This retrospective cohort study included cases admitted to a tertiary general hospital in Guiyang from 31 October 2019, to 31 October 2024. All patients were prospectively followed until 31 October 2024. A total of 100 patients with concurrent lung cancer and pulmonary tuberculosis were ultimately enrolled. We acknowledge that including patients across all stages (I-IV) introduces population heterogeneity. However, this approach reflects the real-world clinical scenario where prognostic tools are needed across different stages, particularly in resource-limited settings managing the complex diagnosis of concurrent lung cancer and tuberculosis, where traditional staging may have diminished predictive value due to confounding inflammatory responses. Inclusion criteria were as follows:1. Age between 18 and 85 years; 2. Histologically confirmed lung cancer with TNM stage I–IV; 3. Newly diagnosed pulmonary tuberculosis, as defined by WHO guidelines, with completion of a standard anti-tuberculosis regimen; 4. No previous anti-tumor therapy (including surgery, radiotherapy, chemotherapy, targeted therapy, or immunotherapy) prior to enrollment; 5. Completion of relevant hematological examinations before anti-tumor treatment; 6. Complete medical records. Exclusion criteria were:1. Concomitant malignancies in other organ systems; 2. Severe cardiac, renal, or hepatic failure; 3. Missing data for exposure or outcome variables. All data, including demographic information, laboratory results, and follow-up records, were extracted from electronic medical records. Data collection was conducted by trained researchers to ensure accuracy and consistency.

The exposure variables included the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII), lymphocyte-to-monocyte ratio (LMR), prognostic nutritional index (PNI), hemoglobin-albumin-lymphocyte-platelet score (HALP), hemoglobin-to-red cell distribution width ratio (HRR), and albumin-to-globulin ratio (ALB/GLB). These variables were measured based on blood samples collected at hospital admission. All variables were treated as continuous variables. For analytical purposes, certain variables were categorized according to clinical relevance or recommendations from the literature. The definitions and calculation methods for these variables strictly followed international standards.

The following formulas were used to calculate the exposure variables. NLR (Neutrophil-to-Lymphocyte Ratio) = Absolute Neutrophil Count (×10⁹/L)÷Absolute Lymphocyte Count (×10⁹/L). LMR (Lymphocyte-to-Monocyte Ratio) = Absolute Lymphocyte Count (×10⁹/L)÷Absolute Monocyte Count (×10⁹/L). PLR(Platelet-to-Lymphocyte Ratio) = Platelet Count (×10⁹/L)÷Absolute Lymphocyte Count (×10⁹/L). SII (Systemic Immune-Inflammation Index) = Platelet Count (×10⁹/L) × Absolute Neutrophil Count (×10⁹/L)÷Absolute Lymphocyte Count (×10⁹/L). HALP (Hemoglobin-Albumin-Lymphocyte-Platelet Score) = [Hemoglobin (g/L) × Albumin (g/L) × Absolute Lymphocyte Count (×10⁹/L)]÷Platelet Count (×10⁹/L). PNI (Prognostic Nutritional Index) = Serum Albumin (g/L)+[5×Absolute Lymphocyte Count (×10⁹/L)]. HRR(Hemoglobin-to-Red Cell Distribution Width Ratio) = Hemoglobin (g/L)÷Red Cell Distribution Width (%). ALB/GLB (Albumin-to-Globulin Ratio) = Serum Albumin (g/L)÷Serum Globulin (g/L). All units and calculation methods strictly adhere to international standards.

The primary outcome was overall survival (OS), defined as the interval from the date of diagnosis to either the date of death or the last follow-up. Outcome status was determined based on follow-up records, including confirmation of death or the last known date alive. All survival outcomes were independently assessed by researchers blinded to study exposure, ensuring objectivity and consistency of results.

Covariates included age (≥60 or <60 years), sex (male or female), body mass index (BMI; ≥25.5 or <25.5), smoking history (yes or no), pathological type (adenocarcinoma, squamous cell carcinoma, small cell carcinoma, or others), lymph node metastasis (yes or no), distant metastasis (yes or no), clinical stage (I–IV), differentiation grade (poor or moderate/well), ECOG performance status (0–1 or ≥2), diabetes (yes or no), hypertension (yes or no), and HIV infection (yes or no). Selection of these covariates was based on prognostic factors identified in previous literature. All variables were extracted from patients’ electronic medical records.

This study was approved by the Ethics Committee (Approval No: Guiyang Public Health Treatment Center 2025–07). As a retrospective cohort study, all patient data were anonymized, and the requirement for informed consent was waived. The research adhered to the Declaration of Helsinki and relevant ethical guidelines to ensure the protection of patient privacy. All data were used solely for scientific research purposes, and the study team is committed to not using patient information for any purposes unrelated to this research. The data of this study are available from the corresponding author upon reasonable request.

The distribution of continuous variables was assessed using the Shapiro-Wilk test. Variables with a normal distribution are presented as mean ± standard deviation, while non-normally distributed data are reported as median (interquartile range). Categorical variables are summarized as counts and percentages. Between-group differences were evaluated using the chi-squared test for categorical variables, the independent-samples t-test for normally distributed continuous variables, and the Mann-Whitney U test for non-normally distributed continuous variables. Survival curves were generated via the Kaplan-Meier method, with differences between groups compared by the log-rank test. The Cox proportional hazards regression model was employed to assess associations between exposure variables and overall survival (OS). Three hierarchical models were constructed: Model 1 was unadjusted; Model 2 adjusted for demographic factors; Model 3 further adjusted for clinicopathological variables. Variable selection was performed via least absolute shrinkage and selection operator (LASSO) regression. The optimal regularization parameter (λ) was determined using ten-fold cross-validation, applying the one-standard-error rule to enhance model generalizability and minimize the partial likelihood deviance, thereby reducing multicollinearity and overfitting risk. A nomogram was developed based on the selected variables to predict 1, 3, and 5years survival probabilities. Model performance was evaluated by time-dependent receiver operating characteristic (ROC) curves (area under the curve, AUC) and concordance index (C-index). Calibration was assessed by calibration curves using 1,000 bootstrap resamples to compare predicted and observed outcomes. The PowerSurvEpi package was used to calculate statistical power for the Cox regression analyses. Sensitivity analyses, including outlier handling, variable stability, interaction assessment, and threshold adjustment, were conducted to ensure data robustness. All statistical tests were two-sided, and significance was set at P < 0.05. Multiple comparisons were adjusted using Bonferroni correction. All statistical analyses were performed using R software (version 4.3.0).

The overall research design and analytical workflow of this study are illustrated in Figure 1.

A nomogram was developed based on six optimal variables—PNI, HRR, diabetes, NLR, RDW and ECOG PS—to predict 1, 3, and 5 years survival rates in patients with lung cancer and concurrent pulmonary tuberculosis. The predictors retained in the final model (diabetes, RDW, NLR, PNI, and HRR), although they did not achieve statistical significance (p > 0.05), were retained based on their contribution to overall model performance and their potential value in clinical practice for inflammation-nutrition assessment. Within the nomogram, PNI values of 38–50 and HRR values of 2.4–3.6 were associated with the most favorable prognosis. An NLR greater than 8 indicated a poorer prognosis. RDW values of 30–65 demonstrated a complex, nonlinear relationship with outcomes. Diabetes was significantly associated with prognosis. ECOG PS emerged as the strongest independent predictor of adverse outcomes. The nomogram demonstrated good discriminatory ability (C-index = 0.71; 95% CI:0.66–0.76). See Figure 6.

Based on the strongest predictor, ECOG PS, Kaplan-Meier survival curves were used to compare OS between the high-risk group (ECOG PS ≥ 2) and the low-risk group (ECOG PS 0–1). The results demonstrated that patients in the high-risk group (ECOG PS ≥ 2) had significantly poorer outcomes (log-rank test, p = 0.0025), especially after 2 years of follow-up. By the end of the 5-year observation period, the survival probability in the high-risk group approached zero, while a small proportion of patients in the low-risk group remained alive. See Figure 7.

Receiver operating characteristic (ROC) curves were used to evaluate the predictive performance of the model at 1, 3, and 5 years. The model demonstrated consistent, moderate predictive ability across all follow-up intervals. The area under the curve (AUC) was 0.656 at 1 year, 0.693 at 3 years, and 0.689 at 5 years, with the best performance observed at the 3-year follow-up (AUC = 0.693). See Figure 8.

Calibration curves were validated using 1,000 bootstrap resamples to reduce overfitting bias. The predicted probabilities showed excellent agreement with the observed outcomes, and the model demonstrated good calibration at all three time points (bootstrap P > 0.05). See Figure 9.

The statistical power of this study, assessed using the PowerSurvEpi package (significance level α = 0.05), was 83%. This result demonstrates that the study design is adequate to support the predictive value of ECOG PS. See Table 3.

Sensitivity analyses revealed several key findings. First, After excluding extreme outliers (NLR >20, n = 1) and HIV-positive cases (n = 1), the model was reconstructed, the C-index decreased slightly from 0.71 to 0.70, with no statistically significant difference, indicating stable model performance. Second, a significant synergistic effect was observed for patients with both ECOG PS ≥ 2 and NLR> 8 (95% CI: 1.12–4.08, p = 0.032), underscoring the importance of functional assessment in highly inflamed, comorbid populations. Third, the NLR> 8 threshold demonstrated the most robust performance in the sensitivity analyses. See Table 4.

This retrospective cohort study investigated the association between eight inflammation-nutrition biomarkers (NLR, PLR, SII, LMR, PNI, HALP, HRR and ALB/GLB) and OS in patients with lung cancer complicated by tuberculosis. A multi-stage modeling approach yielded the following results: 1. LASSO regression identified six optimal variables PNI, HRR, diabetes, NLR, RDW, and ECOG PS for constructing a nomogram model. This model demonstrated robust predictive performance (C-index = 0.71; 95% CI:0.66–0.76). 2. Cox regression revealed that PNI values between 38 and 50 and HRR values between 2.4 and 3.6 were associated with the best prognosis. An NLR>8 indicated worse outcomes, and RDW (30–65) showed a non-linear relationship with prognosis. Diabetes was correlated with survival, and ECOG PS emerged as an independent prognostic risk factor. 3. Kaplan-Meier survival analysis stratified by ECOG PS indicated that patients in the high-risk group (ECOG PS ≥ 2) had significantly poorer outcomes (log-rank p = 0.0025), with a 76% higher risk of death (95% CI:1.02–3.02, p = 0.041). In the high-risk subgroup (ECOG PS ≥ 2 and NLR>8), the 5-year survival rate was nearly zero. 4. The area under the ROC curve (AUC) for 1, 3, and 5 years survival predictions was 0.656, 0.693, and 0.689, respectively, with the model performing best at 3 years (AUC = 0.693). 5. Calibration curves validated with 1,000 bootstrap resamples showed good agreement between predicted and observed outcomes at all three key time points (bootstrap P > 0.05). 6. Power analysis using the PowerSurvEpi package indicated a statistical power of 83% (significance level = 0.05), supporting the adequacy of the study design and the predictive value of ECOG PS. 7. A series of sensitivity analyses were conducted, including outlier handling, variable stability assessment, evaluation of interactions, and determination of optimal cutoff values. In particular, the HIV-positive case (n = 1) was excluded due to insufficient statistical power. These analyses confirmed the robust performance of the model. Notably, the synergistic interaction between ECOG PS and NLR was significant (95% CI:1.12–4.08, p = 0.032), and NLR>8 proved to be the most stable threshold, consistent with previous reports (Cupp et al., 2020). These findings highlight the need for careful assessment of physical function in highly inflamed comorbid populations. This study is the first to systematically integrate multiple inflammatory, immune, and nutritional biomarkers for prognostic evaluation in patients with both lung cancer and tuberculosis, providing a novel biomarker panel for individualized prognosis in this unique patient population. Our study revealed a synergistic effect between ECOG PS and the NLR. Poor ECOG PS suggests worsening malnutrition, chronic inflammation, and immune dysfunction, which may elevate NLR levels. Elevated NLR further reflects tumor-associated inflammation and immune evasion. This synergy indicates diminished host antitumor capacity and a pro-inflammatory environment, both of which are associated with poorer prognosis.

This study demonstrated significant consistency with the findings of Chen et al. in metastatic colorectal cancer (Chen et al., 2019). Specifically, each one-unit increase in NLR was associated with a 7% increase in mortality risk in our cohort, compared to 5% in their study. The threshold for adverse prognosis in our study was NLR>8, higher than the NLR>3 used in Chen et al., suggesting that the prognostic value of inflammatory imbalance may be applicable across different cancer types. Notably, in the context of tuberculosis comorbidity, the discriminatory power of PNI (95% CI:0.91–0.99) in our cohort was markedly reduced compared to the hepatocellular carcinoma cohort reported by Pan et al. (2024) (95% CI: 0.76–0.89). This difference may be attributed to immune-metabolic disturbances unique to Mycobacterium tuberculosis infection. Mechanistic studies have shown that mycobacterial membrane proteins activate the NF-κB signaling pathway via TLR4, resulting in a 3.5-fold increase in IL-6 transcription. Persistent IL-6 stimulation downregulates albumin mRNA expression in hepatocytes and increases Fas receptor expression in lymphocytes, leading to higher rates of apoptosis. These changes ultimately manifest as increased NLR and decreased PNI (O'Garra et al., 2013), reducing the protective impact of nutrition-immune indices. The association between diabetes and poor prognosis observed in our study is consistent with the findings of Donath MY. Hyperglycemia (HbA1c>7%) activates the NLRP3 inflammasome, leading to increased IL-6 secretion from monocytes, reduced efficiency of serum albumin synthesis, and a positive correlation between the homeostasis model assessment of insulin resistance (HOMA-IR) and NLR. This forms a vicious cycle of hyperglycemia, inflammation, and malnutrition (Donath and Shoelson, 2011). Methodologically, although Mazzella et al. (2024) used the LASSO-Cox model in their study of non-small cell lung cancer, the absence of adjustment for tuberculosis infection and the selective inclusion of only NLR and PLR as inflammatory markers resulted in a lower model discrimination (C-index = 0.68) compared to our study (C-index = 0.71). While Yang et al. (2019) confirmed the prognostic value of inflammation markers in a multicenter study, they did not account for the immunosuppressive effects mediated by CD163+M2 macrophages via the PD-L1/IL-10 axis in the tuberculosis microenvironment (Quail and Joyce, 2013), potentially biasing PLR-specific assessments. Importantly, our study is the first to construct a multidimensional predictive model using LASSO-selected variables in patients with comorbid lung cancer and tuberculosis. The high-risk characteristic of ECOG PS ≥ 2 (HR = 1.76) identified in our cohort was notably higher than that reported in the immunotherapy cohort by Ito et al. (2024) (HR = 1.32). This suggests that tuberculosis-associated systemic inflammation may exacerbate organ dysfunction through a TNF-α/IL-1β positive feedback loop, providing new insights for precise interventions in comorbid populations. These differences and mechanistic explanations enhance our understanding of the prognostic roles of inflammation and nutrition in patients with both lung cancer and tuberculosis.

Based on our findings, this is the first study to construct an inflammation-nutrition prognostic multi-parameter model (INMPM) in lung cancer tuberculosis comorbidity. Through the systematic integration of eight inflammation; nutrition biomarkers and optimal variable selection via LASSO regression, the C-index of our model was significantly improved. Methodologically, our study fills an important gap in the field of prognostic evaluation for lung cancer tuberculosis comorbidity and provides practical evidence support for the implementation of targeted nutrition intervention. Our study demonstrated the following main findings: 1. ECOG PS is an independent prognostic risk factor in lung cancer patients with tuberculosis comorbidity. 2. In clinical practice, targeted evaluation of physical function is necessary for all comorbid patients with high inflammation; it is particularly suggested that clinicians should prioritize nutritional intervention, and if necessary, combined anti-infective therapy for high-risk subgroups of comorbidity patients with ECOG PS ≥ 2 and NLR>8. 3. The model exhibited high statistical power (83%) and strong robustness; the model’s prediction accuracy for 3-year OS in comorbidity patients was excellent. These findings could provide clinicians with a quantitative reference for therapeutic decision-making in lung cancer tuberculosis comorbidity, especially for evaluation of candidacy for aggressive anti-tumor therapy and timing of nutrition intervention.

This study exhibited the following methodological innovations and advantages:First, it was the first to clarify the synergistic mechanism between inflammation-nutrition imbalance and lung cancer prognosis in tuberculosis comorbidity. Second, the data analysis avoided the limitations of conventional conventional single-stage modeling and adopted a multi-stage approach. In this approach, LASSO regression was used for variable selection and multivariate Cox regression was used to construct the model. This strategy effectively alleviated the influence of heterogeneous comorbidity and substantially improved the performance of the model (C-index = 0.71 vs 0.63–0.65 for traditional models). The predictive performance was further validated using the calibration curves based on 1,000 bootstrap resamples. The result showed that the prediction error rate was less than 5% at 1, 3, and 5 years time points. Statistical power analysis based on PowerSurvEpi function exhibited that the Cox regression had 83% power (significance level = 0.05) and the sample design was appropriate. Sensitivity analyses demonstrated the robustness of the model. These methods provided clinicians with a highly reliable and quantitative decision-making tool. Finally, this study provided precise nutritional intervention criteria for high-risk subgroups (ECOG PS ≥ 2 and NLR>8) and suggested that combined anti-infective therapy should be adopted when necessary, which facilitated individualized treatment.

This study has several limitations. First, due to the unique characteristics of the study population and resource limitations, this study was conducted as a single-center retrospective analysis with a relatively small sample size and without external validation using an independent cohort. Although a power analysis confirmed the adequacy of the sample size, validation of interaction effects requires larger cohorts, and the retrospective design is prone to limitations in causal inference. Future multicenter studies are needed to enhance the generalizability of our conclusions. Second, the study population was primarily from Guizhou Province, China. Their unique genetic background and tuberculosis exposure characteristics may influence the prognostic thresholds of inflammation-nutrition biomarkers. Thus, the generalizability of the model should be validated in cohorts from different ethnic and geographic backgrounds. Finally, due to the limitations of retrospective data, several key variables [such as adherence to anti-tuberculosis therapy, PD-L1 status, and LIPI score (De Giglio et al., 2024)] could not be comprehensively calculated and were not included in the analysis owing to missing data on specific treatment regimens, therapeutic responses, and baseline LDH values. Future studies should expand sample sizes, validate in stage-specific cohorts, and incorporate additional clinical interventional factors and biomarkers to verify the generalizability of these interactions.