Development and validation of models based on clinical and CT features: multivariate analysis for predicting vascular invasion in non-small cell lung cancer

Introduction

Lung cancer is a topic of great concern worldwide, with the highest incidence and mortality rate of all malignant tumors and one of the leading causes of cancer deaths (1,2). Non-small cell lung cancer (NSCLC) is the most common histopathologic subtype of lung cancer, contributing up to 80–85% of cases, with a five-year survival rate of 58–80% (2-4). Lymphovascular invasion (LVI) is one of the adverse factors related to tumor recurrence, metastasis, and poor prognosis of patients (5,6). The diagnosis of LVI mainly relies on histopathological testing. LVI is defined as tumor cell infiltration in the lumen of arteries, veins, and/or lymphatic vessels identified by hematoxylin and eosin (H&E) staining or elastic staining and immunohistochemistry, but this process is prone to problems such as sample collection limitations and the complexity of histological differential diagnosis (7). There is an urgent need to develop effective preoperative assessment methods.

At present, the prediction methods of non-invasive LVI include radiomics models, 18F-fluorodeoxyglucose (¹⁸F-FDG) positron emission tomography (PET)-computed tomography (CT) multimodal imaging technology, biomarker detection, and so on. However, radiomics models are greatly affected by parameters and their reproducibility is limited. ¹⁸F-FDG PET/CT has radiation exposure and cost limitations. Although biomarker detection is a minimally invasive examination, its sensitivity is insufficient and its clinical promotion is limited (8-10).

CT of the chest is a common and non-invasive diagnostic tool that is commonly used for the diagnosis of lung cancer (1,11). With the development of low-dose chest CT technology and the enhancement of public health awareness, CT scanning has become one of the most popular imaging techniques, which not only helps in the early screening and diagnosis of lung cancer but also provides important support for the long-term health management of patients. Quantitative imaging of CT examination can clearly show the disease characteristics and anatomical structure such as spiculation sign, vacuole sign, and solid component proportion, and can evaluate the anatomical basis of vascular invasion (8,12). Clinical characteristics such as smoking history and tumor markers can reflect the biological characteristics of tumors, with both types of information being highly accessible and complementary. Therefore, we combined CT and clinical features to construct a multidimensional prediction model.

The aim of this study was to develop and validate a predictive model for vascular tumor thrombosis in NSCLC based on CT imaging features and clinically relevant features, and to verify the efficacy of the model through an experimental group to assist clinical identification of high-risk patients and assist pathological diagnosis. We present this article in accordance with the TRIPOD+AI reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-24-1886/rc).

Methods

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study had a multicenter retrospective design and was approved by the Ethics Committees of Tongde Hospital of Zhejiang Province (No. 2022-029-JY, center 1), Anqing Municipal Hospital (No. 83230471, center 2), and Taizhou Municipal Hospital (No. LWYJ2023059, center 3). The need for written informed consent was waived because of the retrospective observational nature of the research.

Patients

A retrospective cohort study was conducted using the hospital electronic medical record system. A total of 2,830 lung cancer patients from centers 1 and 2 from January 2015 to December 2023 were divided into a training group and a validation group according to a 7:3 ratio; further, 275 lung cancer patients from center 3 from January to December 2024 were selected as an external validation group.

The inclusion criteria for this study were as follows (Figure 1): (I) confirmation of malignant tumor through postoperative pathology; (II) patients who underwent a chest CT scan within one month before the operation; (III) patients with complete clinical and pathological data; (IV) patients with disease intervention who did not receive neoadjuvant chemotherapy or radiotherapy before surgery.

Figure 1 Flowchart of patient selection and demographics. (A,D) Joining process diagram for the training, validation, and external validation sets. (B) The numerical statistics between histopathological types and different types of CTR, as well as the average of histopathological types and different types of CTR separately. (C) The statistics with and without LVI in different histopathological types and the variability thereof. Center 1, Tongde Hospital of Zhejiang Province; center 2, Anqing Municipal Hospital; center 3, Taizhou Municipal Hospital. *, P<0.05; ****, P<0.0001. AAH, atypical adenomatous hyperplasia; AIS, adenocarcinoma in situ; ASC, adenosquamous carcinoma; CT, computed tomography; CTR, consolidation/tumor ratio of mixed ground-glass nodule; IA, invasive adenocarcinoma; LUSC, lung squamous cell carcinoma; LVI−, non-lymphovascular invasion; LVI+, lymphovascular invasion; MIA, minimally invasive adenocarcinoma; MPA, mucin producing adenocarcinoma; SCLC, small cell lung cancer.

The exclusion criteria were as follows: (I) the histopathological diagnosis was pulmonary metastasis tumor, atypical adenomatous hyperplasia (AAH)/adenocarcinoma in situ (AIS), minimally invasive adenocarcinoma (MIA), or small cell lung cancer (SCLC); (II) a time interval exceeding one month between imaging examination and pathological results; (III) clinical data could not be found or were incomplete; (IV) CT image quality was unqualified, such as motion artifacts, metal artifacts, and so on.

CT examination

The CT images were acquired using multiple CT scanners, including the Optima CT 680 Quantum (GE Healthcare, Chicago, IL, USA), SOMATOM Definition Flash (Siemens, Erlangen, Germany), LightSpeed 16 (GE Healthcare), and Incisive CT Power (Philips, Amsterdam, Netherlands). All images were reconstructed with a slice thickness ranging from 0.625 to 2 mm. With the patient in the supine position, the scan range was from the apex to the base of the lung, and the scan was taken at the end of inspiration.

The scanning parameters used were as follows: tube voltage, 120 kV; tube current, 250–300 mA; matrix, 512×512; collimation, 0.6 or 0.625 mm; lung window width was set to 1,500 Hounsfield units (HU) with a window level of −600 HU; and the width of the mediastinum window was set to 400 HU with a window level of 40 HU.

Clinical and histopathological data

The electronic medical records of the study population were searched and the following features were recorded: sex, age, smoking status, clinical basic disease hypertension (HBP), clinical basic disease diabetes mellitus (DM), and laboratory tumor marker [including neuron-specific enolase (NSE), squamous epithelial cell carcinoma antigen (SCC), carcinoembryonic antigen (CEA), and cytokeratin-19 fragment (CYFRA21-1) and other abnormal indicators].

The diagnosis of LVI was made by double-blind evaluation. Two experienced pathologists (with 12 and 15 years of experience in pathology, respectively) evaluated the surgically resected NSCLC and determined the presence of LVI sections. All sections were stained with H&E, and the intravascular tumor cell mass was evaluated by two pathologists independently. Lymphatic/vascular invasion was confirmed by immunohistochemical staining in suspected cases. Cases of disagreement were reviewed by a third senior pathologist (20 years of experience) to reach a consensus. Three-level criteria were used: clear tumor emboli (positive), suspicious lesions (uncertain), and no evidence (negative).

CT image data acquisition

Blinded to the clinical and histopathological results, two thoracic radiologists (5 and 18 years of experience in chest imaging diagnosis, respectively) reviewed every image adhering to the inclusion and exclusion criteria of lung and mediastinal window characteristics. Any disagreement in describing semantic features was resolved by a consensus read.

Nodule types were classified visually into three subgroups: pure ground-glass nodule (pGGN: no solid component), mixed ground-glass nodule (mGGN), and solid nodule (SN). mGGN consolidation/tumor ratio (CTR) was calculated as follows: maximum slice of lesion, proportion of solid components in mGGN = (maximum diameter of solid component/maximum diameter of the ground-glass component) ×100%.

CT image evaluation features included the following: (I) location of lesion; (II) mGGN CTR: Type 1 (0%), Type 2 (<25%), Type 3 (<50%), Type 4 (<75%), Type 5 (<100%), and Type 6 (100%); (III) dimensions: long diameter of the largest cross section of the lesion (LD), short diameter (SD), ratio of long and short diameters (LD/SD); (IV) CT value: within the lesion, avoiding the areas of blood vessels, calcification, hemorrhage, and cystic degeneration, draw the region of interest (ROI) to measure the maximum/minimum/mean/standard deviation of CT attenuation values; (V) shape: round (LD/SD ≤1.2), oval (LD/SD >1.2), irregular; (VI) lobulation; (VII) spiculation: short spiculation (spiculate length <5 mm), long spiculation (spiculate length <5 mm); (VIII) vacuole sign: Type 1 (none), Type 2 (size of vacuoles <5 mm), Type 3 (size of vacuoles ≥5 mm), cavity and pulmonary diseases that require evaluation including emphysema, bullae, ventilation-perfusion imbalance, interstitial lung disease, bronchiectasis, or air bronchogram.

Statistical methods

All statistical analyses were performed using the software SPSS 26.0.0 (IBM Corp., Armonk, NY, USA) and GraphPad Prism version 9.0.0 (GraphPad Software, San Diego, CA, USA). In qualitative data, the chi-square test and Fisher’s exact probability test were used to compare differences. In the quantitative data, Kolmogorov-Smirnov test was used to determine the normality of the data. Quantitative data conforming to a normal distribution were expressed as means ± standard deviations, and the differences were compared by independent sample t-test; the quantitative data that did not conform to a normal distribution were expressed as median and quartile, and the Mann-Whitney U-test was used to compare the differences. In univariate variables, statistically significant risk factors with P<0.05 were analyzed and included in binary logistic regression, with P<0.05 indicating statistical significance, and confounding variables were removed to screen out independent risk factors to draw the forest map. The receiver operating characteristic (ROC) curve was used to evaluate the effectiveness of each model, and the area under the ROC curve (AUC) was calculated. The sensitivity and specificity corresponding to the best Youden index of the output results were used to evaluate the prediction accuracy of the model, and the Hosmer-Lemeshow goodness-of-fit test in binary logistic regression was used to evaluate the calibration degree of the model to judge the calibration effect of the model. A nomogram was created by assigning a score to each independent risk factor using R statistical software (The R Foundation for Statistical Computing, Vienna, Austria). The total benefit to the population was assessed using the clinical impact curve (CIC).

Results

Clinical and histopathological characteristics

The training set included 1,981 patients with NSCLC, 1,866 (94.1%) of whom did not have LVI (LVI−; Figure 2A,2B) (1,104 women, 762 men; mean age 61.15±10.90 years) and 115 of whom had LVI (LVI+; Figure 2C,2D) (53 women, 62 men; mean age 62.25±8.91 years). In the univariate analysis, there was no significant difference in the distribution of age between the LVI+ group and the LVI− group (P>0.05); however, there were significant differences in gender (P=0.006), smoking, HBP, DM, and laboratory tumor markers (all P<0.001). After multivariate binary logistic regression analysis, smoking status, DM, and laboratory tumor markers were identified as independent risk factors for LVI+ (all P<0.05). The clinicopathological characteristics of the patients in the training set, validation set, and external validation set are shown in Table 1.

Figure 2 Representative CT and pathological images of LVI− and LVI+ cases. (A,B) Images of a 65-year-old man with histopathologically confirmed LVI−. (A) The axial CT images show an irregular solid nodule (long diameter of 23 mm, short diameter of 16 mm) located in the right upper lung. The nodule has superficial lobulation and short spiculation at the margin, and a border sign is seen within it. (B) H&E staining image shows no cancerous embolism in the blood vessel (green dashed line) (×100). (C,D) Images of a 78-year-old man with histopathologically confirmed LVI. (C) The axial CT images show an irregular solid nodule (long diameter of 22 mm, short diameter of 13 mm) located in the right upper lung. The nodule has superficial lobulation and short spiculation at the margin, and a border sign is seen within it. The nodule is in indirect contact with the pleura, which is pulled by a pleural tag without displacement. (D) H&E staining image shows cancerous emboli in the vasculature (green dotted line) (×100). CT, computed tomography; H&E, hematoxylin and eosin; LVI−, non-lymphovascular invasion; LVI+, lymphovascular invasion.

CT imaging findings

The univariate analysis of CT findings showed no differences in terms of ventilation-perfusion imbalance, interstitial lung disease, bronchiectasis, multiple lung comorbidity, shape, calcification, and LD/SD (all P>0.05); however, there were significant differences in location (P=0.002), air bronchogram (P=0.005), emphysema/bullae, CTR, lobulation, spiculation, vacuole sign, border sign, LD, SD, CT_max, CT_min, CT_mean, and CT_sd (all P<0.001). After multivariate binary logistic regression and removal of confounding factors, CTR and vacuole sign (both P<0.001) were shown to be independent risk factors for LVI+. The CT imaging features of the patients in the training set, validation set, and external validation set are shown in Tables 2,3.

Performance of the model

According to the independent risk factors such as smoking, DM, laboratory tumor markers, CTR, and vacuole sign the model is established and the forest plot is drawn in Figure 3. The ROC curves of the training, validation, and external validation sets are plotted in Figure 4, the calibration curves are plotted in Figure 5, and the CIC curves are plotted in Figure 6. In the training set, the model AUC, accuracy, sensitivity, and specificity were 0.836 [95% confidence interval (CI): 0.806–0.867], 65.2%, 92.1%, and 63.5%, respectively; in the validation set, the model AUC, accuracy, sensitivity, and specificity were 0.803 (95% CI: 0.755–0.852), 71.6%, 82.7%, and 70.9%, respectively; and in the external validation set, the model AUC, accuracy, sensitivity, and specificity were 0.845 (0.775–0.916), 70.9%, 87.8%, and 68.6%, respectively. The model performance is shown in Table 4. The calibration curves of the training set, validation set, and external validation set models are drawn by the x-axis representing the predicted probability and the Y-axis representing the actual probability, the long diagonal dashed line representing the perfect prediction curve of the ideal model, and the short-dashed line representing the actual curve of the model, of which closer fit to the long diagonal dashed line represents a better prediction. The Hosmer-Lemeshow goodness-of-fit test showed χ²=9.765 and P=0.282 for the training set, and χ²=3.882 and P=0.868 for the validation set, indicating that the model was well calibrated. The negative predictive values of the training set, validation set, and external validation set were 99.2%, 98.4%, and 97.6%, respectively (Figure 7). The nomogram was constructed to determine the risk classification for each patient (Figure 8). The value of each variable was located on the corresponding axis, and a line was drawn upwards to determine the corresponding score. All scores were summed up to a line drawn down to the survival axis to determine the probability of LVI of NSCLC in an individual patient.

Figure 3 Forest plots of independent risk factors in the training (A) and validation groups (B). CI, confidence interval; CTR, consolidation/tumor ratio of mixed ground-glass nodule; DM, diabetes mellitus; OR, odds ratio.

Figure 4 AUCs in the training, validation, and ex-validation groups. AUC, area under the curve.

Figure 5 Calibration curves of the prediction model in the training (A), validation (B), and ex-validation groups (C). The x-axis represents the predicted risk, and the y-axis represents the actual probability of LVI in NSCLC; the blue dashed line represents the ideal curve, and the pink dashed line represents the actual curve. LVI, lymphovascular invasion; NSCLC, non-small cell lung cancer.

Figure 6 Clinical impact curve.

Figure 7 Confusion matrices of the prediction models. LVI−, non-lymphovascular invasion; LVI+, lymphovascular invasion.

Figure 8 Nomogram of the probability of LVI in patients with NSCLC. CTR, consolidation/tumor ratio of mixed ground-glass nodule; DM, diabetes mellitus; LVI, lymphovascular invasion; NSCLC, non-small cell lung cancer.

Discussion

In reviewing the literature, we found that previous studies of cancer LVI have focused on gastric cancer (13), breast cancer (14), esophageal squamous cell carcinoma (15), and colorectal cancer (16); as well as studying the prognosis, survival, and risk of recurrence in NSCLC, there are fewer predictive diagnostics regarding preoperative NSCLC.

Accordingly, in the present study, we constructed and validated a model for predicting LVI in lung cancer based on clinical and CT images features, which yielded AUCs of 0.836 and 0.803 for the training and validation sets, respectively, which were higher than those of the utilized conventional model of Nie et al. (9) (AUC of 0.818 and 0.800 for the training and test sets, respectively), and had a higher predictive value and clinical calibration ability.

One of the major causes of lung cancer-related deaths is metastasis, which is a complex process closely related to LVI. Studies have shown that LVI is a high-risk pathological feature of NSCLC, affecting lymph node metastasis (LNM) and tumor lymph node metastasis stage (TNM), as well as overall survival (OS) and recurrence-free survival (RFS), with LVI+ patients having lower OS compared with LVI− patients (17); TNM is an important factor influencing the clinical decision-making (18), which also emphasizes the importance of preoperative prediction of LVI in NSCLC.

In this study, smoking status was found to be an independent risk factor for LVI in NSCLC, which affects the incidence and survival of patients with lung cancer, as reported in a previous article (1). Current and previous history of smoking have always been high-risk factors for lung cancer because tobacco smoke contains more than 7,000 compounds, many of which can increase the risk of lung cancer. In addition, smoking and niacin share an important link—smoking will interfere with the normal metabolism of niacin, and one study has shown that niacin end metabolites can lead to vascular endothelial cell damage, increase the risk of vascular inflammation, exacerbate thrombosis, and increase the risk of LVI, thus affecting morbidity and survival (19).

LVI affects the survival of lung cancer patients and is a poor prognostic factor. The study by Hernandez et al. (20), among others, showed that lung cancer patients with comorbid DM have decreased survival and prognostic outcomes, which may also be related to the fact that DM can exacerbate vascular inflammation as well as thrombosis. The hyperglycemia environment in diabetic patients promotes inflammation and triggers the secretion of vascular endothelial growth factor, which promotes tumor angiogenesis and progression. These changes make the metabolism of NSCLC tumor cells become dysregulated and adapt to the hyperglycemia microenvironment by metabolic reprogramming, which may accelerate the formation of intravascular tumor thrombus (19,21,22). The present study analyzed the effect of underlying diseases on LVI separately, unlike that of Yang et al. (11), who combined the underlying diseases into one analysis; separate analyses are more conducive to clinical precision on the extent of disease impact on LVI.

Abnormalities in tumor markers are associated with the stage and prognosis of lung cancer and can increase the sensitivity of lung cancer diagnosis. Tumor markers are biological substances produced by tumor cells or by the body in response to tumor tissue. The levels of these substances fluctuate abnormally during tumor development due to the genetic activity of the tumor cells. Using the immune response properties of these markers, we can identify and differentiate between different types of tumors and monitor the state of tumor activity (9,23). In this study, we analyzed NSE, SCC, CEA, and CYFRA21-1, which are highly correlated with lung cancer. Abnormalities in these indices are usually related to the expression of glycoproteins, lipoproteins, and other substances that are produced by lung cancer in the process of development; therefore, we recorded analysis of NSE/SCC/CEA/CYFRA21-1 and other abnormal tumor indicators. We believe that serum abnormalities can, to some extent, indicate LVI of the tumor. Abnormalities in SCC and CEA are commonly seen in patients with lung squamous cell carcinoma. The SCC component is more prevalent in the cytoplasm of squamous cells; although it was first isolated from squamous cell carcinoma tissue of the cervix, it has since been found in cancers of the pharynx and other locations. Although not as pronounced as in cervical carcinomas, elevated levels of SCC can help in the detection of NSCLC (23). NSE and CYFRA21-1 have important value in the diagnosis and treatment of lung cancer.

Another important predictor associated with LVI in NSCLC is CTR, and in the present study, the more solid components in patients presenting LVI+, the greater the probability of disease (type 6 in 114 out of 167 cases, or 68.2%, and type 5 in 26 out of 167, or 15.5%), which is well explained by the fact that the degree of the lesion is positively correlated with the proportion of solid components and that the solid component is considered to be an infiltrative lesion and is also considered to be one of the indications for surgery. Additionally, the solid component of mGGN was more prone to and had a higher mutation rate of EGFR, one of the lung adenocarcinoma genes (24,25); this also suggests that LVI is a poor prognostic factor for LVI in NSCLC.

The vacuole sign is defined as a low-density area with a lesion diameter <5 mm, which is an important sign of early lung cancer and malignant ground-glass opacity (24) is also more likely to occur in squamous cell carcinoma (23). Tumor spread through air spaces (STAS) is an aggressive mode of lung cancer spreading that occurs when tumor cells or cell clusters spread along the alveolar walls in the alveolar lumen beyond the margins of the main tumor. STAS is associated with poor prognosis of lung cancer and also plays an important role in lung cancer staging. The diagnosis of STAS mainly relies on pathological tissue examination and its preoperative identification is complicated. There are limitations to intraoperative frozen section analysis; however, we can observe it with the aid of the sign of vacuole sign, which is closely related to STAS. Studies have shown that the vacuole sign is an independent risk factor for NSCLC STAS, which is one of the high-risk factors for recurrent metastasis of lung cancer (26,27), which is also consistent with the findings of this study.

The diagnosis of LVI relies primarily on pathologic tissue results, and there are difficulties in the diagnostic process. In addition to the complexity of the disease’s progression, the compression of tissues during surgery and the preparation of pathologic sections, such as the atrophy of alveolar structures and the increase in the number of alveoli after cryosections, lead to difficulty in achieving accurate diagnostic results. In this case, the pathologist can be alerted to the likelihood of the presence of the disease and improve the pathologic diagnosis by incorporating high-risk factors from the research results. Furthermore, the diagnosis of LVI in combination with high-risk factors can inform clinical treatment. Although the current treatment is mainly focused on surgical resection, the combination of neoadjuvant chemotherapy can improve the prognosis for patients. Neoadjuvant chemotherapy can benefit LVI+ patients preoperatively, and neoadjuvant therapy can reduce tumor size and stage, decrease the risk of recurrence of lung cancer, improve survival, and even, in some patients, enhance OS compared with surgical-only treatments.

This study has some limitations: this is a retrospective study, so there were some inevitable selection biases. There is an urgent need for prospective studies that are strictly adapted. Second, the low positive prediction rate of the confusion matrix in this study may be due to the large gap between the positive and negative sample sizes in the sample. Finally, continuous variables such as LD, SD, CT_max, CT_min, CT_mean, and CT_sd were extracted from CT images, and none of them were ultimately included as independent risk factors for predicting performance after binary logistic regression analyses, which may be related to the artificially selected ROIs for extracting image features. Therefore, in our future research, we will introduce a deep learning-based multimodal method, which is a longitudinal multimodal Transformer model that integrates multi-dimensional data such as imaging data [such as CT, magnetic resonance imaging (MRI), ultrasound, etc.], clinical data (such as symptoms, medical history, laboratory test results, etc.) from electronic medical records, genomic data, and so on (28). We will endeavor to explore the correlation between different modal data, make up for the deficiency of a single modal model, improve the performance and discrimination of the model, and improve the accuracy of disease prediction, so as to better assist clinical practice and help more patients. In conclusion, this study constructed and validated a model to predict LVI in NSCLC by combining clinical indicators and CT image features, a non-invasive examination method, which can reduce the trauma of patients, provide an effective auxiliary diagnostic tool for the clinic, facilitate its preoperative management, effectively improve the pathological diagnosis of lung cancer, and provide an important reference for the clinical treatment.

Conclusions

Our research shows that the model established by combining features such as smoking history, diabetes history, tumor markers, CTR, and vacuole signs can effectively predict LVI in NSCLC and provide an important reference for pathological diagnosis and clinical treatment.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD+AI reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-24-1886/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-24-1886/dss

Funding: This research was supported by the Medical Health Science and Technology Project of Zhejiang Province, China (Grant No. 2022KY702 to J.W.), Scientific Research Project of Anhui Higher Education Institutions, China (Grant No. 2023AH050608 to H.S.), and Key Research and Development Program of Anhui Province, China (Grant No. 2022e07020033 to Z.X.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-24-1886/coif). This research was supported by the Medical Health Science and Technology Project of Zhejiang Province, China (Grant No. 2022KY702 to J.W.), Scientific research project of Anhui Higher Education Institutions, China (Grant No. 2023AH050608 to H.S.) and Key Research and Development Program of Anhui Province, China (Grant No. 2022e07020033 to Z.X.). 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. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments and was approved by the Ethics Committees in Clinical Research of Tongde Hospital of Zhejiang Province (No. 2022029-JY), Anqing Municipal Hospital (No. 83230471), and Taizhou Municipal Hospital (No. LWYJ2023059). The requirement for individual consent was waived due to the retrospective nature of the analysis.

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

References

1. Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, Jemal A. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2024;74:229-63. [Crossref] [PubMed]
2. Thai AA, Solomon BJ, Sequist LV, Gainor JF, Heist RS. Lung cancer. Lancet 2021;398:535-54. [Crossref] [PubMed]
3. Altorki N, Wang X, Kozono D, Watt C, Landrenau R, Wigle D, et al. Lobar or Sublobar Resection for Peripheral Stage IA Non-Small-Cell Lung Cancer. N Engl J Med 2023;388:489-98. [Crossref] [PubMed]
4. Yang G, Nie P, Zhao L, Guo J, Xue W, Yan L, Cui J, Wang Z. 2D and 3D texture analysis to predict lymphovascular invasion in lung adenocarcinoma. Eur J Radiol 2020;129:109111. [Crossref] [PubMed]
5. Choe J, Kim MY, Yun JK, Lee GD, Kim YH, Choi S, Kim DK. Sublobar Resection in Stage IA Non-Small Cell Lung Cancer: Role of Preoperative CT Features in Predicting Pathologic Lymphovascular Invasion and Postoperative Recurrence. AJR Am J Roentgenol 2021;217:871-81. [Crossref] [PubMed]
6. Wang Y, Qian F, Hu M, Chen Y, Yang Z, Han B. Clinical significance of visceral pleural and lymphovascular invasion in surgically resected adenosquamous lung cancer. Eur J Cardiothorac Surg 2021;59:617-23. [Crossref] [PubMed]
7. Li C, Tian Y, Shen Y, Wen B, He Y. Utility of Volumetric Metabolic Parameters on Preoperative FDG PET/CT for Predicting Tumor Lymphovascular Invasion in Non-Small Cell Lung Cancer. AJR Am J Roentgenol 2021;217:1433-43. [Crossref] [PubMed]
8. Wang J, Yang Y, Xie Z, Mao G, Gao C, Niu Z, Ji H, He L, Zhu X, Shi H, Xu M. Predicting Lymphovascular Invasion in Non-small Cell Lung Cancer Using Deep Convolutional Neural Networks on Preoperative Chest CT. Acad Radiol 2024;31:5237-47. [Crossref] [PubMed]
9. Nie P, Yang G, Wang N, Yan L, Miao W, Duan Y, Wang Y, Gong A, Zhao Y, Wu J, Zhang C, Wang M, Cui J, Yu M, Li D, Sun Y, Wang Y, Wang Z. Additional value of metabolic parameters to PET/CT-based radiomics nomogram in predicting lymphovascular invasion and outcome in lung adenocarcinoma. Eur J Nucl Med Mol Imaging 2021;48:217-30. [Crossref] [PubMed]
10. Kinoshita T, Ohtsuka T, Yotsukura M, Asakura K, Goto T, Kamiyama I, Otake S, Tajima A, Emoto K, Hayashi Y, Kohno M. Prognostic impact of preoperative tumor marker levels and lymphovascular invasion in pathological stage I adenocarcinoma and squamous cell carcinoma of the lung. J Thorac Oncol 2015;10:619-28. [Crossref] [PubMed]
11. Yang Y, Xie Z, Hu H, Yang G, Zhu X, Yang D, Niu Z, Mao G, Shao M, Wang J. Using CT imaging features to predict visceral pleural invasion of non-small-cell lung cancer. Clin Radiol 2023;78:e909-17. [Crossref] [PubMed]
12. Wei SH, Zhang JM, Shi B, Gao F, Zhang ZX, Qian LT. The value of CT radiomics features to predict visceral pleural invasion in ≤3 cm peripheral type early non-small cell lung cancer. J Xray Sci Technol 2022;30:1115-26. [Crossref] [PubMed]
13. Sun B, Li H, Gu X, Cai H. Prognostic Implication of Lymphovascular Invasion in Early Gastric Cancer Meeting Endoscopic Submucosal Dissection Criteria: Insights from Radical Surgery Outcomes. Cancers (Basel) 2024;16:979. [Crossref] [PubMed]
14. Kuhn E, Gambini D, Despini L, Asnaghi D, Runza L, Ferrero S. Updates on Lymphovascular Invasion in Breast Cancer. Biomedicines 2023;11:968. [Crossref] [PubMed]
15. Tao Y, Chen S, Yu J, Shen Q, Ruan R, Wang S. Risk factors of lymph node metastasis or lymphovascular invasion for superficial esophageal squamous cell carcinoma: A practical and effective predictive nomogram based on a cancer hospital data. Front Med (Lausanne) 2022;9:1038097. [Crossref] [PubMed]
16. Lin Z, Zheng Y, Yang J, Jin W, Wang J, Wang W, Li S. Prognostic Analysis of Lymphovascular Invasion in Stages I-III Colorectal Cancer: A Retrospective Study Based on Propensity Score Match. Am J Clin Oncol 2023;46:366-73. [Crossref] [PubMed]
17. Yang Z, Li X, Bai J, Li D, Ma Z, Deng C, Fu F, Zhang Y. Prognostic Factors for Survival of Stage IB Non-small Cell Lung Cancer Patients: A 10-Year Follow-Up Retrospective Study. Ann Surg Oncol 2023;30:7481-91. [Crossref] [PubMed]
18. Cai JS, Wang X, Yang F, Li Y, Qiu MT. Lymphovascular invasion: A non-sized T descriptor for stage IA non-small cell lung cancer. Thorac Cancer 2022;13:2413-20. [Crossref] [PubMed]
19. Wen T, Zhang X, Gao Y, Tian H, Fan L, Yang P. SOX4-BMI1 axis promotes non-small cell lung cancer progression and facilitates angiogenesis by suppressing ZNF24. Cell Death Dis 2024;15:698. [Crossref] [PubMed]
20. Hernandez D, Cheng CY, Hernandez-Villafuerte K, Schlander M. Survival and comorbidities in lung cancer patients: Evidence from administrative claims data in Germany. Oncol Res 2022;30:173-85. [Crossref] [PubMed]
21. Lyu X, Wang Y, Xu Y, Zhao Z, Liu H, Hu Z. Metabolomic Profiling of Tumor Tissues Unveils Metabolic Shifts in Non-Small Cell Lung Cancer Patients with Concurrent Diabetes Mellitus. J Proteome Res 2024;23:3746-53. [Crossref] [PubMed]
22. Varlotto J, Medford-Davis LN, Recht A, Flickinger J, Schaefer E, Shelkey J, Lazar M, Campbell D, Nikolov M, Decamp MM. Confirmation of the role of diabetes in the local recurrence of surgically resected non-small cell lung cancer. Lung Cancer 2012;75:381-90. [Crossref] [PubMed]
23. Wu LH, Chen L, Wang QY, Wang YT. Correlation between HRCT signs and levels of CA125, SCCA, and NSE for different pathological types of lung cancer. Eur Rev Med Pharmacol Sci 2023;27:4162-8. [Crossref] [PubMed]
24. Cai Y, Chen T, Zhang S, Tan M, Wang J. Correlation exploration among CT imaging, pathology and genotype of pulmonary ground-glass opacity. J Cell Mol Med 2023;27:2021-31. [Crossref] [PubMed]
25. Park S, Lee SM, Choe J, Choi S, Do KH, Seo JB. Recurrence Patterns and Patient Outcomes in Resected Lung Adenocarcinoma Differ according to Ground-Glass Opacity at CT. Radiology 2023;307:e222422. [Crossref] [PubMed]
26. Yoshida C, Kadota K, Ikeda T, Ibuki E, Go T, Haba R, Yokomise H. Tumor-associated macrophage infiltration is associated with a higher rate of tumor spread through air spaces in resected lung adenocarcinomas. Lung Cancer 2021;158:91-6. [Crossref] [PubMed]
27. Travis WD, Eisele M, Nishimura KK, Aly RG, Bertoglio P, Chou TY, et al. The International Association for the Study of Lung Cancer (IASLC) Staging Project for Lung Cancer: Recommendation to Introduce Spread Through Air Spaces as a Histologic Descriptor in the Ninth Edition of the TNM Classification of Lung Cancer. Analysis of 4061 Pathologic Stage I NSCLC. J Thorac Oncol 2024;19:1028-51.
28. Li TZ, Still JM, Xu K, Lee HH, Cai LY, Krishnan AR, Gao R, Khan MS, Antic S, Kammer M, Sandler KL, Maldonado F, Landman BA, Lasko TA. Longitudinal Multimodal Transformer Integrating Imaging and Latent Clinical Signatures From Routine EHRs for Pulmonary Nodule Classification. Med Image Comput Comput Assist Interv 2023;14221:649-59. [Crossref] [PubMed]