The predictive value of systemic inflammation response index and albumin-to-globulin ratio for prognosis in non-small cell lung cancer Authors: Qiuping Xu, Yinyin Deng, Yuping Cai, Shaobin Yu, Mingqiang Kang, Fei He Published: 2026 ABSTRACT We investigated whether inflammatory markers would be significant prognostic factors for survival in patients with non-small-cell lung cancer (NSCLC). This was a retrospective cohort study including 748 consecutive Chinese NSCLC patients. The relationships of inflammatory markers with clinicopathologic characteristics and prognosis were analyzed by chi-squared test and Cox regression. Restricted cubic splines (RCS) with 4 knots were used to flexibly model non-linear relationships between markers and overall survival (OS). Survival was estimated using Kaplan-Meier curves. In multivariate analysis, the systemic inflammation response index (SIRI) and albumin-to-globulin ratio (AGR) were independently associated with OS (HR: 1.506, 95% CI: 1.224–1.852; HR: 0.749, 95% CI: 0.608–0.923). RCS showed a non-linear association for SIRI (P for non-linear = 0.004) and a linear association for AGR (P for non-linear = 0.258). We constructed a 3-tier SIRI-AGR score: Score 1 (low risk: SIRI ≤ 0.91 and AGR > 1.19), Score 2 (intermediate: either SIRI > 0.91 or AGR ≤ 1.19, but not both), Score 3 (high risk: SIRI > 0.91 and AGR ≤ 1.19). Patients with Score 3 had significantly shorter OS (HR: 1.974; 95% CI: 1.486–2.622). Time-dependent ROC showed stable predictive performance with AUC ≈ 0.70 throughout follow-up. The SIRI-AGR score is an independent, convenient, and low-cost prognostic factor for NSCLC. It can serve as a useful indicator for risk stratification and clinical decision-making.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-55879-x. Introduction Lung cancer is the most common cancer and the first leading cause of cancer death worldwide, and non–small-cell lung cancer (NSCLC) accounts for 80%–85% of lung cancer1. In 2022, there were approximately 1.9 million new lung cancer cases and 1.6 million deaths globally1, and in China, lung cancer remains the leading cause of cancer-related mortality2. More recent estimates confirm that lung cancer continues to pose a major public health challenge, with ongoing efforts to improve early detection and treatment3. In spite of advances made in the early diagnosis and treatment of NSCLC, its prognosis remains unsatisfactory, with five-year survival rates of less than 15%4,5. In recent years, novel biomarkers such as FNDC4 and LMNB2 have been reported to be closely related to the prognosis and immunotherapy response of lung adenocarcinoma, providing new ideas for precision treatment6,7.Therefore, there is an urgent necessity to identify some biomarkers to predict patients’ prognosis and guide personalized treatment. Inflammation plays a major role in the development and progression of cancer by promoting angiogenesis, cancer cell proliferation, tumor invasion, and metastasis8. Thus, inflammatory markers, including the neutrophil-to-lymphocyte ratio (NLR) and monocytes-to-lymphocyte ratio (MLR), have become prognostic factors for various malignancies, including lung cancer9, breast cancer10, gastric cancer11, and colorectal cancer12. Recently, a novel inflammatory indicator based on peripheral neutrophil, lymphocyte, and monocyte counts, termed the systemic inflammation response index (SIRI), has been used to better reflect the status of the local immune response and systemic inflammation, as its prediction ability has been confirmed in a variety of tumors, including esophageal cancer13, gastric cancer14, and oral cancer15. Another interesting parameter is represented by the albumin-to-globulin ratio (AGR). As the AGR involves both albumin and globulin, it combines the two strong prognostic factors, nutritional status and systemic inflammation. A low serum albumin level indicates poor nutritional status and is a prognostic factor in several types of cancer, including NSCLC16–18. Furthermore, globulin, as a carrier of sex hormones, plays an important role in immunity and inflammation. However, both the albumin and globulin levels can be influenced by many factors, such as pressure, illness, and fluid retention, which may limit their application in the clinic. As such, an increasing number of studies have demonstrated that the AGR presents a better prognostic value, which is strongly associated with the poor outcome when it is increased in various tumors, including lung cancer, colorectal cancer, and gastric cancer19–21. To our knowledge, there are currently limited studies on the combined prognostic value of SIRI and AGR in patients with NSCLC. Conventional inflammatory markers, including NLR, merely reflect the innate-adaptive immune axis, whereas the platelet-to-lymphocyte ratio (PLR) mainly represents platelet-related inflammation without incorporating monocyte data. In contrast, SIRI simultaneously integrates three critical peripheral immune components: neutrophils, lymphocytes, and monocytes. As a comprehensive index, SIRI adequately characterizes both innate and adaptive immunity, as well as the complex crosstalk within the tumor microenvironment. In contrast, AGR simultaneously captures nutritional status (albumin) and chronic inflammatory/humoral immune activity (globulin), making it biologically complementary to SIRI. Unlike composite scores such as the Glasgow Prognostic Score (GPS), NLR, and PLR, SIRI-AGR offers a broader view by incorporating cellular and humoral components. Accordingly, we conducted an analysis to determine whether SIRI and AGR could predict the clinical outcomes of NSCLC patients. As part of this research, we assessed both the independent and combined associations of these two prognostic indicators with survival. Methods Patients selection We recruited 748 patients with primary NSCLC diagnosed by pathological histology from two hospitals (the First Affiliated Hospital of Fujian Medical University and Fujian Provincial Cancer Hospital) between January 2012 and May 2017. All NSCLC patients were of the Chinese population. Exclusion criteria were: (1) history of other malignancies; (2) autoimmune diseases; (3) active infection within 2 weeks before blood sampling; (4) use of systemic corticosteroids within 4 weeks prior to blood collection; (5) incomplete clinical or follow-up data. The present study was approved by the ethics committees of Fujian Medical University and was conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent prior to enrollment. Data collection The following clinicopathological variables were collected for each enrolled patient: gender, age, pathological type, clinical stage, tumor size, metastasis, pleural effusion, history of lung disease, surgery, chemotherapy, radiotherapy, date of diagnosis, and date of discharge. The clinical stage was determined using the TNM staging system of the American Joint Commission on Cancer (AJCC), eighth edition, via clinical evaluation and postoperative pathological examination. We categorized cases as the following pathological types: adenocarcinoma, squamous cell carcinoma, or other. Tumor size was defined by the largest diameter of the tumor according to a pathological report or imaging examination. For our purposes, metastasis was used to refer to lymph node metastasis found by imaging or pathological examination, other organ metastasis found by imaging examination, and malignant pleural effusion indicated by cytological examination at the time of the initial diagnosis. Pleural effusion was determined by chest computed tomography at the time of the initial diagnosis. Patients’ history of lung disease might include pulmonary infectious diseases, chronic airway inflammatory diseases, and interstitial lung diseases. The peripheral blood of all patients was collected and assessed for neutrophil, lymphocyte, monocyte, albumin, and globulin counts before therapy. The SIRI was calculated as : SIRI = monocyte × neutrophil/lymphocyte. AGR was calculated as: AGR = serum albumin level/serum globulin level. Follow-up Standard follow-up assessments were conducted by trained investigators using telephonic interviews. The start of follow-up was the date of diagnosis, and follow-up was censored at the date of death or September 20, 2023, whichever occurred first. A follow-up assessment was performed every three months for the first year and every six months thereafter until the end of the study. The median follow-up time in this study was 55 months (range: 1–135 months). At the date of the last follow-up, 57.0% had died of NSCLC. The primary follow-up outcome was overall survival (OS). Progression-free survival (PFS) and disease-free survival (DFS) data were not collected in this retrospective study. Statistical analysis The optimal cutoff values of the inflammatory markers were obtained using the X-tile software (Version 3.6.1; Yale School of Medicine, New Haven, CT, USA). X-tile identifies the optimal cutoff by maximizing the survival difference between patient subgroups via log-rank statistic maximization, implementing a minimum p-value approach across all possible dichotomization thresholds. To address the multiple testing problem inherent to this approach, identified cutoffs were validated using internal bootstrap resampling (1,000 iterations) with Bonferroni correction applied to determine statistical significance. The stability of each cutoff was assessed by the 95% confidence interval derived from the bootstrap distribution, and statistical significance was determined if the Bonferroni-corrected P value was < 0.05. Construction of the SIRI‑AGR score: Based on the optimal cutoffs from X-tile (SIRI ≤ 0.91 vs. > 0.91; AGR ≤ 1.19 vs. > 1.19), we initially stratified patients into four subgroups: Group 1 (SIRI ≤ 0.91 and AGR > 1.19), Group 2a (SIRI > 0.91 and AGR > 1.19), Group 2b (SIRI ≤ 0.91 and AGR ≤ 1.19), and Group 3 (SIRI > 0.91 and AGR ≤ 1.19). To determine whether Groups 2a and 2b could be merged, we compared their survival distributions using Kaplan-Meier curves and the log-rank test, as well as hazard ratios relative to Group 1. If no significant difference in overall survival (OS) was observed between Groups 2a and 2b, they were combined into a single intermediate-risk category (Score 2). Group 1 was defined as Score 1 (low risk), and Group 3 as Score 3 (high risk). The prognostic value of the final 3-tier score was then evaluated using multivariate Cox regression and time-dependent ROC analysis. Restricted cubic splines (RCS): RCS with 4 knots located at the 5th, 35th, 65th, and 95th percentiles of SIRI and AGR were used to flexibly model non-linear relationships with OS. The choice of 4 knots follows Harrell’s guidelines to balance model flexibility and overfitting, and was further validated by AIC comparison. Comparisons between clinicopathological characteristics of the patients and inflammatory markers were performed using the chi-squared test. Survival patterns were identified using the Kaplan-Meier curves. A Cox univariate analysis was carried out to determine the association between the OS rate and baseline parameters. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated. Cox multivariate analysis with a forward conditional stepwise procedure was performed to identify independent predictors of OS among the patients with NSCLC. The covariates included in multivariate Cox models were selected based on univariate analysis (P < 0.10) and clinical relevance from prior literature. To avoid multicollinearity, NLR and MLR were excluded from models containing SIRI, as SIRI incorporates the same cellular components. In addition, the associations between levels of SIRI, AGR, and the overall survival of NSCLC were evaluated on a continuous scale with restricted cubic spline curves based on Cox proportional hazards models. Furthermore, time-dependent receiver operating characteristic (ROC) curve analysis was completed to evaluate the performance of inflammatory markers in predicting patient survival or death based on the duration of time since diagnosis. The Statistical Package for the Social Sciences version 20 software program (IBM Corp., Armonk, NY, USA) and GraphPad Prism 5 (GraphPad Software, Inc., La Jolla, CA, USA) were used for statistical analysis. The R version 4.3.2 software (www.r-project.org; R Foundation for Statistical Computing, Vienna, Austria) was used for performance assessment in time-dependent ROC curve estimation. All p-values were derived from two-tailed tests, with p < 0.05 considered to indicate statistical significance. Results Characteristics of the study participants A total of 748 NSCLC patients were enrolled in this study, with a median OS of 71.1 months. In total, there were 450 males and 298 females, with a median age of 60 years (IQR: 52–68 years; range: 26–87 years), including 344 cases with a tumor size of 3 cm or less and 404 cases with a tumor size of more than 3 cm. Roughly two-thirds of study participants (67.8%) had been diagnosed with adenocarcinoma, whereas the remaining percentage had been diagnosed with squamous cell carcinoma or another tumor type. According to the eighth edition of the AJCC’s TNM staging system, 384 patients were diagnosed with stages I and II tumors, and 364 were diagnosed with stages III and IV tumors. In addition, 385 (51.5%) patients experienced metastasis when diagnosed, and 485 patients underwent surgery. Other clinicopathological variables are shown in Table 1. Table 1Relationship between clinicopathological variables and inflammatory markers in the 748 NSCLC patients.Variablesn(%)NLR≤ 2.37/2.37P valueMLR≤ 0.23/>0.23P valueSIRI≤ 0.91/>0.91P valueAGR≤ 1.19/>1.19P valueGender Male450(60.2)198/252 < 0.001 167/283 < 0.001 155/295 < 0.001 132/3180.147 Female298(39.8)182/116190/108185/11373/225Age (years) <60365(48.8)195/1700.161181/1840.320171/1940.45586/279 0.021 ≥60383(51.2)185/198176/207169/214119/264Pathological type Adenocarcinoma507(67.8)285/222 < 0.001 274/233 < 0.001 268/239 < 0.001 115/392 < 0.001 Squamous cell carcinoma161(21.5)67/9460/10152/10962/99 Others80(10.7)28/5223/5720/6028/52Clinical stage I + II384(51.3)237/147 < 0.001 228/156 < 0.001 223/161 < 0.001 76/308 < 0.001 III + IV364(48.7)143/221129/235117/247129/235Tumor size (cm) ≤3344(46.0)217/127 < 0.001 201/143 < 0.001 199/145 < 0.001 49/295 < 0.001 > 3404(54.0)163/241156/248141/263156/248Metastasis Yes385(51.5)162/223 < 0.001 151/234 < 0.001 137/248 < 0.001 137/248 < 0.001 No363(48.5)218/145206/157203/16068/295Pleural effusion Yes119(15.9)30/89 < 0.001 33/86 < 0.001 33/86 < 0.001 41/780.060 No629(84.1)350/279324/305307/322164/465History of lungdisease Yes275(36.8)120/155 0.003 105/170 < 0.001 107/168 0.006 70/2050.361 No473(63.2)260/213252/221233/240135/338Surgery Yes485(64.8)288/197 < 0.001 268/217 < 0.001 261/224 < 0.001 105/380 < 0.001 No263(35.2)92/17189/17479/184100/163Chemotherapy Yes322(43.0)154/1680.157148/1740.401123/199 0.001 89/2230.901 No426(57.0)226/200209/217217/209116/310Radiotherapy Yes56(7.5)26/300.49617/39 0.007 14/42 0.001 20/360.147 No692(92.5)354/338340/352326/366185/507Bold values indicate statistical significance. Table 1Relationship between clinicopathological variables and inflammatory markers in the 748 NSCLC patients.Variablesn(%)NLR≤ 2.37/2.37P valueMLR≤ 0.23/>0.23P valueSIRI≤ 0.91/>0.91P valueAGR≤ 1.19/>1.19P valueGender Male450(60.2)198/252 < 0.001 167/283 < 0.001 155/295 < 0.001 132/3180.147 Female298(39.8)182/116190/108185/11373/225Age (years) <60365(48.8)195/1700.161181/1840.320171/1940.45586/279 0.021 ≥60383(51.2)185/198176/207169/214119/264Pathological type Adenocarcinoma507(67.8)285/222 < 0.001 274/233 < 0.001 268/239 < 0.001 115/392 < 0.001 Squamous cell carcinoma161(21.5)67/9460/10152/10962/99 Others80(10.7)28/5223/5720/6028/52Clinical stage I + II384(51.3)237/147 < 0.001 228/156 < 0.001 223/161 < 0.001 76/308 < 0.001 III + IV364(48.7)143/221129/235117/247129/235Tumor size (cm) ≤3344(46.0)217/127 < 0.001 201/143 < 0.001 199/145 < 0.001 49/295 < 0.001 > 3404(54.0)163/241156/248141/263156/248Metastasis Yes385(51.5)162/223 < 0.001 151/234 < 0.001 137/248 < 0.001 137/248 < 0.001 No363(48.5)218/145206/157203/16068/295Pleural effusion Yes119(15.9)30/89 < 0.001 33/86 < 0.001 33/86 < 0.001 41/780.060 No629(84.1)350/279324/305307/322164/465History of lungdisease Yes275(36.8)120/155 0.003 105/170 < 0.001 107/168 0.006 70/2050.361 No473(63.2)260/213252/221233/240135/338Surgery Yes485(64.8)288/197 < 0.001 268/217 < 0.001 261/224 < 0.001 105/380 < 0.001 No263(35.2)92/17189/17479/184100/163Chemotherapy Yes322(43.0)154/1680.157148/1740.401123/199 0.001 89/2230.901 No426(57.0)226/200209/217217/209116/310Radiotherapy Yes56(7.5)26/300.49617/39 0.007 14/42 0.001 20/360.147 No692(92.5)354/338340/352326/366185/507Bold values indicate statistical significance. Relationship between clinicopathological variables and inflammatory markers in the 748 NSCLC patients. History of lung disease Bold values indicate statistical significance. Association of the inflammatory markers with clinicopathological variables Optimal cutoff values for systemic inflammatory indexes determined in X-tile were as follows: NLR (≤ 2.37, > 2.37), MLR (≤ 0.23, > 0.23), SIRI (≤ 0.91, > 0.91), and AGR (≤ 1.19, > 1.19) (Fig. 1). To address the multiple testing issue inherent in the minimum p-value approach, we performed internal bootstrap resampling (1,000 iterations) with Bonferroni correction. All four cutoffs remained statistically significant after correction (bootstrap-corrected P < 0.001 for each index), and the 95% confidence intervals for the cutoffs were narrow (e.g., SIRI cutoff: 0.89–0.93; AGR cutoff: 1.16–1.22), indicating excellent stability of the selected thresholds. Detailed bootstrap results are provided in Supplementary Table 1. Table 1 summarizes the associations between clinicopathological variables and systemic inflammatory indices in NSCLC patients. NLR and MLR values were significantly associated with gender, pathological type, clinical stage, tumor size, metastasis, pleural effusion, a history of lung disease, and surgery. Meanwhile, MLR is also associated with radiotherapy. In addition to age, SIRI values were significantly correlated with other clinical features. Moreover, AGR closely correlated with age, pathological type, clinical stage, tumor size, metastasis, and surgery but not the other variables. Fig. 1X-tile analyses of OS based on patient data to determine the optimal cut-off values for NLR (a), MLR (b), SIRI (c), and AGR (d). The optimal cut-off values are shown in histograms of the entire cohort. X-tile analyses of OS based on patient data to determine the optimal cut-off values for NLR (a), MLR (b), SIRI (c), and AGR (d). The optimal cut-off values are shown in histograms of the entire cohort. OS analysis Univariate analysis indicated that gender, pathological type, clinical stage, tumor size, metastasis, pleural effusion, surgery, chemotherapy, radiotherapy, NLR, MLR, SIRI, and AGR all had a statistically significant association with OS, whereas age and history of lung disease displayed no prognostic significance for OS. A higher SIRI value was associated with reduced OS (HR: 2.296; 95% CI: 1.874–2.811). On the contrary, a higher AGR value was associated with prolonged OS (HR: 0.509; 95% CI: 0.417–0.621). Prior to constructing the multivariate Cox proportional hazards model, we explicitly employed the variance inflation factor (VIF) to assess multicollinearity among variables, setting VIF > 10 as the threshold for variable exclusion to prevent highly correlated variables from interfering with the accuracy of model results. Multicollinearity assessment confirmed significant redundancy between SIRI and NLR/MLR (VIF = 10.386 for NLR, 11.569 for MLR), justifying their exclusion from the multivariate Cox model for SIRI/AGR (Supplementary Table 2). The multivariate analysis suggested, after adjusting for other variables, that pretreatment SIRI and AGR were independently associated with OS (HR: 1.506, 95% CI: 1.224–1.852; HR: 0.749, 95% CI: 0.608–0.923) (Table 2). Table 2Univariate and multivariate Cox regression analyses for OS in patients with NSCLC.VariablesUnivariate analysisMultivariate analysisHR (95% CI)HR (95% CI) aGender Male1.000 Female 0.666(0.545–0.814) Age (years) <601.0001.000 ≥601.160 (0.959–1.403) 1.366 (1.125–1.659) Pathological type Adenocarcinoma1.0001.000 Squamous cell carcinoma 1.481 (1.177–1.864) 1.033 (0.814–1.312) Others 3.401 (2.602–4.445) 1.848 (1.402–2.438) Clinical stage I + II1.0001.000 III + IV 5.251 (4.233–6.513) 2.819 (2.144–3.708) Tumor size(cm) ≤31.0001.000 > 3 2.997 (2.433–3.692) 1.610 (1.284–2.019) Metastasis No1.000 Yes 4.289 (3.458–5.320) Pleural effusion No1.000 Yes 2.295 (1.827–2.882) History of lung disease No1.000 Yes1.179 (0.970–1.433)Surgery Yes1.0001.000 No 4.499 (3.699–5.473) 1.800 (1.414–2.290) Chemotherapy Yes1.000 No 0.737 (0.610–0.892) Radiotherapy Yes1.000 No 0.622 (0.458–0.845) NLR ≤ 2.371.000 > 2.37 2.195 (1.805–2.668) MLR ≤ 0.231.000 > 0.23 2.063 (1.693–2.513) SIRI ≤ 0.911.0001.000 > 0.91 2.296 (1.874–2.811) 1.506 (1.224–1.852) AGR ≤ 1.191.0001.000 > 1.19 0.509 (0.417–0.621) 0.749 (0.608–0.923) aAdjusted by gender, metastasis, pleural effusion, history of lung disease, chemotherapy, and radiotherapy; NLR and MLR were excluded to avoid multicollinearity.Bold values indicate statistical significance. Table 2Univariate and multivariate Cox regression analyses for OS in patients with NSCLC.VariablesUnivariate analysisMultivariate analysisHR (95% CI)HR (95% CI) aGender Male1.000 Female 0.666(0.545–0.814) Age (years) <601.0001.000 ≥601.160 (0.959–1.403) 1.366 (1.125–1.659) Pathological type Adenocarcinoma1.0001.000 Squamous cell carcinoma 1.481 (1.177–1.864) 1.033 (0.814–1.312) Others 3.401 (2.602–4.445) 1.848 (1.402–2.438) Clinical stage I + II1.0001.000 III + IV 5.251 (4.233–6.513) 2.819 (2.144–3.708) Tumor size(cm) ≤31.0001.000 > 3 2.997 (2.433–3.692) 1.610 (1.284–2.019) Metastasis No1.000 Yes 4.289 (3.458–5.320) Pleural effusion No1.000 Yes 2.295 (1.827–2.882) History of lung disease No1.000 Yes1.179 (0.970–1.433)Surgery Yes1.0001.000 No 4.499 (3.699–5.473) 1.800 (1.414–2.290) Chemotherapy Yes1.000 No 0.737 (0.610–0.892) Radiotherapy Yes1.000 No 0.622 (0.458–0.845) NLR ≤ 2.371.000 > 2.37 2.195 (1.805–2.668) MLR ≤ 0.231.000 > 0.23 2.063 (1.693–2.513) SIRI ≤ 0.911.0001.000 > 0.91 2.296 (1.874–2.811) 1.506 (1.224–1.852) AGR ≤ 1.191.0001.000 > 1.19 0.509 (0.417–0.621) 0.749 (0.608–0.923) aAdjusted by gender, metastasis, pleural effusion, history of lung disease, chemotherapy, and radiotherapy; NLR and MLR were excluded to avoid multicollinearity.Bold values indicate statistical significance. Univariate and multivariate Cox regression analyses for OS in patients with NSCLC. aAdjusted by gender, metastasis, pleural effusion, history of lung disease, chemotherapy, and radiotherapy; NLR and MLR were excluded to avoid multicollinearity. Bold values indicate statistical significance. Association of the SIRI and AGR with OS As shown in Fig. 2, the relationships between SIRI, AGR, and OS were evaluated on a continuous scale with restricted cubic splines (RCS) based on a multivariable Cox model with four knots at the 5th, 35th, 65th, and 95th percentiles of SIRI and AGR. The risk of mortality in NSCLC was relatively flat until around 0.98 of predicted SIRI and then started to increase rapidly afterwards (P for non-linear = 0.004). On the contrary, the AGR showed a linear relationship with the overall survival of NSCLC (P for non-linear = 0.258). That is to say, the risk of mortality in NSCLC decreased with increasing AGR. Fig. 2Restricted Cubic Spline Plot of the association between SIRI, AGR, and the overall survival of NSCLC. Adjusted by age, pathological type, clinical stage, tumor size, and surgery. Restricted Cubic Spline Plot of the association between SIRI, AGR, and the overall survival of NSCLC. Adjusted by age, pathological type, clinical stage, tumor size, and surgery. The prognostic significance of SIRI and AGR as stratified by patient characteristics The prognostic significance of the SIRI or AGR was further elucidated in subgroups of age, pathological type, clinical stage, tumor size, and surgery. The effects values conferred by SIRI or AGR were more prominent in some subgroups, with an HR ranging from 1.239 to 1.954 or 0.567 to 0.740, respectively. As shown in Fig. 3, we observed that the prognostic role of SIRI seems to be greater in the groups of patients aged 60 years or older, patients with adenocarcinoma or squamous cell carcinoma, and patients with tumor sizes of greater than 3 cm. Meanwhile, the prognostic role of AGR seems to be greater among patients with adenocarcinoma or squamous cell carcinoma, patients with tumor sizes of greater than 3 cm (Fig. 3). Fig. 3Stratified analysis of the association between SIRI, AGR, and the prognosis of NSCLC in the Chinese population. Adjusted by age, pathological type, clinical stage, tumor size, and surgery. Stratified analysis of the association between SIRI, AGR, and the prognosis of NSCLC in the Chinese population. Adjusted by age, pathological type, clinical stage, tumor size, and surgery. Association between survival and SIRI-AGR score When SIRI was combined with AGR, we initially stratified patients into four distinct subgroups based on the pre-specified cutoffs: Group 1 (SIRI ≤ 0.91 and AGR > 1.19), Group 2a (SIRI > 0.91 and AGR > 1.19), Group 2b (SIRI ≤ 0.91 and AGR ≤ 1.19), and Group 3 (SIRI > 0.91 and AGR ≤ 1.19). To rigorously justify merging the two intermediate subgroups into a single Score 2, we performed a dedicated survival analysis of Group 2a and Group 2b. As shown in Supplementary Fig. 1, the Kaplan-Meier survival curves of the two subgroups showed no meaningful separation and followed nearly identical trajectories throughout the entire follow-up period. The Log-rank (Mantel-Cox) test confirmed no statistically significant difference in overall survival (OS) between the two groups (χ2 = 2.367, df = 1, P = 0.124). The mean OS was comparable between Group 2a (60.73 months, 95% CI: 54.52–66.94) and Group 2b (67.81 months, 95% CI: 56.10–79.53), with no clinically meaningful difference. Univariate Cox regression analysis, with Group 1 as the reference, showed comparable hazard ratios (HRs) for Group 2a (HR: 1.172, 95% CI: 0.896–1.531, P = 0.247) and Group 2b (HR: 1.221, 95% CI: 0.928–1.605, P = 0.153). These results provided robust statistical justification for merging the two intermediate subgroups into a single Score 2 (intermediate risk), while retaining Group 1 as Score 1 (low risk) and Group 3 as Score 3 (high risk). The Kaplan-Meier survival analysis of the final 3-tier SIRI-AGR score revealed that patients with a score of 3 had the lowest OS (P < 0.001) (Fig. 4). To validate the prognostic value across different disease settings, we further performed subgroup analyses in the resectable (n = 485) and unresectable (n = 263) cohorts. Consistent results were observed in both subgroups: patients with a score of 3 had significantly shorter OS in the resectable cohort (HR: 2.165, 95% CI: 1.436–3.262, P < 0.001) and the unresectable cohort (HR: 1.495, 95% CI: 1.019–2.192, P = 0.040) (Table 3; Fig. 5). Multivariate analysis confirmed that an SIRI-AGR score of 3 was an independent risk predictor for shorter OS (HR: 1.974; 95% CI: 1.486–2.622; P < 0.001) (Table 4). Therefore, the SIRI-AGR score was considered to be an independent risk predictor of OS in both resectable and unresectable NSCLC patients. Table 3Multivariate Cox regression analysis in resectable and unresectable NSCLC cohorts.VariablesCategoryResectable cohort (n = 485)Unresectable cohort (n = 263)HR (95% CI)P valueHR (95% CI)P valueGenderMale1.0001.000Female0.925 (0.668–1.281)0.6390.800 (0.589–1.085)0.151Age (years)< 601.0001.000≥ 60 1.433 (1.063–1.931) 0.018 1.312 (1.003–1.718) 0.048 Pathological typeAdenocarcinoma1.0001.000Squamous cell carcinoma1.270 (0.890–1.812)0.1870.762 (0.524–1.110)0.157Others2.731 (1.604–4.652) <0.001 1.497 (1.072–2.092) 0.018 Clinical stageI + II1.0001.000III + IV 3.253 (2.388–4.432) <0.001 2.008 (1.153–3.496) 0.014 Tumor size (cm)≤ 31.0001.000> 3 1.613 (1.174–2.217) 0.003 1.378 (0.998–1.904)0.051SIRI-AGR scoreOne1.0001.000Two 1.407 (1.003–1.976) 0.048 0.918 (0.644–1.308)0.634Three 2.165 (1.436–3.262) <0.001 1.495 (1.019–2.192) 0.040 Bold values indicate statistical significance. Table 3Multivariate Cox regression analysis in resectable and unresectable NSCLC cohorts.VariablesCategoryResectable cohort (n = 485)Unresectable cohort (n = 263)HR (95% CI)P valueHR (95% CI)P valueGenderMale1.0001.000Female0.925 (0.668–1.281)0.6390.800 (0.589–1.085)0.151Age (years)< 601.0001.000≥ 60 1.433 (1.063–1.931) 0.018 1.312 (1.003–1.718) 0.048 Pathological typeAdenocarcinoma1.0001.000Squamous cell carcinoma1.270 (0.890–1.812)0.1870.762 (0.524–1.110)0.157Others2.731 (1.604–4.652) <0.001 1.497 (1.072–2.092) 0.018 Clinical stageI + II1.0001.000III + IV 3.253 (2.388–4.432) <0.001 2.008 (1.153–3.496) 0.014 Tumor size (cm)≤ 31.0001.000> 3 1.613 (1.174–2.217) 0.003 1.378 (0.998–1.904)0.051SIRI-AGR scoreOne1.0001.000Two 1.407 (1.003–1.976) 0.048 0.918 (0.644–1.308)0.634Three 2.165 (1.436–3.262) <0.001 1.495 (1.019–2.192) 0.040 Bold values indicate statistical significance. Multivariate Cox regression analysis in resectable and unresectable NSCLC cohorts. Bold values indicate statistical significance. Table 4Multivariate analysis for the prediction of NSCLC survival outcomes when interrelated SIRI and AGR are combined.VariablesCategoryP valueHR (95% CI)GenderMale1.000Female0.1890.862 (0.690–1.076)Age (years)< 601.000≥ 60 0.005 1.321 (1.086–1.606) Pathological typeAdenocarcinoma1.000Squamous cell carcinoma0.8540.976 (0.754–1.263)Others < 0.001 1.727 (1.299–2.296) Clinical stageI + II1.000III + IV < 0.001 2.996 (2.273–3.949) Tumor size (cm)≤ 31.000> 3 < 0.001 1.592 (1.268–2.001) SurgeryYes1.000No < 0.001 1.790 (1.402–2.286) SIRI-AGR scoreOne1.000Two0.1531.196 (0.935–1.530)Three < 0.001 1.974 (1.486–2.622) Bold values indicate statistical significance. Table 4Multivariate analysis for the prediction of NSCLC survival outcomes when interrelated SIRI and AGR are combined.VariablesCategoryP valueHR (95% CI)GenderMale1.000Female0.1890.862 (0.690–1.076)Age (years)< 601.000≥ 60 0.005 1.321 (1.086–1.606) Pathological typeAdenocarcinoma1.000Squamous cell carcinoma0.8540.976 (0.754–1.263)Others < 0.001 1.727 (1.299–2.296) Clinical stageI + II1.000III + IV < 0.001 2.996 (2.273–3.949) Tumor size (cm)≤ 31.000> 3 < 0.001 1.592 (1.268–2.001) SurgeryYes1.000No < 0.001 1.790 (1.402–2.286) SIRI-AGR scoreOne1.000Two0.1531.196 (0.935–1.530)Three < 0.001 1.974 (1.486–2.622) Bold values indicate statistical significance. Multivariate analysis for the prediction of NSCLC survival outcomes when interrelated SIRI and AGR are combined. Bold values indicate statistical significance. Fig. 4Kaplan-Meier survival curves of overall survival in non-small-cell lung cancer patients according to the SIRI-AGR score. Kaplan-Meier survival curves of overall survival in non-small-cell lung cancer patients according to the SIRI-AGR score. Fig. 5Kaplan-Meier survival curves of overall survival in (a) resectable and (b) unresectable NSCLC cohorts according to the SIRI-AGR score. Kaplan-Meier survival curves of overall survival in (a) resectable and (b) unresectable NSCLC cohorts according to the SIRI-AGR score. Figure 6 shows the area under the curve (AUC) for the SIRI-AGR score and OS from 1 to 135 months after the start of follow-up determined using time-dependent ROC analysis. During the whole follow-up period, the AUC curve remained relatively flat, suggesting that the accuracy of the SIRI-AGR score was relatively stable. Later on during follow-up, the accuracy of prediction fluctuated greatly, but the AUC (t) values were all around 0.70, indicating that, at any time during the whole follow-up period, the probability that the SIRI-AGR scores of NSCLC patients who died were higher than those of survivors was about 0.70—that is, the probability that the SIRI-AGR score could correctly predict the outcome of NSCLC patients was about 0.70. Fig. 6Time-dependent AUCs for the SIRI-AGR score and overall survival from 1 to 135 months after the start of follow-up. Time-dependent AUCs for the SIRI-AGR score and overall survival from 1 to 135 months after the start of follow-up. Discussion In the present study, the prognostic values of inflammatory markers were evaluated in an NSCLC patient cohort. As observed, high SIRI values and low AGR were significant independent predictors of poor prognosis among NSCLC patients. Furthermore, the favorable prognostic effect of SIRI was more evident in specific subgroups of NSCLC patients. The results of multivariate analysis demonstrated that the SIRI-AGR score was an independent prognostic factor and could correctly predict the outcome of NSCLC patients after adjusting for gender, age, pathological type, clinical stage, tumor size, surgery, and radiotherapy. To the best of our knowledge, this report suggests that the SIRI-AGR score can serve as an effective prognostic indicator for NSCLC patients. Hence, it can be applied during risk stratification and clinical decision-making. Although these findings require validation before clinical application. Accumulating evidence indicates that there is a correlation between the systemic inflammatory response and tumorigenesis and tumor progression22. Studies have suggested that inflammatory mediators secreted by inflammatory cells, such as cytokines, growth factors, and chemokines, can induce abnormal cell proliferation through inflammatory cascade reactions, further chemotaxis, the recruitment of inflammatory cells, and the production of more inflammatory mediators, thus changing the stability of the intracellular environment23,24. These inflammatory mediators can cause DNA oxidative damage and DNA mutation; inhibit cell apoptosis; and promote angiogenesis, invasion, and metastasis in cancer patients, thus maintaining tumor growth and metastasis25. NLR and PLR, as inflammatory response indices related to the prognosis of various diseases, have the advantages of low cost, simplicity, rapidity, and high repeatability, which have attracted more and more attention from medical scholars in recent years. Meanwhile, SIRI, a novel inflammatory-related marker, has been significantly correlated with the prognosis of various types of tumors15,26,27, consistent with our findings. The biological reasons behind the utility of SIRI in predicting the prognosis of NSCLC may be elucidated by the functions of neutrophils, monocytes, and lymphocytes. Neutrophils promote NSCLC progression by secreting pro-tumorigenic cytokines and enhancing angiogenesis28, while neutrophil extracellular traps further facilitate invasion and metastasis29. Tumor-associated neutrophils suppress antitumor immunity30 and induce an immunosuppressive microenvironment via regulating PAN optosis31. Circulating monocytes can be recruited into the tumor microenvironment and subsequently differentiate into tumor-associated macrophages32. Tumor-associated macrophages, mainly exhibiting an M2-like phenotype, further suppress anti-tumor immune responses and facilitate tumor progression and distant metastasis in NSCLC33. Lymphocytes, especially CD8⁺ cytotoxic T cells and NK cells, serve as the major effector cells of adaptive anti-tumor immunity34. A high peripheral lymphocyte count indicates effective immune monitoring and is correlated with favorable survival outcomes in NSCLC patients35. In contrast, lymphopenia reflects severely compromised immune function and independently predicts shorter overall survival and progression-free survival in patients with NSCLC36. SIRI, as a simple noninvasive prognostic indicator based on peripheral neutrophil, monocyte, and lymphocyte counts, can predict the survival rate of NSCLC patients37. The optimal cutoff value for SIRI identified in our study (0.91) is somewhat lower than the range of 1.0 to 1.5 reported in some previous NSCLC studies38. This discrepancy may be attributable to differences in ethnic composition, clinical stage distribution (a substantial proportion of our cohort had early-stage disease), and the inherent sensitivity of the X-tile algorithm. Similar variability has also been observed for AGR cutoffs in lung cancer39. Therefore, while our bootstrap-corrected cutoffs demonstrate internal robustness, external validation in independent, large-scale cohorts is essential to establish universally applicable thresholds. Although the cutoff value of SIRI in the present study was slightly different from that in previous studies, our study also confirms that SIRI is an effective predictor of NSCLC prognosis. Furthermore, we also found that there was an interaction between SIRI and age. High SIRI values, which may be more serious with inflammation, also suggest a worse prognosis in older NSCLC patients, resulting in a reduced survival rate. In this study, it was found that the high level of AGR before treatment was significantly related to the better prognosis of NSCLC patients, which was consistent with previous research results39–41. A higher AGR indicated higher albumin or lower globulin levels in the peripheral blood of NSCLC patients. Previous studies have found that protein is not only a nutritional marker42 but also an important indicator of inflammation43 The production of inflammatory cytokines such as interleukin-1, interleukin-6, and tumor necrosis factor can inhibit the synthesis of albumin, resulting in a decrease in the albumin level44. A number of studies have shown that low serum albumin is a predictor of poor prognosis in a variety of malignant tumors, including lung cancer16–18. In addition, globulin is a broad carrier of bioactive substances, which plays an important role in immune and inflammatory responses. It has been reported that the presence of high levels of globulin prior to operation is significantly related to the adverse prognosis of gastric cancer45 and rectal cancer46. However, albumin is affected by many factors, such as stress, disease, liver dysfunction, and the consumption or synthesis ability of visceral protein. Therefore, AGR can minimize these influences, which have greater clinical significance. Similar to the role of AGR in colorectal cancer prognosis reported by Li et al.47, our study confirms that AGR, as an immune-nutritional composite ratio, has stable prognostic value in NSCLC. Of particular clinical relevance is the nonlinear association between SIRI and OS identified by restricted cubic spline analysis. Mortality risk remained relatively stable at lower SIRI levels but exhibited a marked acceleration beyond an approximate inflection point of 0.98, suggesting a threshold effect. This finding indicates that patients approaching or exceeding this threshold may face a disproportionately elevated risk of poor outcomes relative to those with only marginally lower SIRI values. From a clinical perspective, baseline screening for elevated SIRI could facilitate the identification of a high-risk subgroup that might benefit from intensified surveillance or early intervention aimed at mitigating systemic inflammation. In contrast, the linear relationship of AGR suggests that every increase in AGR is consistently associated with a reduced risk of mortality, highlighting the continuous protective effect of maintaining a high AGR. This pattern of non-linear and linear associations between inflammatory-nutritional markers and prognosis is consistent with findings in other cancer types48, where composite indices integrating such markers have shown superior prognostic performance. In the current study, we found that the SIRI and AGR were independent prognostic factors for OS in NSCLC patients among pretreatment hematological components. Then, we combined the SIRI and AGR into a novel score, defined as the SIRI-AGR score. The 3-tier SIRI-AGR scoring system was constructed based on rigorous statistical validation. We first confirmed the prognostic equivalence of the two intermediate subgroups (Bad SIRI/Good AGR and Good SIRI/Bad AGR), with no significant difference in OS (Log-rank P = 0.124) or HRs, justifying their merging into a single intermediate risk group. Biologically, both subgroups represent single-dimensional dysfunction (either systemic inflammation or nutritional-immune impairment), with equivalent moderate prognostic impact, while the high-risk group (Bad/Bad) represents dual-dimensional dysfunction, leading to a significantly worse prognosis. This 3-tier system is not only statistically robust but also highly practical for clinical use, enabling rapid risk stratification of NSCLC patients to guide personalized management. The Kaplan-Meier analysis indicated that the survival time was longer in patients with high SIRI values and a low AGR. This means that the synergistic effect of high SIRI values and a low AGR leads to the poor prognosis of patients with NSCLC. Moreover, multivariate analyses revealed that the SIRI-AGR score was an independent predictor of OS in NSCLC patients, and the SIRI-AGR score could correctly predict the outcomes of these individuals. Our study evaluated the effect of the SIRI-AGR score on the prognosis of NSCLC patients and confirmed that this index is significantly associated with survival. The superior prognostic performance of this composite score is consistent with other studies on composite inflammation scores49, which have shown that integrating multiple biological dimensions can improve prognostic accuracy compared with single markers. In the Chinese NSCLC population, EGFR mutations account for 40–50% of adenocarcinomas50, and existing studies suggest EGFR-mutant patients may have distinct baseline inflammatory profiles compared with wild-type patients51. For example, some reports indicate that EGFR-mutant tumors are associated with lower systemic inflammatory responses, including reduced neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR), potentially due to differences in tumor biology and host immune interactions52.This potential confounding was not adjusted in this study due to the lack of genomic data, which may influence the generalizability of SIRI-AGR in targeted therapy populations. This study evaluated the prognostic value of the SIRI-AGR score in patients with NSCLC and identified this novel biomarker as a robust predictor of clinical outcomes. Both SIRI and AGR are derived solely from routine pretreatment complete blood count and serum protein assays, which are universally obtained at initial diagnosis without additional cost or invasive procedures. As such, the SIRI-AGR score could potentially be integrated into standard oncology practice, although this remains speculative pending prospective validation. In clinical settings, baseline assessment of the SIRI-AGR score might hypothetically facilitate the stratification of NSCLC patients into low, intermediate, and high-risk categories. This could have potential implications for guiding the intensity of surveillance and follow-up (e.g., more frequent imaging for higher-risk patients), informing decisions regarding adjuvant or intensified treatment in resectable disease, and supporting eligibility assessment for clinical trials that incorporate inflammatory marker-based stratification. However, these proposed applications are currently hypothetical and require confirmation through well-designed prospective studies before they can be recommended for routine clinical decision-making. Several important limitations of this study should be acknowledged. First, this was a retrospective, single-country observational study from two institutions in China, which is susceptible to selection bias, and the findings may not be generalizable to other populations. Second, although we included 748 patients, the lack of an independent external validation cohort limits the robustness of our results. The optimal cut-offs identified by X-tile are data-dependent and require multi-center confirmation. Third, residual confounding from unmeasured variables such as performance status, detailed medication history, and cause of death cannot be excluded. Fourth, the cohort was recruited from 2012 to 2017, and modern treatments, including immune checkpoint inhibitors and targeted therapies, were not fully captured. The prognostic value of SIRI-AGR in patients receiving these therapies remains unclear. Fifth, data on tumor genomic mutations (e.g.,, EGFR, ALK) and corresponding targeted therapy were unavailable, which may act as important unmeasured confounders. Sixth, the non-linear threshold effect of SIRI is exploratory and needs validation. Finally, the absence of PFS/DFS data limits the assessment of the score’s ability to predict recurrence or progression. Future large-scale, prospective, multi-center studies with diverse ethnic populations are warranted to validate the SIRI-AGR score, establish standardized cut-offs, and evaluate its utility in the era of modern NSCLC treatment. Conclusions In conclusion, this retrospective cohort study demonstrates that pretreatment SIRI and AGR are independent prognostic factors for OS in Chinese NSCLC patients. The novel 3-tier SIRI-AGR score, derived from routine blood tests, is a readily available, low-cost, and stable prognostic biomarker that can effectively stratify patients into low-risk, intermediate-risk, and high-risk groups. These findings support the clinical utility of inflammation-nutritional markers in NSCLC prognosis. Larger prospective and multi-center studies are needed to validate these results. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1