Risk factors and predictive model of lymph node metastasis in clinical stage IA peripheral non-small cell lung cancer: a retrospective study.

Introduction
Lung cancer remains the leading cause of cancer-related mortality worldwide, with non-small cell lung cancer (NSCLC) accounting for approximately 85% of all cases [1]. In recent years, the widespread implementation of low-dose computed tomography (LDCT) for lung cancer screening has significantly increased the detection rate of early-stage lung cancers, particularly clinical stage IA NSCLC. These cases are typically characterized by tumors measuring ≤ 3 cm in maximum diameter without radiological evidence of lymph node or distant metastasis (cT1N0M0) [2]. And peripheral lung cancers account for a substantial proportion of these early-stage lesions. Due to its comparatively low risk of lymph node metastasis, early-stage peripheral NSCLC is increasingly treated with limited surgical resection (e.g. segmentectomy or wedge resection) and targeted lymph node sampling [3]. Given that 15–20% of clinical stage IA NSCLC patients are postoperatively found to have lymph node metastases [4–6], preoperative identification of high-risk individuals is critical for tailoring surgical strategies and improving outcomes. Various radiographic, pathological, and serum tumor marker–based predictors have been proposed in previous studies; however, most studies have focused on all-stage or centrally located NSCLC, and few have specifically addressed peripheral clinical stage IA tumors. In this study, we retrospectively investigated risk factors for lymph node metastasis in clinical stage IA peripheral NSCLC and developed a logistic regression–based nomogram to assist preoperative risk assessment. Model performance was evaluated using receiver operating characteristic (ROC) analysis, calibration, and decision curve analysis (DCA). A subgroup analysis was also conducted for mediastinal (N2) metastasis to inform surgical planning and individualized lymphadenectomy strategies.

Methods

Patients and study design
This retrospective study included consecutive patients with clinical stage IA peripheral non-small cell lung cancer (NSCLC) who underwent surgical resection at the Department of Thoracic Surgery, Peking University First Hospital, between January 2015 and September 2018. The detailed selection process is illustrated in Fig. 1.The inclusion criteria were as follows: No previous history of malignancy; No prior systemic or local antitumor therapy before surgery; Availability of baseline tumor marker data obtained at our center before treatment; Estimated glomerular filtration rate (eGFR) > 60 ml/min; Preoperative contrast-enhanced chest CT showing a peripheral lung lesion (defined as located in the outer one-third of the lung field on axial CT view); Maximum tumor diameter ≤ 3 cm without evidence of hilar or mediastinal lymphadenopathy or distant metastasis on imaging, consistent with clinical stage IA (cT1N0M0) according to the 8th edition of the TNM classification by the Union for International Cancer Control (UICC) and American Joint Committee on Cancer (AJCC) [2]. Lymph nodes were considered suspicious for metastasis if the short-axis diameter exceeded 1 cm on CT or showed hypermetabolic activity on Positron Emission Tomography–Computed Tomography (PET-CT). All patients underwent lobectomy or segmentectomy with either systematic lymph node dissection (SLND) or systematic lymph node sampling (SLNS). Preoperative evaluation and clinical staging were performed using a standardized protocol, including electrocardiogram, contrast-enhanced chest CT, abdominal ultrasound or CT, brain MRI, bone scintigraphy or PET-CT, pulmonary function testing, and transthoracic echocardiography. The exclusion criteria were: receipt of neoadjuvant chemotherapy or radiotherapy, pure ground-glass opacity (GGO) on preoperative CT, non-primary lung cancer, multiple primary lung cancers, a prior history of malignancy or synchronous malignancy, and final pathology confirming carcinoma in situ (Tis). Based on postoperative pathological evaluation, patients were categorized into a lymph node metastasis (LNM) group and a non-metastasis group. Mediastinal lymph node metastasis (N2) was recorded separately and analyzed in a subgroup analysis. The following data were collected from the electronic medical records and imaging archives: Demographic and clinical data: age, sex, smoking history, clinical symptoms assessed at presentation included cough, hemoptysis, chest discomfort or dyspnea, chest pain, and unintentional weight loss; history of chronic pulmonary diseases was recorded, including chronic obstructive pulmonary disease (COPD), pulmonary tuberculosis, and bronchiectasis, etc.; Serum tumor markers: preoperative levels of CEA, SCC, CA19-9, CYFRA21-1, NSE, TPA, and ProGRP; Radiological features: tumor size (maximum tumor diameter), tumor location (left vs. right lung lesion; upper vs. middle/lower lobe), presence of spiculation, pleural indentation, lobulation, and other CT characteristics; Pathological data: histological subtype, lymphovascular invasion, and visceral pleural invasion (VPI). To ensure clinical applicability, only variables obtainable through routine preoperative assessment were incorporated into the logistic regression model. Tumor size was recorded as the largest diameter measured on lung window CT images. This study was approved by the Institutional Review Board of Peking University First Hospital. Owing to the retrospective design, the need for written informed consent from individual patients was waived.

Cut-off values of tumor markers
To ensure consistency, all preoperative tumor marker measurements were performed in the Department of Laboratory Medicine at Peking University First Hospital. Serum tumor markers, including CEA, CYFRA 21 − 1, NSE CA19-9 and TPA were analyzed using a commercial electrochemiluminescence assay (Roche Diagnostics, Mannheim, Germany). Serum levels of SCC and ProGRP were measured using the ARCHITECT automated chemiluminescent immunoassay system (Abbott Laboratories, Chicago, IL, USA). The following reference thresholds were applied based on manufacturer recommendations: CEA: 3.2 ng/mL, CYFRA 21 − 1: 3.0 ng/mL, SCC: 2.0 ng/mL, NSE: 16.3 ng/mL, CA19-9: 37 U/mL; During the study period, minor adjustments were made to the thresholds for TPA and ProGRP due to updates in assay calibration. Therefore, receiver operating characteristic (ROC) analysis was used to determine optimal cutoff values for these markers in our dataset, resulting in thresholds of 130.0 U/L for TPA and 66.0 pg/mL for ProGRP—both closely approximating manufacturer-recommended ranges.

Statistical analysis
All statistical analyses were performed using IBM SPSS version 20.0 (IBM Co., Chicago, IL, USA) and R version 4.3.1. Categorical variables were compared using the chi-square or Fisher’s exact test, and continuous variables were analyzed using Student’s t-test or Mann–Whitney U test, as appropriate. Univariate logistic regression analysis was used to screen for potential risk factors for lymph node metastasis. Variables with p < 0.1 in univariate analysis were entered into multivariate logistic regression to identify independent predictors. A binary logistic regression model was constructed using significant independent variables. The performance of the model was assessed by calculating the area under the receiver operating characteristic (ROC) curve (AUC). A nomogram was developed based on the logistic model to visually estimate the probability of lymph node metastasis. Model calibration was evaluated using a calibration plot and Hosmer–Lemeshow goodness-of-fit test. Clinical utility was assessed using decision curve analysis (DCA). In addition, univariate and multivariate analyses were performed to identify independent predictors of mediastinal lymph node metastasis (N2). A two-tailed p-value < 0.05 was considered statistically significant.

Results

Baseline characteristics
A total of 346 patients with clinical stage IA peripheral non-small cell lung cancer (NSCLC) were enrolled in this study, including 159 males (46.0%) and 187 females (54.0%), with a mean age of 63.03 ± 9.64 years. All patients underwent in-house preoperative chest CT as part of their staging workup. Among them, 305 patients were assessed using CT alone, while 41 patients also underwent PET-CT. At the time of diagnosis, 94 patients (27.2%) presented with lung cancer–related clinical symptoms. A family history of lung cancer was reported in 36 patients (10.4%), and 99 patients (28.6%) were current or former smokers. Twenty-seven patients (7.8%) had a history of chronic pulmonary disease. Among the included patients, 334 (96.5%) were pathologically diagnosed with adenocarcinoma and 12 (3.5%) with squamous cell carcinoma. Right lung tumors were observed in 208 cases (60.1%) and left lung tumors in 138 cases (39.9%). Regarding tumor location within the lung, 216 patients (62.4%) had lesions in the upper lobes, while 130 patients (37.6%) had lesions in the middle or lower lobes. Based on preoperative CT, clinical T staging was as follows: 61 patients (17.6%) were staged as T1a, 172 (49.7%) as T1b, and 113 (32.7%) as T1c. Postoperative pathological examination confirmed lymph node metastasis in 29 patients, yielding an overall metastasis rate of 8.4%. Among these, 9 patients (2.6%) had both N1 and N2 involvement, 11 (3.2%) had only N1 metastasis, and 9 (2.6%) had only N2 metastasis. Baseline clinical, radiological, and pathological characteristics of the entire cohort were detailed in Table 1.

Univariate and multivariate analysis of lymph node metastasis
Univariate logistic regression analysis revealed that the mean age of patients with lymph node metastasis (LNM) was 63.86 ± 7.89 years, compared to 61.86 ± 9.78 years in those without LNM (p = 0.29), with no significant difference observed. Female patients had a significantly lower risk of LNM than male patients (OR = 0.42, 95% CI: 0.19–0.92, p = 0.03). Current or former smokers were at significantly higher risk of LNM compared to non-smokers (OR = 2.55, 95% CI: 1.18–5.50, p = 0.02). Patients with tumors located in the middle or lower lobes had a significantly increased risk of LNM compared to those with upper lobe lesions (OR = 2.19, 95% CI: 1.02–4.72, p = 0.05). The median CT-measured tumor diameter was 2.40 cm (IQR: 1.95–2.80 cm) in the LNM group versus 1.70 cm (IQR: 1.20–2.10 cm) in the non-LNM group. Larger tumor size on CT was significantly associated with increased LNM risk (OR = 5.47, 95% CI: 2.63–11.40, p = 0.00). The presence of spiculation on preoperative chest CT was also a significant predictor of LNM (OR = 2.59, 95% CI: 1.20–5.62, p = 0.02). Among immunohistochemical markers, both P53 and Ki-67 expression levels were significantly associated with LNM (p < 0.05). Among the seven tumor markers analyzed, abnormal carcinoembryonic antigen (CEA) levels showed a borderline association with LNM (p = 0.07) and were thus included in the multivariate analysis. Abnormal carbohydrate antigen 19 − 9 (CA19-9) levels were significantly associated with increased LNM risk (OR = 3.95, 95% CI: 1.01–15.49, p = 0.05). Detailed results are presented in Table 2. These variables were entered into a multivariate logistic regression model. Based on the results of univariate analysis, a total of nine variables were included in the multivariate logistic regression model: sex, smoking history, tumor location (upper vs. middle/lower lobe), CT-measured tumor size, spiculation sign, P53 expression, Ki-67 expression, CEA, and CA19-9. The analysis identified three independent predictors of lymph node metastasis in patients with clinical stage IA peripheral NSCLC: Tumor location in the middle or lower lobe (OR = 2.92, 95% CI: 1.15–7.41, p = 0.02), Larger tumor size on CT (OR = 3.85, 95% CI: 1.67–8.87, p = 0.00), and Elevated CA19-9 levels (OR = 9.88, 95% CI: 1.62–60.09, p = 0.01). These findings suggested that tumor location, size, and CA19-9 may be important factors in preoperative risk stratification for lymph node involvement. Detailed results are presented in Table 3.

Development and validation of the predictive model
The three independent risk factors were incorporated into a binary logistic regression model. The predictive formula was defined as: P = ex/((1 + ex)), x=−4.648+(0.792×tumor location)+(1.360× CA19-9)+(1.686×CT tumor size). Hosmer-lemeshow (H-L) test showed that P value was equal to 0.1, which was greater than 0.05, indicating that the model was established. In this model, tumor location was coded as 0 for upper lobe and 1 for middle/lower lobe; CA19-9 was coded as 0 for normal and 1 for elevated; and CT tumor size represents the maximum diameter of the tumor in centimeters as measured on chest CT imaging. ROC curve analysis showed that the model had good discriminatory power, with an AUC of 0.78 (95% CI: 0.68–0.88, p < 0.00). (Fig. 2.) To determine an appropriate threshold for risk stratification, we calculated the Youden index based on ROC analysis. The optimal cut-off point was identified at a predicted probability (PRE) of 0.143, which yielded the highest Youden index of 0.545, with a sensitivity of 69% and specificity of 85%.

Development and validation of the predictive nomogram
A nomogram was constructed based on the regression model to visualize the prediction of lymph node metastasis. Each predictor was assigned a weighted score, and the total score was used to estimate the probability of LNM. (Fig. 3.) Calibration was evaluated using a calibration curve and the Hosmer–Lemeshow test (p = 0.1), indicating good agreement between predicted and observed outcomes. (Fig. 4.) Decision curve analysis (DCA) demonstrated that the predictive model yielded a net clinical benefit across a wide range of threshold probabilities, supporting its clinical applicability in preoperative risk stratification. (Fig. 5.)

Subgroup analysis: mediastinal lymph node metastasis
In the subgroup analysis for mediastinal lymph node metastasis (N2), univariate analysis identified tumor location and tumor size as significant factors (Table 4). Multivariate analysis confirmed that both were independent predictors: Middle/lower lobe tumor location (OR = 4.20, 95% CI: 1.31–13.51, p = 0.02); Tumor size (OR = 4.60, 95% CI: 1.52–13.87, p = 0.01), (Table 5)

Discussion
With the promotion of lung cancer screening, particularly the implementation of low-dose CT, an increasing number of early-stage lung cancer cases are being detected. These tumors are associated with favorable long-term outcomes, with reported 5-year survival rates of approximately 92%, 83%, and 77% for stages cIA1, cIA2, and cIA3, respectively [2]. Surgery remains the most effective and potentially curative treatment, with lobectomy and systematic lymph node dissection (SLND) as the standard approach. Traditionally, SLND has been regarded as essential for accurate staging, identification of occult nodal metastasis, and improved survival. The increasing number of early-stage cases—particularly peripheral clinical stage IA NSCLC characterized by small tumor size and low rates of lymph node metastasis—has opened the possibility for less extensive surgical approaches, including sublobar resection and limited lymphadenectomy. Applying the same extent of resection and nodal dissection to all patients without risk stratification inevitably leads to overtreatment in a subset of node-negative individuals. Several studies have shown that systematic lymph node dissection in node-negative patients does not confer additional survival benefit, but may increase operative time, blood loss, the incidence of postoperative complications, and length of hospital stay [7, 8]. In addition to increasing operative risks, excessive removal of regional lymph nodes may impair local immune surveillance, as these nodes serve as critical hubs for initiating anti‑tumor immune responses. Emerging preclinical and clinical evidence highlights that tumor-draining lymph nodes (TDLNs) are essential for T‑cell priming and systemic antitumor immunity, which is key to the efficacy of immune checkpoint therapy. Removal of these nodes has been shown to abrogate such immune responses and diminish therapeutic benefit [9]. Despite the relatively low incidence of lymph node metastasis (LNM) in stage IA NSCLC, a notable number of patients are pathologically upstaged to stage IIB (N1 positive) or stage IIIA (N2 positive), which are associated with markedly lower 5-year survival rates of 56% and 41%, respectively. Thus, preoperative identification of patients at high risk of LNM is essential for tailoring surgical strategies and optimizing oncologic outcomes. Preoperative nodal staging is typically performed using imaging modalities such as CT, PET-CT, Diffusion-Weighted Magnetic Resonance Imaging (DWI-MRI), and Endobronchial Ultrasound-Guided Transbronchial Needle Aspiration (EBUS-TBNA). Although PET-CT demonstrates superior sensitivity and specificity compared to CT, its high cost and notable false-positive rate limit its routine use [10]. Invasive modalities like EBUS-TBNA and mediastinoscopy are considered the gold standard for mediastinal staging. However, these procedures are not routinely performed in clinical stage IA patients, thereby highlighting the need for additional risk stratification tools.
In this study, we retrospectively analyzed 346 consecutive patients with clinical stage IA peripheral NSCLC to identify risk factors for LNM and construct a predictive model. We identified three independent predictors of lymph node metastasis: middle or lower lobe tumor location (OR = 2.92, 95% CI: 1.15–7.41, p = 0.02), increased tumor size on CT (OR = 3.85, 95% CI: 1.67–8.87, p = 0.00), and elevated CA19-9 levels (OR = 9.88, 95% CI: 1.62–60.09, p = 0.01).(Table 3. ) Moreover, tumor size on CT (OR = 4.60, 95% CI: 1.52–13.87) and middle/lower lobe location (OR = 4.20, 95% CI: 1.31–13.51) were also identified as independent predictors of mediastinal lymph node metastasis (N2), further reinforcing their clinical relevance (Table 5). Tumor size has long been recognized as a surrogate for tumor burden and biological aggressiveness. Prior studies suggest a sharp increase in LNM risk as tumor size increases beyond 1 cm, even in clinical stage IA disease [11–13]. In our cohort, LNM rates for tumors ≤ 1 cm, 1–2 cm, and 2–3 cm were 1.6%, 4.1%, and 18.6%, respectively. Mediastinal LNM rates in these groups were 1.6%, 2.3%, and 11.5%, respectively, while skip metastases (N2 without N1 involvement) were observed in 0%, 1.7%, and 5.3%, respectively. These findings highlight the need to remain vigilant for LNM even in small, peripheral lesions. Our findings that non‑upper lobe tumors carry a higher risk of lymph node metastasis are consistent with recent clinical data [14]. Liu et al. [15] reported that primary tumors located in the lower lobe were significantly associated with lobar lymph node metastasis in adjacent, non‑primary tumor-bearing lobes (NTBL), having an odds ratio of 2.61 (95% CI: 1.26–5.75, P = 0.01) compared to upper lobe tumors. This observation suggests altered or retrograde patterns of lymphatic drainage from non-upper lobe regions, potentially increasing the involvement of mediastinal or adjacent lobar nodes even in early-stage disease. Our study also examined radiologic features such as spiculation, lobulation, and pleural retraction. Spiculation, characterized by radiating linear strands extending from the tumor margin on lung window CT, was identified in univariate analysis as a risk factor for LNM (OR = 2.59, p = 0.02), though not retained in multivariate analysis. Tumor markers are widely used in the diagnosis and management of malignancies. While CEA and CYFRA21-1 are commonly elevated in NSCLC, their predictive value for early-stage nodal metastasis remains unclear. Our analysis showed that CEA and CYFRA21-1 were not significantly associated with LNM in clinical stage IA peripheral NSCLC. Interestingly, although CA19‑9 is traditionally used as a marker for gastrointestinal malignancies, our study revealed that elevated CA19‑9 was one of the strongest independent predictors of lymph node metastasis (OR = 9.88, 95% CI: 1.62–60.09, p = 0.01). In NSCLC, particularly adenocarcinoma, elevated CA19‑9 has been associated with increased tumor aggressiveness and nodal metastasis, consistent with other studies linking it to lymphatic spread and advanced disease [16–20]. CA19-9 antigen, also known as sialyl Lewis A (sLe^a), is a carbohydrate antigen that serves as a ligand for selectins. It plays a critical role in cell adhesion and tumor cell–endothelial interactions, particularly facilitating the extravasation of cancer cells during the metastatic process [21, 22] CA19-9 is expressed on the surface of various epithelial tumor cells, especially those originating from the gastrointestinal tract, breast, and lungs [17, 23]. Notably, high CA19-9 expression may indicate a more aggressive tumor phenotype and has been linked to poorer clinical outcomes [24, 25] From a clinical perspective, elevated serum CA19-9 levels in NSCLC—especially in adenocarcinoma—may reflect increased tumor invasiveness and a higher risk of nodal or systemic spread, even in early-stage disease. Therefore, integrating CA19-9 into preoperative risk stratification could help identify patients who may benefit from more extensive lymphadenectomy or intensified postoperative monitoring.
In recent years, predictive models have gained traction in oncology for risk stratification and treatment guidance [26–28]. In this study, we identified three independent predictors of LNM—tumor location, CA19-9, and tumor size—and used them to construct a binary logistic regression model. The model was visualized as a nomogram (Fig. 3). This tool allows for personalized risk estimation and may assist surgeons in identifying patients who are more likely to benefit from systematic lymph node dissection, thereby optimizing surgical strategies in early-stage peripheral NSCLC. The predictive model developed in this study demonstrated good calibration performance in internal validation. The calibration curve showed that in the low to intermediate risk intervals (predicted probabilities of 0.0–0.2 and 0.2–0.4), the predicted probabilities closely matched the observed outcomes. While a slight underestimation was observed in the high-risk interval (>0.4), the overall deviation was minor and clinically acceptable. Decision curve analysis (DCA) further confirmed the clinical utility of the model. The nomogram yielded a higher net benefit than either the treat-all or treat-none strategies across a range of threshold probabilities between 0.05 and 0.6, with the greatest net benefit observed in the 0.05–0.3 range. This indicates the model is particularly valuable for identifying early-stage NSCLC patients with low-to-moderate risk of lymph node metastasis. To illustrate the application of the nomogram, consider a hypothetical patient with the following characteristics: Tumor location: middle/lower lobe (coded as 1); CA19-9: elevated (coded as 1); Tumor size on CT: 2.5 cm. Step 1: Locate each predictor on its corresponding axis in the nomogram: Tumor location (1): corresponds to approximately 17 points. CA19-9 (1): corresponds to approximately 28 points. Tumor size (2.5 cm): corresponds to approximately 85 points. Step 2: Add the points: 17 (location) + 28 (CA19-9) + 85 (tumor size) = 130 total points. Step 3: Find the total points on the “Total Points” axis, then map directly down to the LNM Risk (%) axis. A total score of 130 corresponds to an estimated 50–60% risk of lymph node metastasis (LNM). Based on our results using the Youden index, a predicted probability exceeding 14.3% should be considered high risk. Therefore, this patient would be classified as high-risk and should be recommended for systematic lymph node dissection (SLND) to optimize staging and therapeutic outcomes. Nevertheless, several limitations should be acknowledged. First, this was a single-center retrospective study, and despite the use of strict inclusion criteria, selection bias cannot be entirely ruled out. Second, the number of patients with lymph node metastasis was relatively small (~ 8%), which may have limited the model’s sensitivity and reduced its ability to accurately identify high-risk individuals. Third, only internal validation was performed using calibration plots, the Hosmer–Lemeshow test, and ROC analysis; no external dataset or resampling techniques such as cross-validation or bootstrapping were applied. Therefore, the model’s generalizability may be limited, and its performance could differ across populations or under varying imaging and biomarker assessment protocols. Future studies should aim to enhance the model’s robustness by increasing the sample size, including more high-risk cases, integrating additional predictive variables, and conducting external validation using multicenter prospective cohorts.

Conclusion
We identified middle/lower lobe tumors, CT-measured tumor size, and elevated CA19-9 as independent predictors of lymph node metastasis in clinical stage IA peripheral NSCLC. A nomogram based on these factors demonstrated good discrimination and calibration, supporting its potential use in preoperative nodal staging and individualized surgical planning. Further multicenter prospective studies are needed to externally validate and optimize the model.