Construction of a postoperative disease-free survival prediction model for non-small cell lung cancer patients based on dual-energy computed tomography parameters and blood inflammatory indicators

Introduction

Background

Lung cancer is a serious global disease with high mortality and incidence rates in both men and women (1,2). Non-small cell lung cancer (NSCLC) represents the predominant form of lung cancer, comprising approximately 85% of all lung cancer diagnoses (3), and its high mortality rate poses a serious threat to both patients and healthcare systems (4). Surgical treatment is the primary clinical approach for early and mid-stage NSCLC, effectively prolonging survival by completely excising the primary lesion (5). Nevertheless, 30–50% of patients still experience postoperative local recurrence and distant metastasis, resulting in poor long-term prognosis (6).

Currently, the tumor node metastasis (TNM) staging system provides a preliminary prognosis assessment but often does not account for treatment responses and individual prognosis differences in NSCLC at different stages (7,8). Previous studies have reported clinical characteristics such as age, gender, pathological type, and lymph node metastasis as potential predictive factors for NSCLC recurrence, but these also struggle to accurately reflect individual recurrence risks (9,10). Therefore, exploring more comprehensive prognostic indicators and models is crucial for optimizing postoperative management.

Rationale and knowledge gap

Dual-energy computed tomography (DECT) represents a novel imaging modality that, compared to traditional single-layer computed tomography (CT), can obtain lesion images across a wide range of clinical indications. This technique facilitates the generation of virtual monoenergetic images, spectral curves, iodine distribution maps, effective atomic number (Z_eff) maps, and various other multi-parameter imaging outputs through simple post-processing, without increasing radiation exposure (11). DECT-derived quantitative parameters such as iodine concentration (IC) and Z_eff, which are related to tissue iodine content, can serve as biomarkers reflecting pathological and physiological changes in tissue blood volume and vascular permeability, thereby indicating angiogenesis (12,13).

Previous investigations have demonstrated that quantitative parameters, including IC, Z_eff, and relative electron density (Rho) can non-invasively assess tumor angiogenesis and tissue composition. These parameters demonstrate potential in tumor staging, efficacy evaluation, and prognostic prediction (14-16).

For instance, Zheng et al. demonstrated that normalized iodine concentration (NIC) and Z_eff can serve as reliable indicators to differentiate between pre-invasive lesions and invasive adenocarcinoma in research of ground-glass nodules, achieving an area under the curve (AUC) of 0.957 (17). Wu et al. found that DECT volumetric quantitative analysis of Z_eff and IC closely correlated with Ki-67 and TTF-1 expression in NSCLC, providing a quantitative tool for assessing tumor biological behavior (18). In recent years, studies also have shown that NIC can predict early recurrence after surgery for patients with esophageal cancer (19).

Additionally, inflammatory indicators like the neutrophil-lymphocyte ratio (NLR) reflect immune status and have been shown to correlate with NSCLC prognosis in multiple studies (20,21). Li et al. found that combining DECT parameters (arterial enhancement fraction, AEF; extracellular volume, ECV) with blood indicators (lymphocyte-monocyte ratio, LMR; red blood cell, RBC) significantly improved the predictive efficacy for pathological complete response (pCR) (22), highlighting the advantages of multimodal combined models.

Objective

Existing studies have primarily constructed models using either clinical features or imaging parameters alone (23,24), lacking integrated analyses of DECT functional parameters, blood inflammatory indicators, and clinical pathological characteristics. This study aims to construct a combined predictive model through least absolute shrinkage and selection operator (LASSO) regression and validate its predictive efficacy, thereby providing a new tool for accurately predicting disease-free survival (DFS) in resectable NSCLC patients. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2251/rc).

Methods

Study subjects

A retrospective analysis of data from 140 patients who underwent surgical resection for pathologically confirmed NSCLC was conducted from January 2019 to June 2023. Inclusion criteria: (I) DECT examination completed within 2 weeks before surgery; (II) postoperative pathological staging of stage I–III B; (III) complete clinical and follow-up data; (IV) availability of DECT quantitative parameters and blood test data.

Exclusion criteria: (I) multiple primary lung cancers or distant metastasis; (II) preoperative chemotherapy or targeted therapy; (III) presence of other malignant tumors; (IV) pathology indicating in situ carcinoma or minimally invasive adenocarcinoma.

The patients were allocated into two groups: a training set comprising 98 cases and a test set consisting of 42 cases using a random number table method. A detailed overview of the study procedure can be seen in Figure 1.

Figure 1 Participant flow diagram. DECT, dual-energy computed tomography; K-M, Kaplan-Meier; LASSO, least absolute shrinkage and selection operator; NSCLC, non-small cell lung cancer; ROC, receiver operating characteristic.

Sample size consideration: this exploratory study aimed to develop a preliminary predictive model. According to the rule of thumb requiring at least 10–15 events per predictor variable for reliable Cox regression, our final model incorporated three independent predictors. With 34 events observed in the training set (n=98), the event-to-variable ratio was adequate for initial model development and internal validation.

Ethical consideration

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the institutional ethics committee of the First Affiliated Hospital of Bengbu Medical University (No. 2023-440), and individual consent for this retrospective analysis was waived.

Clinical data collection

Clinical and pathological characteristics

Patient age, gender, body mass index (BMI), smoking history, maximum tumor diameter, TNM staging, lymph node metastasis, pathological grading, pathological typing, tumor location, lymphovascular invasion (LVI), perineural invasion (PNI), pleural invasion (VPI), surgical method, postoperative chemotherapy, and targeted therapy were collected through the electronic medical record system. Pathological data were reviewed blindly by two pathologists.

Follow-up and clinical endpoints

All patients were followed up every 3 months for the first year post-surgery, and then every 6 months, with recurrence and metastasis assessed through enhanced CT and/or positron emission tomography (PET)-CT and/or magnetic resonance imaging (MRI) and/or whole-body bone scans. Telephone follow-ups were conducted to confirm postoperative treatment methods (targeted therapy/chemotherapy/radiotherapy), follow-up frequency, and recurrence or metastasis status. The endpoint of this study was DFS, defined as the time from the date of surgery to the first histologically or radiologically confirmed recurrence or distant metastasis. Patients who died without documented recurrence were treated as a competing risk and censored at the date of death in the primary analysis; a sensitivity analysis treating such deaths as events yielded consistent results. The last follow-up ended on June 15, 2025.

Handling of second primary cancers

During follow-up, the occurrence of a new lung lesion was rigorously evaluated by a multidisciplinary team. A second primary lung cancer was defined based on distinct histological type or, in cases of the same histology, on anatomical separation (different lobe) with no evidence of metastatic spread. Confirmed second primaries were not considered as recurrence events for the DFS endpoint; follow-up for the index cancer was censored at the date of the second primary diagnosis.

Diagnostic criteria for postoperative recurrence and metastasis

Evaluations were based on clinical symptoms, signs, and imaging examination results such as chest CT and MRI. Postoperative recurrence and metastasis in NSCLC radical resection were defined as tumor presence at the surgical margin or in the mediastinal, pulmonary, and pleural lymph nodes on the original lesion side as postoperative recurrence; tumor metastasis to the contralateral thoracic cavity or other tissues and organs, including the contralateral adrenal gland, lungs, cervical lymph nodes, liver, etc., as postoperative metastasis (25).

DECT measurement

Instruments and methods

A GE Revolution 256-slice CT scanner was used, with patients instructed to lie supine. Chest scans were performed in spectral imaging mode, with tube voltage set at 80/140 kV, tube current at 320–370 mA, pitch at 0.992:1, slice thickness and interval both at 5 mm, reconstruction slice thickness at 0.625 mm, window width at 400 Hounsfield unit (HU), and window level at 40 HU. After the plain scan, iodine contrast agent (320 mgI/mL, 1.5 mL/kg body weight) was injected at a flow rate of 2.5 mL/s, with delayed scans performed at 25 seconds (arterial phase) and 55 seconds (venous phase).

DECT quantitative parameters

Images were reconstructed using lung algorithms and standard algorithms. Two radiologists with over 5 years of experience used GSIViever software to place circular regions of interest (ROI) on the maximum layer of the lesion displayed in mediastinal window CT for randomly assigned cases, avoiding necrotic, cystic areas, and blood vessels and calcifications. Measurements and calculations included arterial phase normalized iodine concentration (ANIC), venous phase normalized iodine concentration (VNIC), normalized iodine concentration difference (NIC-Diff), normalized iodine concentration ratio (NIC ratio), Z_eff during plain scan, plain CT value (Plan-CT), arterial phase CT value (A-CT), venous phase CT value (V-CT), arterial phase maximum enhancement rate (A-CER), and venous phase maximum enhancement rate (V-CER).

The mean of the measurements was taken, with consistency testing showing an intraclass correlation coefficient (ICC) >0.90. Typical patient DECT parameter measurements are shown in Figure 2.

Figure 2 Representative DECT images from patients with resectable NSCLC. (A) A case from a 51-year-old female with left lower lobe adenocarcinoma, who experienced postoperative recurrence. (B) A case from a 64-year-old male with right lower lobe squamous cell carcinoma, without recurrence or metastasis during follow-up. DECT, dual-energy computed tomography; NSCLC, non-small cell lung cancer.

Calculation formulas:

NIC=IClesion÷ICaorta[1]

NIC-Diff=VNIC−ANIC[2]

NIC-Ratio=VNIC÷ANIC[3]

CER=(CTenhancement−CTplainscan)÷CTplainscan×100%[4]

Blood parameter measurement

Fasting peripheral blood samples were routinely collected within 1 week prior to surgery as part of the preoperative assessment. The quantifiable indicators included serum albumin (ALB), white blood cells (WBCs), neutrophils (NEUs), absolute lymphocyte count (ALC), monocytes (MOs), RBC, NLR, and carcinoembryonic antigen (CEA). The advanced lung cancer inflammation index (ALI) was subsequently calculated as follows: ALI = BMI × ALB (g/dL) ÷ NLR. For all patients, the blood draw occurred after the DECT scan and before the surgical procedure, ensuring both assessments reflected the same preoperative status. Given the small number of patients with missing DECT or blood parameters, this study employed a complete case analysis.

Statistical analysis

R software (version 4.5.0) and SPSS 27.0 were used for analysis. Normally distributed measurement data were expressed as mean ± standard deviation, and intergroup comparisons were performed using t-tests; skewed distribution data were expressed as median (interquartile range) and analyzed using Mann-Whitney U tests. Count data were expressed as n (%), and intergroup comparisons were performed using χ² tests.

Potential predictive variables were screened using LASSO regression (the optimal regularization parameter λ was determined by 10-fold cross-validation). Strict linear hypothesis testing was conducted on all continuous predictive factors selected. The non-linear relationship between each variable and log[hazard ratio (HR)] was examined using restricted cubic splines (RCS) analysis (with three nodes set). Non-linear variables were determined for cut-off values using diagnostic receiver operating characteristic (ROC) curves, whereas linear variables were directly included in subsequent analyses. Univariate Cox regression analysis was used to screen variables with P<0.1, which were included in multivariate Cox regression to determine independent risk factors (P<0.05 was considered statistically significant). Before constructing the final multivariate model, the variance inflation factor (VIF) was calculated for all candidate predictors identified from the univariate analysis to check for multicollinearity. A VIF threshold of <5 was considered acceptable.

Clinical models (including TNM staging, lymph node metastasis, pathological staging), DECT models (including VNIC, Z_eff), blood models (including NLR), and comprehensive nomogram models (combining the above six indicators) were constructed based on independent risk factors. ROC curves were plotted to calculate the AUC to assess model discrimination, calibration curves were used to evaluate the consistency between predicted and actual values, decision curve analysis (DCA) was used to assess clinical net benefit, and C-index and Brier scores were used to evaluate model stability. In addition to the validation using the hold-out test set, internal validation was performed using 1,000-cycle bootstrap resampling on the entire cohort to obtain a bias-corrected (optimism-corrected) C-index, which provides a more robust estimate of model performance and accounts for overfitting.

Survival curves were plotted using the Kaplan-Meier (K-M) method. First, each patient’s risk score was calculated based on the comprehensive nomogram model. The optimal cut-off value for risk scores was determined using the maximally selected rank statistics method. Patients were then divided into high-risk and low-risk groups based on this cutoff value. Finally, the K-M technique was employed to generate DFS curves for both groups, while the log-rank test was used to evaluate the statistical significance of the survival disparities observed between the two groups.

Similarly, K-M curves were plotted for high-risk and low-risk groups, utilizing the optimal cut-off thresholds established for NLR, VNIC, and Z_eff.

Results

Follow-up summary

The median follow-up time for the included samples was 56 months (range, 24 to 76 months). All 140 patients had complete follow-up information. As per our pre-defined inclusion criteria (requiring complete data), any patients who died from non-cancer-related causes or were lost to follow-up were excluded prior to analysis. By the end of the follow-up period, 49 patients (35.0%) had experienced documented recurrence or metastasis. The remaining 91 patients (65.0%) were right-censored and alive without evidence of cancer at their last contact. The median DFS time was 40.5 months. Eleven patients had follow-up times of less than 36 months.

Comparison of baseline data

Notable distinctions were identified between the recurrence group and the non-recurrence group in smoking history, maximum tumor diameter, TNM staging, lymph node metastasis, pathological grading, LVI, ANIC, VNIC, NIC-Diff, NIC ratio, Z_eff, WBC, NEU, NLR, ALI, and CEA indicators (all P<0.05), while no significant differences were observed in additional parameters (all P>0.05) (Tables 1,2).

The distribution of clinical characteristics, DECT parameters, and blood indicators in the training set and test set was balanced except for NIC ratio (all P>0.05), indicating comparability (Tables S1,S2). Importantly, the distribution of TNM stage, a key prognostic factor, was not significantly different between the training and test sets (P=0.34), supporting the validity of the random split for internal validation.

LASSO regression and Cox regression analysis

LASSO regression identified 10 potential predictive factors from 36 variables among 140 patients: smoking history, lymph node metastasis, pathological staging, LVI, VNIC, NIC ratio, Z_eff, NLR, ALI, and CEA (Figure S1).

Based on RCS analysis (Figure 3A-3F), we observed a non-linear association between NIC ratio, NLR, and CEA with recurrence risk. Considering the sample size and exploratory nature of the analysis, we set the significance level for testing non-linear relationships at α =0.10.

Figure 3 Exploration of variable relationships and the final multivariate model. (A) Comparison of linear and nonlinear fits for VNIC in predicting recurrence risk. (B) Comparison of linear and nonlinear fits for Z_eff in predicting recurrence risk. (C) Comparison of linear and nonlinear fits for ALI in predicting recurrence risk. (D) Nonlinear relationship between NIC ratio and recurrence risk. (E) Nonlinear relationship between NLR and recurrence risk. (F) Nonlinear relationship between CEA and recurrence risk. (G) Forest plot of the multivariate Cox regression analysis for key predictors of recurrence risk. ALI, advanced lung cancer inflammation index; CEA, carcinoembryonic antigen; CI, confidence interval; HR, hazard ratio; NIC, normalized iodine concentration; NLR, neutrophil-lymphocyte ratio; VNIC, venous phase normalized iodine concentration; Z_eff, effective atomic number.

Under this standard, the non-linear relationships of NLR (P=0.046) and CEA (P<0.001) were statistically significant, while the non-linear trend of NIC ratio approached significance (P=0.08). The NIC ratio curve in Figure 3D exhibited a clear platform-rising threshold effect with important clinical implications. Therefore, we determined the optimal cut-off values for NIC ratio, NLR, and CEA using the maximized Youden index (1.47, 2.71, and 5.75 ng/mL) and performed binary classification. In contrast, VNIC, Z_eff, and ALI conformed to the linear association hypothesis (P>0.05), so they were modeled in continuous form.

As shown in Table 3, univariate Cox regression analysis of the 98 patients in the training set indicated that lymph node metastasis, pathological staging, LVI, VNIC, NIC ratio, Z_eff, NLR, ALI, and CEA were associated with DFS (all P<0.05). Multivariate Cox regression further confirmed that VNIC [HR =1.129; 95% confidence interval (CI): 1.077–1.184; P<0.001], Z_eff (HR =3.700; 95% CI: 1.046–13.088; P=0.042), and NLR (HR =3.274; 95% CI: 1.478–7.249; P=0.003) were independent risk factors. A forest plot illustrating these findings is shown in Figure 3G.

Assessment of multicollinearity:

Prior to multivariate Cox regression, we assessed potential multicollinearity among the candidate predictors identified by univariate analysis. The VIFs for all variables were well below the threshold of 5 (range, 1.11–1.90), indicating no substantial multicollinearity (Figure S2A). This was further supported by a correlation matrix, which showed a mean correlation coefficient of 0.208 among variables. The maximum correlation observed was 0.679, and no variable pairs exhibited a strong correlation (|r|>0.7) (Figure S2B).

Nomogram model construction and evaluation

Nomogram construction

We incorporated the three variables VNIC, Z_eff, and NLR, as well as other variables with high HR values and clear clinical significance based on prior clinical knowledge, including TNM staging, lymph node metastasis, and pathological staging. These variables were used to construct a nomogram model for the training set, ensuring the model is both clinically comprehensive and practically useful (Figure 4A).

Figure 4 Development, calibration, and clinical utility of the predictive nomogram. (A) Nomogram for predicting 1-, 2-, and 3-year DFS probability in NSCLC patients. (B) Calibration curve of the nomogram in the training set. (C) Calibration curve of the nomogram in the test set. (D) Time-dependent DCA of the nomogram in the test set. DCA, decision curve analysis; DFS, disease-free survival; NLR, neutrophil-lymphocyte ratio; NSCLC, non-small cell lung cancer; TNM, tumor node metastasis; VNIC, venous phase normalized iodine concentration; Z_eff, effective atomic number.

Model calibration assessment

Figure 4B presents the calibration curve for the training set, while Figure 4C shows the calibration curve for the test set. The curves, distinguished by color, represent calibration results at 1-, 2-, and 3-year time points. The results indicate that the calibration curves for both training and test sets closely follow the ideal line, suggesting the nomogram’s strong calibration and accurate survival prediction.

DCA

The time-dependent DCA of the nomogram model demonstrated significant net benefits across appropriate threshold probabilities, indicating good clinical applicability and valuable guidance for clinical decision-making (Figure 4D).

Model discrimination assessment—ROC curve analysis

Training set ROC curves: Figure 5A-5D show the ROC curves of different models in the training set. The clinical model’s AUCs at 1, 2, and 3 years are 0.743, 0.796, and 0.839, respectively. The DECT model’s AUCs are 0.891, 0.873, and 0.895. The blood model shows AUCs of 0.709, 0.789, and 0.803. In contrast, the nomogram model achieves AUCs of 0.896, 0.926, and 0.948, significantly outperforming the other single models. This indicates the nomogram model has superior discrimination ability in the training set, allowing clearer distinction between surviving and deceased patients.

Figure 5 Discriminative performance of individual models and the integrated nomogram. (A-D) ROC curves of the clinical, DECT, blood, and nomogram models in the training set, respectively. (E-H) ROC curves of the clinical, DECT, blood, and nomogram models in the test set, respectively. AUC, area under the curve; DECT, dual-energy computed tomography; ROC, receiver operating characteristic.

Testing set ROC curves: Figure 5E-5H show the AUCs for the clinical model at 1, 2, and 3 years are 0.783, 0.849, and 0.864; for the DECT model, they are 0.855, 0.899, and 0.909; for the blood model, 0.651, 0.726, and 0.760; and for the nomogram model, 0.809, 0.917, and 0.934. The nomogram model’s AUC in the test set is still higher than that of the other individual models, further proving it has a strong ability to distinguish, and it consistently maintains excellent performance at different time points.

Figure 6A shows the time-dependent ROC analysis results for the training set, displaying the AUCs and 95% CIs for four different models at 1, 2, and 3 years. The results show the nomogram model’s AUC is higher than other models at all time points, and over time, from 1 to 3 years, the AUC trends upward, reaching 0.902–0.981 (including 95% CI) at 3 years.

Figure 6 Comprehensive model comparison and ablation analysis. (A) Time‑dependent ROC analysis of different models in the training set. (B) Distribution of linear predictors in the test set for different model configurations in ablation experiments. AUC, area under the curve; CI, confidence interval; DECT, dual-energy computed tomography; ROC, receiver operating characteristic.

Additionally, Table 4 shows the nomogram model’s C-index is 0.893, and the Brier score (which measures prediction error) remains low across all time points—for example, 0.085 at 3 years—suggesting better predictive accuracy and exceptional power for long-term survival in NSCLC patients.

Performance comparison of the model in training and testing sets

The C-index for the nomogram model was 0.893 in the training set and 0.839 in the testing set, with 3-year Brier scores of 0.085 and 0.179, respectively, indicating good model stability and generalizability (Table 5).

Bootstrap validation for overfitting assessment

To quantitatively assess potential overfitting, 1,000-cycle bootstrap resampling internal validation was performed on the entire cohort (n=140). The apparent C-index of the comprehensive nomogram was 0.743. The optimism-corrected C-index was 0.693, with an optimism of 0.050 (Figure S3). This result indicates a manageable degree of overfitting, and the maintained discriminative ability supports the robustness of our model beyond the original sample split.

Ablation experiments

As shown in Table 6, in the training set, excluding clinical indicators from the full model (C-index 0.893, 3-year AUC 0.948) reduced the C-index to 0.889 and the 3-year AUC to 0.942 (P<0.001). Excluding DECT parameters reduced the C-index to 0.797 and the 3-year AUC to 0.886 (P<0.001). Excluding blood indicators reduced the C-index to 0.855 and the 3-year AUC to 0.930 (P<0.001).

In the test set, the full model C-index was 0.872, and the 3-year AUC was 0.934; excluding clinical indicators maintained a C-index of 0.872 and a 3-year AUC of 0.938 (P=0.18); excluding DECT parameters reduced the C-index to 0.835 and the 3-year AUC to 0.903 (P=0.043); and excluding blood indicators reduced the C-index to 0.842 and the 3-year AUC to 0.912 (P=0.01).

In Figure 6B, the distribution of linear predictive values for different models in the test set showed that the full model had the least overlap between event and non-event patients, while excluding any category of indicators increased the overlap area, supporting the importance of integrating multiple indicators in the nomogram model for clearer differentiation of patient outcomes.

Stratified performance of the nomogram

To evaluate the consistency of the model’s predictive performance across clinically relevant subgroups, we conducted stratified analyses on the entire cohort (n=140). This approach ensures adequate sample size within each subgroup for a meaningful comparison. The results illustrated in Figure S1A-S1C and Table S3 demonstrated consistent and high predictive accuracy across histologic subtypes [lung adenocarcinoma (LUAD) vs. lung squamous cell carcinoma (LUSC)], pathological stages (I vs. II–III), and tumor size dichotomized at the cohort median diameter of 3 cm (≤3 vs. >3 cm), with all between-group AUC differences being statistically non-significant (all P>0.05). This consistency suggests that the predictive power of the integrated indicators (VNIC, Z_eff, NLR) is not substantially confounded by or limited to these key clinicopathological factors, supporting the model’s broad applicability within the spectrum of resectable NSCLC.

Stratified survival analysis

The risk scores calculated based on the comprehensive nomogram model shown in Figure 7A determined that the optimal cut-off value was 0.45, dividing the patients included in the study into high-risk group (score ≥0.45) and low-risk group (score <0.45). The DFS curves for the two groups were plotted using the K-M method, and the results in Figure 7B showed that the DFS for the high-risk group was significantly shorter than for the low-risk group, with the log-rank test indicating statistical significance (P<0.001).

Figure 7 Risk stratification and individual predictor performance. (A) Distribution of nomogram scores and maximally selected rank statistics for risk stratification. (B) DFS curves stratified by nomogram-defined risk groups. (C) ROC curve comparison of VNIC and Z_eff in predicting recurrence. (D-F) DFS curves stratified by NLR, VNIC, and Z_eff, respectively. AUC, area under the curve; CI, confidence interval; DFS, disease-free survival; FPR, false positive rate; HR, hazard ratio; NLR, neutrophil-lymphocyte ratio; ROC, receiver operating characteristic; TPR, true positive rate; VNIC, venous phase normalized iodine concentration; Z_eff, effective atomic number.

The survival curves showed that low-risk patients maintained a high DFS rate throughout the 3-year follow-up, whereas high-risk patients experienced a clear decline over time. As the follow-up period went on, the number of at-risk individuals in the high-risk group dropped faster than in the low-risk group. For example, after 1 year of follow-up, there were 54 individuals at risk in the high-risk group and 69 in the low-risk group; at 2 years, 34 in the high-risk group and 63 in the low-risk group; and at 3 years, 26 in the high-risk group and 55 in the low-risk group, further confirming the higher probability of recurrence and metastasis in the high-risk group.

The diagnostic performance of the key quantitative indicators VNIC and Z_eff for predicting recurrence was evaluated using ROC curve analysis (Figure 7C). The AUC was 0.930 for VNIC and 0.695 for Z_eff, with optimal cut-off values determined at 30.08 and 8.065, respectively.

Simultaneously, stratified survival analyses based on cut-off values for individual indicators (such as NLR, VNIC, Z_eff) showed that groups with high NLR (Figure 7D, HR =6.39; 95% CI: 3.41−11.96; P<0.001), high VNIC (Figure 7E, HR =11.59; 95% CI: 6.10−21.99; P<0.001), and high Z_eff (Figure 7F, HR =2.61; 95% CI: 1.48−4.59; P<0.001) had significantly shorter DFS compared to their corresponding low-level groups. These results further validate these indicators as reliable risk factors for postoperative recurrence and metastasis in NSCLC patients. This also supports including these indicators in the comprehensive nomogram model, which integrates multiple risk factors to more accurately differentiate patient groups with varying recurrence risks.

Discussion

Key findings

This study developed and validated a comprehensive predictive model that integrates DECT quantitative parameters (VNIC; Z_eff), a systemic inflammatory marker (NLR), and clinicopathological features for predicting DFS in resectable NSCLC patients. The nomogram demonstrated high discrimination ability in both training and testing sets, with a C-index of 0.85 versus 0.82 and a 3-year AUC of 0.948 versus 0.934, along with good calibration and clinical net benefit. Bootstrap internal validation yielded an optimism-corrected C-index of 0.693, indicating a manageable degree of overfitting. Core findings indicate that VNIC, Z_eff, and NLR are independent predictive factors. These findings suggest that combining DECT functional parameters with inflammatory and clinical indicators may enhances the prediction of 3-year DFS post-surgery, providing a potential tool for tailoring postoperative management strategies.

Clinical significance of independent risk factors

VNIC: as a key DECT parameter, VNIC objectively reflects the iodine uptake in tumors during the venous phase, closely related to tumor vascular density and permeability (26,27). In this study, VNIC of the recurrence group was markedly elevated compared to that of the non-recurrence group, with each 1% increase in VNIC associated with a 12.9% increase in recurrence risk, confirming that high iodine uptake indicates the biological characteristics of high tumor invasiveness. High VNIC values can objectively reflect richer blood supply in tumor tissue, making recurrence and metastasis more likely (28,29). Zheng et al. found that elevated VNIC correlated positively with tumor invasion degree (30), consistent with our findings. The difference in arterial phase NIC was not statistically significant, possibly due to insufficient sample size.

Z_eff: reflecting the effective atomic number of tumor tissue, Z_eff is associated with cellular composition, calcification, and iodine uptake, reflecting the pathological and physiological structure and characteristics of the tissue (15,31). Our study found that patients with high Z_eff had a 3.7-fold increased risk of recurrence, possibly due to the more active proliferation of malignant tumor cells with higher iodine uptake and binding capacity. Existing studies have also confirmed the correlation between Z_eff and Ki-67 expression in NSCLC (18), further supporting its value as a marker of tumor proliferation activity.

NLR: as a marker of inflammation-immune balance, elevated NLR indicates enhanced neutrophil-mediated pro-tumor inflammatory responses and weakened lymphocyte anti-tumor immunity (32,33). In our research, patients with high NLR had a 3.27-fold increased risk of recurrence, confirming the critical role of systemic inflammation in tumor progression.

Interpretation of results and clinical implications

The results of this study have potential clinical significance. Firstly, the constructed nomogram represents a promising tool for personalized risk estimation (34). Clinicians could potentially use this nomogram, incorporating three quantitative imaging and blood parameters (VNIC, Z_eff, NLR) and three routine clinicopathological factors (TNM staging, lymph node metastasis, pathological grading), to estimate the 3-year postoperative recurrence probability. This might help identify “hidden high-risk” patients who have early TNM stages but a high actual risk of recurrence. Such identification could be valuable for optimizing decisions regarding adjuvant therapy or follow-up intensity in future clinical practice or trial settings; for example, for patients with stage IB and very high nomogram scores, more aggressive consideration of adjuvant chemotherapy may be warranted.

Secondly, our risk stratification system (with a cut-off of 145 points on the nomogram) can categorize patients into distinct risk groups, which could form the basis for designing individualized follow-up plans if validated. Low-risk patients could avoid excessive examinations and psychological burdens, while high-risk patients might receive more intensive and advanced follow-up monitoring (such as dynamic ctDNA monitoring).

Finally, the predictive power derived from both local (DECT) and systemic (NLR) parameters aligns with the biological concept that cancer outcomes are influenced by both tumor intrinsic properties (“seeds”) and host immune status (“soil”) (35). Our model provides supporting evidence for this theory.

Strengths and limitations

Study strengths

• This study innovatively combines DECT quantitative parameters with systemic inflammatory and clinicopathological data, enabling a multi-dimensional assessment. This integration achieved a synergistic effect, as evidenced by ablation experiments and the absence of significant multicollinearity (all VIFs <2).
• Methodological rigor: the analysis adhered to the TRIPOD statement, employing LASSO regression for variable selection, addressing non-linearity via RCS, and rigorously validating the model on an independent test set and through bootstrap resampling.
• Clinical applicability: all model variables are routinely accessible in standard clinical practice, and the nomogram format is user-friendly, facilitating its potential future evaluation and use.

Study limitations

• Design and generalizability: the retrospective, single-center design inherently introduces selection bias and limits the generalizability of our findings. The modest overall sample size and the relatively small test set (n=42) may affect the stability of performance estimates and the model’s ability to generalize. The absence of external validation is a key limitation that precludes any claim of readiness for clinical deployment and must be addressed in future studies.
• Follow-up characteristics: although the median follow-up was substantial (56 months), the follow-up duration was heterogeneous (range, 24–76 months). This could theoretically bias long-term DFS estimates, though the observed event rate (35%) within the follow-up period supports the validity of our 3-year predictions.
• Patient population: our model was developed specifically for resectable invasive NSCLC (stages I–IIIB). The exclusion of in situ carcinoma and minimally invasive adenocarcinoma, while methodologically sound for predicting recurrence in aggressive disease, means the model is not directly applicable to screening populations where such early lesions are prevalent.
• Treatment and confounding factors: as a retrospective prognostic study, our model does not explicitly account for the potential confounding effects of heterogeneous adjuvant therapies (chemotherapy/targeted therapy). Although data were collected, these variables were not retained in the final model due to their lack of significant association with DFS in our cohort and the model’s preoperative design intent. Furthermore, while pathological stage—a major determinant of adjuvant therapy—was included based on its fundamental clinical relevance, its predictive contribution was attenuated in the multivariable model alongside the stronger imaging and serological markers. Therefore, unmeasured or residual confounding by treatment factors remains a possibility.
• Technical aspects: in this study, all lesions measured by DECT were manually selected and delineated, which may introduce subjectivity and bias. Future studies could incorporate artificial intelligence (AI)-based automatic segmentation to reduce such errors (36).

Actions needed

Future research will focus on: (I) external validation in prospective multi-center cohorts to confirm generalizability; (II) decision impact studies to evaluate the model’s effect on clinical decisions and patient outcomes; (III) integrating AI radiomics with current quantitative parameters to reveal deeper prognostic insights.

Conclusions

In summary, we developed and internally validated a comprehensive nomogram model based on DECT and blood markers that showed high accuracy in predicting postoperative recurrence risk in NSCLC patients. This model has the potential to assist clinicians in identifying high-risk individuals and could inform the refinement of adjuvant therapy and surveillance strategies in the future, pending external validation. Ultimately, it represents a promising tool for improving personalized prognosis.

Acknowledgments

We would like to express our sincere gratitude to Dr. Shuni Zhang from the Department of Radiology for his invaluable guidance on the analysis of imaging parameters in this study. We also extend our appreciation to Dr. Danni Wang and Dr. Qiong Wu from the Department of Pathology for their meticulous verification and calibration of the pathological information of the enrolled patients. Furthermore, we acknowledge the support from the First Affiliated Hospital of Bengbu Medical University for providing the necessary hardware for CT examinations, as well as the Department of Medical Oncology at Bengbu Medical University for facilitating the patient follow-up environment.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2251/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2251/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2251/prf

Funding: This work was supported by Anhui Province “Jianghuai Talent Cultivation Plan Outstanding Project”; Longhu Talent Project of Bengbu Medical University (No. LH250302002 to Z.X.); Anhui Province Medical Imaging Discipline (Specialty) Leading Talent Cultivation Project (No. DTR2024028 to Z.X.); Health Research Project and Provincial Financial Support Key Projects, Anhui Province (No. AHWJ2023A10110 to F.S.); and Key Research Project of Natural Sciences in Universities, Anhui Province (No. 2024AH051286 to F.S.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-aw-2251/coif). Z.X. reports the funding from Longhu Talent Project of Bengbu Medical University (No. LH250302002) and Anhui Province Medical Imaging Discipline (Specialty) Leading Talent Cultivation Project (No. DTR2024028). F.S. reports the funding from Health Research Project and Provincial Financial Support Key Projects, Anhui Province (No. AHWJ2023A10110); and Key Research Project of Natural Sciences in Universities, Anhui Province (No. 2024AH051286). The other 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 the institutional ethics committee of the First Affiliated Hospital of Bengbu Medical University (No. 2023-440), and individual consent for this retrospective analysis was waived.

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. Fan Y, Jiang Y, Gong L, et al. Epidemiological and demographic drivers of lung cancer mortality from 1990 to 2019: results from the global burden of disease study 2019. Front Public Health 2023;11:1054200. [Crossref] [PubMed]
2. Zheng RS, Chen R, Han BF, et al. Cancer incidence and mortality in China, 2022. Zhonghua Zhong Liu Za Zhi 2024;46:221-31. [Crossref] [PubMed]
3. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
4. Garinet S, Wang P, Mansuet-Lupo A, et al. Updated Prognostic Factors in Localized NSCLC. Cancers (Basel) 2022;14:1400. [Crossref] [PubMed]
5. Lazzari C, Spagnolo CC, Ciappina G, et al. Immunotherapy in Early-Stage Non-Small Cell Lung Cancer (NSCLC): Current Evidence and Perspectives. Curr Oncol 2023;30:3684-96. [Crossref] [PubMed]
6. Lee JM, McNamee CJ, Toloza E, et al. Neoadjuvant Targeted Therapy in Resectable NSCLC: Current and Future Perspectives. J Thorac Oncol 2023;18:1458-77. [Crossref] [PubMed]
7. Wu L, Lou X, Kong N, et al. Can quantitative peritumoral CT radiomics features predict the prognosis of patients with non-small cell lung cancer? A systematic review. Eur Radiol 2023;33:2105-17. [Crossref] [PubMed]
8. Schegoleva AA, Khozyainova AA, Fedorov AA, et al. Prognosis of Different Types of Non-Small Cell Lung Cancer Progression: Current State and Perspectives. Cell Physiol Biochem 2021;55:29-48. [Crossref] [PubMed]
9. Zhang H, Zeng J, Li X, et al. The nomogram for the prediction of overall survival after surgery in patients in early-stage NSCLC based on SEER database and external validation cohort. Cancer Med 2024;13:e6751. [Crossref] [PubMed]
10. Wang C, Wu Y, Shao J, et al. Clinicopathological variables influencing overall survival, recurrence and post-recurrence survival in resected stage I non-small-cell lung cancer. BMC Cancer 2020;20:150. [Crossref] [PubMed]
11. Huang T, Sun H, Luo X, et al. Correlation study between flash dual source CT perfusion imaging and regional lymph node metastasis of non-small cell lung cancer. BMC Cancer 2020;20:547. [Crossref] [PubMed]
12. Zhao L, Zhou W, Fu Y, et al. Diagnostic value of one-stop CT energy spectrum and perfusion for angiogenesis in colon and rectum cancer. BMC Med Imaging 2024;24:116. [Crossref] [PubMed]
13. Chen XH, Ren K, Liang P, et al. Spectral computed tomography in advanced gastric cancer: Can iodine concentration non-invasively assess angiogenesis? World J Gastroenterol 2017;23:1666-75. [Crossref] [PubMed]
14. Sun X, Niwa T, Ozawa S, et al. Detecting lymph node metastasis of esophageal cancer on dual-energy computed tomography. Acta Radiol 2022;63:3-10. [Crossref] [PubMed]
15. Li H, Zhang D, Pei J, et al. Dual-energy computed tomography iodine quantification combined with laboratory data for predicting microvascular invasion in hepatocellular carcinoma: a two-centre study. Br J Radiol 2024;97:1467-75. [Crossref] [PubMed]
16. Liu X, Ouyang D, Li H, et al. Papillary thyroid cancer: dual-energy spectral CT quantitative parameters for preoperative diagnosis of metastasis to the cervical lymph nodes. Radiology 2015;275:167-76. [Crossref] [PubMed]
17. Zheng X, Tian H, Li W, et al. The diagnosis value of dual-energy computed tomography (DECT) multi-parameter imaging in lung adenocarcinoma and squamous cell carcinoma. BMC Pulm Med 2024;24:545. [Crossref] [PubMed]
18. Wu Y, Li J, Ding L, et al. Differentiation of pathological subtypes and Ki-67 and TTF-1 expression by dual-energy CT (DECT) volumetric quantitative analysis in non-small cell lung cancer. Cancer Imaging 2024;24:146. [Crossref] [PubMed]
19. Liu Y, Cheng F, Wang L, et al. Quantitative parameters derived from dual-energy computed tomography for the preoperative prediction of early recurrence in patients with esophageal squamous cell carcinoma. Eur Radiol 2023;33:7419-28. [Crossref] [PubMed]
20. Shimizu K, Okita R, Saisho S, et al. Preoperative neutrophil/lymphocyte ratio and prognostic nutritional index predict survival in patients with non-small cell lung cancer. World J Surg Oncol 2015;13:291. [Crossref] [PubMed]
21. Derman BA, Macklis JN, Azeem MS, et al. Relationships between longitudinal neutrophil to lymphocyte ratios, body weight changes, and overall survival in patients with non-small cell lung cancer. BMC Cancer 2017;17:141. [Crossref] [PubMed]
22. Li M, Cui Y, Yan Y, et al. Dual energy CT-derived quantitative parameters and hematological characteristics predict pathological complete response in neoadjuvant chemoradiotherapy esophageal squamous cell carcinoma patients. BMC Gastroenterol 2025;25:357. [Crossref] [PubMed]
23. Zhao S, Su Y, Duan J, et al. Radiomics signature extracted from diffusion-weighted magnetic resonance imaging predicts outcomes in osteosarcoma. J Bone Oncol 2019;19:100263. [Crossref] [PubMed]
24. Gong JW, Zhang Z, Luo TY, et al. Combined model of radiomics, clinical, and imaging features for differentiating focal pneumonia-like lung cancer from pulmonary inflammatory lesions: an exploratory study. BMC Med Imaging 2022;22:98. [Crossref] [PubMed]
25. Marty-Ané CH, Canaud L, Solovei L, et al. Video-assisted thoracoscopic lobectomy: an unavoidable trend? A retrospective single-institution series of 410 cases. Interact Cardiovasc Thorac Surg 2013;17:36-43. [Crossref] [PubMed]
26. Lin L, Cheng J, Tang D, et al. The associations among quantitative spectral CT parameters, Ki-67 expression levels and EGFR mutation status in NSCLC. Sci Rep 2020;10:3436. [Crossref] [PubMed]
27. Yan WQ, Xin YK, Jing Y, et al. Iodine Quantification Using Dual-Energy Computed Tomography for Differentiating Thymic Tumors. J Comput Assist Tomogr 2018;42:873-80. [Crossref] [PubMed]
28. Liu Y, Yuan M, Zhao Z, et al. A quantitative model using multi-parameters in dual-energy CT to preoperatively predict serosal invasion in locally advanced gastric cancer. Insights Imaging 2024;15:264. [Crossref] [PubMed]
29. Zhang H, Zou Y, Tian F, et al. Dual-energy CT may predict post-operative recurrence in early-stage glottic laryngeal cancer: a novel nomogram and risk stratification system. Eur Radiol 2022;32:1921-30. [Crossref] [PubMed]
30. Zheng Y, Han X, Jia X, et al. Dual-energy CT-based radiomics for predicting invasiveness of lung adenocarcinoma appearing as ground-glass nodules. Front Oncol 2023;13:1208758. [Crossref] [PubMed]
31. González-Pérez V, Arana E, Barrios M, et al. Differentiation of benign and malignant lung lesions: Dual-Energy Computed Tomography findings. Eur J Radiol 2016;85:1765-72. [Crossref] [PubMed]
32. Bar-Ad V, Palmer J, Li L, et al. Neutrophil to lymphocyte ratio associated with prognosis of lung cancer. Clin Transl Oncol 2017;19:711-7. [Crossref] [PubMed]
33. Zahorec R. Neutrophil-to-lymphocyte ratio, past, present and future perspectives. Bratisl Lek Listy 2021;122:474-88. [Crossref] [PubMed]
34. Huang S, Liu S, Tan F, et al. Construction and validation of a risk nomogram model for colorectal sessile serrated lesions. J Int Med Res 2025;53:3000605251337577. [Crossref] [PubMed]
35. Pei W, Wang C, Wei Y, et al. Dual-energy CT-based nomogram for predicting progression-free survival in locally advanced nasopharyngeal carcinoma. Front Oncol 2025;15:1698927. [Crossref] [PubMed]
36. Brendlin AS, Peisen F, Almansour H, et al. A Machine learning model trained on dual-energy CT radiomics significantly improves immunotherapy response prediction for patients with stage IV melanoma. J Immunother Cancer 2021;9:e003261. [Crossref] [PubMed]