Combination of preoperative CA19-9 levels, cell differentiation, and age predicts survival for patients with gastric cancer before surgery

Abstract Gastric cancer (GC) is very common in China, posing a threat to public health, with high morbidity and mortality ranks. Tumor-node-metastasis (TNM) staging system is routinely used to predict prognosis for patients with GC but only available after surgery. Therefore, searching for markers that can predict prognosis of GC patients before surgery is desirable to assist management decisions preoperatively. Among 322 GC patients followed-up for 128 months, the tumor markers alpha fetoprotein, carcinoembryonic antigen, carbohydrate antigen 19-9 (CA19-9), carbohydrate antigen 15-3 and carbohydrate antigen 72-4 of 168 patients were detected before surgery, and their impact on survival was analyzed. Four major findings were revealed: (1) Preoperative examined CA19-9 levels and cell differentiation using endoscopic biopsies were positively correlated with lymphatic metastases and TNM stages obtained after surgery. (2) Kaplan-Meier analyses demonstrated that poor survival of patients with GC was associated with higher CA19-9 levels, poor cell differentiation, and older age. (3) Cox multi-factorial regression analyses indicated that, in terms of predicting overall survival for GC patients, preoperative CA19-9 level, cell differentiation and age were independent factors, respectively, comparable to postoperative TNM staging system. (4) Using receiver operating characteristic curve analysis, we first revealed that preoperative CA19-9 levels and cell differentiation had the impact weights (IW) on survival comparable to postoperative TNM components. These findings suggest that preoperative CA19-9 levels, cell differentiation and age are useful prognostic related markers for GC patients, superior to postoperative TNM system in terms of timing for management. We propose that, assisted by clinical imaging, a comprehensive utilization of these preoperative survival-predictors may help formulate individualized medical management for GC patients such as surgical strategy, optimal chemotherapy and radiotherapy, and appropriate follow-up intervals after surgery.


Introduction
Gastric cancer (GC) is posing a threat to public health worldwide with higher incidence and mortality in less developed countries. GLOBOCAN 2018 estimates 456,124 new GC cases and 390,182deaths occurring in China, accounting for 44.1% and 49.9% of the world, respectively, and the morbidity and mortality rank second in malignant tumors9. [1,2] Furthermore, there have been increased incidence and mortality of gastric cancer in China. [3] Without routine screening, early diagnosis of GC is difficult and treatment of late-stage GC is often unsatisfactory, resulting in poor survival. Cancer invasion and metastasis are the major causes of death for GC patients and surgical removal of in situ tumors is a routine management. Tumor-node-metastasis (TNM) staging system, based on T, lymph node metastasis (N) and distant metastasis (M), is well established to predict prognosis for GC patients. [4] However, pathological TNM staging (pTNM) cannot be obtained before surgery and the prognosis of patients with GC before surgery is evaluated difficultly. Although with lower accuracy, preoperatively defined clinical TNM (cTNM) may be correlated with prognosis of patients with GC. Therefore, Pre-assessment of prognosis for patients who underwent surgery, such as GC patients, has important clinical implications: (1) Pre-assessing of prognosis for patients with GC serves as a warning sign that may assist formulation of surgical planning which in turn may improve the quality of surgery; (2) Predicting prognosis may help design appropriate postoperative radio-chemotherapy [5] and (3) Determine optimal intervals of follow-up. It is desirable to identify markers that can predict poor prognosis before surgery.
There are several categories of markers that can be obtained before surgery, such as age, [6] cell differentiation, [7] tumor markers amongst others. Tumor markers, including specific proteins, glycoproteins, enzymes and hormones, originate from primary neoplasms and occasionally from tissues/organs affected by the cancer. Tumor markers are absent or present in small quantity in healthy tissues. [8] A number of blood derived tumor markers are routinely tested before surgery for GC patients, such as carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), alpha fetoprotein (AFP), carbohydrate antigen 72-4 (CA72-4), and carbohydrate antigen . Serum levels of tumor markers play significant roles in cancer diagnosis, the evaluation of response to chemotherapy, the early detection of recurrence and metastasis. [9][10][11][12][13] In GC patients, adverse prognosis has been suggested to be associated with poorly cell differentiation obtained via gastroscopic biopsy.
Our previous studies have revealed that preoperatively determined body mass index, blood albumin, triglycerides, bilirubin, and aminotransferase can predict survival for GC patients. [14,15] Therefore, We have hypothesized that tumor markers obtained before surgery, individual or combined, may also be able to predict prognosis among GC patients. This study has tested the hypothesis and obtained several interesting findings: (1) Levels of CA19-9 and degree of cell differentiation are positively correlated with the number of metastatic lymph nodes and TNM clinical stages. (2) Poor survival of GC patients was associated with higher CA19-9 levels, poorly cell differentiation, and older age.
(3) In terms of predicting overall survival among GC patients, CA19-9 level, cell differentiation and age are independent factors, respectively, similar to postoperative determinants, such as M and TNM stages. (4) Because CA19-9 levels, cell differentiation and age are obtained before surgery, these findings will not only provide advice for the doctor to choose a treatment strategy for the patient, but also assist in the formulation of individualized medical treatment after surgery.

Ethical standards
This study was approved by the Institutional Ethics Review Board (IERB No. SHZ2011LL10) at our First Affiliated Hospital, Shihezi University School of Medicine. Patients' informed consents were obtained orally by phone during follow-up communications and standard university hospital guidelines in accordance with the Declaration of Helsinki including confidentiality and anonymity were followed in the handling and publication of patients' tissues. IERB board agreed to obtain informed consent by phone.

Informed consent
Patients' informed consent was obtained orally by phone during follow-up communications.

Classifications of gastric cancer
Gastric adenocarcinoma was classified according to the histopathological classification criteria of the World Health Organization (WHO, 2000) as follows: highly, moderately, and poorly differentiated adenocarcinomas. [16] TNM (tumor-node-metastasis) clinical staging (I-IV), depth of invasion (T1-T4), lymphatic and distant metastases were defined according to the American Joint Committee on Cancer (the 8th edition, 2017) [17] (Table 1).

Examination of tumor markers
Before surgery, heparinized blood was drawn from patients after fasting overnight, centrifuged and the serum was stored at -20°C until test. Tumor markers (CEA, CA19-9, AFP, CA72-4, carbohydrate antigen 15-3) were tested using electrochemical luminescence method on a Roche E170 Automatic Analyzer  [18] the cut-off values of normal reference ranges for the tumor markers tested before surgery were shown in Table 1.

Statistical analyses
Data analyses were performed using the statistical software package SPSS (version 20.0, IBM Corporation, Armonk, NY). The spearman ranking correlation method was employed to identify correlations among preoperative and/or postoperative variables. Chi-square (x 2 ) test and Fisher exact test were used to analyze differences between clinicopathological variables. Univariate and multivariate analyses of the relative prognostic importance of the parameters were performed using the Cox proportional hazards model. Kaplan-Meier method was used to calculate and plot survival curves, and a two-sided log-rank test was used to evaluate differences in survival curves. In addition, we introduced the receiver operating characteristic (ROC) curve and the area under the curve (AUC) to quantify the impact weights (or powers) of survival-predicting factors in terms of their differential abilities to predict survival. The larger the AUC, the more powerful the factor can be used to predict prognosis. All P values were 2-sided and differences with P < .05 were considered statistically significant.

GC patients with advanced TNM stages show poor survival prognosis
This cohort included 168 GC patients with 95 alive and 73 dead by the last follow-up. To validate the reliability of clinical data in terms of prognosis of GC, we first analyzed the relationship between overall survival of these GC patients and several wellestablished conventional risk factors that are capable of predicting prognosis using Kaplan-Meier method and Log-rank test. As displayed in Figure 1, it was immediately clear that TNM staging system, including depth of invasion (T), N, M, and TNM stages, significantly impacted on overall survival of these patients with GC. The prognostic value of traditional risk factors that affect prognosis in this study has been verified, we carried on to analyze those less known risk factors or the combination of risk factors for survival as described below. As shown in Figure 2, Kaplan-Meier survival curve indicted that age and cell differentiation significantly affected overall survival ( Fig. 2A, B). Furthermore, poor survival of GC patients was associated with higher CA 19-9 levels (above clinical reference range, Fig. 2C). Similarly, when CA19-9 lenel was grouped by the median (cut-off value 8.3 U/mL), GC patients with higher levels of CA19-9 showed poorer survival than those with lower CA-19-9 levels (Fig. 2D). The above observations suggested that no matter by median grouping or by reference range grouping, the higher the CA19-9 level, the poorer the survival prognosis, that is, higher blood levels of CA19-9 would be a risk factor for survival of patients with GC. In addition, blood levels of tumor markers CEA (Fig. 2E) and AFP (Fig. 2F) were not significantly associated with survival outcomes among these GC patients.

Correlations between CA19-9 levels and postoperative survival predictors
Having demonstrated a role of CA-19-9 in predicting the survival and prognosis of GC patients, we tested the hypothesis that CA-19-9 levels should be relevant to postoperative conventional survival predictors such as metastasis and TNM stage (as shown in Fig. 1). We found that there were 35 patients (19.6%) whose CA 19-9 levels were above the reference range (≥27 U/mL). Indeed as hypothesized, GC patients carrying lymphatic metastasis, M and late TNM stages exhibited higher levels (≥27 U/mL) of CA 19-9 (Table 3), which was shown to be an independent factor affecting survival of GC patients ( Table 2, Fig. 2).

Correlations between preoperative and postoperative survival predictors
Age, cell differentiation and CA-19-9 were associated with the survival and prognosis of GC patients, we further hypothesized that the above preoperative survival predictors would show correlations with conventional postoperative ones, namely the TNM system. Not surprisingly, this hypothesis was validated by the following observations as shown in Table 4   Positive correlations of preoperative cell differentiation were also observed with postoperative predictors including invasion depth (r = 0.223, P < .001), N (r = 0.242, P < .001), and TNM staging (r = 0.268, P < .001).

Quantify predicting powers for survival predictors by using ROC curves
As revealed in Figure 3, the large differences were showed in ROC curves defined by their sensitivity and specificity for 12 potential factors tested before surgery and their corresponding AUCs were displayed in Table 5. In this analysis, the volume of an AUC represents the weight (power) of a potential factor in terms of predicting survival prognosis, which we termed as impact weight (IW) ( Table 5). IW is a function in terms of quantifying the relative power of a survival risk factor in a particular cancer. It was surprising to note that preoperatively determined CA-19-9 lever had a comparable AUC (0.707) to well-established TNM staging (0.717) and lymphatic metastasis (0.712), demonstrating their comparable impact weights in terms of predicting survival (Table 5). Furthermore, preoperatively determined factors, cell differentiation and age, also showed similar AUCs to those of Table 4 Correlations of preoperative prognostic predictors of cell differentiation and CA 19-9 levels with postoperative TNM system.   Table 5. It is very interesting to note that 3 nonconventional survival predictors, namely age, cell differentiation and CA19-9, have AUCs similar to, or even better than, AUCs of some conventional survival predictors of the TNM staging components (see Table 5). In keeping with the survival comparisons in Figures 1 and 2 (Table 5), but uneven distributions of patients in Kaplan-Meier curves may be one, if not the only reason (Fig. 1C).

Discussion
The survival time of GC patients after surgery is a critical barometer evaluating the effectiveness of diagnosis and management. The 5-year survival rates of gastric cancer in Japan and Korea had reached 60.3% and 68.9%, respectively. However, the 5-year survival rate in China is only 35.9%, with a significant gap with the developed countries in the world, [19] Like all other cancers, postoperative survival of GC patients is a complex function determined by comprehensive interactions among multiple factors, known and unknown. Therefore, it is undoubtedly of great significance to carry out research on the factors affecting the prognosis of GC patients. [20] Postoperative TNM staging system, based on T, N, and M is well-established to be reliable prognostic factors (4). In this study, we have first validated the GC patient cohort using TNM staging systemby survival analysis. As can be seen in Figure 1, poor cumulative survival rate of GC patients is correlated with invasive T3+T4, N, M, and advanced TNM staging III+IV, respectively, confirming previous observations. [21] However, TNM staging is only a index that can only be confirmed after surgery, which can only provide a theoretical basis for postoperative treatment strategies, and it is of little significance for individual patients to plan surgical strategies before surgery. Therefore, it is desirable to look for bio-markers before surgery that (i) are routinely tested; (ii) have relationships with TNM staging system; and (iii) are able to predict the survival and prognosis of GC patients. Based on these criteria, five preoperative tumor markers were detected, such as CA19-9, CEA, CA12-5. We analyzed their relationships with conventional TNM staging as well as their abilities to predict the survival and prognosis of patients with GC before surgery.
We have hypothesized that, if preoperative tumor markers and cell differentiation have impact on the survival of patients with GC, they should be correlated with one or more postoperative predictors which are well established to predict survival. Indeed and as expected, we have observed that positive correlations are shown between cell differentiation and T, N, and TNM staging, respectively, and CA19-9 levels are positively correlated with N, M, and TNM staging, respectively. They may suggest there are intrinsic relationships between TNM system and CA19-9 levels and cell differentiation. In other words, the status of CA19-9 and cell differentiation may imply, before surgery, ongoing cancer progression or the potential of cancer progression prior to TNM staging.
These above correlations, on the other hand, further suggest that cell differentiation and CA19-9 levels may be able to predict prognosis of patients with GC just as TNM system (Fig. 1). Indeed as shown in Figure 2, patients with ≥ 60 years of age appear to show inferior survival compared with patients with < 60 years. Patients with poorly cell differentiation, which is diagnosed by gastroscopic biopsy before surgery, show poor survival. Higher blood levels of tumor marker CA19-9, either grouped by the clinical reference range or by the median, render GC patients poorer survival than those with lower levels of CA-19-9 (Fig. 2C, D), in keeping with the correlation analyses described above and with previous studies. [22] Our observations shown CA-19-9 can be used to as a well prognostic marker for GC patients. In addition, the higher the CEA level, the lower the survival rate, but the observation is not statistically significant (Fig. 2F). Tumor marker AFP (Fig. 2E) has failed to show impact on survival of GC patients which is in line with previous findings. [12] In our cohort, 19.6% of GC patients carry elevated levels of CA19-9, which is in the lower end as compared with other GC patient populations (16.0% to 34.6%). [23,24] In general as seen in Table 3, GC Patients with lymphatic metastasis, M, and advanced TNM stages have higher blood levels (≥27 U/mL) of CA19-9. For example, higher levels of CA19-9 are more frequently seen in GC patients diagnosed as having higher TNM stages III and IV (68.5%) compared with GC patients diagnosed as having lower TNM stages I and II (31.5%). Furthermore, Cox proportional hazards regression analysis indicates that GC patients with elevated levels of CA19-9 are at a risk of death 1.73 times higher than GC patients with normal reference levels of CA19-9, which appears to support the survival impact of CA19-9 on GC patients (Fig. 2). These observations corroborates some previous findings. [25,26] As shown in Table 2, it is important to note that multivariate analyses indicate 3 preoperative factors, namely age, cell differentiation, and CA 19-9 to behave as independent prognostic predictors, comparable to postoperative factors of M and TNM staging, further strengthening the case of those survival observations as shown in Figure 2. To compare discriminative powers between or among different categories of risk factors so as to score the differences in their powers of predicting survival prognosis, we have first introduced ROC curve and AUC to quantify (score) the power of a survival predictor in gastric cancer (Fig. 3). As shown in Table 5, It is important to note that, preoperatively obtained marker CA19-9 carries an IW (0.707) to those of well-established TNM staging (0.717) and lymphatic metastasis (0.712), demonstrating CA19-9 to be a comparable factor to TNM staging components in terms of survival prediction. Furthermore, the other two factors obtained before Table 5 Preoperatively determined CA19-9, age and cell differentiation have similar the AUCs or impact weights to postoperative TNM system in predicting survival prognosis of patients with GC.  Figure 3) which represents the impact weight (IW) on survival prediction. 95% CI = 95% confidence interval; N/A = not applicable; P values are the products derived from the comparisons between individual AUCs of the listed survival predictors and the reference AUC (0.500 on top, see red diagonal line in Fig. 3), respectively.
Yin et al. Medicine (2021) 100:49 Medicine surgery, cell differentiation and age, also carry similar IWs/AUCs to those of the TNM components although to a lesser degree. In summary, our study demonstrates that, in gastric cancer, preoperative CA19-9 levels and cancer cell differentiation are positively correlated with the well-established TNM components. Furthermore, it should be emphasized that cancer cell differentiation and, especially CA19-9, have comparable prognostic powers to the components of the golden standard TNM staging system by using ROC/AUC analysis. Accordingly, we introduce the concept of impact weight (IW) in terms of prognosis, which may be useful to compare across a spectrum of individual survival risk factors in a particular cancer that can benefit the clinical selection of the best prognostic factors for cancer patients of interest.
Predicting prognosis for cancer patients before surgery has clinical implications. Because information of CA19-9 levels, cell differentiation, and age is available before surgery, they may serve as useful "pre-warning indicators" in clinical management decisions before surgery. For example, when combined with other examinations such as clinical imaging, these "pre-warning indicators" obtained before surgery can be administered to patients with individualized surgical planning, the best radiotherapy and chemotherapy dose and time, and appropriate follow-up interval can be determined. Although it varies from patient to patient, chemo-radiotherapy has serious side-effects which often affect patients' quality of life and perhaps prognosis of patients. Therefore, if a patient is suggestive of having a poor prognosis by preoperative markers or marker combinations, it may implicate to consider appropriate dosage and course length of chemo-radiotherapy which may differ from routine guidelines. This hypothesis surely warrants further clinical studies in patients with GC. On the other hand, for those patients who are suitable for surgery, detecting these prognostic predictors may have implications in patients' palliative care program.
There exist several limitations in this study. Besides markers examined, there are also other recently proposed candidate substances such as cytokines (IL-6, IL-8, IL-17) that are indicative to have high clinical diagnosis and prognosis value as biomarkers of gastric cancer. [27][28][29][30] With the increase in the typys of detected markers, the establishment of a combined detection and prediction model for patients with gastric cancer will be very necessary in the future. In addition, limited sample size and a single-center design may reduce the power of statistical analysis and the accuracy of results. and as possible confounding factors, such as smoking and infections, may affect some tumor markers, future studies should consider collecting information of possible confounding factors for similar analysis. Furthermore, surgeryrelated factors, such as the quality of lymphadenectomy, surgical radicalness, extent of gastrectomy and associated resections, may affect prognosis in patients with GC, which should be investigated whenever possible in future studies.