Combined laboratory and imaging indicators to construct risk models for predicting immunotherapy efficacy and prognosis in non-small cell lung cancer: An observational study (STROBE compliant)

Abstract

This study aimed to investigate the correlations between short- and long-term efficacy of immune checkpoint inhibitors (ICIs) and pretreatment laboratory/imaging parameters in advanced non-small cell lung cancer (NSCLC), and to construct risk prediction models. We enrolled 137 NSCLC patients with stage IIIB-IV disease who completed 4 cycles of PD-1/PD-L1 inhibitor monotherapy or combination therapy. All participants underwent pretreatment laboratory assessments encompassing inflammatory markers, lymphocyte subsets, tumor biomarkers, coagulation profiles, and contrast-enhanced computed tomography (CE-CT) scans. The primary endpoints were objective response rate (ORR) and overall survival (OS), with progression-free survival (PFS) as the secondary endpoint. Univariate and multivariate logistic regression analyses were performed to identify significant predictors of short-term treatment response and develop an efficacy prediction model. For long-term outcomes, univariate and multivariate Cox proportional hazards regression analyses were conducted to establish a prognostic risk model. The final models were presented as nomograms and validated through receiver operating characteristic (ROC) curve analysis, calibration curves, and decision curve analysis (DCA). CD4⁺ T-cell count (P = .007), fibrinogen (FIB, P = .047), and mediastinal lymph node enlargement (P = .028) emerged as independent predictors of ORR. The prediction model demonstrated an area under the ROC curve (AUC) of 0.838, with bootstrap validation (1000 resamples) yielding a mean AUC of 0.867. Calibration analysis, DCA, and clinical impact curve (CIC) collectively confirmed the model’s robust predictive performance. For OS, metastatic site (P = .007), neutrophil-to-lymphocyte ratio (NLR, P = .025), carbohydrate antigen 125 (CA125, P = .020), cytokeratin 19 fragment (CYFRA 21-1, P = .004), FIB (P < .001), and pleural effusion (P < .001) were identified as significant prognostic determinants. The model achieved AUC values of 0.858 and 0.860 for 1- and 2-year survival prediction, respectively. Calibration plots revealed excellent concordance between predicted and observed survival probabilities at both timepoints. Furthermore, DCA indicated superior net clinical benefit of the prognostic model compared to random chance models across threshold probability ranges. Comprehensive prediction models integrating clinical characteristics, laboratory biomarkers, and imaging parameters were developed for both short- and long-term efficacy evaluation of immunotherapy, offering clinically actionable guidance for personalizing treatment strategies in advanced NSCLC.

1. Introduction

Lung cancer maintains its position as one of the most prevalent malignancies in terms of both global incidence and mortality. Non-small cell lung cancer (NSCLC), encompassing histological subtypes such as adenocarcinoma and squamous cell carcinoma, constitutes approximately 85% of all lung cancer cases.^[1] Despite widespread implementation of early detection programs, the majority of patients present with advanced-stage disease at initial diagnosis, resulting in missed opportunities for optimal surgical intervention and consequently adversely affecting prognosis and survival outcomes. Current therapeutic strategies for advanced NSCLC primarily involve chemotherapy, radiotherapy, targeted therapy, and immunotherapy, with immune checkpoint inhibitors (ICIs) demonstrating particular clinical efficacy.^[2] These agents function by blocking negative regulatory molecules expressed on T lymphocytes, thereby counteracting tumor-immune evasion mechanisms and potentiating antitumor immune responses, ultimately improving patient survival.^[3] Clinical trial evidence consistently demonstrates superior efficacy of ICIs compared to platinum-based chemotherapy regimens when administered as first- or second-line treatment for advanced NSCLC.^[4] However, only 20%–40% of patients derive durable clinical benefit from ICI therapy,^[5] underscoring the critical need for reliable biomarkers to predict treatment response and monitor clinical outcomes.

Currently recognized biomarkers for immunotherapy response encompass tumor mutation burden (TMB), microsatellite instability (MSI), and programmed cell death protein 1/ligand 1 (PD-1/PD-L1) expression.^[6] TMB, quantified as the total somatic mutations per megabase of tumor genomic coding regions (Mut/Mb), has emerged as a promising predictor of ICI therapeutic outcomes.^[7] MSI – a pivotal pan-cancer biomarker characterized by insertions/deletions in short tandem repeat sequences – was initially identified in colorectal carcinoma and subsequently associated with lung cancer pathogenesis.^[8] In 2017, pembrolizumab received US Food and Drug Administration (FDA) approval for microsatellite instability-high (MSI-H) or mismatch repair-deficient (dMMR) unresectable/metastatic solid tumors.^[9] However, the prevalence of MSI-H in lung cancer remains exceptionally low (0.4%–0.8%),^[10] while technical limitations including technically demanding procedures and prolonged turnaround times hinder the clinical adoption of TMB/MSI testing for routine immunotherapy monitoring.^[11] Although PD-L1 expression retains its status as the most validated predictive biomarker, with the 2019 NCCN Guidelines endorsing PD-L1 immunohistochemical testing as a category 1 recommendation for advanced NSCLC, its standalone predictive capacity exhibits well-documented limitations. The KEYNOTE-407 trial demonstrated significantly improved overall survival (OS) and progression-free survival (PFS) with pembrolizumab-chemotherapy combination versus chemotherapy alone in advanced squamous NSCLC patients, irrespective of PD-L1 expression levels.^[12] Given the imperfect predictive performance and associated costs of current approaches, developing clinically applicable and cost-efficient biomarkers remains an important focus in clinical practice.

Laboratory testing remains the most clinically utilized assessment modality due to its inherent advantages of simplicity, cost-effectiveness, and rapid turnaround. Routine laboratory evaluation in cancer patients typically encompasses multiparameter diagnostic panels including systemic inflammatory biomarkers, lymphocyte subpopulation profiles, tumor-associated antigens, and coagulation parameters. 30 a fundamental component of the tumor microenvironment, driving oncogenesis through angiogenesis promotion and cellular proliferation, with direct implications for therapeutic response and clinical outcomes across malignancies.^[13,14] Established inflammatory indices such as neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), lymphocyte-to-monocyte ratio (LMR), and systemic immune-inflammation index (SII) serve as valuable tools for differential diagnosis and prognostic stratification.^[15,16] However, the clinical utility of these inflammatory indices for assessing immunotherapy response remains controversial.^[17,18] Immune cells serve as central mediators of antitumor immunity, where the magnitude and quality of immune responses depend critically on cellular composition, activation status, and functional polarization. Specifically, T lymphocytes recognize tumor-associated antigens through their T-cell receptors, with activated CD8⁺ cytotoxic T cells executing direct tumoricidal activity via granzyme/perforin pathways, while CD4⁺ helper T cells coordinate adaptive immune responses through cytokine signaling.^[19] Consequently, peripheral blood lymphocyte subpopulations – including CD4⁺ T cells, CD8⁺ T cells, B cells, and natural killer (NK) cells – may serve as dynamic biomarkers for monitoring immune status and predicting therapeutic outcomes.^[20] Tumor-derived biomarkers, measurable in serum or other biofluids through minimally invasive assays, provide clinically accessible indicators of tumor burden and biological aggressiveness. Their operational simplicity and low procedural risk underpin widespread applications in cancer screening, diagnostic workups, treatment monitoring, and prognosis. However, the optimal selection of tumor markers for predicting immunotherapy response and survival outcomes requires further investigation.^[21] Emerging evidence indicates that procoagulant factors produced during lung cancer pathogenesis induce a hypercoagulable state, facilitating tumor dissemination and metastatic progression.^[22] This coagulopathy further elevates thrombotic risk, particularly venous thromboembolic events. These pathophysiological interactions collectively contribute to cancer-related mortality, highlighting clinically significant associations between coagulation parameter abnormalities and survival outcomes in lung cancer populations.^[23]

Imaging modalities constitute essential components of diagnostic evaluation and therapeutic monitoring in oncology. Contrast-enhanced computed tomography (CE-CT), involving intravenous administration of iodinated contrast media, enhances tissue contrast between lesions and surrounding structures. This technique improves delineation of lesion morphology and dimensions while detecting occult lymph node metastases or subcentimeter lesions often undetectable on noncontrast CT.^[24] CE-CT captures clinically relevant imaging biomarkers such as lobulation, spiculation, mediastinal lymph node enlargement (MLE), and pleural effusion (PE). The predictive value of these radiologic characteristics for treatment response and survival outcomes remains unexplored.

This study systematically evaluated the association between multidimensional parameters (clinical characteristics, laboratory biomarkers, and radiologic features) and therapeutic outcomes of PD-1/PD-L1 inhibitors in advanced NSCLC, with the dual objectives of identifying robust predictive biomarkers and developing risk stratification models for treatment response and survival. The established predictive framework may offer an evidence-based approach for optimizing immunotherapy efficacy assessment and prognostic evaluation in this patient population.

2. Patients and methods

2.1. Patients

A total of 2486 patients with NSCLC were consecutively enrolled from the Affiliated Hospital of Chengde Medical University between September 2019 and September 2023. Inclusion criteria: Age ≥ 18 years. Histologically or cytologically confirmed NSCLC. Stage IIIB and IV disease (AJCC 8th edition TNM staging). Eastern Cooperative Oncology Group (ECOG) performance status 0 to 2 with anticipated survival exceeding 3 months. At least one measurable tumor lesion meeting immune Response Evaluation Criteria in Solid Tumors (iRECIST). Completion of ≥ 4 cycles (21-day intervals) of PD-1/PD-L1 inhibitor therapy. Exclusion criteria: History of other malignancies. Concurrent autoimmune disorders. Documented EGFR, ALK, KRAS, or MET driver gene mutations. Insufficient clinical documentation for efficacy evaluation. Nonadherence to treatment protocols or follow-up requirements. After applying inclusion/exclusion criteria, 137 patients were ultimately included. All participants provided written informed consent, and this study received ethical approval from the Ethics Committee of Chengde Medical University Affiliated Hospital (Approval No.: CYFYLL2021204).

2.2. Clinical laboratory testing and calculating

All participants received standardized precollection instructions requiring fasting for at least 8 hours with medication restriction. Four fasting venous blood samples (3 mL each) were collected during the 7-day pretreatment period preceding immunotherapy initiation.

One venous blood sample was homogenized by gentle inversion for 3 minutes. Absolute counts of neutrophils, lymphocytes, monocytes, and platelets were quantified using the Sysmex XN-9000 automated hematology analyzer (Sysmex Corporation, Kobe, Japan). Inflammatory indices were derived using the following formulae: NLR = (Neutrophil count)/ (Lymphocyte count), PLR = (Platelet count)/ (Lymphocyte count), LMR = (Lymphocyte count)/ (Monocyte count), SII = (Platelet count × Neutrophil count)/ (Lymphocyte count).

Heparin-anticoagulated plasma (50 μL) was mixed with fluorochrome-conjugated monoclonal antibodies (5 μL; Beckman Coulter, Brea) and transferred to the base of collection tubes. The mixture was incubated at room temperature under light-protected conditions for 30 minutes. Subsequently, 250 μL of hemolysin was added to each tube, followed by an additional 30-minute incubation under identical conditions. Cellular components were then analyzed using a NAVIOS EX 10-color/3-laser flow cytometer (Beckman Coulter, Fullerton). Lymphocyte subpopulations (CD4⁺ T cells, CD8⁺ T cells, B cells, NK cells) were quantified through standardized gating strategies implemented in Beckman Coulter CXP Software (v2.1), with CD4⁺/CD8⁺ ratio calculation based on absolute counts.

Levels of carcinoembryonic antigen (CEA), carbohydrate antigen 125 (CA125), squamous cell carcinoma antigen (SCC), cytokeratin 19 fragment (CYFRA 21-1), neuron-specific enolase, and pro-gastrin releasing peptide (PROGRP) were measured via electrochemiluminescence immunoassay using the Roche 601 automated electrochemical luminescence immunoassay analyzer (Roche, Basel, Switzerland).

A subsequent 3 mL aliquot of venous blood was centrifuged to obtain platelet-poor plasma. Coagulation parameters including D-dimer (D-D), fibrinogen (FIB), thrombin time (TT), prothrombin time activity, and activated partial thromboplastin time were quantified using immunoturbidimetric assays and mechanical clot detection methods on a Sysmex CS5100 coagulation analyzer (Siemens Healthineers, Erlangen, Germany).

2.3. CE-CT scans

CE-CT examinations were performed on a dual-source CT scanner (SOMATOM Drive, Siemens Healthineers, Munich, Germany) with standardized acquisition parameters: 120 kV tube voltage, 180 mAs current-time product, and 3 to 5 mm slice thickness. The scan coverage extended from the pulmonary apex to diaphragmatic surface. Intravenous administration of iohexol (80–100 mL, 350 mgI/mL) was delivered at 3 mL/s flow rate using a dual-chamber power injector. Bolus-triggered arterial phase (30 seconds postinjection) and delayed venous phase (60 seconds postinjection) acquisitions were obtained. All DICOM images were archived in a picture archiving and communication system (PACS) for blinded analysis by 2 board-certified radiologists (>10 years’ experience in thoracic imaging). The imaging indicators of lesions included maximum tumor diameter (baseline tumor burden), lobulation sign, spiculation sign, vessel convergence sign, ground-glass opacity (GGO), MLE (short-axis diameter > 1 cm), pleural thickening, pleural indentation, enhancement degree, pneumonia, and PE.

2.4. Curative effect evaluation and follow-up

Follow-up data were collected through electronic medical record review and structured telephone interviews. All participants underwent multiregional CE-CT (neck/chest/abdominal/pelvic) and cranial MRI surveillance at 2-cycle intervals during treatment. Posttreatment monitoring followed this schedule: quarterly assessments during year 1, biannual evaluations through years 2 to 3, and annual follow-up thereafter until disease progression, death, or study withdrawal. The study endpoint was September 30, 2023. Treatment response was evaluated after 4 therapy cycles using iRECIST criteria: Immune complete response (iCR), Immune partial response (iPR), Immune stable disease (iSD), Immune unconfirmed progressive disease (iUPD), Immune confirmed progressive disease (iCPD). Progression confirmation required ≥ 4-week interval between initial iUPD designation and subsequent imaging validation. Clinical outcomes comprised: Short-term efficacy: Objective response rate (ORR) = (iCR + iPR)/total number of cases, Long-term efficacy: PFS: Time interval from study enrollment to first radiologically confirmed disease progression. OS: Time interval from histopathological diagnosis to death from any cause.

2.5. Statistical methods

Statistical analysis was performed using SPSS 27.0 software (SPSS Inc., Chicago) or R software (R version 4.3.2). Laboratory parameters, as continuous variables, were described using mean ± standard deviation for normally distributed data and median with interquartile range (IQR) for non-normally distributed data, as verified by Shapiro–Wilk tests (α = .05). Clinical characteristics and imaging metrics, being categorical variables, were presented as proportions (percentage).

For analyzing correlations between short-term clinical efficacy (ORR) and clinical/laboratory/imaging indicators, univariable logistic regression was employed. Variables demonstrating statistical significance (P < .05) were subsequently incorporated into multivariable logistic regression models to calculate adjusted odds ratios (OR) with 95% confidence intervals (CI). Regarding long-term clinical outcomes (PFS and OS), univariable Cox proportional hazards regression analyses were performed. Statistically significant predictors (P < .05) were advanced to multivariable Cox regression models, with results expressed as hazard ratios (HR) and 95% CI.

Dichotomous variables including ECOG performance status, BMI, PD-L1 TPS, metastatic site classification, and maximum tumor diameter were stratified using evidence-based cutoff values derived from prior landmark studies.^[25–29] All analyses adhered to STROBE guidelines for observational research reporting. The predictive nomogram construction, receiver operating characteristic curve (ROC) generation (AUC analysis), bootstrap validation (1000 iterations), decision curve analysis (DCA), and clinical impact curve (CIC) analyses were implemented in R software using the following packages: nomogramFormula, pROC/timeROC, bootsrap, rmda, and ggDCA. Statistical significance was defined as two-tailed P < .05.

3. Results

3.1. Clinical characteristics of patients

Baseline clinical characteristics are presented in Table 1. Following 4 treatment cycles, iRECIST responses comprised: 0 iCR, 40 iPR (29.2%), 66 iSD (48.2%), 0 iUPD, and 31 iCPD (22.6%). With median follow-up of 21.8 months (95% CI = 18.9–24.2), 59 deaths (43.1%) occurred. Median PFS was 6.8 months (95% CI = 6.4–7.2) and OS 24.1 months (95% CI = 20.2–26.5).

Table 1.

Clinical characteristics of 137 advanced NSCLC patients.

Characteristics	n (%)	Characteristics	n (%)
Sex		Smoking history
Male	96 (70.1%)	No	55 (40.1%)
Female	41 (29.9%)	Yes	82 (59.9%)
Age		Previous history
<63	62 (45.3%)	No	86 (62.8%)
≧63	75 (54.7%)	Yes	51 (37.2%)
ECOG score		Family history
0–1	107 (78.1%)	No	115 (83.9%)
≧2	30 (21.9%)	Yes	22 (16.1%)
BMI		Operation history
<24	70 (51.1%)	No	121 (88.3%)
≧24	67 (48.9%)	Yes	16 (11.7%)
Histopathology		Metastatic site
Adenocarcinoma	76 (55.5%)	0–1	86 (62.8%)
Squamous carcinoma	56 (40.9%)	≧2	51 (37.2%)
Others	5 (3.6%)	Radiotherapy
TNM stage		No	83 (60.6%)
IIIB–IIIC	56 (40.9%)	Yes	54 (39.4%)
IV	81 (59.1%)	Treatment
PD-L1 TPS		First-line treatment	76 (55.5%)
<50%/NA	114 (83.2%)	second-line treatment	61 (44.5%)
≧50%	23 (16.8%)	Outcome
Clinical efficacy		PFS	6.8
iCR	0 (0%)	OS	24.1
iPR	40 (29.2%)
iSD	66 (48.2%)
iUPD	0 (0%)
iCPD	31 (22.6%)

BMI = body mass index, ECOG = Eastern Cooperative Oncology Group, iCPD = immune confirmed progressive disease, iCR = immune complete response, iPR = immune partial response, iSD = immune stable disease, iUPD = immune unconfirmed progressive disease, NSCLC = non-small cell lung cancer, OS = overall survival, PD-L1 = programmed death-ligand 1, PFS = progression-free survival, TNM = tumor-node-metastasis.

3.2. Correlations of ORR with clinical characteristics, laboratory and imaging indicators

Laboratory parameters and imaging characteristics of the 137 advanced NSCLC patients are detailed in Tables S1 and S2 (Supplemental Digital Content, https://links.lww.com/MD/Q347), respectively. Univariate logistic regression analyses assessed associations between ORR and baseline clinical parameters, laboratory biomarkers, and imaging-derived metrics. Univariate logistic regression identified age (P = .032), TNM stage (P = .033), and radiotherapy combination (P = .006) as significant clinical predictors of ORR. Inflammatory indices demonstrated strong associations: NLR (P = .004), PLR (P = .002), and SII (P = .001). Lymphocyte subpopulation analysis demonstrated significant associations with CD4⁺ T cells (P < .001). Elevated CYFRA 21-1 (P = .001) and coagulation markers D-D (P = .027)/FIB (P < .001) also correlated with reduced response. Imaging predictors included MLE (P = .001), GGO (P = .011), and PE (P = .023) (Fig. 1A–C). Multivariate analysis confirmed 3 independent determinants: CD4⁺ T cells (P = .007), FIB (P = .047), and MLE (P = .028) (Fig. 1D). In this study, representative contrast-enhanced CT features from selected patients are illustrated in Figure 2.

Figure 1.

Correlations of ORR with clinical indicators. (A) Univariable logistic regression analysis of clinical characteristics. (B) Univariable logistic regression analysis of laboratory indicators. (C) Univariable logistic regression analysis of imaging indicators. (D) Multivariate logistic regression analysis of statistically significant clinical indicators. APTT = activated partial thromboplastin time, CA125 = carbohydrate antigen 125, CEA = carcinoembryonic antigen, CYFRA 21-1 = cytokeratin 19 fragment, D-D = D-dimer, ECOG = Eastern Cooperative Oncology Group, FIB = fibrinogen, GGO = ground-glass opacity, LMR = lymphocyte monocyte ratio, MLE = mediastinal lymph node enlargement, NLR = neutrophil-lymphocyte ratio, NSE = neuron-specific enolase, ORR = objective response rate, PE = pleural effusion, PLR = platelet lymphocyte ratio, PROGRP = pro-gastrin releasing peptide, PTA = prothrombin time activity, SCC = squamous cell carcinoma antigen, SII = systemic immune inflammation index, TT = thrombin time.

Figure 2.

CE-CT images with statistically significant imaging indicators in this study. (A) Full-size chest CE-CT image with MLE; short diameter of the lymph node is 12.56 mm (arrow). (B) Full-size Chest CE-CT image with GGO sign (arrow). (C) Full-size chest CE-CT image with bilateral PE (arrow). CE-CT= contrast-enhanced computed tomography, GGO = ground-glass opacity, MLE = mediastinal lymph node enlargement, PE pleural effusion.

3.3. Construction of risk model for predicting short-term efficacy and evaluation

The final multivariable logistic regression model coefficients were used to construct a predictive risk score. The prognostic index (PI) was calculated as: PI = β₀ + Σ(βᵢXᵢ). For this model: PI = −0.03 − (0.007 × CD4⁺ T cells) + (0.694 × FIB) + (1.501 × MLE).

The multivariable risk prediction model was operationalized through a prognostic nomogram constructed with R software rms, nomogramFormula, and forestmodel (Fig. 3A). For clinical implementation, individual patient profiles are translated into point scores for CD4⁺ T cells, FIB, and MLE through nomogram axis alignment. The summation of these variable-specific points yields a total prognostic score, which is subsequently mapped to the risk probability axis to quantify individualized progression risk following immunotherapy.

Figure 3.

Visualization and verification of risk model for predicting ORR. (A) Nomogram model for predicting ORR. (B) ROC curves for the risk model, CD4⁺ T cells, FIB, and MLE. (C) Internal validation of the nomogram model by internal bootstrap analyses with 1000 resamplings. (D) Calibration curve for evaluating the prediction accuracy of the model. (E) DCA for evaluating the practical application ability of the model. (F) CIC for evaluating the clinical value. FIB fibrinogen, MLE mediastinal lymph node enlargement. CIC = clinical impact curve, DCA = decision curve analysis, FIB = fibrinogen, MLE = mediastinal lymph node enlargement, ORR = objective response rate, ROC = receiver operating characteristic.

The predictive performance of the nomogram was systematically evaluated through ROC curve analysis using R software pROC and rmda. The composite model demonstrated an AUC of 0.838 (95% CI = 0.760–0.916) with an optimal cutoff value of 0.644 (Fig. 3B), significantly outperforming individual predictors (CD4⁺ T cells: AUC 0.707; FIB: 0.745; MLE: 0.665). Internal validation via 1000 bootstrap resamples yielded a mean AUC of 0.867, confirming robust discriminative capacity (Fig. 3C). Calibration analysis revealed close concordance between predicted and observed probabilities along the ideal 45-degree line, indicating superior predictive accuracy (Fig. 3D).

DCA quantified the clinical net benefit of the nomogram across threshold probabilities. The model demonstrated superior net benefit compared to “treat all” and “treat none” strategies when the threshold probability exceeded 8% (Fig. 3E). CIC analysis revealed increasing concordance between model-predicted positive cases and true-positive cases with rising thresholds. At threshold probabilities > 75%, model-predicted positive cases closely approximated true-positive counts (Fig. 3F), indicating robust clinical applicability. These validation metrics collectively support the nomogram’s utility for guiding immunotherapy decisions in advanced NSCLC.

3.4. Correlations of PFS with clinical characteristics, laboratory and imaging indicators

Univariate Cox regression identified significant PFS predictors across clinical, laboratory, and imaging domains (Fig. 4A–C). Clinical predictors included operation history (P = .033) and radiotherapy combination (P = .028). Laboratory predictors demonstrated strong associations: NLR (p﹤.001), PLR (P = .005), SII (p﹤.001), CD4⁺ T cells (P = .009), CEA (P = .005), CYFRA 21-1 (p﹤.001), and FIB (p﹤.001). Significant imaging predictors were MLE (P = .004), GGO (P = .001). Multivariate analysis confirmed 3 independent predictors: elevated SII (P = .016), CEA (P = .008), and FIB (P = .004) (Fig. 4D). Increased levels correlated with reduced PFS duration.

Figure 4.

Correlations of PFS with clinical indicators. (A) Univariable logistic regression analysis of clinical characteristics. (B) Univariable logistic regression analysis of laboratory indicators. (C) Univariable logistic regression analysis of imaging indicators. (D) Multivariate logistic regression analysis of statistically significant clinical indicators. PFS = progression-free survival.

3.5. Correlations of OS with clinical characteristics, laboratory and imaging indicators

Univariate Cox proportional hazards regression identified significant predictors of OS across clinical, laboratory, and imaging domains (Fig. 5A–C). Clinical parameters included ECOG score (P = .039) and metastatic site (P = .006). Laboratory biomarkers demonstrated robust associations: NLR (P < .001), PLR (P = .047), SII (P = .044), CD4⁺ T cells (P = .012), CA125 (P = .006), CYFRA 21-1 (P < .001), D-D (P = .001), and FIB (P < .001). Significant imaging predictors comprised GGO (P = .019), pneumonia (P = .004), and PE (P < .001). Significant univariate predictors were advanced to multivariate Cox regression analysis, which identified 6 independent prognostic determinants: metastases involving ≥ 2 sites (P = .007), elevated NLR (P = .025), increased CA125 (P = .020), higher CYFRA 21-1 (P = .004), hyperfibrinogenemia (P < .001), and radiologically confirmed PE (P < .001) (Fig. 5D). These factors demonstrated a cumulative association with reduced overall survival duration.

Figure 5.

Correlations of OS with clinical indicators. (A) Univariable logistic regression analysis of clinical characteristics. (B) Univariable logistic regression analysis of laboratory indicators. (C) Univariable logistic regression analysis of imaging indicators. (D) Multivariate logistic regression analysis of statistically significant clinical indicators. OS = overall survival.

3.6. Construction of risk models for predicting long-term efficacy OS and evaluation

The final multivariable Cox regression coefficients were utilized to construct a prognostic scoring system. The PI was derived as: PI = Σ(βᵢXᵢ) = 0.648 × (metastatic sites) + 0.397 × NLR + 0.002 × CA125 + 0.068 × CYFRA 21-1 + 0.349 × FIB + 1.130 × PE.

The multivariable Cox regression model was translated into a clinically applicable prognostic tool through a nomogram constructed using R software rms, nomogramFormula, and regplot (Fig. 6A and B). This instrument enables clinicians to quantify individual risk profiles by assigning points based on 6 key parameters: metastatic site, NLR value, CA125 level, CYFRA 21-1 concentration, FIB level, and PE. The summation of these weighted scores projects onto 2 clinically critical endpoints: 1-year survival probability and median overall survival duration. For example, a 62-year-old male with advanced NSCLC exhibiting: Multiorgan metastasis (3 sites), NLR 3.2, CA125 25 U/Ml, CYFRA 21-1 13 ng/mL, FIB 4.0 g/L, Absence of malignant PE, The nomogram-derived total score of 193 points corresponds to: Median predicted survival: 18.0 months, 1-year survival probability: 0.757 (1.000–0.243).

Figure 6.

Visualization and verification of risk model for predicting the prognosis of OS. (A). Nomogram model for predicting 1- and 2-yr survival. (B) One patient’s median survival time and survival probability are calculated according to the nomogram model. (C) ROC curves of 1- and 2-yr survival rate. (D) The calibration curves of 1-yr survival. (E) The calibration curves of 2-yr survival. (F). DCA for evaluating the practical application ability of the model. (G). The DCA for 2 random models another_1 (metastatic site + NLR + CA125 + PE) and another_2 (CYFRA 21-1 + FIB + PE). CA125 = carbohydrate antigen 125, CYFRA 21-1 = cytokeratin 19 fragment, DCA = decision curve analysis, FIB = fibrinogen, NLR = neutrophil-lymphocyte ratio, OS = overall survival, PE = pleural effusion, ROC = receiver operating characteristic.

The prognostic performance was evaluated through time-dependent ROC analysis, demonstrating excellent discriminative capacity with AUCs of 0.858 for 1-year survival and 0.860 for 2-year survival (Fig. 6C). Calibration plots revealed minimal deviation from the ideal 45-degree line at both timepoints, confirming high prediction accuracy (Fig. 6D and E). DCA at 18.0 months demonstrated superior net benefit across the clinically relevant threshold probability spectrum (5%–100%) compared to treat-all/treat-none strategies (Fig. 6F). To assess model robustness, we constructed 2 simplified random models through selective predictor exclusion: Another1: Metastatic site + NLR + CA125 + PE, Another2: CYFRA 21-1 + FIB + PE. Decision curve comparison demonstrated the composite nomogram maintained superior net benefit across all clinically relevant threshold probabilities (5%–95%) compared to these submodels (Fig. 6G), validating its prognostic value in advanced NSCLC survival prediction.

4. Discussion

Extensive research has characterized individual biomarkers in cancer immunotherapy. Inomata et al^[30] identified tumor-infiltrating CD4⁺ T cells as significant predictors of prolonged PFS post-ICI initiation in NSCLC. Similarly, Li et al^[31] established that pretreatment NLR levels strongly correlated with therapeutic efficacy of ICIs in advanced NSCLC. Despite these advances, comparative evaluations of biomarker performance remain critically underexplored. The complexity of tumor-immune interactions, particularly the multistep nature of the cancer-immunity cycle, necessitates multidimensional biomarker integration rather than reliance on single parameters.

Current clinical practice predominantly utilizes tumor tissue-based biomarkers. While informative, tissue biopsies carry inherent limitations: procedural invasiveness, sampling heterogeneity, and restricted longitudinal monitoring feasibility. In contrast, peripheral blood biomarkers offer distinct advantages through noninvasive serial sampling, cost-effectiveness, and dynamic treatment response monitoring. CE-CT demonstrates superior diagnostic performance in advanced NSCLC management, particularly in metastatic lymph node detection, tumor burden quantification, lesion characterization, and peritumoral vascular architecture assessment – advantages attributable to its rapid acquisition protocols, superior spatial resolution, and excellent soft-tissue contrast.^[32]This study established multidimensional prediction models integrating clinical profiles, systemic biomarkers, and radiologic parameters to optimize immunotherapy response assessment and survival prediction in advanced NSCLC. The validated superior performance of these composite models demonstrates their potential to inform personalized therapeutic decision-making. To improve clinical accessibility, we developed interpretable nomograms that transform complex multivariable algorithms into visual decision aids. Unlike existing AI-dependent radiomic nomograms requiring laborious tumor segmentation and radiomic feature extraction,^[33,34] our methodology employs clinically accessible imaging parameters (e.g., MLE, GGO) that require neither specialized software nor manual region-of-interest delineation. This pragmatic approach maintains diagnostic accuracy while significantly reducing operational complexity.

Current clinical practice routinely employs 4 key laboratory biomarker categories: systemic inflammatory biomarkers, lymphocyte subpopulations, tumor-associated antigens, and coagulation profiles. The inflammatory tumor microenvironment (TME) plays a pivotal role in oncogenesis and modulates responses to immunotherapy through multifaceted mechanisms.^[35] Within the TME, neutrophils drive cancer progression through multiple protumor mechanisms, including stimulation of cellular proliferation, induction of angiogenesis, facilitation of tumor invasion/metastasis, and suppression of antitumor immunity, emerging as pivotal mediators of malignant evolution.^[36]

Increased tumor-infiltrating lymphocyte density correlates with improved prognosis through enhanced antitumor immunity, whereas lymphopenia reflects impaired cell-mediated immune responses. NLR, PLR, LMR, and SII – derived from peripheral neutrophil and lymphocyte counts – serve as key biomarkers reflecting tumor microenvironment dynamics and predicting therapeutic outcomes in solid tumors.^[37,38] Elevated NLR levels may suggest tumor invasiveness and metastatic potential, with increased peripheral NLR levels showing positive correlations with tumor burden magnitude and proliferative activity. Suh et al^[39] conducted a retrospective analysis of 54 nivolumab-treated NSCLC patients, demonstrating significantly shorter PFS (1.3 vs 6.1 months) and overall survival (OS: 2.1 vs 14.0 months) in patients with elevated posttreatment NLR (≥ 5) compared to those with NLR < 5. Furthermore, Diem et al^[40] reported a significant association between increased NLR and inferior OS (HR = 3.64 for log-transformed NLR) as well as reduced treatment response (OR = 0.17) in 52 nivolumab-treated metastatic NSCLC patients. Our findings corroborate these observations, revealing elevated NLR as an independent predictor of worse OS (HR = 1.486, 95% CI = 1.059–2.08, P = .022) in immunotherapy-treated advanced NSCLC cohorts. Thrombocytosis in cancer patients correlates with metastatic progression through hypercoagulation-mediated tumor thrombi development, facilitating malignant dissemination. The PLR serves as a composite biomarker of systemic inflammatory state and platelet activation dynamics.^[41] Russo et al^[42] demonstrated in a multicenter retrospective analysis that NSCLC patients with baseline PLR ≥ 200 experienced inferior nivolumab outcomes. Our findings identified PLR as a significant determinant of immunotherapy response (ORR: P = .002), PFS (P = .005), and OS (P = .047) in advanced NSCLC. Emerging evidence suggests LMR may serve as a prognostic biomarker in certain malignancies. Elevated baseline LMR was associated with improved disease control rates (DCR: OR = 3.16, 95% CI = 1.70–5.87, P < .001), prolonged PFS (HR = 0.60, 95% CI = 0.49–0.74, P < .001), and superior OS (HR = 0.46, 95% CI = 0.39–0.56, P < .001) in patients receiving ICI therapy.^[43]Notably, our analysis revealed no significant association between LMR and post-immunotherapy prognosis. The SII, integrating neutrophil, lymphocyte, and platelet counts, has demonstrated prognostic relevance across multiple malignancies including lung, breast, gastric, and hepatic cancers.^[44] Wang et al^[45] conducted a meta-analysis of 9 studies (n = 2441) establishing elevated pretreatment SII as a predictor of inferior OS (HR = 1.88, 95% CI = 1.50–2.36; P < .001), PFS (HR = 2.50, 95% CI = 1.20–5.20; P = .014), and cancer-specific survival (CSS: HR = 1.852, 95% CI = 1.185–2.915; P = .007) in NSCLC. Current consensus regarding optimal inflammatory biomarker thresholds remains elusive, with studies employing heterogeneous methodologies including interquartile ranges, mean values, median thresholds, and ROC-derived optimal cutoffs.^[46,47] Although our nomogram models did not stratify variables into high/low groups, the results align with established prognostic patterns. Technical variability across hematological analyzers precludes definitive reference standards; however, comprehensive multicenter datasets incorporating detailed numerical parameters could enhance clinical utility for immunotherapy response prediction.

Lymphocytes constitute critical cellular mediators of adaptive immunity. Functional lymphocyte impairment coupled with immunosuppressive factor overexpression fosters an immunosuppressive tumor microenvironment, facilitating malignant progression through immune evasion mechanisms and neoplastic proliferation. Key lymphocyte subpopulations – including CD4⁺ T cells, CD8⁺ T cells, NK cells, and B lymphocytes – form the predominant tumor-infiltrating lymphocyte populations demonstrating predictive utility for immune checkpoint inhibitor therapeutic response.^[48] Specifically, CD4⁺ T cells orchestrate antitumor immunity through cytokine-mediated immune modulation and immunological memory formation, thereby potentiating tumoricidal activity.^[49]

Yan et al^[50] demonstrated that baseline CD4⁺ T lymphocyte counts (OR = 0.903, 95% CI = 0.853–0.957, P = .001) and NK cell levels (OR = 0.913, 95% CI = 0.860–0.969, P = .003) served as independent prognostic biomarkers for both therapeutic response and PFS in patients with NSCLC undergoing immune checkpoint inhibitor therapy. Liu et al^[51] reported lower baseline absolute counts of CD34⁺ T-cell (P < .001) and CD4⁺ T-cell (P < .001) were associated with for poorer efficacy. Our findings corroborate these observations, showing CD4⁺ T-cell elevation associates with improved ORR (P < .001), prolonged PFS (P = .009), and superior OS (P = .012). CD8⁺ cytotoxic T lymphocytes mediate tumor cell elimination via granzyme-perforin pathways, with clinical benefit observed in 80% of NSCLC patients exhibiting PD-1⁺CD8⁺ T-cell responses post-4-week PD-1 blockade.^[52] Lopez de Rodas et al^[53] further established stromal CD8⁺ T-cell density as an independent survival predictor. NK cells execute antitumor activity through multifaceted mechanisms: direct cytolysis, IFN-γ secretion, FAS/FASL-mediated apoptosis, and ADCC.^[54] Regrettably, our cohort revealed no significant associations between CD8⁺/NK cell levels and immunotherapy outcomes.

Tumor biomarkers constitute biologically active substances aberrantly produced by malignant cells or induced through host-tumor interactions, reflecting disease progression and therapeutic response dynamics. Clinically utilized serum biomarkers in NSCLC include CEA, CA125, SCC-Ag, ProGRP, neuron-specific enolase, and CYFRA 21-1. CEA is normally produced during fetal development and production ceases after birth, therefore, it is usually present at low levels in the peripheral blood of healthy adults. Many studies regard serum CEA as a tumor marker in various solid cancers including NSCLC. Dall’Olio et al^[55] demonstrated that a 20% reduction of CEA was correlated with a longer OS (HR = 0.12; 95% CI = 0.04–0.33; P < .001) after 3 cycles of ICI treatment. This study also revealed similar results that NSCLC patients with high levels of CEA (P = .008) or CYFRA 21-1 (P = .004) had shorter PFS times. CYFRA 21-1, a typical tumor marker of lung cancer, developing upon tumor transdifferentiation, is widely present in alveolar and endometrial epithelial cells. During malignant transformation of epithelial cells, keratin 19 fragments accumulate as cytoplasmic acidic soluble isoforms, subsequently released into circulation via tumor necrosis and apoptosis, ultimately elevating serum CYFRA 21-1 concentrations. Dall’Olio et al^[55] further identified pretreatment CYFRA 21-1 levels > 8 ng/mL as prognostic for inferior OS in ICI-treated NSCLC patients. CA125, a high-molecular-weight glycoprotein secreted into systemic circulation upon neoplastic invasion and tissue remodeling, demonstrates serum concentrations proportional to tumor burden and disease progression. CA125, a glycoprotein biomarker released into systemic circulation during malignant transformation and tumor-mediated tissue destruction, demonstrates serum concentration patterns correlating with disease severity. Elevated CA125 levels are clinically observed in the majority of NSCLC cases, particularly those exhibiting aggressive biological behavior and advanced pathological stages. A multicenter retrospective analysis of 716 immunotherapy-naïve patients demonstrated dose-dependent associations between CA125 levels and survival outcomes, with elevated concentrations predicting reduced PFS and OS in both LUAD and LUSC subtypes.^[56] Our findings corroborate these observations, revealing pretreatment CA125 elevation as an independent predictor of diminished immunotherapy response and shorter OS (P = .018).

Physiological hemostatic homeostasis is orchestrated through coordinated interactions among coagulation cascades, vascular endothelium, and platelet activation. Neoplastic development perturbs this thrombohemorrhagic balance, inducing a systemic prothrombotic state that elevates venous thromboembolism (VTE) risk and facilitates metastatic dissemination through tumor cell-platelet microaggregate formation. Notably, NSCLC exhibits 4% to 6% VTE incidence in treatment-naïve cohorts, with hypercoagulable states independently predicting mortality.^[57] Consequently, longitudinal coagulation profiling provides critical prognostic insights for therapeutic optimization. D-D, a fibrinolysis byproduct of cross-linked fibrin degradation, serves as a validated biomarker of consumptive coagulopathy. İnal et al^[58] reported median plasma D-D concentrations in lung cancer patients exceeding healthy controls. Furthermore, nonsurvivors demonstrated persistently elevated D-D levels posttreatment cycles, establishing its prognostic utility. Chen et al^[59] prospectively analyzed plasma D-D levels in 100 stage IV/recurrent NSCLC patients undergoing anti-PD-1 immunotherapy, identifying a critical threshold of 981 ng/mL that independently predicted radiographic disease progression (RECIST 1.1) and reduced PFS. Our multivariable Cox regression analysis corroborated this association, demonstrating elevated baseline D-D levels as an independent predictor of inferior OS (P = .001). As the predominant coagulation factor, FIB facilitates tumor progression through paracrine signaling mechanisms involving vascular endothelial growth factor-A (VEGF-A) and platelet-derived growth factor receptor-β (PDGFR-β). Li et al^[60] reported pretreatment hyperfibrinogenemia (≥ 4.0 g/L) as an adverse prognostic factor in lung cancer (HR = 1.43, 95% CI = 1.00–2.04, P = .049). A pooled meta-analysis of 16 cohorts (n = 6881) confirmed FIB’s prognostic value, revealing significant associations with reduced OS (pooled HR = 1.38, 95% CI = 1.22–1.55, P < .001) and PFS (pooled HR = 1.29, 95% CI = 1.01–1.65, P = .042). Our data further substantiate these findings, showing advanced NSCLC patients with hyperfibrinogenemia had diminished ORR (P = .047), shorter median PFS (P = .004), and inferior OS (P < .001).

CE-CT reveals diagnostic imaging hallmarks in lung cancer, lobulation, spiculation, MLE, and PE. Lymph node metastasis induces immunosuppressive niche formation via Treg proliferation and DC functional suppression,^[61] reflecting established immunoevasive mechanisms. N2 involvement consequently predicts diminished therapeutic efficacy. Ma et al^[62] documented lower pCR rates (P = .001) in stage IIA-IIIB NSCLC with N2 disease undergoing neoadjuvant chemoimmunotherapy. Our analysis identified MLE as an independent predictor of rapid immunotherapy resistance (P = .028). Malignant PE, originating from tumor-driven inflammation and lymphatic metastasis, correlates with catastrophic prognosis in advanced NSCLC.^[63] Nishimura et al^[64] observed marked survival deficits in 478 chemoimmunotherapy-treated PE-positive cases (median PFS: 6.2 months vs 9.1 months; P < .001; median OS: 16.4 months vs 27.7 months; P < .001). Our findings corroborate previous observations demonstrating PE as a significant prognostic determinant for reduced survival outcomes in PD-1/PD-L1 inhibitor-treated NSCLC patients. Our findings corroborate previous observations demonstrating PE as a significant prognostic determinant for reduced survival outcomes in PD-1/PD-L1 inhibitor-treated NSCLC patients. Multivariable Cox regression confirmed PE’s independent association with inferior overall survival (P < .001). This prognostic correlation may be mechanistically linked to PD-1 overexpression on tumor-infiltrating lymphocytes within malignant PE, potentially driving T-cell exhaustion and secondary resistance to PD-1/PD-L1 blockade. Current evidence regarding CE-CT-derived quantitative imaging biomarkers and their predictive utility for immunotherapy outcomes in NSCLC remains preliminary, requiring prospective multicenter studies.

This study has several inherent limitations due to its retrospective design. Primarily, the single-center retrospective nature and modest sample size precluded external validation cohort establishment. Second, the abbreviated follow-up duration precluded comprehensive survival analysis, with censored cases persisting at data cutoff. Mature OS data remain pending, necessitating extended follow-up for 3-/5-year survival maturation. Third, novel predictive biomarkers (e.g., genomic signatures or multiplex immunohistochemical profiles) were not incorporated. Future investigations should prioritize multicenter prospective trials with adequately powered cohorts to validate these findings.

5. Conclusions

In conclusion, this study identified CD4⁺ T cells (P = .007), FIB (P = .047), and MLE (P = .028) as independent predictors of ORR in advanced NSCLC patients receiving PD-1/PD-L1 inhibitors. For OS, metastatic site (P = .007), NLR (P = .025), CA125 (P = .020), CYFRA 21-1 (P = .004), FIB (P < .001), and PE (P < .001) emerged as significant prognostic determinants. By integrating clinical parameters, laboratory biomarkers, and radiologic features, we developed 2 distinct risk stratification models demonstrating clinically relevant predictive accuracy for both short-term treatment response and long-term survival outcomes following 4 cycles of immunotherapy. These models exhibit potential clinical utility for prognostic evaluation and therapeutic decision-making in PD-1/PD-L1 inhibitor-based regimens.

Author contributions

Conceptualization: Xinyu Bai.

Data curation: Xinyu Bai.

Methodology: Xin Wang, Yiying Bai, Qianhui Chen.

Investigation: Hailan Xu.

Resources: Sheng Bi, Senyang Chen, Hongbin Yang.

Validation: Xiaotong Zhang, Fan Li.

Writing – review & editing: Lei Liu, Li Zhang.