prior to treatment might shed light on the diversity of quality of health care delivery in the USA. This is the first and largest database-based epidemiological study quantifying the prevalence of C1 cancers in the colorectal cancer population.

ACKNOWLEDGMENTS
Statistical analyses were performed by personnel from the Oklahoma Shared Clinical & Translational Resources Institute at the University of Oklahoma Health Sciences Center. This institute is supported by the National Institutes of Health, National Institute of General Medical Sciences (U54GM104938).

Declaration of Interests
Authors declare no conflicts of interest