Development and validation of a multi-center prognostic model for predicting survival in non-small cell lung cancer using pulmonary and hematological data

Introduction

Epidemiological data from authoritative sources, including the Global Burden of Cancer database, reveal the highest incidence and mortality due to lung cancer (1). In China, this malignancy imposes a substantial public health burden (2). Five-year survival rates for advanced stages are extremely low (3). Over the past decade, multiple prognostic models have been developed based on clinicopathological variables such as tumor stage, histological subtype, performance status, and smoking history. Nevertheless, these models generally demonstrate only moderate predictive accuracy and limited external validation, restricting their clinical utility. To enhance prediction performance, molecular and genetic biomarkers [such as EGFR and KRAS mutations, ATR expression, and programmed death ligand 1 (PD-L1) expression] have been incorporated into prognostic models (4,5). While biologically informative, such models face barriers to implementation due to cost, availability, and heterogeneity in testing platforms. More recently, studies have explored hematological and lymph node skip metastasis as prognostic indicators, given their accessibility and biological relevance (6,7). However, no existing model combines preoperative pulmonary function parameters with hematological variables.

Pulmonary function parameters, including vital capacity (VC), forced vital capacity (FVC), and forced expiratory volume in one second (FEV1), maximum voluntary ventilation (MVV), serve as critical biomarkers for respiratory health assessment. These metrics quantify physiological lung performance and are associated with both lung cancer development and clinical outcomes (8-11). These findings highlight the potential utility of pulmonary function metrics as prognostic indicators in oncology practice.

Likewise, blood tests, a promising approach, may theoretically overcome tumor heterogeneity issues and provide comprehensive tumor information, especially in advanced metastatic non-small cell lung cancer (NSCLC) with further research (12). Several recent studies have attempted to establish prognostic models based on hematological factors, yet their discriminative power has often been modest. For instance, Ma and Wang constructed a prognostic model for NSCLC integrating inflammation and nutritional indexes, yet the area under the curve (AUC) was about 0.7 (13). Xie et al. aimed to develop a novel inflammatory and nutritional index to predict pathological complete response and survival prognosis in patients with NSCLC. However, although the index showed potential in predicting immunochemotherapy response, its ability to discriminate survival outcomes remained suboptimal (AUC =0.68) (14). Total serum bilirubin (TBIL) is a clinical indicator for evaluating hepatic function and biliary disorders (15). Notably, emerging evidence highlights significant associations between dysregulated TBIL levels and colorectal carcinogenesis (16), suggesting its potential as a prognostic biomarker in oncology.

In this study, using the clinically established data analysis database, a new clinical evaluation framework has been proposed based on the combined application of the respiratory function index and blood test indicators in the prognosis of lung cancer. We aimed to provide precision medicine strategies for patients in different strata and lay a theoretical foundation for further research. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-700/rc).

Methods

Participant screening and study design

In this retrospective study, we enrolled patients with NSCLC diagnosed between January 2014 and December 2020 at Sichuan Cancer Hospital and Institute. The sample size was determined to ensure statistical power based on previous studies and aimed to reach a comprehensive representation of the NSCLC population. We included cases with pathological confirmation through biopsy or surgical resection. Pathological staging was determined based on the 8th edition National Comprehensive Cancer Network Lung Cancer Classification criteria, patients with clinical stage I–IV or pathological stage I–IV NSCLC were included. Regarding pathologic type, patients diagnosed with adenocarcinoma, squamous cell carcinoma were included for analyses. Exclusion criteria included: (I) history of bronchial asthma; (II) history of chronic obstructive pulmonary disease (a heterogeneous lung condition characterised by chronic respiratory symptoms due to abnormalities of the airways and/or alveoli that cause persistent, often progressive, airflow obstruction); (III) comorbid diabetes mellitus, cardiac dysfunction (a complex clinical syndrome that results from any structural or functional impairment of ventricular filling or ejection of blood, the cardinal manifestations are dyspnea and fatigue), or severe hepatic impairment; (IV) non-pulmonary primary tumors or concurrent malignancies; and (V) loss to follow up. For external validation, cohorts from West China Hospital and Dazhu County People’s Hospital were established under identical criteria to verify model generalizability. Follow-up procedures entailed telephone interviews, aiming to capture overall survival (OS) as the primary endpoint. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by Ethics Committee of Sichuan Cancer Hospital and Institute (No. SCCHEC-02-2021-064). All participating hospitals were informed of and agreed to the conduct of this study. Informed consent was taken from all the patients.

Clinical data were retrieved from electronic medical records, comprising demographic characteristics (sex, age), medical history, smoking status, and oncological profiles including lesion dimensions, histopathological classification, and staging criteria. Respiratory functional parameters were analyzed using standardized pulmonary function tests, while hematological biomarkers were quantified by biochemical assays of blood specimens. Pulmonary function tests and hematologic tests were conducted before the initiation of any treatment, with samples collected in a fasting state between 7 a.m. and 9 a.m. (fasting was defined as refraining from eating after 10 p.m. the previous evening until blood sample collection was completed).

Variable selection

Variables were selected through backward stepwise regression guided by the Akaike information criterion (AIC). This method starts with a full model and, at each step, removes the least significant variable until no further variables can be removed. The modeling protocol was initiated by incorporating all candidate predictors (including age, sex, clinicopathological characteristics, pulmonary function parameters, and serum biochemical biomarkers) into the preliminary model. The model’s goodness-of-fit (quantified by log-likelihood) and corresponding AIC value were computed as baseline metrics. Subsequent iterations involved sequential exclusion of individual predictors, and each modified model was re-evaluated for AIC performance. The variable demonstrating maximal AIC reduction during this elimination process was removed. This selection cycle was iterated until no further AIC improvement was achieved by variable exclusion. The final model configuration, stabilized when AIC minimization plateaued, represented the optimal predictive feature subset. This systematic approach ensured retention of variables with significant prognostic relevance while maintaining model parsimony.

Development and validation of the prediction model

The cohort was segmented into training and testing sets in a 7:3 ratio for model construction and internal validation. External validation deployed independent datasets from collaborating hospitals. Model efficacy was assessed using Harrell’s concordance index (C-index), complemented by 500 bootstrap iterations generating time-dependent Receiver operating characteristic (ROC) curves and respective AUC across datasets. These analyses provided 3- and 5-year survival prediction performance insights. Model calibration and practical utility were further evaluated using decision curve analyses (DCAs) and calibration plots. Risk stratification, achieved through model-derived scores, segregated patients for survival comparison via Kaplan-Meier curves, applying two-sided log-rank tests for statistical rigor. A nomogram was constructed to visualize model utility in clinical settings.

Statistical analysis

Statistical analyses were conducted using R software (version 4.4.1). Continuous variables deviating from normal distribution criteria (assessed by normality tests) were reported as medians with interquartile ranges (IQRs) (25th–75th percentiles) [M (Q1–Q3)]. The pathological stage was analyzed as an ordinal categorical variable, comparing groups using the Kruskal-Wallis test. Categorical data were expressed as n (%) and analyzed through Chi-squared tests. Statistical significance was considered for two-tailed P values <0.05.

Results

Baseline characteristics

Figure 1 illustrates the research flow, with 1,013 eligible patients with lung cancer meeting the inclusion criteria. The cohort comprised 732 cases from Sichuan Cancer Hospital and Institute (divided into 513 training and 219 testing samples using a 7:3 ratio) and 281 external validation cases from West China Hospital and Dazhu County People’s Hospital. Median ages were 59 (IQR, 51–66), 58 (IQR, 49–65), and 61 (IQR, 53–67) years for training, testing, and validation cohorts, respectively. Time refers to survival time. If the patient has died, it is calculated as the duration from the initial medical visit to the time of death (in months). If the patient is still alive, it is calculated as the duration from the initial medical visit to the last follow-up (in months). Demographic data revealed female sex predominance (70.2%, n=711). Tumor staging distribution showed 40.6% (n=411) stage I, 15.8% (n=160) stage II, 29.3% (n=297) stage III, and 14.3% (n=145) stage IV malignancies. Complete clinical characteristics are detailed in Table 1.

Figure 1 Flow chart of the study. The figure is created via Figdraw with credit. AUC, area under the curve; NSCLC, non-small cell lung cancer.

Backward stepwise regression combined with the AIC criteria for variable selection

Clinical characteristics, pathological characteristics, blood biochemical test results, and pulmonary function test results of the patients were included in the initial model, specifically, sex, age, pathological stage (stage), maximum tumor diameter (size), ground-glass nodule (GGN), tumor classification, and smoking status (smoking), TBIL, direct bilirubin (DBIL), free bilirubin (FBIL), total protein (TP), albumin (ALB), globulin (GLB), alanine aminotransferase (ALT), aspartate aminotransferase (AST), blood urea nitrogen (UREA), creatinine (CREA), glucose (GLU), total cholesterol (TC), triglycerides (TGs), FVC, FEV1, peak expiratory flow (PEF), maximal mid-expiratory flow (MMEF), MVV, VC, diffusing capacity of the lung for carbon monoxide to alveolar volume ratio (DLCO/VA). According to the AIC criteria, FEV1, MVV, smoking, stage, TBIL, and tumor classification were selected. Table 2 shows the AIC screening process. Table 3 presents the hazard ratios and P values for the six included variables. Pathological stage, smoking status, FEV1, and MVV were statistically significant variables. Pathological stage and FEV1 were the two greatest risk factors, while smoking status also demonstrated a considerable effect. Tumor classification did not reach statistical significance. The 95% confidence intervals (CIs) for pathological stage, smoking status, FEV1, TBIL, and MVV did not cross 1, indicating that the association between these variables and increased risk was both significant and reliable. In contrast, the association shown for tumor classification was not significant. Table S1 presents the variance inflation factors (VIFs) for the six variables. Pathological stage, tumor classification, smoking status, and TBIL exhibited negligible collinearity. FEV1 and MVV showed mild collinearity. Table S2 shows the C-index values of the adjusted model after modifying variables. We removed FEV1 and re-modeled with the remaining variables, the C-indices were 0.732 (training), 0.724 (test), and 0.735 (validation). Similarly, after removing MVV, the re-modeled C-indices were 0.733 (training), 0.729 (test), and 0.733 (validation), both of which were lower than when both were included. Table S3 shows the tests for equal proportional hazards were performed with Schoenfeld residuals. As can be seen from the Table S3, the tests for each covariate were not statistically significant, and the global tests were not statistically different; therefore, we can assume equal proportional risks. Figure S1 shows the Schoenfeld residual test results for the six variables, used to test the proportional hazards assumption in the Cox model. Figure S2 shows the plot of deviance residuals. We found that the regression coefficients changed very little after removing each observation, indicating that no single observation had a particularly influential effect on the model. Overall, all variables satisfied the proportional hazards assumption at conventional significance levels.

Establishing a prognostic model using six variables

A Cox proportional hazards model was constructed using six variables selected by backward stepwise regression: FEV1, MVV, TBIL, smoking status, tumor classification, and pathological stage. The model achieved C-index of 0.740 (95% CI: 0.707–0.773), 0.734 (95% CI: 0.681–0.787), and 0.746 (95% CI: 0.701–0.791) in the training set, test set, and external validation set, respectively. Figure 2A-2C display the time-dependent ROC curves, with AUC values exceeding 0.7 across all three cohorts, indicating robust predictive performance of the model. In the training set (Figure 2D), the AUC values for predicting 3- and 5-year OS were 0.770 and 0.774, respectively. Similarly, in the test set (Figure 2E), the model showed AUCs of 0.770 and 0.789 for 3- and 5-year OS predictions, respectively. External validation (Figure 2F) confirmed the model’s reliability, with AUCs of 0.763 and 0.783 for 3- and 5-year OS, respectively.

Figure 2 Establishing the prediction model. (A) Time-dependent ROC curve for the training set. (B) Time-dependent ROC curve for the test set. (C) Time-dependent ROC curve for the validation set. (D) AUC curves for the training set for predicting 3- and 5-year overall survival. (E) AUC curves for the test set for predicting 3- and 5-year overall survival. (F) AUC curves for the validation set for predicting 3- and 5-year overall survival. AUC, area under the curve; ROC, receiver operating characteristic.

The Cox model exhibits strong calibration and clinical utility

Calibration plots for 3- and 5-year OS predictions in patients with NSCLC are presented in Figures 3A (training cohort), Figure 3B (test cohort), and Figure 3C (external validation cohort). The model consistently demonstrated robust calibration accuracy across all cohorts. To evaluate clinical utility, DCA was performed. Figure 3D-3F illustrate clinically actionable threshold probabilities for 3-year OS predictions in the training, test, and external validation cohorts, respectively, while Figure 3G-3I display corresponding thresholds for 5-year OS predictions. Notably, the model exhibited enhanced clinical applicability for 5-year OS risk stratification compared to shorter-term predictions.

Figure 3 Evaluation of the model. (A) Calibration curves for 3- and 5-year survival rates in the training set. (B) Calibration curves for 3- and 5-year survival rates in the test set. (C) Calibration curves for 3- and 5-year survival rates in the validation set. (D) Decision curve for 3-year survival prediction in the training set. (E) Decision curve for 3-year survival prediction in the test set. (F) Decision curve for 3-year survival prediction in the validation set. (G) Decision curve for predicting 5-year survival in the training set. (H) Decision curve for the test set predicting 5-year survival. (I) Decision curve for the validation set predicting 5-year survival.

Constructing a visual nomogram

FEV1, MVV, TBIL, smoking status, tumor classification, and pathological stage data of 1,013 patients with NSCLC included in the study were integrated to generate a nomogram. According to the contribution of the variables in the Cox model to the outcome, the value of the variables was assigned, and the scores of each variable were added to obtain the total score. The mapping relationship between the total score and the outcome was used to predict the final result (Figure 4). For example, a patient with stage II adenocarcinoma and no history of smoking and a TBIL value of 35, a FEV1 value of 4, and an MVV value of 80 had a total score of 103. Consequently, his 3- and 5-year survival rates were more than 90%.

Figure 4 Integrating six variables to generate a nomogram. FEV1, forced expiratory volume in one second; LSCC, lung squamous cell carcinoma; LUAD, lung adenocarcinoma; MVV, maximum voluntary ventilation; TBIL, total serum bilirubin.

Comparative predictive performance of the risk stratification model

Patients were stratified into high- and low-risk cohorts based on the median predicted OS derived from the model. Across all cohorts—training (Figure 5A), test (Figure 5B), and validation (Figure 5C)—the low-risk category exhibited significantly higher OS rates compared to the high-risk group (P<0.001). Subsequent stratified analyses were used to evaluate the model’s discriminative capacity across clinical stages of NSCLC. In both the modeling (Figure 5D) and external validation (Figure 5E) cohorts, the model effectively differentiated survival outcomes among patients with stage I–V NSCLC, demonstrating robust performance across disease severities, P<0.001.

Figure 5 Analysis in different groups. (A) Survival curves for high- and low-risk groups in the training set. (B) Survival curves for the high- and low-risk groups in the test set. (C) Survival curves for the high- and low-risk groups in the validation set. (D) Survival outcomes of different pathological stages in the modeling cohort. (E) Survival outcomes of different pathological stages in the validation cohort.

Discussion

In this study, we developed a Cox proportional hazards model integrating pulmonary function parameters (FEV1 and MVV), hematological indices (TBIL), and clinicopathological data to predict survival outcomes in patients with NSCLC, highlighting the synergistic impact of respiratory function and systemic metabolic status on prognosis. This multidimensional approach offers novel insights for personalized risk stratification in clinical practice.

FEV1, a cornerstone metric for assessing chronic obstructive pulmonary disease severity, is a critical predictor of surgical tolerance and postoperative pulmonary complications in lung resection candidates. Wang et al. (17) demonstrated that optimizing FEV1 reduces postoperative atelectasis risk. Emerging evidence further implicates reduced FEV1 as an independent risk factor for lung cancer incidence (18), with meta-analyses indicating that a decrease in FEV1 by 10% is associated with a 20% (95% CI: 17–23%) increase in lung cancer risk (19). FEV1 is a prognostic factor for lung cancer (20) but its clinical utility remains controversial due to inconsistent findings from underpowered studies. Our analysis of 1,013 cases across multiple centers provides robust validation of FEV1’s predictive value. Notably, the inclusion of non-surgical patients (diagnosed via biopsy without operative indications) suggests that FEV1 may worsen prognosis through the airway inflammatory microenvironment.

As a core parameter of pulmonary ventilatory reserve, MVV was first identified in our study as a prognostic indicator for lung cancer. MVV reflects integrated respiratory mechanics, including respiratory muscle strength, thoracic compliance, lung elasticity, and airway resistance. Targeted respiratory muscle training may enhance MVV (21,22). We hypothesized that reduced MVV may indicate respiratory muscle atrophy, potentially exacerbating tumorigenesis through hypoxia-driven pathways. Early respiratory rehabilitation in patients with lung cancer and impaired MVV could improve survival outcomes. Another point we need to be mindful of in the study is the collinearity issue between FEV1 and MVV. We found that regardless of whether one of them is removed or both are excluded and the model is restructured, the C-index value of the model is inferior compared to when both are included. In fact, FEV1 and MVV are both indicators of lung function tests obtained through the same means, so including both does not incur any additional economic cost. We also consulted clinical experts, who indicated that it is difficult to prioritize the importance of one over the other. After comprehensive consideration, we decided to include both in our model.

TBIL, a key hepatic function parameter, is typically elevated in hepatic injury, biliary obstruction, or hemolytic disorders. In the course of immunotherapy for lung cancer, some patients develop immune checkpoint inhibitor-related hepatotoxicity due to the accumulation of immune checkpoint inhibitors, with its incidence ranging from 1% to 20% (23,24). TBIL is an important monitoring index. The survival rate of patients with NSCLC and liver injury after immunotherapy is reduced (25). Emerging evidence links TBIL to prognosis across diverse diseases. For instance, Cao et al. (26) identified TBIL as a prognostic marker of colorectal cancer. Similar associations have been observed in bladder cancer (27), idiopathic pulmonary fibrosis (28), and ovarian cancer (29), underscoring its broader biological significance. Notably, Atasoy et al. (30) demonstrated a three-fold survival advantage in patients with lung cancer with high versus low TBIL levels. A prospective study conducted in Belgium by Temme et al. demonstrated a significant inverse association between elevated total bilirubin (TBIL) levels and cancer-related mortality (31). This finding may be attributed to the antioxidant, anti-inflammatory, and antiproliferative properties of bilirubin, as oxidative stress is a key contributor to carcinogenesis (32). Mechanistically, bilirubin has been shown to suppress the mammalian target of rapamycin (mTOR) signaling pathway by modulating adenosine monophosphate activated protein kinase (AMPK) activity, thereby exerting antiproliferative effects. AMPK plays a pivotal role in maintaining cellular energy homeostasis, and its regulation under elevated bilirubin levels may contribute positively to these outcomes (33).

Our findings corroborate TBIL as a predictor of NSCLC prognosis. Mechanistically, this may relate to bilirubin’s role as an endogenous antioxidant—low TBIL levels may compromise antioxidant defenses, exacerbating oxidative stress, deoxyribonucleic acid (DNA) damage, and carcinogenesis. This metric could aid in immunotherapy decision-making. However, TBIL can be influenced by various physiological and pathological factors, such as liver diseases, biliary obstruction, iron-deficiency anemia, and others. In clinical practice, we need to differentiate among patients and exclude these confounding factors.

Traditional prognostic prediction for NSCLC predominantly relies on clinical tumor node metastasis (TNM) staging. However, real-world clinical observations reveal limitations to this approach because factors such as severe pulmonary dysfunction may portend poor outcomes even in early-stage disease, further underscoring the need for biomarker-integrated models. Our study addresses this gap by incorporating readily accessible clinical parameters—including TNM stage, histological subtype, smoking status, TBIL levels, FEV1, and MVV—into a machine learning-derived prognostic tool trained on a large-scale cohort. Compared to TNM staging alone, the model is more sophisticated in risk stratification of patients, exemplified by a 5-year survival rate <30% for patients with nomogram scores >180. Such stratification supports tailored management: low-risk cohorts may benefit from quality-of-life optimization, while high-risk groups warrant early multidisciplinary intervention.

We excluded patients with chronic obstructive pulmonary disease to mitigate confounding effects. The model’s utility extends beyond prognosis prediction; for instance, the assessment of lung function can guide the exercise plan for patients with lung cancer, and the quality of life can be improved following the improvement in lung function through respiratory muscle training and other approaches.

There are some limitations in this study. The model was based on baseline data, but the prognosis of patients with lung cancer may change dynamically with treatment response and complications. Secondly, our dataset did not include detailed treatment-related information such as specific surgical techniques, postoperative outcomes, and systemic therapy regimens. While these are important factors that can influence patient prognosis, the absence of such detailed data may introduce some residual confounding. The fluctuations in TBIL levels during treatment are not captured by our current static model, which could affect the understanding of how these dynamic changes impact patient outcomes. However, the framework established in our study provides a foundational analysis that future research can build upon by incorporating more dynamic and comprehensive datasets. This will enable a deeper exploration of how detailed treatment modalities and biomarker fluctuations influence long-term prognostic outcomes, potentially enhancing the precision and applicability of prognostic models in clinical practice. Although our study included data from three research centers, all of these data were collected within China. We are acutely aware of the importance of ethnic diversity in research outcomes. In the future, we will promote collaboration with more international institutions and include participants from different ethnicities and regions to validate and broaden the generalizability of our research conclusions.

Conclusions

In summary, FEV1, MVV, TBIL, smoking status, tumor classification, and pathological stage data can be used to construct a clinical prognosis prediction model for NSCLC. It is a low-cost and easy-to-obtain prognostic tool that can help clinicians make decisions and achieve more accurate individualized treatment. Future studies should elucidate its biological mechanism and explore its translational value in clinical practice.

Acknowledgments

We thank https://app.editage.com/ for guidance on the language of the article.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-700/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-700/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-700/prf

Funding: This study was funded by Sichuan Medical Association Research Project (No. 2024HR13).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-700/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by Ethics Committee of Sichuan Cancer Hospital and Institute (No. SCCHEC-02-2021-064). All participating hospitals were informed of and agreed to the conduct of this study. Informed consent was taken from all the patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Laversanne M, Sung H, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Ji YT, Liu SW, Zhang YM, et al. Comparison of the latest cancer statistics, cancer epidemic trends and determinants between China and the United States. Zhonghua Zhong Liu Za Zhi 2024;46:646-56. [Crossref] [PubMed]
3. Arbour KC, Riely GJ. Systemic Therapy for Locally Advanced and Metastatic Non-Small Cell Lung Cancer: A Review. JAMA 2019;322:764-74. [Crossref] [PubMed]
4. Jiang L, Luo C, Zhang P, et al. ATR expression as a prognostic biomarker in KRAS-mutated non-small cell lung cancer. J Thorac Dis 2025;17:5197-209. [Crossref] [PubMed]
5. Sisca L, Cascetta P, Aijaz A, et al. KRAS and STK11 co-mutations in resectable non-small cell lung cancer: enduring prognostic value and impaired immunotherapy response. Transl Lung Cancer Res 2025;14:2374-82. [Crossref] [PubMed]
6. Liang S, Yang Y, Tian J, et al. Prognostic impacts of skip metastasis in N2a and N2b subgroups of non-small cell lung cancer: insights from a large cohort. Transl Lung Cancer Res 2025;14:2537-59. [Crossref] [PubMed]
7. Suwinski R, Kysiak M, Cybulska-Stopa B, et al. Assessing the prognosis of metastatic or recurrent non-small cell lung cancer in the era of modern systemic therapies: a multivariable analysis of 343 patients treated in Poland. Transl Lung Cancer Res 2025;14:2688-99. [Crossref] [PubMed]
8. Kyaw TW, Tsai MK, Wen CP, et al. Impaired lung function and lung cancer risk in 461 183 healthy individuals: a cohort study. BMJ Open Respir Res 2024;11:e001936. [Crossref] [PubMed]
9. Kang HS, Park YM, Ko SH, et al. Impaired Lung Function and Lung Cancer Incidence: A Nationwide Population-Based Cohort Study. J Clin Med 2022;11:1077. [Crossref] [PubMed]
10. Su Z, Jiang Y, Li C, et al. Relationship between lung function and lung cancer risk: a pooled analysis of cohorts plus Mendelian randomization study. J Cancer Res Clin Oncol 2021;147:2837-49. [Crossref] [PubMed]
11. Ju JW, Suh M, Choi H, et al. Clinical Factors Affecting Discrepancy Between Predicted and Long-term Actual Lung Function Following Surgery. Clin Nucl Med 2024;49:e574-9. [Crossref] [PubMed]
12. Nagasaka M, Uddin MH, Al-Hallak MN, et al. Liquid biopsy for therapy monitoring in early-stage non-small cell lung cancer. Mol Cancer 2021;20:82. [Crossref] [PubMed]
13. Ma S, Wang L. Prognostic factors and predictive model construction in patients with non-small cell lung cancer: a retrospective study. Front Oncol 2024;14:1378135. [Crossref] [PubMed]
14. Xie Y, Shi Z, Chen T, et al. The Lung Cancer Immune Prognostic Score predicts pathologic complete response and survival in NSCLC patients receiving neoadjuvant immunochemotherapy. Front Immunol 2025;16:1567565. [Crossref] [PubMed]
15. Vitek L, Hinds TD Jr, Stec DE, et al. The physiology of bilirubin: health and disease equilibrium. Trends Mol Med 2023;29:315-28. [Crossref] [PubMed]
16. Li W, Shen L, Yang Q, et al. Oxidative stress index-based scoring for prediction of long-term prognosis in patients with colorectal cancer with liver metastases. Sci Rep 2025;15:5253. [Crossref] [PubMed]
17. Wang J, Deng N, Qi F, et al. The effectiveness of postoperative rehabilitation interventions that include breathing exercises to prevent pulmonary atelectasis in lung cancer resection patients: a systematic review and meta-analysis. BMC Pulm Med 2023;23:276. [Crossref] [PubMed]
18. Murison KR, Warkentin MT, Khodayari Moez E, et al. Respiratory Function as a Prognostic Factor for Lung Cancer in Screening and General Populations. Ann Am Thorac Soc 2025;22:591-7. [Crossref] [PubMed]
19. Fry JS, Hamling JS, Lee PN. Systematic review with meta-analysis of the epidemiological evidence relating FEV1 decline to lung cancer risk. BMC Cancer 2012;12:498. [Crossref] [PubMed]
20. Cho O, Oh YT, Chun M, et al. Prognostic implication of FEV1/FVC ratio for limited-stage small cell lung cancer. J Thorac Dis 2018;10:1797-805. [Crossref] [PubMed]
21. Bhagel P, Saha M. Effects of yogic intervention on pulmonary function and respiratory muscle strength parameters: A systematic literature review and meta-analysis. J Biosci 2021;46:76.
22. Sales AT, Fregonezi GA, Ramsook AH, et al. Respiratory muscle endurance after training in athletes and non-athletes: A systematic review and meta-analysis. Phys Ther Sport 2016;17:76-86. [Crossref] [PubMed]
23. Remash D, Prince DS, McKenzie C, et al. Immune checkpoint inhibitor-related hepatotoxicity: A review. World J Gastroenterol 2021;27:5376-91. [Crossref] [PubMed]
24. Triantafyllou E, Gudd CLC, Possamai LA. Immune-mediated liver injury from checkpoint inhibitors: mechanisms, clinical characteristics and management. Nat Rev Gastroenterol Hepatol 2025;22:112-26. [Crossref] [PubMed]
25. Swanson LA, Kassab I, Tsung I, et al. Liver injury during durvalumab-based immunotherapy is associated with poorer patient survival: A retrospective analysis. Front Oncol 2022;12:984940. [Crossref] [PubMed]
26. Cao Y, Deng S, Yan L, et al. A nomogram based on pretreatment levels of serum bilirubin and total bile acid levels predicts survival in colorectal cancer patients. BMC Cancer 2021;21:85. [Crossref] [PubMed]
27. Zhang X, Ma L. Predictive Value of the Total Bilirubin and CA50 Screened Based on Machine Learning for Recurrence of Bladder Cancer Patients. Cancer Manag Res 2024;16:537-46. [Crossref] [PubMed]
28. Shi S, Liu Y, Qiu X, et al. Correlation between serum bilirubin levels and the severity as well as the prognosis of idiopathic pulmonary fibrosis. Chron Respir Dis 2020;17:1479973120957676. [Crossref] [PubMed]
29. Xi XX, Wang HL, Chen T, et al. Prognostic value of preoperative serum bilirubin levels in ovarian cancer. Am J Transl Res 2020;12:2267-80.
30. Atasoy O, Cini N, Erdogan MA, et al. Radiotherapy and high bilirubin may be metformin like effect on lung cancer via possible AMPK pathway modulation. Bratisl Lek Listy 2022;123:100-9. [Crossref] [PubMed]
31. Temme EH, Zhang J, Schouten EG, et al. Serum bilirubin and 10-year mortality risk in a Belgian population. Cancer Causes Control 2001;12:887-94. [Crossref] [PubMed]
32. Kawanishi S, Wang G, Ma N, et al. Cancer Development and Progression Through a Vicious Cycle of DNA Damage and Inflammation. Int J Mol Sci 2025;26:3352. [Crossref] [PubMed]
33. Vítek L. Bilirubin as a signaling molecule. Med Res Rev 2020;40:1335-51. [Crossref] [PubMed]