Clinical overall survival prediction and disease characteristics of locally advanced non-small cell lung cancer: an integrated analysis based on the SEER and TCGA databases Authors: Yuanhui Tian, Yan Deng, Xiaoli Liao, Shutao Yuan, Ke Li Published: 2026 ABSTRACT BackgroundLung cancer is the second most commonly diagnosed malignancy worldwide and remains the leading cause of cancer-related death. Non-small cell lung cancer (NSCLC) accounts for approximately 80–85% of lung cancers. Given the low rate of early detection and substantial interpatient heterogeneity, management of NSCLC, particularly locally advanced disease, remains challenging. This study aimed to characterize clinical heterogeneity and progression-associated molecular features of locally advanced NSCLC and to identify candidate biomarkers and therapeutic targets.MethodsPatients with locally advanced NSCLC diagnosed between 2010 and 2017 were identified in the Surveillance, Epidemiology, and End Results (SEER) database. Eligible stage III NSCLC cases were randomly divided in a 7:3 ratio into a training cohort (n=11,459) and an internal validation cohort (n=4,911). Univariable and multivariable Cox proportional hazards models were used to identify independent prognostic factors, which were integrated into a nomogram to predict 1-, 3-, and 5-year overall survival (OS). Model performance was evaluated using the concordance index (C-index), time-dependent area under the curve (AUC), and calibration plots. In parallel, transcriptomic ribonucleic acid sequencing (RNA-seq) and clinical data were retrieved from The Cancer Genome Atlas (TCGA). Differential expression was found using DESeq2 by comparing stage III tumors with normal tissues and with early-stage tumors; the intersection of differentially expressed genes (DEGs) was used to capture progression-associated signals. Functional enrichment was conducted with clusterProfiler. Candidate genes were further selected using least absolute shrinkage and selection operator (LASSO) regression and random forest (RF), followed by immune infiltration analysis (CIBERSORT).ResultsIn total, 16,370 patients with stage III NSCLC were included from SEER. A nomogram incorporating age, sex, race, marital status, histologic subtype, grade, T stage, N stage, tumor size, and treatment modalities demonstrated stable discrimination and good calibration in both cohorts for predicting 1-, 3-, and 5-year OS. TCGA analysis yielded 130 intersecting DEGs. Enrichment analyses highlighted extracellular matrix remodeling, immune response, and metabolic reprogramming as key processes associated with locally advanced progression. The dual-method feature-selection strategy identified three core genes (SFTPC, GKN2, and CLDN18), its expression levels of which were associated with differences in immune cell infiltration, including macrophages, dendritic cells, and mast cells.ConclusionsThis study integrates SEER-based prognostic modeling with TCGA-based transcriptomic analyses to characterize heterogeneity in stage III NSCLC. The nomogram may support individualized OS estimation, and the identified genes warrant further external validation and functional investigation. Highlight box Key findings • We developed and validated a robust prognostic nomogram for stage III non-small cell lung cancer (NSCLC) using a large Surveillance, Epidemiology, and End Results (SEER) cohort, demonstrating stable discrimination and calibration for predicting 1-, 3-, and 5-year overall survival. • Integrated transcriptomic analyses identified 130 progression-associated differentially expressed genes (DEGs), mainly enriched in extracellular matrix remodeling, immune response, and metabolic pathways. • Three core genes (SFTPC, GKN2, and CLDN18) were identified and found to be closely associated with distinct immune infiltration patterns. What is known and what is new? • Stage III NSCLC is highly heterogeneous, and existing prognostic models are limited in capturing both clinical and molecular complexity. • This study provides a dual-database, multi-scale framework integrating population-level clinical modeling with transcriptomic and immune microenvironment analyses. We identified novel progression-associated genes and linked their expression to immune landscape alterations in locally advanced NSCLC. What is the implication, and what should change now? • The proposed nomogram may assist clinicians in individualized risk stratification and survival prediction for stage III NSCLC. • The identified genes (SFTPC, GKN2, and CLDN18) may serve as potential biomarkers or therapeutic targets and warrant further validation. • Future studies should integrate molecular features into clinical models and validate these findings in prospective and contemporary cohorts. Introduction Lung cancer is the second most commonly diagnosed cancer worldwide and the deadliest malignancy, accounting for the highest number of cancer-related deaths (1). Non-small cell lung cancer (NSCLC) is the most common histologic type, representing approximately 80–85% of all lung cancers. NSCLC development is closely associated with multiple factors; tobacco smoking is the most important risk factor, with about 85% of patients reporting a smoking history, and risk increases with smoking intensity and duration (2). Genetic factors also contribute to NSCLC susceptibility and pathogenesis (3). Only around 20% of NSCLC patients are diagnosed at an early stage and are therefore candidates for curative resection (4). In early-stage NSCLC, surgical resection is the preferred treatment, and adjuvant chemotherapy and targeted therapy can further improve outcomes. By contrast, locally advanced NSCLC, which typically corresponds to stage III disease, is characterized by complex local invasion and/or regional lymph node metastasis, and often requires multimodality treatment integrating radiotherapy, chemotherapy, and immunotherapy. Consequently, the management of NSCLC poses substantial challenges for both clinicians and patients (5). Stage III NSCLC represents a clinically heterogeneous group of locally advanced disease, in which tumors remain confined to the thorax but have invaded adjacent structures or regional lymph nodes and are often not amenable to complete surgical resection (6). Stage III NSCLC is clinically heterogeneous and encompasses diverse presentations (7). Patients with locally advanced NSCLC generally experience a high recurrence risk and poor survival, with a 5-year survival rate of approximately 20–30% (8). Under the 8th edition tumor-node-metastasis (TNM) classification, the 5-year survival rates for clinical/pathological stage IIIA are 36%/41%, and for stage IIIB are 26%/24% (9). Approximately 70% of NSCLC patients present with metastatic disease, which contributes to unfavorable outcomes (10). Even within locally advanced disease, prognosis varies markedly with tumor burden, extent of nodal involvement, and treatment patterns (7). Current strategies often apply broad coverage multimodality regimens, which may not adequately address differences in treatment sensitivity across patient subtypes. Importantly, NSCLC is a highly heterogeneous disease. Significant genomic differences exist between early- and late-stage NSCLC (11,12). Such alterations can ultimately reshape the tumor microenvironment, promote metastasis, and drive therapeutic resistance. Mutations in EGFR, KRAS, ALK, and TP53 are common in NSCLC and play critical roles in progression (13). During cancer evolution, intratumoral heterogeneity can be observed as phenotypic variation driven by propagated genomic instability, leading to mutations, somatic copy number alterations, and epigenomic changes (14). Accordingly, accurately predicting progression and identifying genes influencing NSCLC progression are essential for diagnosis and treatment optimization (15). Here, we propose and implement a dual-database strategy that integrates macroscopic clinical risk stratification with exploratory microscopic biological interpretation. Specifically, we developed and validated an overall survival (OS) nomogram for TNM stage III NSCLC using the Surveillance, Epidemiology, and End Results (SEER) database as a clinically applicable risk prediction tool. To complement these clinical findings, we further investigated transcriptomic alterations associated with locally advanced progression using The Cancer Genome Atlas (TCGA). We performed pathway enrichment analysis, identified key signature genes through the intersection of least absolute shrinkage and selection operator (LASSO) and random forest (RF) approaches, and explored their associations with tumor immune infiltration. Together, these analyses aim to provide a more comprehensive understanding of stage III NSCLC by linking population-level clinical risk patterns with potential underlying molecular and immunological mechanisms. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0612/rc). Methods Data sources This study used the SEER database, which collects information from U.S. SEER cancer registries and covers approximately 28% of the U.S. population. Using SEER*Stat, version 8.3.9, we identified patients diagnosed with stage III NSCLC between 2010 and 2017. In addition, ribonucleic acid sequencing (RNA-seq) transcriptomic expression data and detailed clinicopathological information for normal tissues and locally advanced NSCLC were obtained from TCGA. According to the original database reports, informed consent had been obtained from all participants included in the original studies. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Case selection All stage III NSCLC patients from SEER diagnosed between 2010 and 2017 were screened. Clinical information was collected. Figure 1 presents the inclusion and exclusion criteria. Inclusion criteria were: (I) diagnosis between 2010 and 2017; (II) histopathological confirmation of lung cancer; (III) lung cancer as the only malignancy; (IV) only one primary tumor; (V) stage III disease [American Joint Committee on Cancer (AJCC) 7th edition]; and (VI) eligible histology codes including squamous cell carcinoma (8052, 8070–8074), adenocarcinoma (8140, 8250–8255, 8260, 8290, 8310, 8323, 8430, 8480, 8481, 8490, 8550, 8551), adenosquamous carcinoma (8560), large cell carcinoma (8012, 8013), and other NSCLC (8046, 8010, 8050, 8575). Exclusion criteria were: (I) missing key variables (race, treatment pattern, or survival time); and (II) follow-up shorter than 1 month. A total of 16,370 patients met the criteria. Flowchart of participant inclusion and exclusion. DEG, differentially expressed gene; SEER, Surveillance, Epidemiology, and End Results; TCGA, The Cancer Genome Atlas; TNM, tumor-node-metastasis. From TCGA, RNA-seq and clinicopathological data included 168 stage III NSCLC samples, 803 early-stage lung cancer samples, and 110 normal tissue samples. Variable definition Variables in the SEER analysis included demographics (age, sex, race), tumor-related factors (grade, size), and treatments (chemotherapy, radiotherapy, surgery, lymph node dissection). Categorical variables were defined as follows: age (<70 vs. ≥70 years), sex (male vs. female), race (White, Black, others), marital status (married, unmarried/divorced, unknown), laterality (left vs. right), histology (squamous cell carcinoma, adenocarcinoma, other NSCLC), surgery (yes vs. no/unknown), radiotherapy (yes vs. no/unknown), chemotherapy (yes vs. no/unknown), lymph node surgery (yes vs. no/unknown), tumor size (≤3 cm, 3–5 cm, 5–7 cm, >7 cm, unknown), primary site (upper, middle, lower lobe), and tumor grade (I–II vs. III–IV). OS was the primary endpoint and was calculated from diagnosis to death from any cause. In TCGA, clinical variables included demographics (age, sex, race), histology, laterality, American Joint Committee on Cancer (AJCC) stage, and primary site. Overlapping variable definitions followed those used in SEER. Statistical analysis Eligible stage III NSCLC cases were randomly divided in a 7:3 ratio into a training cohort and an internal validation cohort. Descriptive statistics were used to summarize demographic and clinical characteristics. Univariable and multivariable Cox proportional hazards regression analyses were performed in the training cohort to identify independent prognostic factors for OS. These variables were incorporated into a nomogram to estimate 1-, 3-, and 5-year OS. Model discrimination and calibration were evaluated using time-dependent ROC curves [area under the curve (AUC)] and calibration plots. Identification of differentially expressed genes (DEGs) and functional enrichment To identify DEGs between locally advanced NSCLC and early-stage/normal samples, DESeq2 was applied in the TCGA cohort. Significance thresholds were set at |log2 fold change| >1 and adjusted P value (Padj) <0.05. Volcano plots and heatmaps were generated to visualize significant DEGs. Functional enrichment analyses were performed using the clusterProfiler package, including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses (16-19). A significance threshold of P<0.05 was applied for enrichment results. Machine learning to identify disease-specific signature genes To identify disease signature genes among the DEGs, LASSO regression was first applied using the glmnet package. Ten-fold cross-validation was used to determine the optimal regularization parameter, and genes corresponding to the minimum error were selected. RF analysis was then performed using the randomForest package, and the top five genes by importance were selected. Finally, the intersection of genes identified by LASSO and RF was defined as the final signature gene set for locally advanced NSCLC. Immune infiltration analysis Using R and the CIBERSORT algorithm, we analyzed gene expression matrices to estimate the relative proportions of 22 immune cell types in each sample. Samples with CIBERSORT P<0.05 were considered reliable. The limma package was used to compare immune cell fractions between high- and low-expression groups for each signature gene. All analyses were conducted using statistical software and R (version 4.0.2). Results Univariable and multivariable Cox regression analyses A total of 16,370 stage III NSCLC patients were randomly assigned to the training cohort (n=11,459) and validation cohort (n=4,911). Baseline characteristics were comparable between cohorts (Table 1). Univariable Cox regression identified age, sex, race, marital status, histology, surgery, radiotherapy, chemotherapy, lymph node surgery, tumor size, tumor stage, AJCC stage, T stage, and N stage as factors associated with OS (all P<0.05). On multivariable analysis, age, sex, race, marital status, histology, surgery, radiotherapy, chemotherapy, lymph node surgery, tumor size, grade, T stage, and N stage remained independent prognostic factors for OS (Table 2). Table 1Baseline characteristics of patients with stage III NSCLCCharacteristicsTraining (n=11,459)Validation (n=4,911)PAge, years0.49 <706,212 (54.2)2,692 (54.8) ≥705,247 (45.8)2,219 (45.2)Sex0.81 Male6,383 (55.7)2,725 (55.5) Female5,076 (44.3)2,186 (44.5)Race0.29 White9,047 (79.0)3,902 (79.5) Black1,463 (12.8)586 (11.9) Others949 (8.3)423 (8.6)Marital status0.18 Married5,898 (51.5)2,595 (52.8) Unmarried5,145 (44.9)2,129 (43.4) Unknown416 (3.6)187 (3.8)Laterality0.75 Left4,657 (40.6)1,982 (40.4) Right6,802 (59.4)2,929 (59.6)Hist0.83 Adenocarcinoma5,522 (48.2)2,371 (48.3) Squamous cell carcinoma4,919 (42.9)2,091 (42.6) Others1,018 (8.9)449 (9.1)Surgery0.77 No/unknown7,851 (68.5)3,377 (68.8) Yes3,608 (31.5)1,534 (31.2)Radiation0.03 No/unknown4,633 (40.4)1,895 (38.6) Yes6,826 (59.6)3,016 (61.4)Chemotherapy0.67 No/unknown3,465 (30.2)1,468 (29.9) Yes7,994 (69.8)3,443 (70.1)LN surgery0.70 No/unknown7,727 (67.4)3,296 (67.1) Yes3,732 (32.6)1,615 (32.9)Tumor size, cm0.059 ≤32,844 (24.8)1,288 (26.2) 3–53,259 (28.4)1,365 (27.8) 5–72,381 (20.8)938 (19.1) >7 cm2,353 (20.5)1,043 (21.2) Unknown622 (5.4)277 (5.6)Primary site0.52 Lower3,462 (30.2)1,440 (29.3) Middle534 (4.7)234 (4.8) Upper7,463 (65.1)3,237 (65.9)Grade0.57 I–II4,735 (41.3)2,005 (40.8) III–IV6,724 (58.7)2,906 (59.2)AJCC0.94 IIIA8,438 (73.6)3,613 (73.6) IIIB3,021 (26.4)1,298 (26.4)T stage0.15 T11,412 (12.3)656 (13.4) T23,545 (30.9)1,456 (29.6) T33,045 (26.6)1,333 (27.1) T43,457 (30.2)1,466 (29.9)N stage0.61 N01,238 (10.8)511 (10.4) N11,200 (10.5)526 (10.7) N27,525 (65.7)3,202 (65.2) N31,496 (13.1)672 (13.7) Survival months31.56±34.0131.22±33.830.55Data are presented as mean ± standard deviation or n (%). AJCC, American Joint Committee on Cancer; LN, lymph node; N, node; NSCLC, non-small cell lung cancer; T, tumor. Baseline characteristics of patients with stage III NSCLC Data are presented as mean ± standard deviation or n (%). AJCC, American Joint Committee on Cancer; LN, lymph node; N, node; NSCLC, non-small cell lung cancer; T, tumor. Table 2Univariable and multivariable Cox regression for overall survivalCharacteristicsUnivariable Cox regressionMultivariable Cox regressionHR (95% CI)PHR (95% CI)PAge, years <701 (reference)1 (reference) ≥701.43 (1.38–1.49)<0.0011.28 (1.23–1.34)<0.001Sex Male1 (reference)1 (reference) Female0.77 (0.74–0.8)<0.0010.81 (0.78–0.86)<0.001Race White1 (reference)1 (reference) Black1.02 (0.96–1.09)0.450.94 (0.89–1.00)0.049 Others0.77 (0.71–0.83)<0.0010.81 (0.75–0.87)<0.001Marital status Married1 (reference)1 (reference) Unmarried1.18 (1.13–1.22)<0.0011.10 (1.06–1.15)<0.001 Unknown1.12 (1.01–1.25)0.041.01 (0.91–1.13)0.84Laterality Left1 (reference) Right1.03 (0.99–1.07)0.16Hist Adenocarcinoma1 (reference)1 (reference) Squamous cell carcinoma1.52 (1.46–1.59)<0.0011.16 (1.11–1.22)<0.001 Others1.65 (1.54–1.78)<0.0011.25 (1.16–1.34)<0.001Surgery No/unknown1 (reference)1 (reference) Yes0.45 (0.43–0.47)<0.0010.59 (0.54–0.65)<0.001Radiation No/unknown1 (reference)1 (reference) Yes0.92 (0.88–0.96)<0.0010.78 (0.74–0.82)<0.001Chemotherapy No/unknown1 (reference)1 (reference) Yes0.58 (0.56–0.60)<0.0010.62 (0.59–0.65)<0.001LN surgery No/unknown1 (reference)1 (reference) Yes0.46 (0.44–0.48)<0.0010.80 (0.73–0.88)<0.001Tumor size, cm ≤31 (reference)1 (reference) 3–51.30 (1.23–1.37)<0.0011.16 (1.08–1.24)<0.001 5–71.53 (1.44–1.62)<0.0011.30 (1.21–1.40)<0.001 >71.72 (1.62–1.82)<0.0011.47 (1.36–1.58)<0.001 Unknown1.97 (1.80–2.16)<0.0011.50 (1.36–1.66)<0.001Primary site Lower1 (reference) Middle0.93 (0.84–1.02)0.14 Upper0.97 (0.93–1.02)0.23Grade I–II1 (reference)1 (reference) III–IV1.20 (1.16–1.25)<0.0011.11 (1.07–1.16)<0.001AJCC IIIA1 (reference)1 (reference) IIIB1.41 (1.34–1.47)<0.0010.90 (0.81–1.00)0.056T stage T11 (reference)1 (reference) T21.25 (1.17–1.34)<0.0011.01 (0.93–1.11)0.75 T31.38 (1.29–1.48)<0.0011.04 (0.95–1.14)0.42 T41.49 (1.39–1.59)<0.0011.23 (1.09–1.38)<0.001N stage N01 (reference)1 (reference) N10.97 (0.89–1.06)0.541.32 (1.19–1.47)<0.001 N21.10 (1.03–1.17)0.0061.55 (1.39–1.73)<0.001 N31.41 (1.30–1.52)<0.0011.75 (1.46–2.10)<0.001AJCC, American Joint Committee on Cancer; CI, confidence interval; HR, hazard ratio; LN, lymph node; N, node; T, tumor. Univariable and multivariable Cox regression for overall survival AJCC, American Joint Committee on Cancer; CI, confidence interval; HR, hazard ratio; LN, lymph node; N, node; T, tumor. Nomogram construction and validation Based on independent prognostic factors identified in multivariable Cox analysis, a nomogram was developed to predict 1-, 3-, and 5-year OS for stage III NSCLC (Figure 2). Each level of every variable was assigned a score; the total score was calculated by summing the scores for all variables, and the corresponding predicted probabilities provided individualized estimates of OS at 1, 3, and 5 years. Nomogram for predicting 1-, 3-, and 5-year OS in locally advanced (stage III) NSCLC. LN, lymph node; N, node; NSCLC, non-small cell lung cancer; OS, overall survival; T, tumor. AUCs were compared between cohorts (Figure 3). Receiver operating characteristic (ROC) curves plot sensitivity against 1 − specificity across different probability thresholds, and the AUC summarizes discrimination (0.5 indicates no discriminative ability and 1.0 indicates perfect discrimination). In the training cohort, the AUCs for predicting 1-, 3-, and 5-year OS were 0.742, 0.746, and 0.743, respectively. In the validation cohort, the corresponding AUCs were 0.740, 0.727, and 0.721, indicating good discrimination. Calibration plots demonstrated high agreement between predicted and observed survival in both cohorts (Figure 4). ROC curves evaluating nomogram accuracy at 1, 3, and 5 years. AUC, area under the curve; ROC, receiver operating characteristic. Calibration curves for the nomogram in the training and internal validation cohorts. OS, overall survival. Differential gene expression analysis In the TCGA cohort, 168 eligible stage III (locally advanced) NSCLC (LA-NSCLC) cases were identified. Comparing stage III tumors with normal tissues yielded 6,378 DEGs, including 3,945 upregulated and 2,433 downregulated genes (Padj <0.05 and |log2FC| >1; Figure 5A). Comparing stage III tumors with early-stage tumors yielded 175 DEGs, including 15 upregulated and 160 downregulated genes (Padj <0.05 and |log2FC| >1; Figure 5B). The intersection of these comparisons resulted in 130 DEGs, which were used as the focus gene set for downstream analyses (Figure 5C). Volcano plots of DEGs between stage III tumors and (A) normal tissues, (B) early-stage tumors, and (C) intersection of DEGs. DEG, differentially expressed gene. Functional enrichment analysis To further investigate the biological functions of the 130 intersecting DEGs in LA-NSCLC, GO and KEGG enrichment analyses were performed using clusterProfiler, and results were visualized as bubble plots (Figure 6). Table 3 summarizes the top five GO terms ranked by gene counts. Top enriched GO terms (top 5) and KEGG pathways (top 10) among DEGs. BP, biological process; CC, cellular component; DEG, differentially expressed gene; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function. Table 3GO enrichment results for DEGs between LA tumors and early-stage/normal tissuesCategoryTermCountGene ratioP valueBPVascular process in circulatory system77/1100.001Cell-cell adhesion via plasma-membrane adhesion molecules77/1100.001Defense response to bacterium77/1100.003Amide transport77/1100.005Forebrain development77/1100.01CCSecretory granule lumen77/1210.003Cytoplasmic vesicle lumen77/1210.004Vesicle lumen77/1210.004Collagen-containing extracellular matrix66/1210.047Primary lysosome44/1210.01MFProtein tyrosine kinase binding55/1160.001Hormone activity55/1160.001Endopeptidase inhibitor activity55/1160.004Heparin binding55/1160.005Peptidase inhibitor activity55/1160.005BP, biological process; CC, cellular component; DEG, differentially expressed gene; GO, Gene Oncology; LA, locally advanced; MF, molecular function. GO enrichment results for DEGs between LA tumors and early-stage/normal tissues BP, biological process; CC, cellular component; DEG, differentially expressed gene; GO, Gene Oncology; LA, locally advanced; MF, molecular function. In GO biological processes (BPs), DEGs were enriched in vascular processes in the circulatory system, cell-cell adhesion via plasma membrane adhesion molecules, defense response to bacteria, and amide transport, suggesting abnormal immune responses, disrupted adhesion, and enhanced transport/metabolic activity, consistent with invasive and metastatic potential. In cellular components (CCs), enrichment included secretory granule lumen, cytoplasmic vesicle lumen, vesicle lumen, and collagen-containing extracellular matrix, indicating active secretion/exocytosis and extracellular matrix remodeling, which may contribute to tumor microenvironment reconstruction and immune evasion. In molecular functions (MFs), DEGs were enriched in protein tyrosine kinase binding, hormone activity, endopeptidase inhibitor activity, and heparin binding, suggesting active signaling, immune modulation, and proteolysis-related regulation in stage III NSCLC. KEGG analysis (Figure 6B) showed that the top enriched pathways involved cornified envelope formation (lowest P value), fat digestion and absorption, glycosphingolipid biosynthesis (lacto and neolacto series), linoleic acid metabolism, neuroactive ligand-receptor interaction, salivary secretion, neutrophil extracellular trap formation, virion-hepatitis viruses, and estrogen signaling. These findings collectively suggest coordinated alterations in extracellular structure remodeling, metabolic regulation, innate immune activation/chronic inflammation, and hormonal/neuroactive signaling during locally advanced progression (Table 4). Table 4KEGG enrichment results for DEGs between locally advanced tumors and early-stage/normal tissuesCategoryTermCountGene ratioP valuehsa04382Cornified envelope formation77/500.0001hsa04975Fat digestion and absorption33/500.001hsa00601Glycosphingolipid biosynthesis-lacto and neolacto series22/500.01hsa00591Linoleic acid metabolism22/500.01hsa04080Salivary secretion66/500.01hsa04970Neuroactive ligand-receptor interaction33/500.01hsa05143African trypanosomiasis22/500.02hsa04613Neutrophil extracellular trap formation44/500.02hsa03272Virion-hepatitis viruses22/500.03hsa04915Estrogen signaling pathway33/500.04DEG, differentially expressed gene; KEGG, Kyoto Encyclopedia of Genes and Genomes. KEGG enrichment results for DEGs between locally advanced tumors and early-stage/normal tissues DEG, differentially expressed gene; KEGG, Kyoto Encyclopedia of Genes and Genomes. Signature genes of locally advanced NSCLC Using LASSO regression (glmnet), five genes were selected as candidate signature genes, and corresponding plots were generated (Figure 7A,7B). RF analysis identified the top five genes by importance (Figure 7C,7D). The intersection of the two methods yielded three core genes: SFTPC, GKN2, and CLDN18. Machine learning approaches for identifying disease-specific genes: (A) LASSO coefficient profiles; (B) cross-validation for optimal λ; (C) random forest results; (D) gene importance ranking. LASSO, least absolute shrinkage and selection operator. Immune infiltration analysis Immune infiltration analysis based on 22 immune cell types revealed the immune landscape in stage III NSCLC. Figure 8 shows the overall distribution of immune cell subsets across LA-NSCLC samples. The most abundant infiltrating cells were M2 macrophages, followed by M0 macrophages. These results indicate a macrophage- and T-cell-dominated tumor microenvironment, with M2 macrophages representing an immunosuppressive phenotype potentially related to immune escape. Overall infiltration abundance distribution of 22 immune cell subsets in tumor tissue of LA-NSCLC patients. LA-NSCLC, locally advanced non-small cell lung cancer. To evaluate associations between the three core genes and immune infiltration, patients were stratified into high- and low-expression groups for each gene. As shown in Figure 9), proportions of monocytes, macrophages M0, dendritic cells resting, and resting mast cells differed significantly between expression groups (P<0.05) for SFTPC, GKN2, and CLDN18. Specifically, macrophages M2 were uniquely different between SFTPC expression groups; neutrophils were uniquely different between GKN2 expression groups; and naive B cells, resting memory CD4 T cells, and eosinophils were uniquely different between CLDN18 expression groups. Immune infiltration analysis: (A-C) comparisons of immune cell fractions between high- and low-expression groups for SFTPC, GKN2, and CLDN18. Discussion This study employs a complementary dual-database, multi-scale framework to interrogate heterogeneity in locally advanced NSCLC at two levels: population-based clinical prognostic modeling and molecular/immune characterization. Specifically, we constructed and validated an OS nomogram for stage III NSCLC in the SEER cohort to quantify the prognostic contributions of routinely available clinical variables and treatment factors and to enable risk stratification. In parallel, using TCGA transcriptome data, we identified molecular pathway features and immune infiltration patterns associated with locally advanced progression, providing biological context for the observed clinical heterogeneity and generating testable hypotheses for future integrative studies. First, in SEER, we developed a prognostic model based on survival analyses and Cox regression in 16,370 stage III NSCLC cases. Multiple factors, including age, sex, tumor stage, and surgery, were significantly associated with OS in clinical subgroups. The nomogram demonstrated good predictive performance in both the training and validation cohorts, supporting its potential use for clinical prognostic assessment. Multivariable Cox regression confirmed that age, sex, marital status, TNM stage, and treatment modalities were independent prognostic factors; notably, surgery was strongly associated with improved survival (P<0.001), suggesting that aggressive local treatment may still provide benefit even in locally advanced disease, consistent with prior studies (20-23). Age was a key determinant of prognosis, consistent with previous findings by Sun et al. (24). Female patients exhibited better outcomes than male patients, potentially related to hormonal and immune response differences as reported previously (25). Laterality and primary tumor site were incorporated into the nomogram, and SEER-based multivariable validation supported their prognostic value (26,27). Over the past decade, prognostic models have been increasingly used in oncology, and nomograms have become a preferred approach (28). The present nomogram integrates demographic, tumor, and treatment variables to support baseline prognostic evaluation and risk stratification for stage III NSCLC. Compared with earlier models that included limited variables, our model covers a broader range of clinically relevant features, enabling a more complete depiction of real-world prognostic heterogeneity (23). ROC analysis showed AUCs of 0.742, 0.746, and 0.743 for 1-, 3-, and 5-year OS in the training cohort, and 0.740, 0.727, and 0.721 in the validation cohort, indicating stable discrimination across time horizons. Calibration plots demonstrated close agreement between predicted and observed outcomes, supporting model reliability. Second, TCGA-based analyses identified key DEGs and highlighted molecular and immune microenvironment alterations associated with locally advanced progression. Unlike conventional comparisons limited to tumor vs normal, we additionally compared locally advanced vs. early-stage tumors and focused on intersecting DEGs, thereby enriching for progression-related signals. Enrichment of the 130 intersecting DEGs implicated extracellular matrix remodeling, abnormal cell-cell adhesion, immune response dysregulation, and metabolic reprogramming (29,30). These molecular and immune microenvironment findings highlight BPs associated with locally advanced progression in NSCLC and provide biological context for the observed heterogeneity of locally advanced disease. Using an intersection strategy combining LASSO and RF screening, we identified three core genes: SFTPC, GKN2, and CLDN18. Notably, these genes were globally downregulated in progression comparisons (locally advanced vs. normal/early-stage), reflecting cross-group average differences. In contrast, immune infiltration analyses were performed within the locally advanced subgroup by stratifying cases into relatively higher and lower expression groups; thus, “high expression” denotes a relative gradient within stage III tumors and should not be interpreted as inherently beneficial or detrimental. SFTPC encodes surfactant protein C, primarily secreted by alveolar type II cells and essential for normal lung function. Recent studies suggest that SFTPC may also relate to differentiation status and microenvironment remodeling in lung cancer (31,32). In our study, SFTPC was overall downregulated in LA-NSCLC; within LA tumors, lower SFTPC expression was associated with a higher proportion of macrophages M0, suggesting a myeloid milieu with limited polarization or effective antitumor activity. Higher SFTPC expression was associated with increased monocytes, resting dendritic cells and resting mast cells, along with a modest increase in M2 macrophages. This pattern may reflect co-occurrence of epithelial differentiation/repair programs with tissue repair- and matrix remodeling-associated myeloid states (including M2 macrophages and mast cells), rather than a direct causal “good/bad prognosis” relationship. Prior evidence indicates that SFTPC expression can be regulated by inflammatory contexts and may relate to dedifferentiation and immune ecological remodeling (33-36). GKN2 is a secreted protein that has been studied primarily in digestive system tumors, but evidence also links it to lung cancer prognosis and immune infiltration (37,38). Within LA-NSCLC, lower GKN2 expression was associated with an increased proportion of macrophages M0, whereas higher expression correlated with higher proportions of monocytes, resting dendritic cells, resting mast cells and differences in neutrophils. Mast cells can exert both pro-tumor and immune-modulatory effects in NSCLC, and their association with GKN2 suggests potential involvement in shaping immune contexture (39,40). CLDN18 encodes a tight-junction protein involved in maintaining epithelial polarity and barrier function and exhibits subtype-specific expression across solid tumors (41-43). Within LA-NSCLC, lower CLDN18 expression was associated with an increased proportion of macrophages M0, whereas higher expression correlated with higher proportions of monocytes, resting dendritic cells, resting mast cells, and certain resting lymphocyte components (e.g., naïve B cells, resting memory CD4 T cells, and eosinophils), suggesting an immune ecosystem characterized by immune cell presence but relatively quiescent activation. Given accumulating evidence on CLDN18 (including CLDN18.2) in tumorigenesis, barrier disruption, and as a potential immunotherapeutic target (44-50), we speculate that CLDN18-related adhesion/barrier states may influence immune cell entry and spatial distribution, thereby shaping microenvironment composition. Overall, the global downregulation of these three genes during progression suggests attenuation of epithelial differentiation and barrier-related transcriptional programs in LA-NSCLC. Nevertheless, expression gradients persist within stage III tumors and are consistently associated with immune infiltration patterns: lower expression aligns with a myeloid-dominant, insufficiently polarized milieu dominated by macrophages M0, whereas higher expression aligns with an ecosystem featuring monocytes, resting dendritic cells and resting mast cells. Because CIBERSORT provides relative fractions and may be influenced by tumor purity and tissue composition (51,52), these findings should be interpreted as hypotheses generating progression-linked transcriptional features-immune ecology differences and warrant further validation with tumor-purity adjustment and functional immune indices. This study has both clinical and basic research implications. Clinically, the nomogram may facilitate individualized prognostic evaluation and risk stratification to support decision-making for stage III NSCLC. Molecularly, the identified signature genes may serve as candidate biomarkers or therapeutic targets to improve precision oncology. From a mechanistic perspective, our results highlight extracellular matrix remodeling and immune ecosystem changes as key features of locally advanced progression, providing directions for future immunotherapy- and targeted-therapy-oriented research. Limitations This is a retrospective database study. SEER lacks molecular information (e.g., actionable alterations such as EGFR/ALK/KRAS) and provides limited granularity of treatment variables; therefore, we could not account for molecularly defined subtypes or targeted therapies in the nomogram. In addition, the SEER cohort (2010–2017) largely predates recent advances in stage III management (e.g., consolidation immunotherapy after chemoradiotherapy) (53), and the nomogram should be externally validated in contemporary cohorts before routine clinical use. TCGA has a relatively small sample size and a population composition different from SEER; thus, our approach primarily reflects complementary evidence rather than direct cohort-level integration. Moreover, we did not formally adjust for potential confounders (e.g., histology, smoking status, and other clinicopathologic differences) when comparing stage III tumors with early-stage tumors or normal tissues in TCGA, which may influence the observed transcriptomic signals. Immune infiltration results are algorithm-derived and require validation. Future work should validate the associations between the three genes and survival/treatment response in independent cohorts, assess whether these genes improve the nomogram, and perform in vitro and in vivo experiments to verify causal roles in adhesion/matrix remodeling and immune regulation. Conclusions In summary, this study integrates population-based clinical modeling with transcriptomic analysis to comprehensively characterize heterogeneity in stage III NSCLC. The proposed nomogram demonstrates reliable performance for individualized survival prediction and may assist clinical decision-making. Additionally, the identification of SFTPC, GKN2, and CLDN18 highlights potential molecular mechanisms underlying disease progression and immune microenvironment remodeling. These findings provide a foundation for future validation and translational research in precision oncology. Supplementary The article’s supplementary files as Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0612/prf