Development of the G-Risk scoring system utilizing inflammatory and tumor biomarkers to improve prognostic accuracy in gastric cancer patients without lymphovascular invasion

Abstract

Objective

This study aimed to establish the G-Risk scoring system, a prognostic model based on inflammatory and tumor biomarkers, to enhance survival predictions for gastric cancer patients without lymphovascular invasion (LVI) and guide more tailored treatment strategies.

Methods

The key biomarkers associated with survival outcomes were identified using univariate and multivariate Cox regression analyses. These biomarkers were selected from a range of inflammatory and tumor markers to construct the G-Risk scoring system, which was specifically developed to improve prognostic accuracy in patients without LVI.

Results

The G-Risk score effectively stratified patients into high-risk and low-risk groups, achieving an AUC of 0.660, demonstrating strong predictive performance. Further multivariate analysis validated the G-Risk score as an independent prognostic factor with significant implications for patient survival.

Conclusion

The G-Risk score was successfully developed and validated as a reliable prognostic tool for gastric cancer patients without LVI. Its clinical implementation offers enhanced precision in assessing prognostic risk, facilitating personalized treatment planning, improving therapeutic outcomes, and reducing unnecessary medical interventions.

Introduction

Gastric cancer remains a significant health burden globally, particularly in parts of East Asia, Eastern Europe, and South America, where incidence and mortality rates are high [1]. According to global cancer statistics, gastric cancer is the fifth most common cancer worldwide and the fourth leading cause of cancer death [2]. Often diagnosed at a late stage, the complexity and limited options for treatment highlight the importance of accurately predicting disease progression and patient prognosis [3, 4]. In the treatment and prognosis assessment of gastric cancer, specific biomarkers and clinical features are used to evaluate the disease’s aggressiveness and expected outcomes, among which lymphovascular invasion (LVI) is a key factor [5, 6].

LVI is considered a poor prognostic biomarker in gastric cancer, typically associated with higher recurrence rates and lower survival rates. Compared to patients with LVI, those without LVI are often considered to have a better prognosis [7, 8]. However, even among this seemingly better prognosis group, patient survival outcomes exhibit significant heterogeneity. This variability suggests that there are risk stratification factors in patients without LVI that have not been fully identified and assessed. Accurate prognostic assessment in these patients not only helps identify high-risk patients who may require more aggressive treatment but is also crucial for low-risk patients to avoid unnecessary treatment burdens and associated side effects.

In clinical and research practices, inflammation and cancer biomarkers have been shown to correlate closely with the prognosis of various cancers. Inflammation plays a critical role in cancer development, invasion, and metastasis, and inflammatory markers like the fibrinogen-to-lymphocyte ratio (FLR) and the lactate dehydrogenase-to-albumin ratio (LAR) reflect the host’s inflammatory state and immune response, significantly correlating with patient survival prognosis [9, 10]. Additionally, specific tumor markers such as CEA and CA199 are widely used to assess gastric cancer's biological behavior and treatment response [11,12,13].

Therefore, the aim of this study is to develop a comprehensive prognostic scoring system based on inflammatory markers and tumor markers to provide precise prognostic analysis tools for patients with gastric cancer who have not undergone LVI. The application of this tool will help clinicians make more targeted treatment decisions based on the specific condition of the patient, thereby maximizing treatment effectiveness, reducing unnecessary medical interventions, and ensuring the quality of life for patients.

Patient selection

In this study, we selected gastric cancer patients treated at Harbin Medical University Cancer Hospital from January 2014 to December 2017. A total of 1475 patients met the inclusion criteria, which included: (1) histologically or surgically confirmed gastric cancer; (2) absence of lymphovascular invasion (LVI); (3) completion of gastric cancer-related surgical treatment; and (4) substantially complete clinical and pathological information, with missing data not exceeding 30%. Excluded were patients who: (1) had previously undergone neoadjuvant chemotherapy or radiotherapy; (2) had a history of residual gastric cancer or gastric surgery; (3) died within one month postoperatively.

At Harbin Medical University Cancer Hospital, patient data were managed and collected through the Gastric Cancer Information Management System v1.2 (copyright registration No. 2013SR087424). The types of information collected included demographic characteristics, treatment methods, laboratory test results, and pathology reports. Tumor staging was defined according to the 8th edition of the AJCC/UICC gastric cancer TNM staging system.

In this study, we defined tumor cell invasion into the vascular wall or presence in the endothelial gap as lymphovascular invasion (LVI), regardless of whether it was in blood vessels or lymphatics. Cases not detecting LVI were marked as L0 and V0, i.e., L(−) and V(−), representing no invasion; cases detecting L1, V1, or V2 were marked as L(+) and V(+), indicating presence of invasion. In short, cases without invasion in this study were uniformly classified as LVI negative [LVI(−)], and those with invasion as LVI positive [LVI(+)].

The blood samples from patients were collected within 1 week before surgery. LMR was the ratio of lymphocytes to monocytes; PLR was the ratio of platelets to lymphocytes; NLR was the ratio of neutrophils to lymphocytes; PNI was calculated as 5*lymphocytes + serum albumin; APRI as aspartate aminotransferase to platelets ratio; AAR as aspartate aminotransferase to alanine aminotransferase ratio; LAR as lactate dehydrogenase to serum albumin ratio; DIR as direct to indirect bilirubin ratio; FLR as fibrinogen to lymphocyte ratio.

Follow-up

After discharge, we conducted biannual checks on the patients’ survival status. This study followed the patients for 5 years until the end of the follow-up period or until the patient's death.

Statistical methods

All statistical analyses were performed using Python 3.9 and R language 4.2.1. The missing values were handled using the KNNImputer method, applied to variables with a missing rate of less than 30%.

The group differences were analyzed using the Pearson chi-square test or Fisher’s exact test, depending on sample size and data distribution.

Survival analysis was conducted using the Kaplan–Meier method, with differences between survival curves compared using the log-rank test.

The Cox proportional hazards model was applied for univariate and multivariate survival analyses, with variables showing a p value < 0.05 in univariate analysis included in the multivariate model. The final multivariate model was selected using a stepwise backward elimination approach based on the Akaike Information Criterion (AIC), and results were reported as hazard ratios (HRs) with 95% confidence intervals (CIs).

The receiver operating characteristic (ROC) curve was used to evaluate the predictive performance of the G-Risk score, the pROC package was used to perform ROC analysis on the data and the results were visualized using ggplot2. The pROC package automatically adjusts the outcome order to ensure that the ROC curve is convex upward by default. The Youden index was used to determine the optimal cutoff value for stratifying patients into high-risk and low-risk groups.

A nomogram was constructed based on significant variables identified in the multivariate Cox regression model to facilitate individualized survival predictions.

Additionally, subgroup analyses were performed to assess the robustness of the G-Risk score, stratifying patients by pTNM stage (I + II vs. III + IV), Borrmann classification (Type IV vs. non-Type IV), tumor location, sex and age. The statistical significance was set at p < 0.05 for all analyses.

Results

Study population

In this study, we tracked the 5-year survival of 1475 gastric cancer patients without LVI. Among them, 71.9% were male (1060 patients), and 28.1% were female (415 patients). The median age of the participants was 59 years, with an age range from 52 to 65 years. The most common site of cancer occurrence was the antrum of the stomach, accounting for 67.5% of cases (995 patients), followed by the body of the stomach at 20.2% (298 patients). By the end of the study, 387 patients had passed away, comprising 26.2% of the total cohort (Table 1).

Table 1 Clinical and pathological characteristics of gastric cancer patients without LVI

Complement of missing values

For variables with a missing data rate below 30%, we applied the KNNImputer method to perform data imputation. Among the 22 variables, 17 variables had a missing data rate of less than 5%, three variables had a missing rate between 5 and 7%, and two variables had a missing rate between 10 and 20% (Additional file 1: Table S1).

Biomarker screening

The preliminary assessments using a univariate Cox regression risk model for nine inflammatory markers and four tumor markers showed that all indicators except DIR had p-values < 0.05. These markers were further evaluated using a multivariate Cox regression risk model, which identified PNI (HR = 0.98; 95% CI [0.95–1]; p = 0.0316), AAR (HR = 1.3; 95% CI [1.03–1.64]; p = 0.0255), LAR (HR = 1.11; 95% CI [1.04–1.19]; p = 0.0011), FLR (HR = 1.17; 95% CI [1.01–1.36]; p = 0.0349), CEA (HR = 1; 95% CI [1]; p = 0.0032), CA199 (HR = 1; 95% CI [1]; p = 0.002), and CA125 (HR = 1.01; 95% CI [1–1.01]; p = 0.019) as significant predictors of prognosis in patients without LVI (Table 2).

Table 2 Survival analysis of blood indicators using univariate and multivariate Cox regression analyses

The predictive capability of these seven factors was assessed using the area under the curve (AUC) of the ROC curve, and optimal cutoff values for these markers were determined. The AUCs for CA125, FLR, CA199, PNI, CEA, LAR, and AAR were 0.618, 0.595, 0.591, 0.587, 0.583, 0.572, and 0.537, respectively; optimal cutoffs were 9.894, 1.2993, 19.35, 49.975, 3.395, 4.1174, and 1.2158. The survival analysis indicated that high levels of CA125, FLR, CA199, CEA, LAR, and AAR were associated with poor prognosis, while a high PNI was associated with better prognosis, all with p values < 0.001 (Fig. 1).

Fig. 1

The ROC curve and survival curve of the selected indicators. (A) The ROC curve of CA125. (B) The ROC curve of FLR. (C) The ROC curve of CA199. (D) The ROC curve of PNI. (E) The ROC curve of CEA. (F) The ROC curve of LAR. (G) The ROC curve of AAR. (H) The Kaplan-Meier survival curve of CA724. (I) The Kaplan-Meier survival curve of FLR. (J) The Kaplan-Meier survival curve of CA199. (K) The Kaplan-Meier survival curve of PNI. (L) The Kaplan-Meier survival curve of CEA (M) The Kaplan-Meier survival curve of LAR. (N) The Kaplan-Meier survival curve of AAR

Construction of prognostic models

We initially constructed a basic prognostic scoring system using the two indicators with the highest area under the curve (AUC) values, which we named the A-Risk score. Subsequently, we added additional indicators in descending order of their AUC values. This process began with the introduction of the third highest AUC indicator, followed by the fourth, fifth, sixth, and seventh highest. After each new indicator was added, we re-evaluated and renamed the prognostic scoring systems as B-Risk score, C-Risk score, D-Risk score, E-Risk score, and F-Risk score, respectively.

The AUC values for these six prognostic scoring systems were 0.631, 0.655, 0.642, 0.647, 0.655, and 0.657. We observed that the addition of the PNI indicator led to a decrease in the AUC value, suggesting that it might not contribute positively to the accuracy of the prognosis prediction. Therefore, we decided to exclude this feature and only use the remaining six features to rebuild the prognostic scoring system.

Ultimately, we developed a new prognostic scoring system named the G-Risk score, which showed the highest predictive accuracy with an AUC of 0.660 (Table 3). Survival analysis indicated that patients in the high-risk group defined by the G-Risk score had significantly worse prognosis compared to those in the low-risk group (Fig. 2). These findings demonstrate the effectiveness and reliability of the G-Risk score in predicting the survival prognosis of gastric cancer patients without lymphovascular invasion.

Table 3 The changes in AUC with an increase in the number of features

Fig. 2

The capabilities of the G-risk Score model. (A) the nomogram, (B) the calibration curve. (C) the decision curve. (D) the ROC curve. (E) the survival curve

Univariate and multivariate Cox regression analysis

In the survival analysis of gastric cancer patients without lymphovascular invasion (LVI), both univariate and multivariate Cox regression analyses were conducted. Initial results revealed significant correlations between survival duration and factors such as Age, pTNM staging, Borrmann classification, location, tumor size, and G-Risk score. Further multivariate Cox regression analysis determined that Age (HR = 1.01; 95% CI [1–1.02]; p = 0.0208), pTNM staging (HR = 16.41; 95% CI [8.72–30.89]; p < 0.0001), Borrmann classification (HR = 3.8; 95% CI [1.78–8.1]; p = 0.0004), Location (HR = 2.56; 95% CI [1.66–3.94]; p < 0.0001), and G-Risk score (HR = 1.64; 95% CI [1.31–2.04]; p < 0.0001) are independent prognostic factors significantly affecting the survival prognosis of patients without LVI. This finding underscores the importance of the G-Risk score in assessing the prognosis of such gastric cancer patients (Table 4).

Table 4 Survival analysis of gastric cancer patients without LVI using univariate and multivariate Cox regression analyses

Subgroup analysis

We plan to conduct more detailed subgroup analyses aimed at evaluating the impact of the G-Risk score on the survival prognosis of gastric cancer patients across different subgroups. In the analysis, patients will be categorized based on cancer staging (Stages I + II vs. III + IV), Borrmann classification (Type IV vs. non-Type IV), and the primary tumor location. Additionally, we performed subgroup analyses based on sex (male vs. female) and age (≥ 60 years vs. < 60 years) to further assess the consistency of our findings. The results indicate that regardless of I + II or III + IV stage, Borrmann Type IV or non-Type IV, tumor location, sex, or age group, patients in the high-risk group consistently exhibit significantly worse survival outcomes compared to those in the low-risk group (Fig. 3).

Fig. 3

Subgroup analysis of the G-risk Score in gastric cancer patients. (A) Stage I+II. (B) Stage III+IV. (C) Borrmann IV. (D) Non-Borrmann IV. (E) Location-Whole stomach. (F) Location-Non-Whole stomach. (G) less than 60 years old. (H) greater than or equal to 60 years old. (I) Male. (J) Female

Comparison of the traditional TNM staging model and the G-Risk score model

Our analysis showed that the TNM staging system demonstrated higher prognostic accuracy (AUC = 0.771) compared to the G-Risk score (AUC = 0.660). However, the G-Risk score provides additional prognostic stratification within TNM-defined stages, particularly in patients without lymphovascular invasion (LVI). This suggests that inflammatory and tumor biomarkers contribute to risk assessment and may serve as a complementary tool to traditional staging models, helping to refine prognosis and personalize treatment strategies (Additional file 2: Fig. S1).

Discussion

In this study, we focused on the survival prognosis of gastric cancer patients without lymphovascular invasion (LVI) and developed a prognostic scoring system based on multiple biomarkers, termed the G-Risk score. By meticulously tracking and analyzing 1475 patients, we discovered that inflammatory and tumor markers such as CA125, FLR, CA199, CEA, LAR, and AAR significantly predict survival outcomes, providing crucial biomarkers for personalized treatment and clinical prognosis assessment in gastric cancer.

The potential mechanisms of these biomarkers are as follows: tumor-associated antigens like CA125, CA199, and CEA, which are typically linked with tumor cell expression, secretion, or interaction with the extracellular matrix. In gastric cancer, elevated levels of these antigens may reflect tumor burden, aggressiveness, and metastatic potential [14, 15]. For instance, increases in CA125 and CA199 could be associated with enhanced invasiveness through changes in cell adhesion, angiogenesis promotion, or immune evasion mechanisms [16]. Furthermore, inflammation is a critical factor in cancer development, promoting malignant processes within the tumor microenvironment [17, 18]. Elevated FLR might reflect an activation of the body’s inflammatory and coagulation systems, contributing to tumor cell adhesion, invasion, and metastasis [19]. An increase in LAR could indicate high metabolic activity and poor nutritional status, thereby affecting overall prognosis [20, 21].

In this study, the G-Risk score demonstrated commendable accuracy in distinguishing between high-risk and low-risk gastric cancer patients. It achieved an area under the curve (AUC) value of 0.660, which underscores the reliability of this scoring system in predicting survival outcomes among this patient population. This level of predictive accuracy suggests that the G-Risk score can be a valuable tool in clinical settings, aiding oncologists in tailoring treatment approaches based on individual risk assessments.

Further analysis within various subgroups emphasized the G-Risk score’s applicability and utility across different clinical scenarios. Such analyses ensure that the scoring system remains robust and versatile, capable of providing reliable prognostic information in diverse clinical conditions, and patient subpopulations.

To further enhance predictive accuracy and optimize clinical application, we plan to explore machine learning techniques, such as random forests, gradient boosting models, and deep learning algorithms, to refine the risk prediction model. These approaches can capture non-linear relationships between biomarkers and survival outcomes, potentially improving prognostic performance. Additionally, instead of a static scoring system, we aim to develop a dynamic prognostic model that integrates longitudinal biomarker trends. The time-dependent factors and sequential monitoring of biomarker levels may allow for more precise risk stratification compared to a single preoperative assessment. By adopting these advanced methodologies, we seek to refine the G-Risk scoring system and enhance its clinical utility.

Additionally, in the multivariate Cox regression analysis, the G-Risk score emerged as an independent prognostic factor for patients who do not exhibit lymphovascular invasion (LVI). This finding is particularly significant as it highlights the G-Risk score’s potential in enhancing clinical decision-making processes.

This ability to function independently of other known prognostic markers not only solidifies the G-Risk score’s standalone utility but also enhances its potential integration into existing prognostic models. By doing so, it could significantly improve the predictive accuracy of these models, thereby optimizing treatment strategies and ultimately impacting patient management in a positive manner.

The G-Risk score holds significant dual value in clinical settings. For high-risk patients with poorer prognosis, this scoring system can guide physicians to adopt more aggressive treatment and management strategies, such as intensified monitoring and early intervention, to improve survival outcomes. For low-risk patients, physicians might consider more conservative treatment approaches to avoid unnecessary aggressive treatments, thereby reducing economic and psychological burdens while maintaining the quality of life [22].

The specific guidance measures for the G-Risk scores are as follows:

For high-risk patients (high G-Risk scores): recommend the use of more aggressive treatment approaches, such as combination chemotherapy or targeted therapy, and possible surgical interventions; increase the frequency of follow-ups to detect recurrence or disease progression early; and consider experimental treatment options or participation in clinical trials to explore new therapeutic possibilities; for low-risk patients (low G-Risk scores): consider more conservative treatment strategies, such as solitary surgical intervention or accompanied by milder adjuvant therapy; reduce unnecessary invasive treatments to avoid the side effects and medical costs associated with overtreatment. The follow-up interval can be appropriately extended to reduce the economic and psychological burden on patients; for patients with uncertain risk: recommend more detailed diagnostic assessments, such as molecular biomarker testing or genotyping, to determine a more precise risk assessment and adjust the treatment plan based on further risk assessment results.

The G-Risk score, developed to predict the prognosis of gastric cancer patients without lymphovascular invasion (LVI), incorporates inflammatory markers and tumor markers, which have shown promise in predicting survival outcomes in gastric cancer. However, its applicability to adenocarcinoma of the gastroesophageal junction (AEG), a cancer subtype with distinct clinical characteristics, warrants further consideration. While both gastric cancer and AEG share some common pathophysiological features, AEG is clinically and biologically different, especially in its response to treatments such as adjuvant chemotherapy.

In studies, adjuvant chemotherapy has demonstrated significant survival benefits for AEG patients, particularly those with pathological stage II/III disease. The G-Risk score, which integrates biomarkers like CA199, CEA, FLR, and LAR, is potentially well-suited for AEG patients because these markers also correlate with prognosis in this group, reflecting tumor burden, inflammatory responses, and metabolic status. The G-Risk score could, therefore, serve as a supplementary tool to help predict the outcomes of AEG patients who may be candidates for postoperative adjuvant chemotherapy, assisting in risk stratification and individualized treatment decisions [23, 24].

The G-Risk score may play a crucial role in multiple clinical areas, particularly in postoperative adjuvant therapy decisions and follow-up management, significantly enhancing the precision of treatment plans and patient management. The primary goal of postoperative adjuvant therapy is to reduce the risk of recurrence and improve overall survival. In traditional treatment decisions, whether to administer adjuvant therapy is based on the TNM staging and other known clinical factors. However, this approach may not fully account for individual patient differences, leading to some patients receiving unnecessary overtreatment, while high-risk patients may not receive sufficient treatment. The G-Risk score, combining inflammatory markers and tumor markers, provides a more detailed and personalized risk assessment, helping clinicians tailor treatment plans based on each patient’s specific condition. By incorporating the G-Risk score into postoperative treatment decisions, the physicians can more accurately identify high-risk patients and design more aggressive adjuvant therapies for them, while avoiding overtreatment for low-risk patients, ensuring personalized treatment and optimizing therapeutic outcomes [25, 26].

The purpose of postoperative follow-up management is to detect signs of recurrence or metastasis early and ensure timely intervention. Traditional follow-up management usually relies on TNM staging and routine clinical assessments, leading to standardized follow-up intervals, with all patients undergoing the same frequency of checks. However, this approach does not consider each patient’s individualized recurrence risk, which may result in over-monitoring for some patients or insufficient monitoring for high-risk patients. The G-Risk score provides a personalized risk assessment for each patient, allowing clinicians to adjust follow-up strategies based on the patient's specific risk level. For high-risk patients, more frequent and thorough follow-up checks, such as regular imaging and biomarker monitoring, can be implemented. For low-risk patients, unnecessary tests can be minimized, reducing healthcare resource consumption and alleviating patients' psychological burdens. In this way, the G-Risk score not only helps reduce healthcare costs but also effectively reduces patient anxiety, while still ensuring adequate surveillance for recurrence.

The application of the G-Risk score in postoperative adjuvant therapy and follow-up management significantly enhances the ability to provide personalized treatment and precise management. By developing treatment plans and follow-up strategies based on individual risk assessments, the G-Risk score holds the potential to help clinicians optimize treatment outcomes, reduce unnecessary treatment and over-monitoring, thereby improving the patient's treatment experience and quality of life. This precise risk assessment system shows immense potential and broad applications in the treatment of gastric cancer and other tumors.

This study has some limitations; the data were exclusively from patients at Harbin Medical University Cancer Hospital, which may introduce selection bias. Additionally, the retrospective data collection could include incompleteness and recording biases. Although we used the KNNImputer method to address missing data, the incompleteness might still impact the study's outcomes.

Future studies will validate the effectiveness of the G-Risk scoring system by expanding to multiple centers, including hospitals in different regions, to enhance the representativeness and extrapolation of the research results. Furthermore, future studies should be designed as prospective studies to reduce these biases and improve the quality and reliability of data. While our study focuses on inflammatory and tumor markers due to their established prognostic significance and clinical accessibility, we acknowledge that incorporating genetic and molecular biomarkers could further enhance the predictive accuracy of the G-Risk scoring system. Emerging evidence suggests that genomic alterations, transcriptomic signatures, and molecular classifications (e.g., MSI status, HER2 expression, and PD-L1 levels) play crucial roles in gastric cancer prognosis. Future studies will explore a multi-omics approach, integrating genomic, proteomic, and metabolomic data to refine risk stratification and improve the model’s clinical utility. By expanding the biomarker panel to include molecular signatures, we aim to develop a more comprehensive and precise prognostic tool while maintaining its feasibility for routine clinical application.

Conclusion

This research provides significant insights into prognostic biomarkers for gastric cancer and demonstrates the potential value of these markers in clinical prognosis assessment through the construction and validation of the G-Risk score. It lays a foundation for future research directions and clinical practice, especially in the context of personalized treatment and precision medicine.