Gastric cancer, one of the most common malignant tumors globally, ranks fourth among the causes of cancer-related deaths worldwide (1). According to statistics, approximately half of the GC deaths worldwide occur in China, imposing a heavy burden (1). Due to the lack of specific clinical symptoms and signs in the early stages, as well as the absence of effective biomarkers and screening methods, most GC patients are diagnosed in late or metastatic stages, with low resection rates and poor prognosis (2, 3). Therefore, it is crucial to conduct early screening, diagnosis, and treatment in order to improve the survival rate and prognosis of patients with GC.

Currently, endoscopic examination combined with histopathological evaluation of tissues is the gold standard for clinical diagnosis of GC (4). However, endoscopic examination is invasive and has a certain rate of missed diagnosis, and the tolerance and compliance of the population are relatively poor, limiting its routine use in GC screening (5, 6). Serum biomarker detection has the advantages of non-invasiveness, minimally invasive procedures, ease of operation, and convenient dynamic monitoring, making it easily accepted by subjects and widely used clinically (7). Currently, the sensitivity and specificity of single indicators cannot meet the demands (8), and the clinical diagnostic performance of multiple routine blood indicators combined detection remains unsatisfactory. In recent years, research on tumor diagnostic methods has shifted from searching for single biomarkers to finding a specific group of markers, also known as “characteristic patterns” (9, 10). The establishment of biomarker characteristic patterns for early diagnosis of GC can provide new ideas for research on tumor diagnostic methods.

Nomograms, as a reliable and convenient tool for quantifying significant risk factors, have been widely used in clinical practice (11, 12). Currently, there are few reports on the application of Nomogram models for GC diagnosis, with more studies focusing on prognosis and metastasis of GC (13, 14). Based on large-sample case data, this study constructs and validates an early GC diagnosis model, screening out a sensitive and specific group of early GC diagnosis biomarkers, aiming to provide effective references for clinical screening and early diagnosis of GC.

Gastric cancer imposes a substantial disease burden globally (1, 20). It is essential to identify high-risk populations early and detect GC to improve patient prognosis. Recently, an increasing number of studies have focused on the early prediction and diagnosis of GC, exploring and developing risk prediction methods and diagnostic models. The “ABC method” developed by MIKI et al. combines serum anti-Helicobacter pylori (Hp) IgG antibody and serum pepsinogen (PG) levels to identify individuals at high risk of developing GC in the future (21). The method (22) developed by Tu et al., which includes five biomarkers—serum G-17, PG I, PG II, PG I/II ratio, and anti-Hp IgG antibody—along with the afore mentioned “ABC method,” has shown limited predictive performance, with AUCs of less than 0.60. A multicenter cross-sectional study in China developed a GC risk prediction method incorporating seven variables (sex, age, G-17 level, PG I/II ratio, H. pylori infection, pickled food, and fried food), demonstrating good discriminatory ability with the AUC of 0.76 (23). Furthermore, most current studies tend to explore GC predictive factors from a micro perspective. The detection of indicators included in these models often requires significant time and may increase the financial burden on patients, affecting the clinical practicality of the models (24–26).

Currently, serum biomarkers are a minimally invasive, cost-effective, convenient, and repeatable tool for tumor diagnosis, which can detect disease progression relatively quickly. However, due to the insufficient sensitivity and specificity of a single biomarker for the diagnosis of GC, many studies often employ multiple biomarkers for combined detection to effectively enhance diagnostic efficacy. A nomogram integrates multiple predictive indicators to construct a multifactorial regression model, presenting the predicted probability of a clinical event as a score in a graphical format. This approach has been widely used to evaluate disease prognosis or predict disease diagnosis (27). In this study, we initially collected clinical data from 4283 patients who met the inclusion criteria and initially incorporated 36 variables, encompassing not only major laboratory test results but also the ratios of certain indicators. Through logistic regression and Lasso regression analysis, we finally identified 11 independent indicators associated with GC, including CEA, CA199, LYM, HGB, MCH, MCHC, PLT, ALB, TG, HDL, and AFR, and incorporated these indicators to develop a nomogram model for the diagnosis of GC.

Serum markers for tumor are widely used in the diagnosis, prognosis evaluation, and monitoring of tumors (28). CEA is a serum glycoprotein polymer primarily present in the human digestive system, playing an important role in regulating tumor cell proliferation and differentiation (29, 30). Elevated CEA levels are closely associated with tumor burden and are commonly used for predicting and diagnosing malignant tumors of the digestive tract (31). Additionally, CA19-9 levels are associated with tumor depth, lymph node metastasis (LNM), and tumor staging (32). Studies have shown that serum CA19-9 levels are significantly higher in GC patients with metastases. In our study, CEA and CA19-9 were identified as independent indicators associated with GC. The AUC for diagnosing GC was 0.618 for CEA and 0.619 for CA19-9, with a combined AUC of 0.720, indicating a higher diagnostic efficacy.

Inflammatory responses play a crucial role in various stages of tumor development, including growth, infiltration, invasion, and metastasis (33). Lymphocytes are a vital component of the body’s immune response and exert anti-tumor immune effects (34). Elevated lymphocyte counts have been associated with favorable prognoses in various cancers (35). In this study, LYM in GC patients were significantly lower than the control group. LYM was identified as an independent indicator of GC, with an AUC of 0.576 for diagnosing GC. Additionally, the inflammatory conditions may inhibit bone marrow hematopoietic function, potentially leading to reduced HGB levels (36). Our study found that the combined diagnostic AUC for GC of HGB (AUC=0.577), MCH(AUC=0.627), MCHC (AUC=0.631) and LYM was 0.655, which is higher than the diagnostic efficacy of each of these four indicators alone.

In the coagulation function indicators, PLT and AFR were included in the nomogram model of this study. During the development of malignant tumors, tumor cell infiltration, destruction, and metastasis can lead to a hypercoagulable state (37). Studies have shown that FIB levels of GC patients are associated with clinical stage, lymph node metastasis, and local infiltration depth (38). Our reach revealed that serum PLT levels in GC patients were notably higher than in those with benign gastric diseases, indicating a high risk of hypercoagulability in GC patients. The AUC for diagnosing GC was 0.600 for PLT and 0.701 for AFR, with a combined AUC of 0.705, indicating a higher diagnostic efficacy when both indicators are used together.

The development and progression of tumors are closely linked to lipid metabolism abnormalities (18). Studies have found that serum levels of LDL-C, HDL-C and TG in GC patients are higher than in normal individuals, whereas HDL-C levels are lower (39). However, our study indicates that serum TG levels in GC patients are lower than in patients with benign gastric diseases. This could be associated with inadequate consumption, increased tumor consumption, and a continuous decline in nutritional status as GC progresses. We identified TG (AUC=0.568) and HDL (AUC=0.560) as independent indicators associated with GC, with a combined AUC of 0.615 for diagnosing GC. Additionally, in this study, serum albumin (ALB) levels, which reflect the nutritional status of the body, were significantly lower in GC patients than those with benign gastric diseases. The AUC for diagnosing GC using ALB alone was 0.655.

Our nomogram model showed high diagnostic performance for GC in both the training set (AUC=0.803) and validation set (AUC=0.797), which was significantly higher than the diagnostic efficacy of individual or combined indicators included in the model. An article published in JAMA in 2017 emphasized that calibration or goodness of fit is often considered the most important feature of predictive models, as it reflects the model’s ability to accurately estimate absolute risk (40). In this study, calibration curves were plotted for the model in both the training and validation sets, showing good model fit and calibration ability, which indicates excellent performance. The clinical utility of predictive models has also been widely discussed as an important feature for evaluating models in recent years. This evaluation is primarily based on decision curve analysis to assess whether the model can benefit patients by influencing clinical decisions (41). The clinical utility of predictive models is mainly evaluated through decision curve analysis, and we also plotted decision curves for both sets, demonstrating substantial net benefit produced by the model. The main strengths of this study lie in its large sample size and the inclusion of key laboratory indicators. The variables included in the model are generally obtainable in most hospitals, and test results can be obtained within 24 hours of admission, ensuring the practical application of the diagnostic model in clinical settings.

However, this study also has some limitations. The study was retrospective and could introduce bias that may affect the causality and interpretation of the results. Some cases and other potentially meaningful indicators, such as PG I, PG II, PG I/II, CA724, NSE, D-dimer, CK-MB, CK, LDH, and HCY were excluded due to data loss. While the absence of these indicators may underestimate the predictive power of the model, inclusion of these indicators and filling in the missing data may affect the authenticity of the data, thereby affecting the predictive power of the model. Additionally, this study is single-center, and both the training and validation datasets are from one hospital, with only internal validation conducted, lacking external validation. Therefore, based on the existing results, our next step is to conduct a multi-center prospective study, gathering comprehensive clinical data of patients, to further improve the early diagnosis of GC.

Based on the analysis of large sample size, we constructed a nomogram model with 11 routine laboratory indicators, which showed good diagnostic efficacy and calibration, providing a convenient visualization tool and new possibilities for the early diagnosis of GC.