Dynamic monitoring of serum tumor markers in predicting recurrence after radical gastrectomy for gastric cancer

Abstract

BACKGROUND

Postoperative recurrence remains a major challenge following radical gastrectomy for gastric cancer, with 5-year recurrence rates reaching 30%-50%. Serum tumor markers, including carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), carbohydrate antigen 72-4 (CA72-4), and carbohydrate antigen 125 (CA125), are widely used in postoperative surveillance due to their non-invasive and cost-effective nature. However, previous studies have predominantly focused on single preoperative measurements, which demonstrate limited sensitivity and specificity for predicting recurrence. Emerging evidence suggests that dynamic changes in tumor marker levels may better reflect tumor burden and biological behavior than static measurements, yet systematic evaluation of different dynamic patterns remains lacking.

AIM

To investigate the value of dynamic monitoring of serum tumor markers in predicting recurrence after radical gastrectomy for gastric cancer.

METHODS

A retrospective analysis was conducted on clinical data of 213 patients who underwent radical gastrectomy between January 2020 and December 2021. Serum levels of CEA, CA19-9, CA72-4, and CA125 were monitored at preoperative and various postoperative time points. Six dynamic change patterns were defined: Persistently normal, postoperative normalization, persistently elevated, recurrent elevation, progressive elevation, and delayed elevation. Receiver operating characteristic curves were used to evaluate predictive value, Cox proportional hazards regression model was used to analyze independent risk factors for recurrence, and Kaplan-Meier method was used for survival analysis.

RESULTS

The median follow-up time was 36.5 months, with 68 patients (31.9%) experiencing recurrence at a median time of 14.5 months. The recurrence group had significantly higher preoperative and 1-month postoperative marker levels compared to the non-recurrence group (P < 0.001). The area under the curve (AUC) of combined preoperative four markers for predicting recurrence was 0.825. The progressive elevation pattern of CEA, CA19-9, CA72-4, and CA125 showed AUCs of 0.856, 0.842, 0.838, and 0.831, respectively. The combined four-marker progressive elevation pattern achieved an AUC of 0.912 with 85.3% sensitivity and 89.0% specificity, superior to preoperative single detection (P = 0.003). Multivariate analysis revealed that tumor size ≥ 5 cm [hazard ratio (HR) = 2.156], poorly differentiated-undifferentiated (HR = 2.024), tumor-node-metastasis stage III (HR = 3.245), vascular invasion (HR = 1.847), neural invasion (HR = 1.726), progressive elevation of CEA (HR = 3.124), CA19-9 (HR = 2.856), CA72-4 (HR = 2.634), CA125 (HR = 2.487), recurrent elevation of CEA (HR = 2.145), and CA19-9 (HR = 1.967) were independent risk factors for recurrence (P < 0.05). The 36-month disease-free survival (DFS) rate was 92.6%-95.7% in the persistently normal group, while the median DFS was 9.0-12.0 months in the progressive elevation group (P < 0.001).

CONCLUSION

Dynamic monitoring of serum tumor markers has important value in predicting recurrence after radical gastrectomy for gastric cancer. Progressive elevation and recurrent elevation patterns can identify high-risk patients early, and combined multi-marker dynamic monitoring can improve predictive accuracy, providing a scientific basis for individualized follow-up strategies.

Key Words: Gastric cancer; Radical surgery; Tumor markers; Dynamic monitoring; Recurrence prediction; Disease-free survival

Core Tip: Dynamic post-gastrectomy monitoring of serum markers (carcinoembryonic antigen, carbohydrate antigen 19-9, carbohydrate antigen 72-4, carbohydrate antigen 125) facilitates early identification of high-risk recurrence in gastric cancer. Progressive and recurrent elevation patterns markedly outperform single preoperative marker values (area under the curve up to 0.912), offering a streamlined, cost-effective strategy to refine follow-up and personalise surveillance.

INTRODUCTION

Gastric cancer is one of the most common malignant tumors of the digestive tract worldwide, ranking among the top in both incidence and mortality rates[1]. Despite continuous advances in surgical techniques and perioperative management, radical surgery combined with adjuvant chemotherapy remains the primary treatment modality for gastric cancer. However, postoperative recurrence and metastasis remain key factors affecting patients' long-term survival[2]. Studies have shown that the 5-year recurrence rate after radical gastrectomy can reach 30%-50%, with most recurrences occurring within 2 years postoperatively[3]. Therefore, establishing an effective recurrence prediction system to achieve early identification of high-risk patients and timely intervention is of significant clinical importance for improving patient prognosis.

Serum tumor markers are widely used in the diagnosis and treatment of gastric cancer due to their advantages of being non-invasive, economical, and easily repeatable. Carcinoembryonic antigen (CEA), carbohydrate antigen 19-9 (CA19-9), carbohydrate antigen 72-4 (CA72-4), and carbohydrate antigen 125 (CA125) are commonly used gastric cancer-related tumor markers in clinical practice[4]. However, previous studies have mainly focused on the prognostic evaluation value of single preoperative detection, with limited sensitivity and specificity[5]. In recent years, increasing evidence suggests that dynamic change trends of tumor markers better reflect tumor burden and biological behavior than single detection values[6]. Persistent elevation, recurrent elevation, or progressive elevation of postoperative tumor markers may indicate the presence of minimal residual disease or early signals of tumor recurrence. However, there is currently a lack of large-sample prospective studies systematically evaluating the clinical significance of different dynamic change patterns[7].

This study retrospectively analyzed clinical data from 213 patients who underwent radical gastrectomy, systematically monitoring serum CEA, CA19-9, CA72-4, and CA125 levels at preoperative and various postoperative time points. This study aims to systematically investigate whether dynamic monitoring patterns of multiple serum tumor markers provide superior predictive accuracy for postoperative recurrence compared to conventional single preoperative measurements, and to establish clinically actionable risk stratification criteria. By defining and comparing different tumor marker dynamic change patterns, we aim to provide a scientific basis for developing individualized follow-up strategies[8].

MATERIALS AND METHODS

Study subjects

A retrospective analysis was conducted on clinical data of patients who underwent radical gastrectomy in our hospital’s general surgery department between January 2020 and December 2021. Inclusion criteria: (1) Gastric adenocarcinoma confirmed by postoperative pathology; (2) Underwent radical gastrectomy (R0 resection) with negative postoperative pathological margins; (3) No neoadjuvant chemotherapy or radiotherapy before surgery; (4) Completed at least 3 serum tumor marker tests postoperatively; (5) Complete clinical pathological and follow-up data; and (6) Follow-up time ≥ 24 months. Exclusion criteria: (1) Concomitant other malignant tumors; (2) distant metastasis before surgery; (3) Severe heart, liver, or kidney dysfunction; (4) incomplete follow-up data or missing tumor marker data at key time points; and (5) Did not receive standardized adjuvant therapy postoperatively [standardized therapy defined as: Stage II-III patients received at least 6 cycles of SOX or XELOX regimen adjuvant chemotherapy; stage I patients could not receive adjuvant chemotherapy]. Finally, 213 patients who underwent radical gastrectomy and met the criteria were included. This study was approved by the hospital ethics committee, and patient informed consent was waived for this retrospective study.

Clinical data collection

Through the hospital information system, clinical pathological data were collected, including: (1) general information: Gender, age, body mass index (BMI, kg/m²), underlying diseases (hypertension, diabetes, coronary heart disease, etc.); (2) tumor-related information: Tumor location, tumor size, histological type, degree of differentiation, depth of invasion (T stage), lymph node metastasis status (N stage), tumor-node-metastasis (TNM) stage (according to the 8^th edition of American Joint Committee on Cancer staging criteria), vascular invasion, neural invasion, etc.; and (3) Treatment-related information: Surgical method (total gastrectomy or distal subtotal gastrectomy), extent of lymph node dissection (D1+ or D2).

Serum tumor marker detection

All patients had 5 mL of fasting venous blood collected preoperatively, at 1 month postoperatively, and at every 3-month follow-up visit. Blood samples were allowed to stand at room temperature (20-25 °C) for 30 minutes, then centrifuged at 1500-2000 × g for 15 minutes to separate serum. Serum samples were tested within 2 hours of collection using electrochemiluminescence immunoassay for real-time detection of serum CEA, CA19-9, CA72-4, and CA125 levels. The detection instrument was a Roche Cobas e602 fully automated electrochemiluminescence immunoanalyzer, with reagent kits provided by Roche Diagnostics. Reference normal value ranges: CEA < 5.0 ng/mL, CA19-9 < 37.0 U/mL, CA72-4 < 6.9 U/mL, CA125 < 35.0 U/mL. All tests were performed by the same laboratory technician according to standard operating procedures, with quality control performed for each batch of tests. To minimize treatment-induced fluctuations, all postoperative blood samples were collected before each chemotherapy cycle (at least 7 days prior to chemotherapy administration) and before any intravenous treatments. For patients experiencing chemotherapy-related toxicity requiring dose modifications, the marker monitoring schedule was adjusted accordingly, and these exceptions were documented in the medical records.

Assessment of dynamic changes in tumor markers

Preoperative detection values and 1-month postoperative detection values were jointly used as the evaluation baseline. Based on tumor marker test results at various postoperative time points, the following dynamic change patterns were defined: (1) Persistently normal: Preoperative and all postoperative tests within normal range; (2) Postoperative normalization: Elevated preoperatively, returned to normal at 1 month postoperatively and remained normal; (3) Persistently elevated: Elevated preoperatively, continuously higher than normal postoperatively; (4) Recurrent elevation: Normalized postoperatively, elevated again during follow-up; (5) Progressive elevation: Continuous upward trend postoperatively (defined as two consecutive test values increasing ≥ 20% compared to each previous test value, and the last test value increasing ≥ 50% compared to the 1-month postoperative baseline); and (6) Delayed elevation: Normal preoperatively, elevated postoperatively. The dynamic change rate of each marker was calculated as: Change rate = (test value at each follow-up time point - 1-month postoperative baseline value)/1-month postoperative baseline value × 100%. For a single marker, values exceeding the upper limit of normal were determined as positive; for combined detection, at least one positive marker determined combined positivity.

Follow-up and recurrence diagnosis

The follow-up cutoff date was December 31, 2024, or the time of patient death. Tumor recurrence was defined as meeting any of the following conditions: (1) New lesions found on imaging examination (solid lesions with diameter ≥ 1 cm, or progressive enlargement in a short period with abnormal enhancement characteristics); (2) Pathological confirmation by gastroscopic biopsy; and (3) Pathological confirmation by repeat surgery. Recurrence types included local recurrence, regional lymph node recurrence, peritoneal metastasis, and distant metastasis (liver, lung, bone, etc.). Non-recurrence was defined as no evidence of tumor recurrence found through the above examinations during follow-up. Disease-free survival (DFS) was defined as the time interval from surgery date to confirmed recurrence or last follow-up.

Statistical analysis

SPSS 26.0 software was used for statistical analysis. Normally distributed measurement data were expressed as mean ± SD, and independent sample t-test was used for comparison between groups; non-normally distributed measurement data were expressed as median and quartiles [M (Q₁, Q₃)], and Mann-Whitney U test was used for comparison between groups. Count data were expressed as n (%), and χ² test was used for comparison between groups. Receiver operating characteristic (ROC) curves were used to evaluate the predictive value of tumor markers for recurrence, calculating area under the curve (AUC) and 95% confidence interval (95%CI). DeLong test was used to compare differences between different ROC curves, with Bonferroni method for multiple comparison correction of significance level. Cox proportional hazards regression model was used for multivariate analysis, variance inflation factor (VIF) was used to test collinearity between variables, and stepwise backward method was used to screen variables (exclusion criterion P > 0.10). Kaplan-Meier method was used to plot survival curves, and log-rank test was used to compare DFS differences between different groups. All tests were two-sided, with P < 0.05 considered statistically significant.

RESULTS

Basic patient characteristics

This study included 213 patients who underwent radical gastrectomy, with a median follow-up time of 36.5 (28.0, 45.0) months. During follow-up, 68 patients (31.9%) experienced tumor recurrence, with a median recurrence time of 14.5 (9.0, 21.0) months. Recurrence types included: Local recurrence in 12 cases (17.6%), regional lymph node recurrence in 15 cases (22.1%), peritoneal metastasis in 28 cases (41.2%), and distant metastasis in 13 cases (19.1%). Comparison of clinical pathological characteristics between recurrence and non-recurrence groups is shown in Table 1. There were no statistically significant differences in age, gender, BMI, underlying diseases, tumor location, surgical method, extent of lymph node dissection, or histological type between the two groups (P > 0.05). The distributions of tumor size, degree of differentiation, T stage, N stage, TNM stage, vascular invasion, and neural invasion showed statistically significant differences between the two groups (P < 0.05).

Table 1 Comparison of clinical pathological characteristics between recurrence and non-recurrence groups, n (%).

Characteristic	Recurrence group (n = 68)	Non-recurrence group (n = 145)	Test statistic	P value
Age (years)	61.8 ± 10.2	60.5 ± 11.4	t = 0.816	0.416
Gender			χ² = 0.125	0.724
Male	46 (67.6)	95 (65.5)
Female	22 (32.4)	50 (34.5)
BMI (kg/m²)	22.8 ± 3.1	23.2 ± 2.9	t = 0.925	0.356
Underlying diseases	28 (41.2)	53 (36.6)	χ² = 0.486	0.486
Tumor location			χ² = 2.815	0.245
Upper stomach	15 (22.1)	25 (17.2)
Body of stomach	23 (33.8)	48 (33.1)
Lower stomach	30 (44.1)	72 (49.7)
Histological type			χ² = 1.847	0.397
Papillary adenocarcinoma	8 (11.8)	22 (15.2)
Tubular adenocarcinoma	35 (51.5)	81 (55.9)
Mucinous adenocarcinoma	15 (22.1)	26 (17.9)
Signet ring cell carcinoma	10 (14.7)	16 (11.0)
Tumor size			χ² = 8.927	0.003
< 5 cm	28 (41.2)	88 (60.7)
≥ 5 cm	40 (58.8)	57 (39.3)
Degree of differentiation			χ² = 11.456	< 0.001
Well-moderately differentiated	18 (26.5)	71 (49.0)
Poorly differentiated-undifferentiated	50 (73.5)	74 (51.0)
T stage			χ² = 18.245	< 0.001
T1-T2	12 (17.6)	62 (42.8)
T3-T4	56 (82.4)	83 (57.2)
N stage			χ² = 24.186	< 0.001
N0	8 (11.8)	52 (35.9)
N1-N3	60 (88.2)	93 (64.1)
TNM stage			χ² = 21.347	< 0.001
Stage I	5 (7.4)	38 (26.2)
Stage II	22 (32.4)	58 (40.0)
Stage III	41 (60.3)	49 (33.8)
Vascular invasion	42 (61.8)	55 (37.9)	χ² = 11.258	< 0.001
Neural invasion	38 (55.9)	48 (33.1)	χ² = 10.525	0.001
Surgical method			χ² = 0.846	0.358
Distal subtotal gastrectomy	46 (67.6)	107 (73.8)
Total gastrectomy	22 (32.4)	38 (26.2)
Extent of lymph node dissection			χ² = 1.124	0.289
D1+	15 (22.1)	42 (29.0)
D2	53 (77.9)	103 (71.0)

TNM: Tumor-node-metastasis; BMI: Body mass index.

Comparison of preoperative and postoperative tumor marker levels

Comparison of preoperative and 1-month postoperative serum tumor marker levels between recurrence and non-recurrence groups is shown in Table 2. Preoperatively, the recurrence group had significantly higher CEA, CA19-9, CA72-4, and CA125 levels than the non-recurrence group (P < 0.05). At 1 month postoperatively, both groups showed decreased levels of CEA, CA19-9, CA72-4, and CA125 compared to preoperative levels (P < 0.01), but the recurrence group still had higher marker levels than the non-recurrence group (P < 0.05).

Table 2 Comparison of preoperative and 1-month postoperative serum tumor marker levels between recurrence and non-recurrence groups [M (Q1, Q3)].

Marker	Time point	Recurrence group (n = 68)	Non-recurrence group (n = 145)	Z value	P value
CEA (ng/mL)	Preoperative	8.5 (4.2, 15.8)	3.8 (2.1, 6.5)	-5.126	< 0.001
	1 month postoperative	4.2 (2.5, 7.8)	2.3 (1.4, 3.8)	-4.285	< 0.001
CA19-9 (U/mL)	Preoperative	52.3 (24.5, 98.6)	21.8 (12.3, 42.5)	-4.863	< 0.001
	1 month postoperative	28.6 (15.2, 52.3)	14.5 (8.3, 25.6)	-4.157	< 0.001
CA72-4 (U/mL)	Preoperative	12.8 (5.6, 28.4)	4.8 (2.5, 9.2)	-5.347	< 0.001
	1 month postoperative	6.5 (3.2, 14.5)	2.8 (1.5, 5.2)	-4.526	< 0.001
CA125 (U/mL)	Preoperative	48.5 (22.3, 86.7)	18.6 (10.2, 32.5)	-5.694	< 0.001
	1 month postoperative	24.8 (12.5, 45.6)	11.3 (6.8, 18.5)	-4.872	< 0.001

CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125.

Analysis of dynamic change patterns of tumor markers

Based on tumor marker test results during postoperative follow-up, the distribution of dynamic change patterns of each marker in recurrence and non-recurrence groups is shown in Table 3. To ensure the mutual exclusivity and clarity of dynamic change pattern classification, when patients met multiple pattern definitions simultaneously, classification was performed according to the following priority: Progressive elevation > recurrent elevation > persistently elevated > delayed elevation > postoperative normalization > persistently normal. The proportions of abnormal patterns such as progressive elevation, persistently elevated, recurrent elevation, and delayed elevation in the recurrence group were higher than those in the non-recurrence group (P < 0.001). Specifically, progressive or recurrent elevation of CEA accounted for 50.0% in the recurrence group, compared to only 5.5% in the non-recurrence group; progressive or recurrent elevation of CA19-9 accounted for 47.1% in the recurrence group and 6.2% in the non-recurrence group; progressive or recurrent elevation of CA72-4 accounted for 44.1% in the recurrence group and 4.8% in the non-recurrence group; progressive or recurrent elevation of CA125 accounted for 42.6% in the recurrence group and 5.5% in the non-recurrence group.

Table 3 Comparison of dynamic change patterns of tumor markers between recurrence and non-recurrence groups, n (%).

Marker	Dynamic change pattern	Recurrence group (n = 68)	Non-recurrence group (n = 145)	χ²	P value
CEA	Persistently normal	13 (19.1)	82 (56.6)	65.382	< 0.001
	Postoperative normalization	15 (22.1)	55 (37.9)
	Persistently elevated	8 (11.8)	2 (1.4)
	Recurrent elevation	17 (25.0)	5 (3.4)
	Progressive elevation	14 (20.6)	3 (2.1)
	Delayed elevation	1 (1.5)	1 (0.7)
	Subtotal	68 (100.0)	145 (100.0)
CA19-9	Persistently normal	16 (23.5)	88 (60.7)	58.642	< 0.001
	Postoperative normalization	16 (23.5)	48 (33.1)
	Persistently elevated	6 (8.8)	1 (0.7)
	Recurrent elevation	16 (23.5)	6 (4.1)
	Progressive elevation	16 (23.5)	3 (2.1)
	Delayed elevation	2 (2.9)	1 (0.7)
	Subtotal	68 (100.0)	145 (100.0)
CA72-4	Persistently normal	19 (27.9)	92 (63.4)	52.847	< 0.001
	Postoperative normalization	17 (25.0)	46 (31.7)
	Persistently elevated	6 (8.8)	1 (0.7)
	Recurrent elevation	15 (22.1)	4 (2.8)
	Progressive elevation	15 (22.1)	3 (2.1)
	Delayed elevation	4 (5.9)	2 (1.4)
	Subtotal	68 (100.0)	145 (100.0)
CA125	Persistently normal	21 (30.9)	95 (65.5)	50.126	< 0.001
	Postoperative normalization	16 (23.5)	42 (29.0)
	Persistently elevated	6 (8.8)	2 (1.4)
	Recurrent elevation	14 (20.6)	5 (3.4)
	Progressive elevation	15 (22.1)	3 (2.1)
	Delayed elevation	4 (5.9)	1 (0.7)
	Subtotal	68 (100.0)	145 (100.0)

Percentages were recalculated to ensure subtotals equal 1000% for each group, using rounding to one decimal place. CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125.

Comparison of dynamic change rates of tumor markers at different postoperative follow-up time points between recurrence and non-recurrence groups is shown in Table 4. At 3, 6, 12, and 18 months postoperatively, the change rates of CEA, CA19-9, CA72-4, and CA125 in the recurrence group were significantly higher than those in the non-recurrence group (P < 0.001). Patients in the recurrence group showed a continuous upward trend in markers, while the non-recurrence group tended to be stable or slightly decreased. The median change rate of tumor markers in the recurrence group at 18 months postoperatively was relatively high (92.5% for CA72-4 and 88.6% for CA125), suggesting that tumor marker change rates in recurrent patients showed a significantly accelerated upward trend 3-6 months before recurrence, with marker levels in some patients nearly doubling compared to the 1-month postoperative baseline. Upon examination of the raw data distribution, this change rate is consistent with the biological characteristics of recurrent patients, but there are some outliers, suggesting that clinical practice should closely monitor dynamic changes in markers.

Table 4 Comparison of dynamic change rates of tumor markers at different postoperative time points between recurrence and non-recurrence groups [M (Q1, Q3), %].

Marker	Time point	Recurrence group (n = 68)	Non-recurrence group (n = 145)	Z value	P value
CEA	3 months postoperative	12.5 (-8.5, 35.8)	-8.2 (-18.6, -2.3)	-6.847	< 0.001
	6 months postoperative	28.6 (5.2, 68.9)	-10.5 (-22.1, -3.6)	-7.562	< 0.001
	12 months postoperative	52.8 (18.5, 125.6)	-12.3 (-25.8, -4.2)	-8.324	< 0.001
	18 months postoperative	85.6 (32.4, 186.5)	-14.6 (-28.5, -5.5)	-8.891	< 0.001
CA19-9	3 months postoperative	10.8 (-10.2, 32.5)	-6.8 (-15.3, -1.8)	-6.524	< 0.001
	6 months postoperative	25.3 (3.8, 62.5)	-8.5 (-18.9, -2.5)	-7.215	< 0.001
	12 months postoperative	48.5 (15.6, 115.8)	-10.6 (-21.2, -3.3)	-7.968	< 0.001
	18 months postoperative	78.9 (28.5, 168.7)	-12.5 (-24.6, -4.8)	-8.456	< 0.001
CA72-4	3 months postoperative	15.6 (-6.8, 38.9)	-7.5 (-16.8, -2.6)	-6.685	< 0.001
	6 months postoperative	32.5 (8.5, 72.6)	-9.8 (-20.5, -3.9)	-7.328	< 0.001
	12 months postoperative	58.6 (20.8, 132.5)	-11.5 (-23.6, -4.5)	-8.125	< 0.001
	18 months postoperative	92.5 (35.6, 195.8)	-13.8 (-26.9, -5.8)	-8.652	< 0.001
CA125	3 months postoperative	13.8 (-9.5, 36.5)	-6.6 (-14.5, -2.2)	-6.458	< 0.001
	6 months postoperative	30.5 (6.8, 65.8)	-8.2 (-17.8, -3.5)	-7.142	< 0.001
	12 months postoperative	55.8 (18.2, 120.5)	-10.5 (-20.8, -4.8)	-7.895	< 0.001
	18 months postoperative	88.6 (30.5, 178.6)	-12.2 (-23.5, -5.6)	-8.358	< 0.001

Change rate = (test value at each follow-up time point - 1-month postoperative baseline value)/1-month postoperative baseline value × 100%. The higher change rate in the recurrence group at 18 months postoperatively reflects the significant upward trend in markers in recurrent patients. CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125.

Predictive value of single and combined tumor markers for recurrence

ROC curve analysis showed the predictive efficacy of preoperative single tumor markers for recurrence as shown in Table 5 and Figure 1. The AUCs of preoperative CEA, CA19-9, CA72-4, and CA125 for predicting recurrence were 0.742, 0.718, 0.756, and 0.738, respectively, with sensitivities of 67.6%, 64.7%, 70.6%, and 66.2%, and specificities of 72.4%, 70.3%, 75.2%, and 73.1%, respectively. The AUC of combined four-marker detection was 0.825, with 80.9% sensitivity and 76.6% specificity. DeLong test comparison showed that the predictive efficacy of combined detection was superior to each individual marker (vs CEA: P = 0.032; vs CA19-9: P = 0.015; vs CA72-4: P = 0.041; vs CA125: P = 0.028). Given that 4 multiple comparisons were performed, the significance level was set at α = 0.0125 (0.05/4) after Bonferroni correction. After correction, the comparison of combined detection superior to CA19-9 still approached statistical significance (P = 0.015, marginally above α = 0.0125 but P < 0.05), while other comparison P values were all > 0.0125, suggesting that combined detection showed a better overall trend in predictive efficacy.

Figure 1 Receiver operating characteristic curves for predicting postoperative recurrence after radical gastrectomy. The area under the curve (AUC) values for individual tumor markers were 0742 for carcinoembryonic antigen, 0.718 for carbohydrate antigen 19-9, 0.756 for carbohydrate antigen 72-4, and 0.738 for carbohydrate antigen 125. The combined model incorporating all four markers demonstrated superior predictive performance with an AUC of 0.825. AUC: Aea under the curve; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125.

Table 5 Predictive value of preoperative tumor markers for recurrence.

Marker	Cutoff value	AUC	95%CI	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	vs four combined P value	Significance after correction
CEA	6.2 ng/mL	0.742	0.673-0.811	67.6	72.4	53.6	82.5	0.032	NS (α = 0.0125)
CA19-9	38.5 U/mL	0.718	0.646-0.790	64.7	70.3	50.8	80.7	0.015	NS (α = 0.0125)
CA72-4	8.5 U/mL	0.756	0.688-0.824	70.6	75.2	56.9	84.2	0.041	NS (α = 0.0125)
CA125	32.5 U/mL	0.738	0.668-0.808	66.2	73.1	52.7	82.8	0.028	NS (α = 0.0125)
Four combined	-	0.825	0.765-0.885	80.9	76.6	62.8	89.2	Reference	Reference

Bonferroni method was used to correct for multiple comparisons, with corrected significance level α = 0.0125 (0.05/4). AUC: Area under the curve; CI: Confidence interval; NS: Not significant; PPV: Positive predictive value; NPV: Negative predictive value; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125.

The predictive value of postoperative tumor marker dynamic change patterns for recurrence is shown in Table 6. The AUCs of progressive elevation pattern for CEA, CA19-9, CA72-4, and CA125 in predicting recurrence were 0.856, 0.842, 0.838, and 0.831, respectively, all higher than preoperative single detection (P < 0.001). The AUC of combined four-marker progressive elevation pattern for predicting recurrence reached 0.912, with 85.3% sensitivity and 89.0% specificity. According to DeLong test, the predictive efficacy of combined progressive elevation pattern was significantly superior to preoperative combined detection (P = 0.003); compared with progressive elevation pattern of each individual marker, P values were: vs CEA P = 0.028, vs CA19-9 P = 0.021, vs CA72-4 P = 0.019, vs CA125 P = 0.015. After Bonferroni correction with significance level α = 0.0125 (0.05/4), comparisons with CA19-9, CA72-4, and CA125 remained statistically significant, but the comparison with CEA was no longer significant (P = 0.028 > 0.0125).

Table 6 Predictive value of dynamic change patterns of tumor markers for recurrence.

Change pattern	Marker	AUC	95%CI	Sensitivity (%)	Specificity (%)	PPV (%)	NPV (%)	vs preoperative combined P value	vs four combined progressive elevation P value	Significance after correction
Progressive elevation	CEA	0.856	0.802-0.910	75	85.5	72.9	86.8	< 0.001	0.028	NS (α = 0.0125)
	CA19-9	0.842	0.786-0.898	72.1	84.1	70.6	85.2	< 0.001	0.021	Significant (α = 0.0125)
	CA72-4	0.838	0.781-0.895	69.1	86.9	73.4	84.3	< 0.001	0.019	Significant (α = 0.0125)
	CA125	0.831	0.773-0.889	67.6	85.5	71.9	83.2	< 0.001	0.015	Significant (α = 0.0125)
	Four combined	0.912	0.869-0.955	85.3	89	80.6	91.5	0.003	Reference	Reference
Recurrent elevation	CEA	0.801	0.740-0.862	73.5	78.6	64.1	85	< 0.001	< 0.001	-
	CA19-9	0.794	0.732-0.856	70.6	77.2	62.3	83.3	< 0.001	< 0.001	-
	CA72-4	0.787	0.724-0.850	67.6	79.3	63.9	82.1	< 0.001	< 0.001	-
	CA125	0.778	0.714-0.842	64.7	78.6	61.1	81	< 0.001	< 0.001	-
	Four combined	0.868	0.818-0.918	79.4	83.4	73	87.8	0.015	0.008	-

Bonferroni method was used to correct for multiple comparisons, with corrected significance level α = 0.0125 (0.05/4). Recurrent elevation pattern was not analyzed with multiple comparison correction. AUC: Area under the curve; CI: Confidence interval; NS: Not significant; PPV: Positive predictive value; NPV: Negative predictive value; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125.

Multivariate analysis of factors affecting recurrence after radical gastrectomy

Variables with P < 0.10 in univariate analysis (including tumor size, degree of differentiation, T stage, N stage, TNM stage, vascular invasion, neural invasion, and progressive and recurrent elevation patterns of each marker) were included in the Cox proportional hazards regression model for multivariate analysis. Given the high collinearity between T stage, N stage, and TNM stage (VIF > 5), when using stepwise backward method to screen variables, TNM stage was prioritized to comprehensively reflect tumor burden, while T stage and N stage were removed to avoid the impact of multicollinearity on model stability. Results are shown in Table 7. Tumor size ≥ 5 cm [hazard ratio (HR) = 2.156, 95%CI: 1.287-3.613, P = 0.004], poorly differentiated-undifferentiated (HR = 2.024, 95%CI: 1.168-3.508, P = 0.012), TNM stage III (HR = 3.245, 95%CI: 1.856-5.672, P < 0.001), vascular invasion (HR = 1.847, 95%CI: 1.098-3.108, P = 0.021), neural invasion (HR = 1.726, 95%CI: 1.045-2.852, P = 0.033), progressive elevation of CEA (HR = 3.124, 95%CI: 1.687-5.786, P < 0.001), CA19-9 (HR = 2.856, 95%CI: 1.542-5.291, P = 0.001), CA72-4 (HR = 2.634, 95%CI: 1.418-4.895, P = 0.002), CA125 (HR = 2.487, 95%CI: 1.325-4.667, P = 0.005), recurrent elevation of CEA (HR = 2.145, 95%CI: 1.158-3.972, P = 0.015), and CA19-9 (HR = 1.967, 95%CI: 1.054-3.672, P = 0.034) were independent risk factors for recurrence after radical gastrectomy. Recurrent elevation of CA72-4 and CA125 did not reach statistical significance in multivariate analysis (CA72-4: HR = 1.756, 95%CI: 0.912-3.382, P = 0.092; CA125: HR = 1.684, 95%CI: 0.875-3.242, P = 0.118) and were removed from the final model after stepwise backward selection.

Table 7 Multivariate Cox regression analysis of factors affecting recurrence after radical gastrectomy.

Variable	HR	95%CI	P value
Tumor size ≥ 5 cm (vs < 5 cm)	2.156	1.287-3.613	0.004
Poorly differentiated-undifferentiated (vs well-moderately differentiated)	2.024	1.168-3.508	0.012
TNM stage III (vs stage I-II)	3.245	1.856-5.672	< 0.001
Vascular invasion (yes vs no)	1.847	1.098-3.108	0.021
Neural invasion (yes vs no)	1.726	1.045-2.852	0.033
Progressive elevation of CEA (yes vs no)	3.124	1.687-5.786	< 0.001
Progressive elevation of CA19-9 (yes vs no)	2.856	1.542-5.291	0.001
Progressive elevation of CA72-4 (yes vs no)	2.634	1.418-4.895	0.002
Progressive elevation of CA125 (yes vs no)	2.487	1.325-4.667	0.005
Recurrent elevation of CEA (yes vs no)	2.145	1.158-3.972	0.015
Recurrent elevation of CA19-9 (yes vs no)	1.967	1.054-3.672	0.034

T stage and N stage were removed due to collinearity with tumor-node-metastasis stage (variance inflation factor > 5). Recurrent elevation of carbohydrate antigen 72-4 [hazard ratio (HR) = 1.756, 95%CI: 0.912-3.382, P = 0.092] and carbohydrate antigen 125 (HR = 1.684, 95%CI: 0.875-3.242, P = 0.118) did not reach statistical significance (P > 0.05) in multivariate analysis and were excluded from the final model after stepwise backward selection. Stepwise backward method was used with exclusion criterion P > 0.10. CI: Confidence interval; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125; TNM: Tumor-node-metastasis; HR: Hazard ratio.

Survival analysis of different dynamic change patterns of tumor markers

Kaplan-Meier survival analysis showed differences in DFS among patients with different tumor marker dynamic change patterns, as shown in Table 8. Taking CEA as an example, median DFS in the persistently normal group was not reached [median follow-up time 36.5 months, 36-month survival rate 92.6% (95%CI: 85.3%-96.5%)], postoperative normalization group was 42.0 months, persistently elevated group was 15.5 months, recurrent elevation group was 18.0 months, progressive elevation group was 12.0 months, and delayed elevation group was 20.0 months, with log-rank test P < 0.001. DFS differences among different dynamic change patterns of CA19-9, CA72-4, and CA125 were all statistically significant (P < 0.001). Median DFS in patients with combined four-marker progressive elevation was 10.5 months, while median DFS in patients with all four markers persistently normal was not reached [median follow-up time 36.5 months, 36-month survival rate 94.2% (95%CI: 87.6%-97.5%)], with significant difference between the two groups (P < 0.001).

Table 8 Median disease-free survival of patients with different dynamic change patterns of tumor markers.

Marker	Dynamic change pattern	Number of cases	Median DFS (months)	χ²	P value
CEA	Persistently normal	95	Not reached [median follow-up 36.5 months, 36-month survival rate 926% (85.3-96.5%)]	86.524	< 0.001
	Postoperative normalization	70	42.0 (35.0, 48.0)
	Persistently elevated	10	15.5 (12.0, 19.0)
	Recurrent elevation	22	18.0 (15.0, 22.0)
	Progressive elevation	14	12.0 (9.0, 16.0)
	Delayed elevation	2	20.0 (16.0, 25.0)
	Subtotal	213
CA19-9	Persistently normal	104	Not reached [median follow-up 36.5 months, 36-month survival rate 933% (86.8-96.8%)]	78.326	< 0.001
	Postoperative normalization	64	40.0 (34.0, 46.0)
	Persistently elevated	7	14.0 (11.0, 18.0)
	Recurrent elevation	22	17.0 (14.0, 21.0)
	Progressive elevation	19	11.0 (8.0, 15.0)
	Delayed elevation	3	19.0 (15.0, 24.0)
	Subtotal	213
CA72-4	Persistently normal	111	Not reached [median follow-up 36.5 months, 36-month survival rate 946% (88.5-97.6%)]	74.185	< 0.001
	Postoperative normalization	63	39.0 (33.0, 45.0)
	Persistently elevated	7	13.5 (10.0, 17.0)
	Recurrent elevation	19	16.0 (13.0, 20.0)
	Progressive elevation	18	10.0 (7.0, 14.0)
	Delayed elevation	6	18.0 (14.0, 23.0)
	Subtotal	213
CA125	Persistently normal	116	Not reached [median follow-up 36.5 months, 36-month survival rate 957% (90.2-98.2%)]	70.847	< 0.001
	Postoperative normalization	58	38.0 (32.0, 44.0)
	Persistently elevated	8	12.5 (9.0, 16.0)
	Recurrent elevation	19	15.0 (12.0, 19.0)
	Progressive elevation	17	9.0 (7.0, 13.0)
	Delayed elevation	5	17.0 (13.0, 22.0)
	Subtotal	213

Median disease-free survival is expressed as median (quartiles) [M (Q1, Q3)]. DFS: Disease-free survival; CEA: Carcinoembryonic antigen; CA19-9: Carbohydrate antigen 19-9; CA72-4: Carbohydrate antigen 72-4; CA125: Carbohydrate antigen 125.

DISCUSSION

Previous studies have mainly focused on the prognostic evaluation value of single preoperative tumor marker detection, but its predictive efficacy is limited[9]. This study found that the AUCs of preoperative CEA, CA19-9, CA72-4, and CA125 single detection for predicting recurrence were 0.742, 0.718, 0.756, and 0.738, respectively, with sensitivity and specificity in the 60%-75% range, which is basically consistent with previously reported results[10]. However, when dynamic monitoring strategies were adopted, especially when identifying progressive elevation patterns, predictive efficacy was significantly improved, with single marker AUCs reaching 0.831-0.856, and combined detection AUC as high as 0.912. These results fully demonstrate the obvious advantages of dynamic monitoring over single detection[11].

The biological basis of tumor marker dynamic monitoring lies in its ability to reflect the presence of minimal residual disease and tumor cell proliferative activity[12]. Persistent or recurrent elevation of postoperative tumor markers suggests subclinical recurrence, while progressive elevation reflects rapid tumor burden growth. In this study, the median change rate of tumor markers in the recurrence group at 18 months postoperatively was as high as 78.9%-92.5%, while the non-recurrence group showed negative growth, providing an important biological window for early identification of recurrence[13]. Notably, some recurrent patients showed accelerated upward trends in markers 3-6 months before imaging findings, creating a time advantage for early clinical intervention.

This study systematically defined six dynamic change patterns of tumor markers and found significant differences in recurrence risk among different patterns. Progressive elevation and recurrent elevation are the two patterns with the most clinical warning significance. Multivariate analysis showed that progressive elevation of CEA, CA19-9, CA72-4, and CA125 were all independent risk factors, with HR values of 2.487-3.124, suggesting that recurrence risk increased 2-3 times in these patients[14]. Recurrent elevation pattern also deserves attention, with HR values of 2.145 and 1.967 for recurrent elevation of CEA and CA19-9, respectively, indicating that re-elevation after postoperative normalization is an important signal of recurrence[15].

Survival analysis further validated the prognostic value of different dynamic change patterns. The 36-month survival rate in the persistently normal group was 92.6%-95.7%, while the median DFS in the progressive elevation group was only 9.0-12.0 months, with significant differences between the two. This is consistent with previous research conclusions that persistent positivity of postoperative tumor markers indicates poor prognosis[16]. Notably, although the prognosis of the postoperative normalization group was better than the abnormal elevation group, it was still not as good as the persistently normal group, with median DFS of 38.0-42.0 months, suggesting that preoperative marker elevation itself may reflect aggressive characteristics of the tumor[17].

Delayed elevation pattern accounted for a small proportion (0.7%-5.9%) in this study but still has clinical significance. These patients had normal preoperative markers but developed elevation postoperatively, possibly indicating manifestation of occult metastatic lesions or appearance of new lesions, requiring high clinical vigilance[18]. Importantly, the delayed elevation pattern represents a distinct clinical scenario that warrants a different diagnostic approach compared to progressive elevation. While progressive elevation primarily indicates rapidly growing residual disease requiring shortened surveillance intervals, delayed elevation may signify activation of dormant micrometastases or development of de novo lesions. For patients exhibiting delayed elevation, we recommend a more aggressive imaging workup including whole-body positron emission tomography/computed tomography (PET/CT) to detect occult metastases and consideration for second-look endoscopy to identify potential locoregional recurrence. This pattern should prompt clinicians to broaden their differential diagnosis beyond simple disease progression and consider the possibility of new primary lesions or metachronous tumors. Although persistently elevated pattern had small numbers, it had the worst prognosis, with median DFS of only 12.5-15.5 months, indicating that failure of postoperative marker normalization may reflect incomplete tumor removal by surgery or early dissemination[19].

This study confirmed that combined multi-marker dynamic monitoring can significantly improve the accuracy of recurrence prediction. The AUC of combined four-marker progressive elevation pattern reached 0.912, with 85.3% sensitivity and 89.0% specificity, significantly superior to single marker detection. This result echoes previous research on multi-marker combined detection improving diagnostic efficacy[20]. Different tumor markers reflect different biological characteristics of tumors. CEA is mainly related to tumor differentiation degree, CA19-9 is closely related to mucinous tumors, CA72-4 has high specificity for gastric cancer, and CA125 often indicates peritoneal dissemination[21]. Combined detection can comprehensively reflect tumor heterogeneity characteristics and improve detection sensitivity and specificity.

However, this study also found that after strict Bonferroni multiple comparison correction, the advantage of combined detection compared to some single detections was somewhat weakened. This suggests that in clinical practice, detection strategies need to be comprehensively considered based on patient-specific circumstances and economic costs[22]. For high-risk patients (such as stage III, poorly differentiated, positive vascular or neural invasion), combined dynamic monitoring of four markers is recommended; while for low-risk patients, selective detection of CEA and CA72-4 may already meet clinical needs[23].

This study found that clinical pathological factors such as tumor size ≥ 5 cm, poorly differentiated-undifferentiated, TNM stage III, vascular invasion, and neural invasion were closely related to recurrence, consistent with numerous previous research results[24]. Notably, these traditional pathological factors and tumor marker dynamic change patterns both had independent predictive value in multivariate analysis, suggesting that they reflect tumor biological behavior from different dimensions. Pathological factors mainly reflect local invasiveness and metastatic potential of tumors, while marker dynamic changes reflect postoperative residual tumor burden and disease progression status[25].

The results of this study showed that preoperative tumor marker levels in the recurrence group were significantly higher than those in the non-recurrence group, and although levels decreased at 1 month postoperatively, they remained relatively high. This is consistent with reported views that high preoperative marker levels suggest large tumor burden and malignant biological behavior[26]. Insufficient postoperative marker decrease may suggest incomplete surgical resection or presence of micrometastases, and these patients require more intensive follow-up and more aggressive adjuvant therapy.

The results of this study have important guiding significance for postoperative follow-up strategies for gastric cancer. First, it is recommended to include tumor marker dynamic monitoring in routine follow-up protocols, with recommended monitoring frequency of: Every 3 months during the first 2 years postoperatively, and every 6 months during years 3-5[27]. Second, special attention should be paid to high-risk dynamic change patterns such as progressive elevation and recurrent elevation. Once discovered, comprehensive imaging evaluation should be initiated immediately for early diagnosis and treatment. Third, for patients with abnormal marker elevation but negative imaging, it is recommended to shorten follow-up intervals to 1-2 months and consider using sensitive imaging methods such as PET-CT to find occult lesions. Based on our findings, we propose the following pattern-specific surveillance protocols: (1) For progressive elevation pattern: Initiate immediate comprehensive imaging evaluation (contrast-enhanced computed tomography of chest, abdomen, and pelvis) within 2 weeks of detection, followed by monthly tumor marker monitoring until imaging findings are confirmed and treatment is initiated; (2) For recurrent elevation pattern: Repeat marker measurement within 4 weeks to confirm the trend, and if elevation persists or increases, proceed with imaging evaluation within 6 weeks; (3) For persistent normalization pattern: Maintain standard surveillance intervals of every 3 months for the first 2 years postoperatively, which can be extended to every 6 months thereafter; and (4) For delayed elevation pattern: As discussed above, pursue aggressive workup including PET-CT and second-look endoscopy to differentiate between recurrence and new primary lesions. These stratified approaches allow for resource-efficient surveillance while ensuring timely detection in high-risk patients.

It should be emphasized that tumor marker dynamic monitoring should be combined with clinical symptoms, physical examination, and imaging examination, and clinical decisions should not be made solely based on marker levels. For patients with mild marker elevation but no other abnormal manifestations, excessive examination and treatment should be avoided to prevent increasing patients' psychological and economic burden[28].

This study has the following limitations: First, as a single-center retrospective study with relatively limited sample size, results need further validation through multicenter large-sample prospective studies. Additionally, while our standardized adjuvant therapy protocol (SOX or XELOX regimen for stage II-III patients) ensured internal validity and reduced treatment-related confounding, it may limit the direct generalizability of our findings to clinical settings with varying treatment regimens or patient populations with different baseline characteristics. Future multicenter studies encompassing diverse treatment protocols and heterogeneous patient populations would enhance the external validity and broader applicability of these dynamic monitoring strategies. Second, there is some inter-batch variation and individual variation in tumor marker detection, and although quality control has been performed, it may still affect result stability. Third, the median follow-up time of this study was 36.5 months, which is not sufficient for evaluating long-term prognostic value beyond 5 years. Fourth, this study did not include novel tumor markers and molecular markers, such as circulating tumor DNA (ctDNA) and circulating tumor cells (CTC), which may provide more precise recurrence prediction information.

Future research should focus on the following directions: First, conduct multicenter prospective studies to establish recurrence risk prediction models based on tumor marker dynamic monitoring and perform external validation; second, explore the combined application value of tumor markers with novel markers such as ctDNA and CTC; third, utilize artificial intelligence and machine learning technology to integrate clinical pathological, imaging, and molecular marker information to construct individualized recurrence prediction systems; fourth, evaluate the impact of early intervention strategies based on marker dynamic monitoring on patient survival to provide evidence-based support for precision medicine.

CONCLUSION

In summary, dynamic serum tumor marker monitoring has important value in predicting recurrence after radical gastrectomy for gastric cancer, especially abnormal dynamic change patterns such as progressive elevation and recurrent elevation can identify high-risk patients early. Combined multi-marker dynamic monitoring can significantly improve predictive accuracy, providing a scientific basis for developing individualized follow-up and precision treatment strategies. It is recommended that tumor marker dynamic monitoring be made an important component of routine postoperative follow-up for gastric cancer, combined with clinical pathological characteristics and imaging examination, to achieve early detection and timely intervention of recurrence, ultimately improving patient prognosis.