The Association of Neutrophil-To-Lymphocyte and Platelet-To-Lymphocyte Ratios With Distant Metastasis in Vietnamese Gastric Cancer Patients

Abstract

Background

Gastric cancer (GC) remains a leading cause of mortality in Vietnam, where distant metastasis is frequently present at diagnosis. Since accessible biomarkers are scarce, we evaluated the association of Neutrophil-to-Lymphocyte Ratio (NLR) and Platelet-to-Lymphocyte Ratio (PLR) with distant metastasis in a Vietnamese cohort to develop a model for patient stratification.

Methods

This retrospective study analyzed 114 patients, categorizing into metastatic (Stage IV) and non-metastatic groups per AJCC 8^th criteria. The optimal cutoffs were determined using receiver operating characteristic (ROC) curve analysis. The diagnostic efficacy between models was compared by DeLong’s test. LASSO regression was employed to identify stable indicators. A multivariable logistic regression model was constructed, and its performance was evaluated using 1,000 bootstrap resamples. Clinical utility was assessed via Decision Curve Analysis (DCA).

Results

Distant metastasis was 44.7% of cases. Optimal cutoffs of 2.0 for NLR and 181.5 for PLR were identified. Both markers were significantly higher in metastatic group and positively correlated with disease progression. LASSO identified PLR, NLR, and tumor location as the most robust determinants. In multivariable analysis, only multiple tumor location and high PLR remained independent prognostic factors. The combined model integrating clinical factors with NLR and PLR significantly outperformed clinical features alone (AUC: 0.766 vs. 0.619, p=0.0036), with an optimism-corrected C-index of 0.708. At a 0.586 threshold, the model achieved 82.5% specificity and 62.7% sensitivity, supporting a conservative “rule-out” strategy. Calibration slope was 0.659. DCA demonstrated superior net benefit across a 15%–90% risk range; specifically, at a 40% threshold, model spared 19 per 100 patients from unnecessary intervention.

Conclusion

This study establishes the first baseline for NLR and PLR in the Vietnamese cohort, showing an association of these biomarkers with GC distant metastasis. A model integrating PLR, NLR and tumor location optimizes patient stratification and resource allocation in resource-constrained healthcare environments.

Plain Language Summary

The Challenge

Gastric cancer (stomach cancer) is a leading cause of cancer-related deaths worldwide. In Vietnam, many patients are diagnosed only after the cancer has already spread to distant parts of the body (metastasis). When cancer spreads, it becomes much harder to treat. Doctors need affordable and reliable ways to identify which patients are at a higher risk of metastasis to improve how they manage the disease.

The Study

Researchers studied 114 gastric cancer patients at the Ho Chi Minh City Oncology Hospital. They looked at two specific markers found in routine, inexpensive blood tests: the Neutrophil-to-Lymphocyte Ratio (NLR) and the Platelet-to-Lymphocyte Ratio (PLR). These markers measure the balance of different white blood cells and platelets, which partially reflect the body’s “inflammation” levels in response to a tumor.

The Findings

The study found that patients with advanced, metastatic cancer had significantly higher NLR and PLR levels compared to those in earlier stages. By combining these blood markers with other clinical information—such as the patient’s and tumor’s data —the researchers created a diagnostic model. This combined model was much more accurate at identifying patients with distant metastasis than using clinical information alone.

The Impact

Because these blood tests are simple, low-cost, and already widely available in hospitals, they offer a practical way for doctors to monitor cancer progression. Using the NLR and PLR ratios can help healthcare providers in Vietnam and elsewhere better identify high-risk patients and personalize their treatment plans more effectively.

Keywords

gastric cancer distant metastasis systemic immune-inflammatory markers diagnostic value

Introduction

Gastric cancer (GC) is one of the most common cancers in the world, ranking fifth in incidence and fourth in mortality among all cancer types. In Vietnam, GC is among the five most prevalent cancers in both sexes, accounting for 9.8% of all cancer diagnoses.¹

Despite significant advances in detection and disease management through screening recommendations and Helicobacter pylori eradication strategies, the survival prognosis for GC remains poor. This is largely due to the fact that most cases present with distant metastases at the time of diagnosis,^2,3 underscoring the need for thorough investigation of distant metastasis at initial consultation.

While clinical and pathological features such as age, sex, ethnicity, location of the primary tumor, depth of invasion, lesion size, and histological differentiation have been reported to contribute to metastasis risk in GC,⁴ traditional tumor markers like CEA, CA 19-9, and CA 72-4 are primarily used for monitoring and have variable diagnostic performance for initial staging.^5-8 Advanced imaging like PET/CT is effective but hindered by high cost and limited accessibility, particularly in resource-constrained settings like Vietnam.⁹ This situation creates a critical need for accessible, cost-effective biomarkers to identify patients with distant metastasis, especially considering the disparity in advanced screening tools available in Vietnam compared to countries like Japan and Korea, which have established national screening programs.^10,11

Malignancy is increasingly recognized not merely as a localized cellular dysfunction, but as a state of chronic systemic inflammation.¹² This inflammatory environment is pivotal in cancer development and progression, where a persistent immune response facilitates tumor proliferation, angiogenesis, and metastasis.¹³ Consequently, systemic inflammatory-immune markers derived from routine blood counts, such as the neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR), have emerged as promising indicators of disease status.

The elevated NLR, reflecting increased neutrophils, is associated with pro-inflammatory cytokine secretion that facilitates tumor proliferation and metastasis, while concurrent lymphopenia indicates compromised anti-tumor immunity.^14,15 Similarly, increased PLR suggests enhanced platelet activation, which supports metastasis by protecting circulating tumor cells, promoting angiogenesis, and aiding tumor cell dissemination.^16,17 These markers have shown prognostic value in various solid tumors,^18-21 including Gastric Cancer.^22,23

The biological relevance and universal reliability of NLR and PLR are further underscored by their validation across a diverse spectrum of non-oncological conditions characterized by inflammatory dysregulation. In the field of autoimmune and chronic inflammatory disorders, these markers have been established as effective indicators of disease activity in Hashimoto’s thyroiditis,²⁴ inflammatory bowel disease (IBD),^25,26 and the progression of liver fibrosis.²⁷ Their clinical utility extends to acute viral settings, most notably during the COVID-19 pandemic, where elevated NLR and PLR values served as critical predictors of mechanical ventilation requirements and overall disease severity.^28,29 Furthermore, significant correlations have been observed in cardiac conditions and metabolic syndromes such as Type 2 Diabetes Mellitus, where these ratios reflect the underlying low-grade systemic inflammation associated with insulin resistance.^30,31

The consistent correlation of NLR and PLR with disease severity across these disparate medical fields—ranging from metabolic and viral to autoimmune and neoplastic—proves their status as highly reliable “proof-of-concept” markers. By integrating data from both oncological and non-oncological literature, it is evident that these indices offer a practical and universally accessible method for monitoring the systemic inflammatory state, providing critical prognostic insights without the need for high-cost specialized testing.

Despite the globally recognized prognostic value of NLR and PLR, a significant “clinical gap” remains regarding their universal application. Evidence suggests that the optimal cutoffs for NLR and PLR are likely population-specific, influenced by ethnicity backgrounds, environmental exposures, and regional dietary patterns.^32-34 In the context of Vietnam—a region characterized by a high incidence of gastric cancer and a high prevalence of late-stage presentations—there is a clinical need for local calibration of these immune-inflammatory markers to accurately identify distant metastasis in GC.

Therefore, the objective of this study is to investigate the association of NLR and PLR with distant metastasis in Vietnamese GC patients, seeking to establish them as accessible, cost-effective tools for early identification of high-risk individuals and to visualize treatment strategies for improved clinical management.

Materials and Methods

Study Population and Data Collection

We collected the clinical data of gastric cancer patients at Ho Chi Minh City Oncology Hospital between January and June 2023. Patients were excluded if they met one of the following exclusion criteria: 1) concurrent or secondary malignant tumors at other sites. 2) Infection confirmed by laboratory tests. 3) Patients with systemic inflammatory conditions, history of autoimmune diseases, or currently undergoing anti-inflammatory treatment. 4) Patients diagnosed with hematological disorders or those who had received blood transfusions within the past three months. 5) Patients with organ failure (liver, kidney, heart, or other organs). 6) Patients with incomplete or unavailable pre-treatment medical records.

Ultimately, 114 patients with GC were retrospectively enrolled in this study. We collected data on age, sex, ECOG performance status, BMI (Body Mass Index), tumor features (primary tumor location, macroscopic morphology of the lesion, and histological differentiation grade of tumor cells), and disease stage according to the American Joint Committee on Cancer (AJCC) 8^th edition. Pre-treatment laboratory results included total white blood cell (WBC, 10⁹ cells/L), absolute neutrophil count (ANC, 10⁹ cells/L), absolute lymphocyte count (ALC, 10⁹ cells/L), red blood cell count (RBC, 10¹² cells/L), hemoglobin level (g/L), platelet count (PLT, 10⁹ cells/L), and other tumor markers such as CEA, CA19-9, and CA72-4, if available.

The NLR was calculated as the absolute neutrophil count divided by the absolute lymphocyte count. Similarly, PLR was calculated as the absolute platelet count divided by the absolute lymphocyte count.

All personal information of the patients included in the study was kept confidential and was used solely for research purposes. This study was approved by the hospital’s Ethics Committee of Ho Chi Minh City Oncology Hospital (Approval No. 292/BVUB-HĐĐĐ dated April 26, 2023).

Descriptive Statistics and Preliminary Comparisons

We first divided the patients into two groups: those with distant metastasis (stage IV) (n = 51) and those without distant metastasis (n = 63). Categorical variables are presented as percentages and compared between the two groups using the chi-square test (χ²) or Fisher’s exact test. The Kolmogorov-Smirnov test was used to test the normality of the continuous variables. Normally distributed continuous variables were expressed as mean ± standard deviation (SD) and compared between groups using Student’s t-test. In cases where continuous variables did not follow a normal distribution, variables were presented as median and interquartile range (IQR) and compared using the Wilcoxon-Mann-Whitney test.

To account for the non-normal distribution of hematological markers, the 95% confidence intervals (CIs) for the median differences in NLR and PLR between groups were estimated using non-parametric bootstrapping. We performed 1,000 bootstrap resamples to generate a distribution of the median differences. The 95% CIs were determined using the percentile method, defined as the 2.5^th and 97.5^th percentiles of the bootstrap distribution.

The correlation between two continuous variables was determined using Pearson’s correlation test, whereas Spearman’s rank correlation test was used to assess correlations between non-parametric data.

The optimal cut-off values for NLR, PLR and predicted probability of models were estimated as the value with the highest using the Youden index (J = sensitivity + specificity – 1).

Logistic Regression

Univariate and multivariate logistic regression analyses were conducted to identify independent predictors of distant metastasis. To address potential confounding and avoid the biases associated with univariate screening, we employed a forced-entry multivariable logistic regression model. All clinically relevant variables—including Age, Gender, BMI, Tumor Grade, Tumor Location, NLR, and PLR—were included in the initial model regardless of their univariate significance. This approach ensures that the independent diagnostic value of systemic inflammatory markers is assessed while adjusting for known clinical and pathological factors.

The odds ratio (OR) estimated from the regression analysis is reported with a 95% confidence interval (95% CI).

Variable Selection and Model Stability (LASSO)

To minimize overfitting and improve model stability, Least Absolute Shrinkage and Selection Operator (LASSO) penalized regression was performed. We utilized 10-fold cross-validation to identify the optimal penalty parameter (λ) that maximized the area under the receiver operating characteristic curve (AUC). Variables with non-zero coefficients at the minimum λ were considered the most robust predictors for the final model.

Sensitivity Analysis for Unmeasured Confounding (E-Value Analysis)

To assess the potential impact of unmeasured confounding and selection bias inherent in the retrospective study design, we conducted a sensitivity analysis by calculating E-values for the primary predictors in our multivariable logistic regression model. The E-value quantifies the minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need to have with both the exposure and the outcome to fully nullify the observed odds ratios. Calculations were performed following the methodology of VanderWeele and Ding (2017),³⁵ using the point estimates and the lower limits of the 95% confidence intervals to determine the robustness of our findings.

Internal Validation and Calibration Analysis

Internal validation was conducted using a bootstrapping procedure with 1,000 iterations to estimate and correct for optimism bias in model performance.

Receiver operating characteristic (ROC) analysis was used to evaluate the discriminative performance of the binary logistic regression models for distant metastasis. Comparisons between two AUC (Area Under the Curve) values were performed using the DeLong test.

Model calibration was evaluated using calibration plots and the Brier score to assess the agreement between predicted probabilities and observed outcomes.

Threshold Optimization and Diagnostic Performance

To translate the model’s performance into clinical utility, we performed a sensitivity and specificity trade-off analysis. The optimal clinical threshold was determined using the Youden Index with formula described above. Predictive values (PPV, NPV) and accuracy were calculated across a range of probability thresholds with the aim of characterizing the model’s performance in different clinical screening and diagnostic scenarios.

Post-Hoc Power Analysis

A post-hoc power analysis was performed using the pwr package in R.

Clinical Utility Analysis (Decision Curve Analysis, DCA)

To evaluate the clinical utility of the proposed model. The clinical net benefit was calculated across a range of threshold probabilities to compare the model’s performance against two baseline strategies: “Treat All” (intervention for all patients) and “Treat None” (no intervention). The potential clinical impact was further quantified by calculating the net reduction in unnecessary staging procedures per 100 patients. The analysis was conducted using R software using the dcurves package as described by Vickers et al.³⁶

General Statistical Considerations and Software

A p-value < 0.05 was considered at significant for all comparisons, correlations, and regression analyses. All statistical analyses and visualizations were performed using the R software (version 4.4.1).

The reporting of this study conforms to the STROBE checklist,³⁷ which was included in Table S1.

Results

Patient Characteristics

For the entire cohort in our study, the mean age of the patients was 58 years (range 48–68 years). Males accounted for 62.28% of the GC cases, and the male-to-female ratio was 1.6:1. One-third of the patients (30.7%) had a BMI < 18.5.

Regarding tumor location, the antrum was the most common site (31.57%), followed by multiple locations (24.56%), and the corpus (20.17%). In terms of histopathological grade, more than half of the cases (57.89%) were diagnosed as poorly differentiated adenocarcinoma. The predominant macroscopic tumor type was exophytic mass (31.58%), followed by ulcerative (17.54%). Most patients were diagnosed at an advanced stage, with 31.57% classified as stage III and 44.73% as stage IV. Among the patients with distant metastases, peritoneal metastasis was the most frequent (18 cases, 22.5%), followed by mesenteric lymph node metastasis (15 cases, 18.75%), para-aortic lymph node metastasis (14 cases, 17.5%), and hepatic metastasis (12 cases, 15%). Most metastatic cases occurred at a single site (56.86%), whereas metastases affecting three or more sites were observed in 11.76% of cases (Table 1).

Table 1.

Baseline Clinicopathological and Hematological Characteristics of Patients

Characteristics	Total (n= 114)	Non- distant metastasis (n= 63)	Distant metastasis (n= 51)	p-value
Sex (Case, %)				0.379
Male	71 (62.28%)	42 (66.67%)	29 (56.86%)
Female	43 (37.72%)	21 (33.33%)	22 (43.14%)
Age (Year, Mean ± SD)	58.281 ± 10.10	58.16 ± 10.25	58.43 ± 10.01	0.887
BMI (Median, IQR)	19.87 (18.05 – 22.28)	19.82 (17.68 – 22.27)	19.98 (18.26–22.30)	0.7
Primary tumor location (Case, %)				0.152
Antrum (%)	36 (31.57%)	24 (38.09%)	12 (23.52%)
Multiple sites (%)	28 (24.56%)	12 (19.04%)	16 (31.37%)
Body (%)	23 (20.17%)	11 (17.46%)	12 (23.52%)
Cardia region (%)	14 (12.28%)	7 (11.11%)	7 (13.73%)
Pylorus (%)	11 (9.64%)	9 (14.28%)	2 (3.92%)
Entire stomach (%)	1 (0.87%)	0	1 (1.96%)
Fundus (%)	1 (0.87%)	0	1 (1,96%)
Pathological grading (Case, %)				0.972
Grade 1 (Well differentiated)	2 (1.75%)	1 (1.58%)	1 (1.96%)
Grade 2 (Moderate differentiated)	46 (40.35%)	25 (39.68%)	21 (41.18%)
Grade 3 (Poor differentiated)	66 (57.89%)	37 (58.73%)	29 (56.86%)
Lesion morphological (Case, %)				0.157
Mass, Type 1	36 (31.58%)	18 (28.57%)	18 (35.29%)
Ulcerative, Type 2	20 (17.54%)	14 (22.22%)	6 (11.76%)
Infiltrative ulcerative, Type 3	18 (15.78%)	8 (12.70%)	10 (19.60 %)
Diffuse infiltrative, Type 4	10 (8.77%)	4 (6.34%)	6 (11.76%)
Unclassifiable, Type 5	18 (15.78%)	9 (14.28%)	9 (17.64%)
No information	12 (10.52%)	10 (15.87%)	2 (3.92%)
TNM Stage (American Joint Committee on Cancer 8 ^th , 2017)
I	6 cases (5.26%)	I-B: 6 cases
II	21 cases (18.42%)	II-A: 7 cases II-B: 14 cases
III	36 cases (31.57%)	III-A: 29 cases III-B: 5 cases III-C: 2 cases
IV	51 cases (44.73%)
Distant metastatic sites (Case, %)
Peritoneum			18 (22.5%)
Mesenteric lymph nodes			15 (18.75%)
Para-aortic lymph nodes			14 (17.5%)
Liver			12 (15%)
Left supraclavicular lymph nodes			9 (11.25%)
Lungs			3 (3.75%)
Abdominal wall Pelvic (iliac) lymph node Ovaries			2 (2.5%)
Carcinomatosis Adrenal gland Mediastinum lymph node			1 (1.2%)
Number of distant metastatic sites (Case, %)
1 site			29 (56.86%)
2 sites			16 (31.37%)
≥ 3 sites			10 (11.76%)
Hematological parameters
Red blood cell count (10¹²/l, median, IQR)	4.31 (4.03 – 4.63)	4.28 (4.02 – 4.58)	4.35 (4.03 – 4.8)	0.41
White blood cell count (10⁹/l, median, IQR)	7.36 (6.27 – 9.20)	6.98 (6 – 9.1)	7.52 (6.31 – 9.43)	0.57
Platelet (10⁹/l, median, IQR)	332.5 (243.8 – 434.3)	327 (236– 420)	356 (277 – 454)	0.13
Absolute Neutrophil count (10⁹/l, median, IQR)	4.20 (3.22 – 5.80)	4.0 (2.80 – 5.83)	4.62 (3.40 – 6.20)	0.039
Absolute Lymphocyte count (10⁹/l, median, IQR)	2.04 (1.60 – 2.62)	2.26 (1.90 – 2.72)	1.70 (1.50 – 2.50)	0.0023
NLR (median, IQR)	2.08 (1.41 –2.83)	1.91 (1.14 – 2.82)	2.47 (1.67 – 3.82)	0.00072
PLR (median, IQR)	168.78 (110.82-236.40)	147.81 (99.47 – 189.52)	186.0 (130.5–266.7)	0.0039

Continuous variables are presented as mean ± standard deviation (SD) for normally distributed data or median (interquartile range [IQR]) for non-normally distributed data. Categorical variables are reported as percentages (%). Abbreviations: BMI, Body Mass Index; TNM, tumor-node-metastasis; AJCC, American Joint Committee on Cancer; NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; ANC, absolute neutrophil count; ALC, absolute lymphocyte count; RBC, red blood cell count; PLT, platelet count. Bold values indicate statistically significant results (P < 0.05).

Hematological parameters showed that no significant differences were observed in red blood cell, platelet, or total white blood cell counts between two groups of patients with and without distant metastases (Wilcoxon test: red blood cells, p = 0.41; total white blood cells, p = 0.57; platelets, p = 0.13) (Table 1). However, we noticed that patients with distant metastases displayed a higher absolute neutrophil count compared to those without metastases, with median values of 4.62 × 10⁹cells/L (IQR: 3.40–6.20) for metastasis group and 4.0 × 10⁹cells/L (IQR: 2.80–5.83) for non-metastasis group, median difference 0.62 × 10⁹cells/L (95% CI: 0.022–1.250), p = 0.039 (Wilcoxon test). Conversely, the absolute lymphocyte count was statistically lower in the metastatic group (1.70 × 10⁹cells/L, IQR: 1.50–2.50) compared to the non-metastatic group (2.26 × 10⁹cells/L, IQR: 1.90–2.72), median difference 0.56 × 10⁹cells/L (95% CI: 0.095–0.594), p = 0.0023 (Wilcoxon test) (Figure 1).

Therefore, the neutrophil-to-lymphocyte ratio (NLR) was significantly elevated in the group of patients with distant metastases, median difference 0.56 (95% CI: 0.367–1.237), p = 0.00072 (Wilcoxon test). Similarly, a significant difference in the platelet-to-lymphocyte ratio (PLR) was observed between the two patient groups, median difference 38.19 (95% CI: 17.78–89.95), p = 0.0039 (Wilcoxon test) (Figure 1).

Significant differences were observed in inflammatory markers based on metastatic status. The median NLR was higher in the metastatic group, with a median difference of 0.57 (95% bootstrapped CI: 0.12–0.97) compared to the non-metastatic group. Similarly, the median PLR showed a significant elevation in patients with metastasis, with a median difference of 38.19 (95% bootstrapped CI: 6.84–77.62). Because the 95% CIs for both parameters did not include zero, these differences were considered statistically significant, reinforcing the association between elevated systemic inflammatory ratios and tumor progression. (Table S2).

Increased NLR and PLR Correlated With Disease Progression

Our study next investigated the dynamic change of NLR and PLR across different stages. Our study here showed that the median NLR was 1.47 in stage I gradually rising to 1.84, stage II, 1.98 and stage III, and reached the highest value of 2.47 when distant metastasis was present (stage IV). It has been shown that the NLR is significantly different between stages II and IV (p = 0.00055, Wilcoxon test) and between stages III and IV (p = 0.031, Wilcoxon test), but no difference was observed in stage I and IV (Figure 2).

Similarly, PLR displayed a stage-dependent increase. The median PLR was 150.48 in stage I, 173.07 in stage II, 172.14 in stage III, and 186.00 in stage IV. When comparing the PLR values between stage II and stage IV, as well as between stage III and stage IV, the results also indicated a statistical elevation (Wilcoxon test, p = 0.012 and p = 0.035, respectively) (Figure 2).

Our study also demonstrated that both NLR and PLR had a positive correlation with the progression of disease stages, with Spearman’s correlation coefficients of r = 0.33 and r = 0.27, respectively (p < 0.05), indicating that NLR and PLR tended to increase with disease advances. Additionally, NLR was positively correlated with PLR (r = 0.63). Furthermore, no correlation was observed between NLR or PLR and physiological factors, such as age or BMI, in gastric cancer patients in this study (Figure 3).

The Association of NLR and PLR With Gastric Cancer Distant Metastasis

ROC curve analysis and the Youden index were used to determine the optimal cut-off values and discriminative efficiency of NLR and PLR for GC with distant metastasis. The results indicated that, with a cut-off value of 2.0, NLR exhibited a sensitivity of 64.7% and specificity of 63.5% in predicting distant metastasis in GC (Figure 4A). For PLR, the optimal cutoff value was 181.5 with a sensitivity of 60.8% and a specificity of 73.0% (Figure 4B).

The probability of predicting distant metastasis in GC was assessed using binary logistic regression for two models: one incorporating only clinical factors and another integrating both clinical factors with the NLR and PLR. The clinical factors analyzed in the regression analysis included age, sex, BMI, primary tumor location, and histological differentiation. The variables were categorized as follows: age (>60 vs. ≤60 years), sex (male vs. female), BMI (≥18.5 vs. <18.5 kg/m²), primary tumor location (multi-site vs. single-site), histological differentiation (poorly differentiated vs. moderately to well differentiated), NLR (≥2.0 vs. <2.0), and PLR (≥181.5 vs. <181.5).

The optimal cutoff value for the model incorporating only clinical factors (Model P1) was 0.4, with a sensitivity of 66.7% and a specificity of 50.8%. The discriminatory power of the final model was evaluated using Receiver Operating Characteristic (ROC) curve analysis (see Figure 4D). The model of NLR and PLR combined with clinical features (Model P2) demonstrated noticeable diagnostic performance, with an optimal trade-off point identified at a threshold of 0.586. At this value, the model achieved a specificity of 82.5% and a sensitivity of 62.7%. Of note, the positive predictive value (PPV) and negative predictive value (NPV) at this threshold for combined model were 74.4% and 73.2%, respectively. (Table 2).

The AUC for NLR in predicting distant metastasis in GC was 0.685 (95% CI: 0.587–0.782), while that for PLR was 0.658 (95% CI: 0.556–0.759). The AUC for the clinical factor-based model was 0.619 (95% CI: 0.515–0.723), whereas the combined model achieved an AUC of 0.766 (95% CI: 0.676–0.857) (Figure 4E).

The AUC of Model P1 (clinical factors only) was used as a reference to compare the discriminative performance of NLR, PLR, and the combined model (Model P2). The results demonstrated that incorporating NLR and PLR into the clinical model significantly improved the diagnostic accuracy for distant metastasis (DeLong test, p = 0.00356) (Table 2).

Table 2.

Diagnostic Performance and Predictive Accuracy of Inflammatory Markers and Clinical Models for Distant Metastasis

	AUC	CI 95%	Cutoff	Sen	Spec	PPV	NPV	p-value
NLR	0.685	0.587 – 0.782	2.0	0.647	0.635	0.589	0.689	0.339
PLR	0.658	0.556 – 0.759	181.5	0.608	0.730	0.646	0.697	0.573
P2	0.766	0.676 – 0.857	0.59	0.627	0.825	0.744	0.732	0.00356
P1	0.619	0.515 – 0.723	0.41	0.667	0.508	0.523	0.653	Ref

Comparison of the area under the curve (AUC), optimal cut-off values (determined by the Youden index), sensitivity, specificity, and predictive values. Model P1 includes age (categorized as ≤60 vs. >60 years), sex (female vs. male), BMI (≥18.5 vs. <18.5 kg/m²), primary tumor location (single-site vs. multi-site), histological differentiation (poorly differentiated vs. moderately to well differentiated). Model P2 integrates these clinical features with NLR (<2.0 vs. ≥2.0) and PLR (<181.5 vs. ≥181.5). P values for AUC comparisons were calculated using DeLong’s test. Abbreviations: CI, confidence interval; Sen, sensitivity; Spe, specificity; PPV, positive predictive value; NPV, negative predictive value; Ref, reference. Bold values indicate statistically significant results (P < 0.05).

In the combined model, univariate logistic regression analysis indicated that GC tumors with multiple locations (OR: 7.44; 95% CI: 1.84–50.0, p = 0.012), high NLR (≥2.0) (OR: 3.19; 95% CI: 1.49–7.00, p = 0.003), and high PLR (≥181.5) (OR: 4.19; 95% CI: 1.93–9.44, p < 0.001) were associated with distant metastasis. Next, to minimize the residual confounding, we utilized a forced-entry approach that integrates all variables, regardless of their univariate significance (including Age, Gender, BMI, Tumor histological grade, Primary tumor location, NLR and PLR), into multivariate logistic regression model. The results showed that two independent predictors of metastasis were identified. Patients with multiple tumor locations exhibited significantly higher odds of metastasis compared to those with single-site involvement (OR: 7.04; 95% CI: 1.50–53.5; p = 0.026). Additionally, patient with PLR ≥ 181.5 was significantly associated with nearly three-fold increased odds of metastatic disease (OR: 2.95; 95% CI: 1.12–7.98; p = 0.03). Other factors, including Age, Gender, BMI, Tumor Grade, and NLR, did not maintain independent statistical significance within the multivariable framework (p > 0.05). (Table 3).

Table 3.

Univariate and Multivariate Binary Logistic Regression Analyses of Risk Factors Associated With Distant Metastasis

Characteristics	Univariate analysis			Multivariate analysis
Characteristics	OR	95% CI	p-value	OR	95% CI	p value
Age
≤ 60yrs	1.00	—	—	1.00	—	—
> 60yrs	1.20	0.57–2.50	0.63	1.11	0.46–2.66	0.8
Sex
Female	1.00	—	—	1.00	—	—
Male	0.66	0.31–1.41	0.28	0.82	0.34–2.01	0.7
BMI
≥ 18.5	1.00	—		1.00	—	—
< 18.5	0.76	0.33–1.69	0.49	0.98	0.39–2.46	>0.9
Histological grading
Intermediate/Well differentiation	1.00	—	—	1.00	—	—
Poor differentiation	0.93	0.44–1.96	0.84	0.83	0.35–1.97	0.7
Tumor location
Single site	1.00	—	—	1.00	—	—
Multiple sites	7.44	1.84–50.0	0.012	7.04	1.50–53.5	0.026
NLR
Low risk (NLR <2.0)	1.00	—	—	1.00	—	—
High risk (NLR ≥2.0)	3.19	1.49–7.00	0.003	1.92	0.72–5.07	0.2
PLR
Low risk (PLR <181.5)	1.00	—	—	1.00	—	—
High risk (PLR ≥181.5)	4.19	1.93–9.44	<0.001	2.95	1.12–7.98	0.03

Odds Ratios (ORs) and 95% Confidence Intervals (CIs) are presented. Only variables significant in univariate analysis were included in the final multivariate model. Bold values indicate statistically significant results (P < 0.05).

To mitigate the risk of overfitting and to validate the predictors identified in the forced-entry multivariate model, we further performed a Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis. The LASSO model identified three primary predictors with non-zero coefficients: Low risk PLR (PLR <181.5) (Coefficient: -0.359), Tumor Location at single site (Coefficient: -0.260), and Low-risk NLR (Low risk (NLR < 2.0s) (Coefficient: -0.118), confirming their robust association with the study outcome of distant metastasis (Table 4). The negative coefficient means that being in the “low risk” PLR and NLR groups is strongly associated with a reduction in the odds of metastasis. Similarly, having a tumor at a single site reduces the risk of metastasis compared to the multiple locations.

Table 4.

Variable Selection for Gastric Cancer Metastasis Using LASSO Regression

Rank	Variable	Coefficient	Importance	Selected
1	PLR low risk (PLR <181.5)	-0.3596206	0.3596206	Yes
2	Tumors location at multiple sites	-0.2608228	0.2608228	Yes
3	NLR low risk (NLR < 2.0)	-0.1187308	0.1187308	Yes
4	Age ≤ 60yrs	0.0000000	0.0000000	No
5	Sex: Male	0.0000000	0.0000000	No
6	BMI < 18.5	0.0000000	0.0000000	No
7	Tumors with poor differentiation	0.0000000	0.0000000	No

This table presents the coefficients and importance rankings of potential predictors evaluated via the Least Absolute Shrinkage and Selection Operator (LASSO) method. Variables with non-zero coefficients were selected as the most robust predictors. Importance was calculated based on the absolute value of the LASSO coefficients. Abbreviations: NLR, neutrophil-to-lymphocyte ratio; PLR, platelet-to-lymphocyte ratio; BMI, body mass index. Note: Variables marked as “No” in the Selected column had their coefficients reduced to zero during the regularization process, indicating they were not significant predictors in this penalized model.

By contrast, variables such as Age, Gender, BMI, and Tumor Grade were excluded by the LASSO algorithm as they all have coefficients of 0, suggesting that they did not contribute significant independent diagnostic value in this specific cohort.

To further scrutinize the stability of our multivariate model, against the potential impact of unmeasured confounding, we performed a sensitivity analysis using E-values for the primary predictors (e.g., Tumor location, PLR and NLR) in our logistic regression model. The analysis revealed that the association between tumor location (multiple sites) and distant metastasis was exceptionally robust, yielding an E-value of 13.56 (lower 95% CI: 2.37). For patients classified as high-risk via PLR, the E-value was 5.35 (lower 95% CI: 1.49). Conversely, while NLR showed a point-estimate E-value of 3.25, its lower confidence limit reached 1.00, reflecting the statistical non-significance observed in the final adjusted model (Table S3).

Model Performance and Internal Validation

To assess the internal validity and generalizability of the multivariable model, we performed a bootstrap validation using 1,000 resamples. With the bootstrap-corrected Somer’s Dxy of 0.4155, the original C-index (AUC) of 0.766 was adjusted for optimism to a corrected C-index of 0.707, indicating robust discriminative ability. This drop in discriminative performance (optimism = 0.1176) reflects the expected variance when applying the model to independent datasets. Furthermore, the bootstrapping validation displayed a corrected R² of 0.1096 and a maximum calibration error (E_max) of 0.147 (Table S4).

The calibration slope was calculated at 0.659. (Table S4). The calibration plot (Figure 5) indicated that the model is well-calibrated in the low-to-middle risk predictions (0.4–0.6). However, in the high-risk spectrum (>0.6), the model tended to overestimate the probability of metastasis, consistent with the calculated calibration slope of 0.659. The bootstrap-corrected curve remained within the 95% confidence intervals of the ideal prediction line for the majority of the patient distribution.

The Apparent Brier score was 0.1970, indicating that diagnosis model better than chance, but once we accounted for overfitting, the score rose to 0.227 (The Optimism-corrected Brier score).

Clinical Threshold Optimization and Diagnostic Performance

Trade-off analysis at varying probability levels revealed that lowering the threshold to 0.20 increased sensitivity to 98.0% at the cost of significantly reduced specificity (6.3%). Conversely, a strict threshold of 0.80 provided near-perfect specificity (96.8%) but caught only 9.8% of metastatic cases. The optimal 0.586 threshold defined by Youden index reflects a balanced clinical strategy prioritizing the avoidance of false positives (Specificity: 82.5%) while maintaining moderate sensitivity (62.7%) (Figure 4D, Table 5).

Table 5.

Predictive Performance of the Model Across Various Probability Thresholds for Metastasis

Threshold	Specificity	Sensitivity	PPV	NPV
0.2	0.06349206	0.98039216	0.4587156	0.8000000
0.4	0.71428571	0.70588235	0.6666667	0.7500000
0.585797 (optimal threshold by Youden Index)	0.8253968	0.627451	0.744186	0.7323944
0.6	0.87301587	0.56862745	0.7837838	0.7142857
0.8	0.96825397	0.09803922	0.7142857	0.5700935

The table illustrates the trade-off between sensitivity and specificity at different probability cut-off points. The optimal threshold of 0.586 was determined using the Youden index (J = Sensitivity + Specificity - 1), achieving a balance with a specificity of 82.5% and a sensitivity of 62.7%. Lower thresholds (e.g., 0.2) prioritize high sensitivity to minimize missed metastatic cases, whereas higher thresholds (e.g., 0.8) maximize specificity to avoid false-positive results. Abbreviations: PPV, positive predictive value; NPV, negative predictive value.

Statistical Power and Study Robustness

To evaluate the adequacy of the study’s sample size (n = 114) for detecting significant differences between the metastatic (n = 51) and non-metastatic (n = 63) cohorts, a post-hoc power analysis was performed for the primary inflammatory markers. For the NLR index, the difference between groups represented a medium-to-large effect size (Cohen’s d=0.705), achieving a statistical power of 96.0% at a significance level of α = 0.05. For the PLR index, a medium effect size was observed (Cohen’s d=0.569), yielding a power of 85.0%. These results indicate that the study was sufficiently powered to detect clinically meaningful differences, as both markers surpassed the conventional 80% power threshold (Table S5).

Clinical Utility and Decision Curve Analysis

The clinical utility of the diagnostic model was evaluated using Decision Curve Analysis (DCA) to estimate the net benefit across a range of threshold probabilities (Figure 6). Compared to the ‘Treat All’ and ‘Treat None’ strategies, the combined model of PLR, NLR and clinical features demonstrated superior clinical net benefit across nearly the entire range of threshold probabilities, from approximately 15% to 90%.

At a conservative threshold of 10%, the model’s net benefit (0.386) was identical to the ‘Treat All’ strategy. However, as the threshold increased, the model’s utility became more pronounced. At a 20% threshold, the model (Net Benefit = 0.322) outperformed ‘Treat All’ (Net Benefit = 0.309), avoiding 5.3 unnecessary staging procedures per 100 patients without missing any true metastatic cases. This benefit increased significantly at higher thresholds; at a 30% risk threshold, the model avoided 8.5 procedures per 100 patients. At the 40% threshold, where the ‘Treat All’ strategy’s utility declined sharply (0.079), the model maintained a high net benefit (0.208), effectively sparing 19.3 patients per 100 from unnecessary invasive staging while correctly identifying those at risk (Table 6).

Table 6.

Clinical Utility and Net Reduction of Unnecessary Procedures Based on Decision Curve Analysis (DCA)

Threshold probability (%)	Net benefit (Model)	Net benefit (Treat all)	Net reduction	Net reduction in interventions per 100 patients*	Interpretation
10%	0.386	0.386	0	0.0	No added benefit yet.
20%	0.322	0.309	5.26	5.3	Benefit begins. The model avoids ∼5.3 unnecessary staging procedures per 100 patients without missing any true metastasis.
30%	0.247	0.211	8.48	8.5	Strong benefit. The model avoids ∼8.5 unnecessary staging procedures per 100 patients.
40%	0.208	0.079	19.3	19.3	Maximum utility. The model avoids ∼19.3 unnecessary staging procedures per 100 patients.

This table quantifies the clinical impact of the predictive model across various threshold probabilities. “Net Benefit” represents the balance between true-positive and false-positive predictions. “Net Reduction” indicates the number of unnecessary staging procedures that could be avoided per 100 patients by using this model instead of a “treat-all” strategy.

*Net reduction in interventions represents the number of unnecessary staging procedures avoided per 100 patients without missing any cases of metastasis, compared with the “Treat All” strategy.

Discussion

Biological Significance and Diagnostic Utility of NLR and PLR for Distant Metastasis in Gastric Cancer

Our study demonstrated that NLR and PLR were significantly elevated in patients with gastric cancer with distant metastasis, which is consistent with previous global studies.^22,38,39 Likewise, our study agreed with previous report that increased PLR in GC patients was linked to a higher risk of advanced disease stage compared to those with lower PLR levels.²³

The use of bootstrapped confidence intervals confirms that the observed elevations in NLR and PLR are not merely artifacts of sample outliers but represent robust statistical differences. These findings align with the role of systemic inflammation in the pre-metastatic niche, where elevated neutrophil and platelet counts relative to lymphocytes signify a pro-tumorigenic environment.^12,16,40,41 By utilizing a non-parametric bootstrap approach, we have provided a more rigorous validation of these ratios as stable biomarkers for identifying metastatic risk in this patient cohort.

In our study, we used the Youden index to determine the optimal NLR and PLR thresholds for diagnosing distant metastasis in a cohort of 114 patients with GC. The results indicated that both NLR and PLR possess moderate diagnose value for distant metastasis in gastric cancer (GC) patients, with optimal cut-off values of 2.0 for NLR and 181.5 for PLR.

The consistency of our NLR and PLR cut-off values with established literature underscores the robustness of these inflammatory biomarkers in assessing clinical outcomes. Our NLR threshold aligns closely with the values reported by Fang et al (2020) (NLR _{optimal cut-off} =2.258),⁴² Zhang et al (2022) (NLR _{optimal cut-off} =2.91),³⁹ Jeong et al (2012) (NLR _{optimal cut-off} =3.0),⁴³ Jung et al (2011) (NLR _{optimal cut-off} =2.17),⁴⁴ and Kim et al (2018) (NLR _{optimal cut-off} =2.88),⁴⁵ Regarding PLR, Kim et al (2018) also indicated the cutoff level for the PLR was 172.⁴⁵ In a meta-analysis by Zhang et al (2020), including 49 studies comprising 51 cohorts, the authors noted that the PLR cutoff values ranged from 10.1 to 350, with a threshold of >150 being associated with poorer overall survival in GC patients.²³ Collectively, these results suggest that NLR and PLR are reliable markers of the systemic inflammatory response, with our specific thresholds supported by high-impact studies across various clinical settings in the specific context of gastric cancer.

Furthermore, we observed a progressive increase in NLR and PLR values, which correlated with disease stage advancement. The significant differences in NLR between stages II and IV, as well as between stages III and IV, underscore the potential utility of NLR in distinguishing locally advanced disease from metastatic gastric cancer. In contrast, the absence of a significant difference between stages I and IV suggests that NLR alone may have limited sensitivity in capturing early stage disease possibly due to the relatively preserved immune homeostasis in localized tumors or may also be partially attributable to the limited number of patients with stage I disease in our cohort (6 cases). Collectively, these results support the role of NLR as a marker of disease progression rather than early detection, highlighting its potential value in advanced-stage risk stratification and prognosis assessment.

The Additive Value of Integrative Models in the Vietnamese Cohort

The ROC analysis revealed that NLR had a sensitivity of 64.7% and specificity of 63.5%, while PLR showed slightly lower sensitivity (60.78%) but higher specificity (73.02%).

The AUC for NLR and PLR in predicting distant metastasis in GC were 0.685 and 0.658, respectively. Integrating NLR and PLR with established clinical factors (age, sex, BMI, primary tumor location, histological differentiation–categorized as binary groups) significantly improved the discriminative accuracy for distant metastasis compared to clinical factors alone. The combined model (Model P2) achieved a fair-to-good discriminatory capacity, with an AUC of 0.76, outperforming the clinical-only model (Model P1; AUC 0.619) and individual biomarkers. This enhancement underscores the additive value of systemic inflammatory markers in risk stratification frameworks, highlighting their potential utility in clinical decision-making to identify high-risk GC patients more reliably.

Numerous studies have investigated the prognostic and predictive value of NLR, PLR, or their combination in the survival outcomes of patients with GC^46-48 However, the potential application of NLR and PLR in combination with clinical factors for predicting distant metastasis, a critical stage that requires well-planned treatment strategies, remains relatively underexplored. To the best of our knowledge, this is the first investigation on prognostic value of NLR and PLR in a Vietnamese gastric cancer patients cohort using a population-specific cutoff, thereby shaping a framework for future disease management but also facilitating methodological alignment with international studies. By integrating systemic inflammatory markers with conventional clinical parameters, our study not only establishes the first population-specific diagnostic framework for Vietnamese gastric cancer patients but also demonstrates a significant enhancement in discriminative accuracy, outperforming models based on clinical factors alone.

Independent Drivers of Metastasis and Assessment of Model Stability

Multivariable regression analysis identified multiple tumor locations and elevated PLR as independent prognostic factors for distant metastasis in this Vietnamese GC cohort. While high NLR was significant in univariate analysis (OR = 3.19; p = 0.003), it lost independent significance when adjusted for other variables (OR = 1.92; p = 0.2). Moreover, the analysis showed that PLR remained an independent prognostic factor upon multivariate logistic regression (OR = 2.95; p = 0.03), suggesting that PLR, rather than NLR, is the primary independent inflammatory indicator for metastatic risk in our multivariate model. The significance of multiple tumor locations further emphasizes the complexity of tumor biology influencing metastatic spread (OR = 7.04; p = 0.026).

Beyond standard performance metrics, the clinical reliability of our markers is further supported by the E-value sensitivity analysis. The notably high E-value for multiple tumor locations (13.56) suggests that an unmeasured confounder would need an association magnitude nearly fourteen-fold higher than the observed factors to nullify this finding—a scenario highly improbable in clinical gastric cancer pathology. This indicates that tumor location remains a dominant and independent biological driver of metastatic spread. While the E-value for high-risk PLR (5.35) was more modest, it remains substantially higher than the typical impact of common clinical confounders, reinforcing its utility as a stable biomarker. By contrast, the vulnerability of NLR to unmeasured bias (lower E-value limit of 1.00) suggests its role may be more adjunctive than primary. From a strategic perspective, prioritizing markers with high E-values like tumor location and PLR allows for a more resilient triage system. The robustness of PLR as an independent metastatic driver indicated by its high E-value and biological role in platelet-mediated immune evasion suggest that it provides a more stable and resilient signal for risk stratification than NLR in the multivariable context.

Although NLR demonstrated superior discriminative power for distant metastasis in univariate analysis, its loss of statistical significance in the multivariable model–while PLR remained significant–likely stems from the moderate correlation between these two markers (r=0.63). This suggests a degree of multicollinearity and a substantial overlap in the physiological information they provide, which may have attenuated the independent diagnostic value of NLR. From a biological perspective, while NLR reflects general systemic inflammation, the independent significance of PLR may specifically highlight the critical role of platelet-mediated tumor angiogenesis and immune evasion—mechanisms that appear to be more direct drivers of distant spread in gastric cancer than the pathways represented by NLR alone.

The results of the LASSO regression mitigated concerns of optimism bias and provide critical validation for our primary findings, thus emphasizing the biological importance of systemic inflammatory markers. The retention of PLR and NLR as top predictors aligns with growing oncological evidence suggesting that the systemic inflammatory response plays a pivotal role in tumor progression and patient outcomes.^12,16 NLR was selected by LASSO even though it wasn’t statistically significant in previous multivariable logistic regression model, suggesting that NLR still has some minor indicative value that should not be completely ignored. By penalizing less impactful variables, the LASSO method effectively isolated these inflammatory ratios and tumor location as the most stable determinants of the outcome, even when demographic factors such as age and BMI were considered. The exclusion of Age, BMI, and Tumor Grade by the LASSO model is particularly noteworthy. While these factors are traditionally considered relevant in clinical assessments, our analysis suggests that in the presence of biochemical markers like PLR and NLR, their relative diagnostic weight diminishes.

While our model achieved an AUC of 0.766—representing moderate discrimination—this performance is consistent with previous studies using inexpensive systemic inflammatory markers in gastric cancer, such as those by Zhang et al (2022) (AUC=0.679, 95% CI: 0.653–0.705), Zhao et al, (2022) (AUC = 0.699, 95% CI: 0.675-0.721), Jiang et al (2016) (AUC=0.694), which found similar utility for these ratios in diagnosis and preoperative staging.^39,49,50

The clinical strength of this model lies in its high specificity of 82.5% at the 0.586 threshold, offering a conservative clinical strategy. By serving as a robust filter during initial assessment for GC patients, the model mitigates the risk of false positives and effectively “rule-out” metastasis in low-risk cases, thereby averting unnecessary costly imaging or invasive interventions and optimizing the use of diagnostic infrastructure, especially in resource-limited settings.

Methodological Rigor: Internal Validity, Calibration, and Limitations

The discriminative capacity of our diagnostic model was robust, as evidenced by an initial apparent C-index of 0.76. To rigorously assess generalizability and account for potential optimism, we performed internal validation using 1,000 bootstrap resamples, which yielded an optimism-corrected C-index (AUC) of 0.708. While this slight attenuation in performance is expected, it confirms that the model maintains moderate-to-good discriminative power for predicting distant metastasis. However, this drop in performance is intrinsically linked to the calibration of the model, which revealed more nuanced patterns of diagnostic behavior.

Specifically, the calibration slope of 0.6593 indicates a characteristic degree of overfitting, suggesting that the model tends to generate “extreme” predictions. This was most evident in the calibration plot, where we observed a distinct pattern of underestimating risk within the low-probability range (0.2–0.4) and overestimating it when predicted probabilities exceeded 0.6. This instability at the distribution tails is likely a direct consequence of data sparsity in the extreme risk categories (below 0.2 and above 0.8), where limited sample sizes resulted in wider confidence intervals and less stable estimates. Consequently, while the model is highly reliable within the mid-probability spectrum, its estimates at the ends of the risk spectrum should be interpreted with caution.

These calibration challenges are reflected in the overall measures of diagnostic precision and explained variance. The optimism-corrected Brier score of 0.227 remains below the 0.25 threshold—the level of a non-informative prediction—thereby confirming that the model provides genuine clinical value over a null baseline. Nevertheless, the corrected R2 of 0.1096 indicates that approximately 11% of the total variance in metastatic outcomes is accounted for by our current parameters and in the worst-case scenario, the model’s discriminative probability could be off by as much as 14.7% (Emax = 0.147).

This modest explanatory power suggests that while hematological and clinical markers are significant prognostic indicators, they represent only a portion of the complex biological architecture driving metastasis. The remaining variance likely stems from genomic or molecular drivers not yet captured in the current clinical framework, highlighting a clear opportunity for future biomarker integration.

Clinical Implementation and Triage in Resource-Limited Settings

The clinical utility of our model is underscored by its performance across multiple validation frameworks. In the sensitivity-specificity trade-off analysis, the selection of a 0.586 threshold prioritized high specificity (82.5%), reflecting a conservative clinical strategy essential for minimizing false-positive diagnoses in gastric cancer—a critical requirement in resource-constrained environments where over-staging leads to significant healthcare burden.

However, this high-specificity approach results in a moderate sensitivity, meaning approximately one-third of metastatic cases may not be flagged at this threshold. Consequently, the model should be viewed as a supplementary decision-support tool rather than a standalone diagnostic; its results must be integrated with clinical judgment and high-resolution imaging when clinical suspicion persists. However, the model’s inherent flexibility allows for threshold adjustment based on specific clinical objectives: an aggressive screening approach (threshold 0.20) could achieve 98.0% sensitivity to capture nearly all metastases, while a more stringent threshold (0.80) maximizes specificity (96.8%) to avoid overtreatment in high-morbidity scenarios.

The reliability of these clinical inferences is supported by the statistical robustness of our primary inflammatory predictors. Post-hoc power analysis confirmed that the study was adequately powered to detect real biological effects, with observed medium-to-large effect sizes for both NLR (Cohen’s d = 0.71; power = 96.0%) and PLR (Cohen’s d = 0.57; power = 85.0%). These metrics significantly exceed the conventional 80% power threshold required to minimize Type II errors, ensuring that the identified associations between systemic inflammation and metastatic risk are biologically grounded rather than transient statistical findings. While our post-hoc analysis confirmed the study was adequately powered (>80%) to detect differences in NLR and PLR, larger prospective multicenter studies would further validate these findings.

Finally, Decision Curve Analysis (DCA) demonstrated that the model provides a superior net benefit over “Treat All” or “Treat None” strategies across a broad probability range (15%–90%). Specifically, at a 40% risk threshold—where clinical uncertainty is often highest—the model maintains a high net benefit (0.208), effectively sparing approximately 19 patients per 100 from unnecessary invasive staging without compromising the identification of metastatic cases. This translates to a more precise allocation of resources, aligning with the principles of value-based oncology.

Limitations and Future Directions

Despite the clinical significance of our findings, several inherent constraints of this study warrant consideration. First, the single-center, retrospective design introduces a potential risk of selection bias and limits the immediate external generalizability of our findings beyond the Vietnamese population. While constrained by a single-center design, the recruitment from a high-volume national oncology center ensures a cohort that encompasses diverse regional cases. The results establish an initial baseline for cost-effective, minimally invasive biomarkers in resource-limited settings, paving the way for broader multi-center research and clinical validation across the region.

Second, the cross-sectional nature of our data collection means that NLR and PLR were assessed concurrently with the metastatic status at diagnosis. Consequently, the observed relationships are strictly associational rather than predictive in a temporal sense. These markers should be viewed as indicators of current disease status rather than longitudinal predictors of future metastatic progression.

Third, the correlation analysis between NLR and PLR suggests that their diagnostic values are not entirely independent, leading to the attrition of NLR in our multivariable model. This indicates that neutrophil-driven inflammation may be fundamentally channeled through platelet-mediated pathways, an area that warrants further causal analysis.

Future research should focus on validating these localized cut-off values (NLR 2.0 and PLR 181.5) through multi-center prospective trials and longitudinal studies. Such investigations are required to determine if these inflammatory ratios can predict the subsequent development of metastasis in patients initially diagnosed with localized disease. Ultimately, exploring the mechanistic link between systemic inflammation and the metastatic cascade may reveal novel prevention strategies and therapeutic targets for gastric cancer patients in Southeast Asia and beyond.

Conclusion

In conclusion, our findings suggest that NLR and PLR increase with disease progression in patients with gastric cancer. Of note, while both NLR and PLR are associated with metastasis in univariate logistic regression analysis, only PLR (and Tumor Location) remained independent prognostic factors when integrating these inflammatory indices with standard clinical features (Age, Sex, BMI, Tumor Histological Grade) into the multivariate regression. The combined model possesses significant discriminative performance for distant metastasis detection in gastric cancer, offering a cost-effective and accessible tool that requires no additional specialized equipment. Such a model holds substantial potential for the initial assessment and risk stratification of patients, particularly in clinical environments where advanced imaging may be constrained. Importantly, this study represents the first investigation in Vietnam to comprehensively evaluate the clinical relevance of these markers. These results lay a foundation for future multicenter, prospective studies to validate biomarker-based models, with the ultimate goal of optimizing management strategies for gastric cancer within resource-limited healthcare settings.

Footnotes

Acknowledgements

We extend our deepest appreciation to the Board of Directors of the Department of General Internal Medicine, as well as Ho Chi Minh City Oncology Hospital, for granting approval and providing essential support for this study.

ORCID iD

N. Tran

Ethical Considerations

This study was conducted in accordance with the Declaration of Helsinki and received formal approval from the Ethics Committee of Ho Chi Minh City Oncology Hospital (Approval No. 292/BVUB-HĐĐĐ, dated April 26, 2023).

Consent to Participate

As a retrospective chart review, the requirement for informed consent was waived. All patient data were anonymized and kept strictly confidential, used solely for the purposes of this research.

Consent for Publication

As this was a retrospective study utilizing de-identified secondary data from medical records, individual patient consent for publication was not required. No personally identifiable information, clinical photographs, or identifying details are included in this manuscript.

Author Contributions

Nhan Tran: Conceptualization; Data curation; Formal analysis; Visualization; Writing – original draft; Writing – review & editing.

Dang Ngoc Linh Tran: Data curation; Investigation; Formal analysis; Writing – review & editing.

Thi Hong Duc Phan: Data curation.

Hoang Quy Nguyen: Conceptualization; Methodology; Writing – review & editing; Supervision; Validation.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.*

References

Sung

Ferlay

Siegel

, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209-249.

Zhang

Lin

Duan

Mao

Wang

. A Population-Based Analysis of Distant Metastasis in Stage IV Gastric Cancer. Med Sci Monit. 2020;26:e923867.

Lee

, et al. Natural History of Gastric Cancer: Observational Study of Gastric Cancer Patients Not Treated During Follow-Up. Ann Surg Oncol. 2019;26(9):2905-2911.

Chen

Zhang

, et al. A clinical model to predict distant metastasis in patients with superficial gastric cancer with negative lymph node metastasis and a survival analysis for patients with metastasis. Cancer Med. 2021;10(3):944-955.

Zhang

Liu

Hong

. Combined use of AFP, CEA, CA125 and CAl9-9 improves the sensitivity for the diagnosis of gastric cancer. BMC Gastroenterol. 2013;13:87.

Lai

Lee

Huang

Lin

. Comparison of serum CA72-4, CEA, TPA, CA19-9 and CA125 levels in gastric cancer patients and correlation with recurrence. Hepatogastroenterology. 2002;49(46):1157-1160.

Emoto

Ishigami

Yamashita

Yamaguchi

Kaisaki

Kitayama

. Clinical significance of CA125 and CA72-4 in gastric cancer with peritoneal dissemination. Gastric Cancer. 2012;15(2):154-161.

Yang

Shen

. Predictive value of serum CEA, CA19-9 and CA72.4 in early diagnosis of recurrence after radical resection of gastric cancer. Hepatogastroenterology. 2011;58(112):2166-2170.

Kawanaka

Kitajima

Fukushima

, et al. Added value of pretreatment (18)F-FDG PET/CT for staging of advanced gastric cancer: Comparison with contrast-enhanced MDCT. Eur J Radiol. 2016;85(5):989-995.

10.

Hamashima

. Current issues and future perspectives of gastric cancer screening. World J Gastroenterol. 2014;20(38):13767-13774.

11.

Cho

Kang

Choi

Suh

Jun

Park

. Cost-effectiveness outcomes of the national gastric cancer screening program in South Korea. Asian Pac J Cancer Prev. 2013;14(4):2533-2540.

12.

Balci

Aktas

. A Comprehensive Review of the Role of Hemogram Derived Inflammatory Markers in Gastrointestinal Conditions. Iran J Colorectal Res. 2022;10:75-86.

13.

Balkwill

Mantovani

. Inflammation and cancer: back to Virchow? Lancet. 2001;357(9255):539-545.

14.

Koenderman

Vrisekoop

. Neutrophils in cancer: from biology to therapy. Cell Mol Immunol. 2025;22(1):4-23.

15.

Ménétrier-Caux

Ray-Coquard

Blay

Caux

. Lymphopenia in Cancer Patients and its Effects on Response to Immunotherapy: an opportunity for combination with Cytokines? J Immunother Cancer. 2019;7(1):85.

16.

Borsig

. The role of platelet activation in tumor metastasis. Expert Rev Anticancer Ther. 2008;8(8):1247-1255.

17.

Xue

Deng

Qin

, et al. The interaction of platelet-related factors with tumor cells promotes tumor metastasis. J Transl Med. 2024;22(1):371.

18.

Gui

Wang

Luo

Wang

. Platelet to lymphocyte ratio as a prognostic factor in patients with advanced colorectal cancer undergoing palliative treatment. Ann Palliat Med. 2020;9(5):3271-3277.

19.

Guo

Liu

, et al. Prognostic value of neutrophil-to-lymphocyte ratio and platelet-to-lymphocyte ratio for breast cancer patients: An updated meta-analysis of 17079 individuals. Cancer Med. 2019;8(9):4135-4148.

20.

Steinberg-Vorhoff

Tropotel

Theysohn

, et al. Evaluation of Inflammatory Markers as Prognostic Factors in the Treatment of Hepatocellular Carcinoma (HCC) with Degradable Starch Microspheres by Transarterial Chemoembolization (DSM-TACE). Cancers (Basel). 2025;17(4):647.

21.

Templeton

McNamara

Šeruga

, et al. Prognostic role of neutrophil-to-lymphocyte ratio in solid tumors: a systematic review and meta-analysis. J Natl Cancer Inst. 2014;106(6):dju124.

22.

Sahin

Aydin

Unver

Pehlivanoglu

. Predictive Value of Preoperative Neutrophil Lymphocyte Ratio in Determining the Stage of Gastric Tumor. Med Sci Monit. 2017;23:1973-1979.

23.

Zhang

Zhao

, et al. Clinicopathological and prognostic significance of platelet-lymphocyte ratio (PLR) in gastric cancer: an updated meta-analysis. World J Surg Oncol. 2020;18(1):191.

24.

Erge

Kiziltunc

Balci

, et al. A Novel Inflammatory Marker for the Diagnosis of Hashimoto’s Thyroiditis: Platelet-Count-to-Lymphocyte-Count Ratio. Diseases. 2023;11(1):15.

25.

Aktas

Duman

Atak

, et al. Irritable bowel syndrome is associated with novel inflammatory markers derived from hemogram parameters. Family Medicine & Primary Care Review. 2020;22(2):107-110.

26.

Posul

Yilmaz

Aktas

Kurt

. Does neutrophil-to-lymphocyte ratio predict active ulcerative colitis? Wiener klinische Wochenschrift. 2015;127(7):262-265.

27.

Kosekli

. Mean platelet volume and platelet to lymphocyte count ratio are associated with hepatitis B-related liver fibrosis. Eur J Gastroenterol Hepatol. 2022;34(3):324-327.

28.

Aktas

. Hematological predictors of novel Coronavirus infection. Rev Assoc Med Bras (1992). 2021;67Suppl 1(Suppl 1):1-2.

29.

Khalid

Ali Jaffar

Khan

, et al. Hematological and biochemical parameters as diagnostic and prognostic markers in SARS-COV-2 infected patients of Pakistan: a retrospective comparative analysis. Hematology. 2021;26(1):529-542.

30.

Basaran

Aktas

. The relationship of vitamin D levels with hemogram indices and metabolic parameters in patients with type 2 diabetes mellitus. AIMS Medical Science. 2024;11:47-57.

31.

Atak

Aktas

Duman

Erkus

Kocak

Savli

. Diabetes control could through platelet-to-lymphocyte ratio in hemograms. Revista da Associação Médica Brasileira. 2019;65:38-42.

32.

Kweon

Lee

Kim

Chung

Choi

Kim

. Neutropenia and neutrophil-to-lymphocyte ratio in a healthy Korean population: race and sex should be considered. Int J Lab Hematol. 2016;38(3):308-318.

33.

Lee

Kim

Youn

Shin

. Reference values of neutrophil-lymphocyte ratio, lymphocyte-monocyte ratio, platelet-lymphocyte ratio, and mean platelet volume in healthy adults in South Korea. Medicine (Baltimore). 2018;97(26):e11138.

34.

Azab

Camacho-Rivera

Taioli

. Average values and racial differences of neutrophil lymphocyte ratio among a nationally representative sample of United States subjects. PLoS One. 2014;9(11):e112361.

35.

VanderWeele

Ding

. Sensitivity Analysis in Observational Research: Introducing the E-Value. Ann Intern Med. 2017;167(4):268-274.

36.

Vickers

Elkin

. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making. 2006;26(6):565-574.

37.

von Elm

Altman

Egger

Pocock

Gøtzsche

Vandenbroucke

. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Bmj. 2007;335(7624):806-808.

38.

Abu-Shawer

Haimour

, et al. Hematologic markers of distant metastases in gastric cancer. J Gastrointest Oncol. 2019;10(3):529-536.

39.

Zhang

Wang

Sun

Diao

Dang

. Predictive value of neutrophil-to-lymphocyte ratio for distant metastasis in gastric cancer patients. Sci Rep. 2022;12(1):10269.

40.

Jia

Wang

, et al. Neutrophils in the premetastatic niche: key functions and therapeutic directions. Mol Cancer. 2024;23(1):200.

41.

Wang

Jia

Wang

, et al. Pre-metastatic niche: formation, characteristics and therapeutic implication. Signal Transduct Target Ther. 2024;9(1):236.

42.

Fang

Wang

Yin

, et al. Diagnostic Sensitivity of NLR and PLR in Early Diagnosis of Gastric Cancer. J Immunol Res. 2020;2020:9146042.

43.

Jeong

Lim

Yun

, et al. Comparison of two inflammation-based prognostic scores in patients with unresectable advanced gastric cancer. Oncology. 2012;83(5):292-299.

44.

Jung

Park

Jeong

, et al. Elevated preoperative neutrophil to lymphocyte ratio predicts poor survival following resection in late stage gastric cancer. J Surg Oncol. 2011;104(5):504-510.

45.

Kim

Yang

, et al. The neutrophil-to-lymphocyte ratio prechemotherapy and postchemotherapy as a prognostic marker in metastatic gastric cancer. Korean J Intern Med. 2018;33(5):990-999.

46.

Wang

Tong

Sun

, et al. Predictive value of NLR and PLR in response to preoperative chemotherapy and prognosis in locally advanced gastric cancer. Front Oncol. 2022;12:936206.

47.

Hirahara

Arigami

Yanagita

, et al. Combined neutrophil-lymphocyte ratio and platelet-lymphocyte ratio predicts chemotherapy response and prognosis in patients with advanced gastric cancer. BMC Cancer. 2019;19(1):672.

48.

Wang

Ding

, et al. Prognostic Value of Neutrophil-Lymphocyte Ratio, Platelet-Lymphocyte Ratio, and Combined Neutrophil-Lymphocyte Ratio and Platelet-Lymphocyte Ratio in Stage IV Advanced Gastric Cancer. Front Oncol. 2020;10:841.

49.

Zhao

Dong

Liang

, et al. Evaluation of multiple biological indicators for combined diagnosis of gastric cancer: A retrospective analysis. Medicine (Baltimore). 2022;101(47):e31904.

50.

Jiang

Ding

Zheng

. Pretreatment neutrophil-lymphocyte count ratio may associate with gastric cancer presence. Cancer Biomark. 2016;16(4):523-528.