Development of a risk scoring model for predicting visceral pleural invasion in clinical stage T1N0M0 lung adenocarcinoma.

Introduction

Background
Lung cancer remains one of the leading causes of cancer-related mortality in the United States, with non-small cell lung cancer (NSCLC) being the most common subtype (1,2). Accurate clinical staging of lung cancer is essential for determining the appropriate treatment strategy and predicting patient outcomes. In this context, visceral pleural invasion (VPI) has emerged as a significant factor that substantially impacts the staging and prognosis of early lung cancer. Additionally, VPI is critical for surgical planning, as recent studies have linked VPI to the occurrence of skip N2 (3-8).

Rationale and knowledge gap
VPI is defined as the extension of the tumor beyond the elastic layer of the visceral pleura, indicating a higher degree of invasiveness. The presence of VPI in tumors upstages clinical T1N0M0 lung adenocarcinoma to a higher stage, reflecting its association with poorer outcomes and an increased likelihood of recurrence. The eighth edition of the tumor-node-metastasis classification system recognizes the importance of VPI, classifying it as a T2 descriptor even for tumors smaller than 3 cm (3). This reclassification highlights the necessity for reliable preoperative and intraoperative assessment methods to accurately identify VPI.
Despite its clinical significance, the preoperative and intraoperative diagnosis of VPI remains challenging (9-12). Traditional imaging techniques, such as computed tomography (CT) and positron emission tomography (PET), provide valuable information but often fail to reliably predict VPI. Studies have attempted to predict VPI using features such as pleural contact, pleural tags, and pleural retraction observed on CT scans; however, the accuracy of these predictions varies due to the subjective nature of these factors (9,11-13). Consequently, more precise biomarkers and diagnostic tools are required to improve the preoperative and intraoperative assessment of VPI (9-12).

Objective
In this context, we confirmed in a previous study that pleural carcinoembryonic antigen (pCEA), maximum standardized uptake value (SUVmax) on PET/CT, and the consolidation/tumor ratio are significant predictors of VPI in patients with pleural contact on CT (14). This study utilized prior findings to develop a more intuitive risk score model for predicting VPI, which was subsequently validated in patients with clinical stage I adenocarcinoma. The objective was to enhance clinical decision-making to facilitate personalized surgical approaches that minimize the risk of recurrence and optimize patient outcomes. By validating the association between VPI and these factors, this study sought to provide clinicians with robust and reliable indicators to guide surgical planning and improve the prognosis for patients with early-stage lung adenocarcinoma. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1-2630/rc).

Methods

Patients
This study retrospectively analyzed the medical records of patients who underwent surgical treatment for NSCLC at a single tertiary hospital (Seoul St. Mary’s Hospital, College of Medicine, The Catholic University of Korea) between January 2017 and June 2024 and had intraoperative pCEA collected. The exclusion criteria were suspected nodal metastasis at the preoperative stage, tumor size greater than 3 cm on preoperative CT, pathologically confirmed non-adenocarcinoma, and non-pleural contact on preoperative CT (Figure 1). Patients who underwent surgery from January 2017 to December 2022 constituted the development cohort, whereas those who underwent surgery from January 2023 to June 2024 formed the validation cohort.

Method for intraoperative pCEA sampling
As detailed in previous studies, sampling for pCEA was performed in patients with at least 2 mL of pleural effusion identified following creation of camera and working port incisions for multiport VATS (14). pCEA sampling was primarily conducted below the inferior pulmonary ligament, near the costodiaphragmatic recess, within the azygo-esophageal recess, or in the dependent portion around the aortic arch (Figure 2). Sampling was not performed when contamination was suspected due to bleeding during port creation or when adhesions made effusion sampling difficult immediately after port creation. The pCEA levels were measured using an electrochemiluminescence immunoassay with the Elecsys CEA kit (Roche Diagnostics, Penzberg, Germany). Results were typically available within approximately 30 minutes after sampling. Therefore, the pCEA value could be utilized as a reference for surgical planning.

Definition of contact on CT
All patients underwent preoperative CT imaging, with or without contrast enhancement. The CT images were reconstructed with a slice thickness of 1–3 mm in the axial, coronal, and sagittal planes and were subsequently reviewed by two thoracic surgeons. Cases were included in the study if pleural contact was observed, defined as direct contact with the pleural surface or interlobar fissure on lung window settings at 600 Hounsfield units (Figure 3).

PET/CT image acquisition
Preoperative 18F-fluorodeoxyglucose (18F-FDG) PET/CT images were acquired using a Discovery 710 device (GE Healthcare, Milwaukee, WI, USA). Patients fasted for a minimum of 6 hours before the examination. Images were captured 60 minutes after intravenous injection of 18F-FDG (0.12 mCi/kg), with scanning intervals of 1.5–2 minutes per bed position.

Definition of VPI
The resected lung specimens were fixed in formalin, deparaffinized, and stained with hematoxylin and eosin. Elastic staining was used to pathologically confirm VPI, which was assessed using the modified Hammar classification system. PL0 represented tumors located within the subpleural lung parenchyma or those superficially invading connective tissue beneath the elastic layer; PL1 represented tumors invading beyond the elastic layer; PL2 referred to tumors invading the pleural surface; and PL3 represented tumors invading the parietal pleura (15). The presence of PL1, PL2, and PL3 was classified as VPI-positive findings, with PL1 and PL2 categorized as T stage 2, and PL3 as T stage 3 according to the National Comprehensive Cancer Network (NCCN) guidelines (3).

Statistical analysis
Baseline characteristics are presented as frequencies and percentages for categorical variables, and as means ± standard deviations for continuous variables. The Mann-Whitney U test was employed for comparisons of continuous variables, whereas categorical variables were analyzed using Chi-squared tests or Fisher’s exact tests. The relationship between variables and VPI was assessed using receiver operating characteristic (ROC) curve analysis, with the area under the curve (AUC) calculated for each index. Logistic regression was conducted to identify risk factors for VPI, incorporating variables that exhibited P values below 0.2 into the multivariable analysis.
The variables initially examined included age, sex, smoking status, serum CEA, pleural CEA, SUVmax, tumor size, and nodule type. All statistical analyses were conducted using R version 4.0.4 (R Foundation for Statistical Computing, Vienna, Austria) along with the Epi, ggplot2, and moonBook packages. A P value of less than 0.05 was deemed statistically significant. A risk-scoring model to predict VPI was developed based on variables identified through multivariable logistic regression analysis. The risk-scoring model was developed by assigning points in proportion to the magnitude of each variable’s odds ratio (OR). Consequently, pleural CEA (OR 3.00) was assigned a score of 0 or 1, SUVmax (OR 5.25) a score of 0 or 2, and nodule type (OR 3.89) on a scale of 0 to 2, with the final score reflecting the cumulative total of these components. This model was validated using data from the same population collected between January 2023 and June 2024.

Ethical statement
This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The Institutional Review Board of the Catholic University of Korea, College of Medicine, approved the study (No. KC24RASI0504), and the requirement for individual consent for this retrospective analysis was waived.

Results

Patient characteristics
In the development cohort, 38 of 186 patients (20.4%) were identified as VPI-positive. The VPI-positive group exhibited a significantly higher proportion of patients with elevated pCEA levels compared to the VPI-negative group (63.2% vs. 22.3%; P<0.001). Similarly, a greater proportion of patients with increased SUVmax was observed in the VPI-positive group (77.8% vs. 21.9%; P<0.001). This group also underwent lobectomy more frequently (78.9% vs. 58.8%; P=0.03), had larger tumor sizes (P<0.001), and was more likely to present with solid-type nodules (78.9% vs. 33.1%; P<0.001). In the validation cohort, VPI was identified in 24 out of 95 patients (25.3%). Consistent with the development cohort, the VPI-positive group in the validation cohort demonstrated statistically significant differences from the VPI-negative group in similar parameters, including pleural CEA levels, SUVmax, extent of resection, tumor size, and nodule type (Table 1).

Development of the VPI prediction model
We conducted a retrospective analysis of 186 patients with pleural contact from a previous study (14). In this analysis, the cutoff values for elevated pCEA and SUVmax were set at 2.565 ng/mL and 4.25, respectively, based on the optimal thresholds determined in that previous study (14). The multivariate analysis identified elevated pCEA [OR, 3.00; 95% confidence interval (CI): 1.14–7.87; P=0.02], elevated SUVmax (OR, 5.25; 95% CI: 1.90–14.50; P=0.001), and nodule type (OR, 3.89; 95% CI: 1.42–10.68; P=0.008) as independent risk factors for VPI (Table 2).
Based on this information, we developed a model that predicts the VPI to have a value of 0 or 1 based on whether pCEA levels increased, a value of 0 or 2 based on whether SUVmax increased, and a value ranging from 0 to 2 depending on the type of nodule (Table 3).
The VPI prediction risk score model was applied to the development cohort. The proportion of patients with VPI positivity increased with higher scores: 0% for a score of 0, 4.3% for a score of 1, 12.5% for a score of 2, 25.0% for a score of 3, 41.7% for a score of 4, and 73.1% for a score of 5 (Figure 4). Additionally, the ROC curve demonstrated strong discriminatory performance of the risk score model, with an AUC of 0.849 (95% CI: 0.7801–0.9183) [sensitivity: 76.3% (95% CI: 0.6309–0.8684); specificity: 80.6% (95% CI: 0.7405–0.8706); positive predictive value (PPV): 9.4%; negative predictive value (NPV): 42.0%] (Figure 5).

Internal (temporal) validation of risk scoring model
The VPI prediction model developed using the training set was applied to the validation set. Among the 95 patients, the presence of VPI was observed as follows: 0% for a score of 0, 3.6% for a score of 1, 16% for a score of 2, 26.7% for a score of 3, 54.5% for a score of 4, and 69.2% for a score of 5 (Figure 6). The proposed VPI prediction model was validated through ROC curve analysis. The model demonstrated its effectiveness in predicting VPI, achieving an AUC of 0.829 (95% CI: 0.7378–0.9200) with a sensitivity of 79.2% (95% CI: 0.6250–0.9177), a specificity of 71.8% (95% CI: 0.6056–0.8310), a PPV of 8.9%, and an NPV of 51.3% (Figure 7).

Discussion

Key findings
As previously established, VPI can lead to upstaging of the T stage, making it a critical factor in determining the stage of NSCLC. Furthermore, if VPI is associated with the potential for skip N2 metastases, it becomes an important consideration in the assessment of the N stage. Therefore, accurate prediction of VPI is essential for effective surgical planning. Previous studies have identified various factors that can predict VPI (9-12). In this study, we developed and validated a risk prediction model, contributing to a more intuitive clinical assessment of VPI. These advancements may assist surgeons in making more informed decisions regarding surgical strategy.

Strengths and limitations
The factor “nodule type”, utilized in this risk prediction model, has been identified as a significant variable in a recent large-scale randomized controlled trial examining the extent of surgical resection in NSCLC (16,17). This suggests that the consolidation/ tumor ratio may correlate with tumor aggressiveness. Similarly, SUVmax on PET/CT, an established indicator of tumor activity, was identified as a significant factor in this study (14,18). At our institution, preoperative PET/CT is routinely performed for most patients suspected of lung cancer to ensure accurate nodal staging and assess tumor activity. This systematic approach enhances the accuracy of predicting VPI. However, in settings where preoperative PET/CT is not universally performed, our findings indicate that intraoperative pCEA levels may serve as an effective alternative for predicting VPI prior to pulmonary resection (14).
Compared to previous studies focusing on radiological findings associated with VPI, our study presents a more streamlined and clinically applicable risk model by incorporating intuitive and practical parameters. This simplicity enhances its potential utility in clinical practice. However, the study has several limitations. First, as a single-center retrospective study, there is an inherent risk of bias, although internal (temporal) validation was conducted to mitigate this concern. External validation of the model is necessary to confirm its generalizability. Second, further investigation is required to ascertain the clinical significance of VPI in guiding surgical strategies. Recent studies have indicated potential associations with skip N2 metastasis, which, if validated, would further emphasize the importance of accurately predicting VPI (4). Lastly, while CEA was the sole tumor marker analyzed in this study due to its practicality and widespread use, future research exploring additional tumor markers may yield further insights. However, practical limitations such as time, availability, and cost must be taken into account.

Explanation of findings and comparison with similar research
Several studies have been conducted to predict VPI preoperatively (10,19-22). Most of these studies relied on imaging findings, particularly from CT and PET scans. Several studies have sought to predict VPI using CT- or PET-based imaging features, such as pleural tags, retraction, and contact length, or by employing complex machine-learning algorithms on radiologic data (10,19-22). However, these approaches are inherently constrained by interobserver variability in interpreting subtle pleural findings and the lack of transparency and reproducibility often associated with deep-learning models. Our study mitigated these sources of subjectivity by utilizing objective and quantifiable parameters—pleural CEA, SUVmax, and nodule type—to develop a straightforward and clinically applicable scoring model. Notably, the discriminative performance of our model was comparable to those of previous studies, highlighting its significant clinical relevance (10,19-22).
A prediction model for VPI was proposed by Iizuka et al. (11). However, their model incorporated subjective factors, such as tumor contact length, and included a total of six variables. Despite this, the predictive performance of their model was suboptimal, with an AUC of 0.68, which is inferior to the performance demonstrated in our study.
CEA is a well-established tumor marker for NSCLC (23). While numerous studies have investigated its role in NSCLC, few have focused specifically on pCEA, as in our study (24-26). Additionally, most existing research on pCEA has focused on malignant pleural effusion, with limited attention given to its association with VPI.
As stated in our previous study, the underlying mechanisms linking elevated pCEA levels to pathologic VPI remain poorly understood (14). CEA, a glycoprotein that plays a crucial role in cell adhesion, is typically produced during fetal development and ceases production prior to birth (23). With respect to the increase in pCEA levels in NSCLC associated with VPI, we hypothesize that tumors exhibiting invasion beyond the elastic layer, as classified by the VPI criteria, are positioned closer to the pleural surface than tumors without VPI. This proximity may facilitate the release of these cell surface proteins or their leakage through the lymphatic system (24,25). However, further studies are needed to confirm this hypothesis.

Implications and actions needed
Although external validation is required, the VPI risk prediction model proposed in our study enables a more accurate estimation of the clinical stage and facilitates patient-specific surgical planning. Furthermore, this study highlights the clinical utility of pleural CEA, a parameter that has not been extensively investigated, thereby providing valuable insights into its potential significance.

Conclusions
VPI is a critical factor in the staging of NSCLC, as it significantly influences decisions regarding the extent of surgical resection and lymph node dissection, particularly in cases involving nodal metastasis, such as skip N2. Accurate prediction of VPI is essential for enabling surgeons to plan tailored surgical approaches and optimize clinical outcomes. Building upon previously identified predictive factors, including pCEA, SUVmax, and nodule type, this study developed and validated a VPI prediction model. This model provides a practical and comprehensive framework to support individualized surgical planning and enhance decision-making in clinical practice. External validation, along with future prospective multi-center studies, is necessary to confirm the clinical utility of this model and to ascertain whether its application can improve surgical decision-making and oncologic outcomes.

Supplementary
The article’s supplementary files as