Machine Learning–Based Prediction of Recurrence After Curative Resection in Non–Small Cell Lung Cancer

Machine Learning–Based Prediction of Recurrence After Curative Resection in Non–Small Cell Lung Cancer | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Machine Learning–Based Prediction of Recurrence After Curative Resection in Non–Small Cell Lung Cancer Ugur Ozberk, Selin Akturk Esen, Hilal Arslan, Oznur Bal, Efnan Algın, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8912361/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 09 Apr, 2026 Read the published version in Scientific Reports → Version 1 posted 11 You are reading this latest preprint version Abstract Recurrence after curative resection remains a major clinical challenge in non–small cell lung cancer (NSCLC), and improved postoperative risk stratification is needed. Machine learning (ML) approaches may enhance recurrence prediction using routinely available clinicopathologic data. We analyzed 265 patients who underwent curative lung cancer surgery. Recurrence was the primary endpoint. Seventeen clinical, pathological, and treatment-related variables were evaluated. Multiple supervised ML classifiers were trained using the full dataset and reduced feature sets generated by ANOVA, chi-square, and Kruskal–Wallis methods. Model performance was assessed using accuracy, area under the curve (AUC), and F1 score. Prognostic factors were examined with Cox regression, and model interpretability was explored through feature importance and SHAP analysis. Recurrence occurred in 82 patients (30.9%). AdaBoost achieved the highest accuracy (0.79) and F1 score (0.87), whereas SVC-RBF showed the highest AUC (0.81). Performance remained stable across feature-selection strategies. Histologic subtype, tumor size, tumor grade, and ECOG performance status were consistently influential variables, with ECOG status and tumor size dominating SHAP-based predictions. These findings indicate that ML models using routine clinicopathologic variables can reliably predict recurrence after NSCLC surgery and support individualized postoperative risk assessment. Biological sciences/Cancer Health sciences/Oncology Non–small cell lung cancer recurrence machine learning Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Lung cancer is among the most commonly diagnosed cancers and remains a leading cause of cancer-related mortality. Each year, an estimated two million new cases and 1.76 million deaths occur worldwide [ 1 ]. Despite these concerning statistics, survival rates are favorable in early-stage disease [ 2 ]. However, disease recurrence remains a major clinical challenge. Despite curative resection, approximately 30–55% of patients with non–small cell lung cancer (NSCLC) develop recurrence and ultimately die from their disease, with the recurrence risk varying by pathologic stage, nodal status, extent of resection, and histologic subtype [ 3 ]. Artificial intelligence (AI) has advanced oncologic care, with benefits spanning diagnosis through treatment, supported by techniques such as machine learning (ML) and deep learning [ 4 ]. In support of this concept, a large population-based study demonstrated the potential of machine learning to predict lung cancer risk prior to clinical diagnosis [ 5 ]. Using routinely collected clinical and laboratory data from more than 6,500 patients with non–small cell lung cancer and nearly 190,000 control subjects, the authors developed a machine learning model that outperformed established risk-prediction tools, including the mPLCOm2012 model, in identifying individuals at high risk up to 9–12 months before diagnosis. The machine learning approach achieved superior discriminative performance and sensitivity at high specificity, highlighting the added value of data-driven models for early risk stratification in lung cancer. In another study, supervised machine learning techniques applied to the SEER database demonstrated that lung cancer survival time could be predicted using routinely collected clinical variables, with gradient boosting and ensemble models achieving performance comparable to classical Cox proportional hazards models [ 6 ]. AI may also play an important role in predicting lung cancer recurrence. AI-enabled estimation of lung cancer recurrence risk can optimize remission-phase management and facilitate earlier detection, allowing clinicians to concentrate resources on higher-risk patients and thereby improving overall care. This study aimed to resolve persistent limitations in postoperative recurrence prediction for NSCLC by employing ML approaches. The specific aims were: (i) to evaluate the ability of diverse ML algorithms to predict postoperative recurrence using a comprehensive set of clinicopathological and treatment-related variables; (ii) to identify key clinicopathologic, perioperative, and laboratory determinants of recurrence using feature-selection procedures (χ² tests, ANOVA, Kruskal–Wallis); and (iii) to assess the concordance between ML–based predictions and prognostic factors identified by Cox proportional hazards analysis, while improving model interpretability through feature importance and SHAP analyses to elucidate the drivers of individual-level predictions. Through this study, identifying patients at high risk of recurrence in non–small cell lung cancer, a disease with high relapse rates, may enable more individualized postoperative management and improved treatment and surveillance strategies. Methods This retrospective study was conducted at the Department of Medical Oncology, Ankara Bilkent City Hospital. Ethical approval was obtained from the Institutional Review Board of Ankara Bilkent City Hospital (Decision No: TABED 1/1772/2025, Date: 22/10/2025), and the study was performed in accordance with the principles of the Declaration of Helsinki. Owing to the retrospective nature of the study, the requirement for written informed consent was waived. Patients with histopathologically confirmed NSCLC who underwent curative-intent surgical resection and were subsequently followed for disease recurrence between January 2020 and June 2025 were included in the study. Patients with incomplete clinical or pathological data, missing follow-up information, or non-curative surgical intent were excluded. Clinical, pathological, and treatment-related data were collected retrospectively from the hospital’s electronic medical record system and archival files. Recorded variables included age at diagnosis, sex, ECOG performance status, histological subtype, tumor size, pathological T and N stage classified according to the New Ninth Edition TNM Classification for Lung Cancer [ 7 ], tumor grade, visceral pleural invasion, surgical margin status, vascular and lymphatic invasion, type of surgical resection, adjuvant chemotherapy regimen and number of cycles, receipt of adjuvant radiotherapy or chemoradiotherapy, date of diagnosis, and date of recurrence. Recurrence was defined as radiological or histopathological evidence of locoregional or distant disease relapse following curative resection. Patients were followed until documented recurrence, death, or last clinical follow-up. 1. SPSS Software Version 25.0 Analysis Data analysis was conducted using SPSS software version 25.0 (IBM Corp., Armonk, NY, USA). Continuous variables were summarized as median values (minimum and maximum ranges), while categorical variables were presented as frequencies and percentages. Recurrence-free survival (RFS) was defined as the time from curative resection to the date of first documented disease recurrence or death, whichever occurred first. Factors associated with RFS were initially evaluated using univariate Cox proportional hazards regression analysis and multivariate Cox regression model to identify independent prognostic factors. A p-value < 0.05 was considered statistically significant. 2. Feature Selection Process Feature importance for predicting patient outcomes was assessed using three complementary statistical methods: Chi-square (Chi²), Analysis of Variance (ANOVA), and the Kruskal–Wallis test. These techniques were selected to capture different statistical properties of categorical and continuous variables and to improve model interpretability and performance. Features with statistically significant results (p < 0.05) were considered informative and retained. 2.1. Chi-square test The Chi-square test was applied to evaluate the association between categorical features and the target variable by comparing observed and expected frequencies under the assumption of independence [ 8 ]. 2.2. Analysis of variance test Analysis of variance test was used to identify continuous variables showing significant mean differences across outcome groups [ 9 ]. Features with p-values below 0.05 were interpreted as having discriminative power and were included in the reduced feature set. 2.3. Kruskal–Wallis test For variables that violated normality assumptions, the Kruskal–Wallis test, a non-parametric alternative to ANOVA, was employed [ 9 ]. Features with significant median differences across groups were selected, enabling a more robust and assumption-free feature selection strategy. 3. Machine Learning Methods A comprehensive comparative analysis of multiple supervised ML algorithms was conducted to evaluate classification performance on the study dataset. Model performance was evaluated using standard classification metrics derived from the confusion matrix, which consists of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). Precision was defined as TP/(TP + FP), representing the proportion of predicted positive cases that were correctly identified. Recall (sensitivity) was calculated as TP/(TP + FN) to quantify the model’s ability to detect actual positive cases. The F1 score, computed as the harmonic mean of precision and recall, provided a balanced measure of predictive performance. Overall accuracy was determined using (TP + TN)/(TP + TN + FP + FN), reflecting the total proportion of correct predictions. Additionally, the area under the receiver operating characteristic curve (AUC) was calculated using the trapezoidal rule to assess the threshold-independent discriminative capacity of each model. All metrics ranged from 0 to 1, with higher values indicating superior classification performance. 3.1. Decision Tree Models Decision Trees were implemented to derive interpretable rules and partition the dataset into homogeneous subgroups [ 10 ]. The study evaluated multiple tree-based variants including Fine, Medium, Coarse Trees, Boosted Trees, Bagged Trees, and RUSBoosted Trees. A grid search was conducted to optimize key hyperparameters such as minimum leaf size (tested at 1, 5, 20, and 50) and the Gini index as the splitting criterion. The best performance was achieved with a minimum leaf size of 1, indicating that finer granularity provided superior classification capability. 3.2. Regression-Based Models Regression models, particularly Binary GLM Logistic Regression and Efficient Logistic Regression, were adopted to capture linear relationships between predictors and recurrence outcomes [ 11 ]. These models estimate the probability of class membership through logistic functions, enabling clear interpretability and strong baseline predictive performance. Their inclusion ensured that both parametric and non-parametric modeling perspectives were represented within the study. 3.3. Support Vector Machines Support Vector Machines (SVM) were used as robust classifiers capable of handling high-dimensional and non-linear feature interactions [ 12 ]. Seven SVM configurations were examined, including Linear, Efficient Linear, Quadratic, Cubic, and Gaussian (Fine, Medium, Coarse) kernels. Hyperparameter optimization was conducted using grid search across values of the regularization constant C (0.1, 1, 10, 100, 1000). The optimal performance was consistently obtained at C = 1, balancing margin maximization and misclassification tolerance effectively. 3.4. Naive Bayes Models Naive Bayes classifiers, including Gaussian and Kernel-based variants, were utilized owing to their computational efficiency and probabilistic interpretability [ 13 ]. These models assume conditional independence among features, enabling fast and effective classification, especially for high-dimensional datasets. Their inclusion provided a lightweight yet powerful comparative baseline against more complex methods. 3.5. Neural Network Models Artificial Neural Networks were applied to model complex, non-linear relationships within the dataset [ 14 ]. Five architectures were employed: Trilayered Neural Network, Narrow Neural Network, Wide Neural Network, Bilayered Neural Network, and Medium Neural Network. Hidden layers used sigmoid activation functions, while output layers used softmax activation. Network structures varied from shallow (5 neurons in a single layer) to deep (three layers with 10 neurons each), enabling evaluation of how architectural complexity influences predictive performance. 3.6. Kernel-Based Methods Kernel transformation techniques were incorporated to address non-linear classification challenges [ 15 ]. Two kernelized models were evaluated: SVM Kernel and Logistic Regression Kernel. These methods map input features into higher-dimensional spaces where linear separation is more feasible, improving classification power for datasets with inherently non-linear decision boundaries. 3.7. K-Nearest Neighbor (KNN) K-Nearest Neighbor was employed as a distance-based, instance-learning algorithm [ 16 ]. Several variants including Weighted, Cubic, Medium, Cosine, Coarse, and Fine KNN were tested. Hyperparameters evaluated via grid search included the number of neighbors (k ranging from 1 to 30) and distance metrics (Euclidean, Minkowski, Chebyshev). Optimal performance occurred at k = 1 with Euclidean distance, indicating that local neighborhood similarity was highly informative for classification in this dataset. 3.8. Ensemble Classifiers Ensemble classifiers combine the predictive capabilities of multiple base learners to improve model performance and robustness beyond that of individual algorithms [ 17 ]. In this study, several ensemble methods were implemented, including Bagged Trees, Subspace KNN, Boosted Trees, RUSBoosted Trees, and Subspace Discriminant models. Additionally, two advanced gradient boosting algorithms, Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM), were incorporated to further enhance predictive accuracy. 3.9. XGBoost and LightGBM XGBoost employs decision trees as base learners and builds them sequentially in a boosting framework, where each new tree focuses on correcting the errors of preceding trees [ 18 ]. This iterative refinement process increases accuracy and robustness, making XGBoost one of the most powerful boosting algorithms available. LightGBM, an optimized gradient boosting framework, utilizes a leaf-wise tree growth strategy, enabling faster processing and reduced memory usage compared to traditional boosting methods [ 19 ]. Its ability to handle large datasets efficiently contributes to its growing popularity in ML applications. Results A total of 265 patients were included in the study. The majority were male (n = 220, 83%), while 45 (17%) were female. The median age at diagnosis was 63 (39–85) years. Most patients had an ECOG performance status of 1 (n = 208, 78.5%), followed by ECOG 2 (n = 35, 13.2%) and ECOG 0 (n = 22, 8.3%). The most frequent histological subtype was adenocarcinoma (n = 144, 54.3%), followed by squamous cell carcinoma (n = 105, 39.6%). T-stage distribution was as follows: T1a 5 (1.9%), T1b 51 (19.2%), T1c 32 (12.1%), T2a 72 (27.2%), T2b 45 (17%), T3 42 (15.8%), and T4 18 (6.8%). Nodal status showed N0 in 176 (66.4%), N1 in 54 (20.4%), and N2 disease in 35 (13.2%). Visceral pleural invasion was present in 104 (39.2%), while surgical margins were negative in 244 (92.1%). Tumor grade distribution included Grade 1: 37 (14%), Grade 2: 93 (35.1%), and Grade 3: 63 (23.8%). Vascular invasion was detected in 109 (41.1%). Wedge resection was performed in 48 (18.1%), and lymphatic invasion was present in 89 (33.6%). Mutation testing revealed EGFR positivity in 11 (4.2%), ALK in 7 (2.6%), ROS1 in 1 (0.4%), and BRAF in 3 (1.1%), while 69 (26%) were mutation-negative. Adjuvant chemotherapy was administered to 177 (66.8%), adjuvant radiotherapy to 31 (11.7%), and adjuvant chemoradiotherapy to 12 (4.5%). Recurrence occurred in 82 patients (30.9%) (Table 1 ). Table 1 Baseline clinicopathologic and treatment characteristics of the patients Variable Median (Min-Max) N (%) Age at diagnosis (years) 63 (39–85) - Tumor size (cm) 3.2 (0.8–16) - Number of cycles 4 (0–6) - Sex (Male) 220 (83) ECOG performance score ECOG score 0 22 (8.3) ECOG score 1 208 (78.5) ECOG score 2 35 (13.2) Histologic subtype Adenocarcinoma 144 (54.3) Squamous cell carcinoma 109 (41.1) Large-cell carcinoma 6 (2.3) Pleomorphic carcinoma 4 (1.5) NOS (Not otherwise specified) 2 (0.8) T stage T1a 5 (1.9) T1b 51 (19.2) T1c 32 (12.1) T2a 72 (27.2) T2b 45 (17) T3 42 (15.8) T4 18 (6.8) N stage N0 176 (66.4) N1 54 (20.4) N2 35 (13.2) Visceral pleural invasion Yes 104 (39.2) No 158 (59.6) Unknown 3 (1.1) Surgical margin Negative 244 (92.1) Positive 17 (6.4) Unknown 4 (1.5) Tumor grade Grade 1 37 (14) Grade 2 93 (35.1) Grade 3 63 (23.8) Unknown 72 (27.2) Vascular invasion Yes 109 (41.1) No 153 (57.7) Unknown 3 (1.1) Lymphatic invasion Yes 89 (33.6) No 174 (65.7) Unknown 2 (0.8) Wedge resection Performed 48 (18.1) Not performed 215 (81.1) Unknown 2 (0.8) Mutation status Negative 69 (26) EGFR-mutant 11 (4.2) ALK-mutant 7 (2.6) ROS1-mutant 1 (0.4) BRAF-mutant 3 (1.1) Unknown 174 (65.7) Adjuvant chemotherapy None 88 (33.2) Cisplatin–vinorelbine 125 (47.2) Cisplatin–gemcitabine 20 (7.5) Carboplatin–paclitaxel 15 (5.6) Cisplatin–pemetrexed 9 (3.4) Carboplatin–vinorelbine 5 (1.9) Carboplatin–gemcitabine 2 (0.8) Carboplatin–pemetrexed 1 (0.4) Adjuvant radiotherapy Yes 31 (11.7) No 234 (88.3) Adjuvant chemoradiotherapy Yes 12 (4.5) No 253 (95.5) Recurrence Yes 82 (30.9) No 183 (69.1) In the univariate Cox regression analysis, several variables were significantly associated with recurrence-free survival (RFS). Higher ECOG performance status, larger tumor size, advanced T stage, higher tumor grade, presence of vascular invasion, lymphatic invasion, visceral pleural invasion, and receipt of adjuvant radiotherapy were all associated with poorer RFS. Female sex was associated with better RFS compared with male sex. Other variables were not significantly associated with RFS (Table 2 ). In the multivariate Cox regression analysis, ECOG performance status remained an independent predictor of RFS (p = 0.012), with ECOG 2 showing a markedly increased risk compared with ECOG 0 (HR 3.84, 95% CI 1.28–11.52). Advanced T stage also retained significance (p = 0.012), with T4 disease associated with a substantially elevated recurrence risk (HR 3.24, 95% CI 1.51–6.93). Higher tumor grade remained independently associated with poorer RFS (p = 0.042), particularly for grade 3 tumors (HR 3.62, 95% CI 1.15–11.39). Absence of vascular invasion continued to confer a protective effect (HR 0.49, 95% CI 0.26–0.91, p = 0.021). Additionally, patients who did not receive adjuvant radiotherapy demonstrated better RFS than those who did (HR 0.41, 95% CI 0.22–0.78, p = 0.012). Other variables were not independently associated with RFS in the multivariate model (Table 2 ). Table 2 Univariate and Multivariate Cox Regression Analysis for Recurrence-Free Survival Variable HR (95% CI) p-value HR (95% CI) p-value Age at diagnosis (years) 1.02 (0.99–1.05) 0.080 Sex Male 1.00 0.047 Female 0.49 (0.24–0.99) ECOG performance score ECOG score 0 1.00 0.003 1.00 0.012 ECOG score 1 1.64 (0.59–4.52) 2.09 (0.73–6.02) ECOG score 2 3.54(1.22–10.30) 3.84 (1.28–11.52) Histologic subtype Adenocarcinoma 1.00 0.659 Squamous cell carcinoma 0.99 (0.63–1.57) Other 1.47 (0.62–3.46) Tumor size (cm) 1.16 (1.07–1.26) < 0.001 T stage T1 1.00 < 0.001 1.00 0.012 T2 1.17 (0.68–2.01) 1.08 (0.62–1.88) T3 1.45 (0.73–2.86) 1.15 (0.57–2.32) T4 5.14 (2.50-10.58) 3.24 (1.51–6.93) N stage N0 1.00 0.568 N1 0.86 (0.52–1.42) N2 1.08 (0.61–1.92) Visceral pleural invasion Yes 1.00 0.019 No 0.59 (0.38–0.91) Surgical margin Negative 1.00 0.052 Positive 1.99 (0.99–3.98) Tumor grade Grade 1 1.00 < 0.001 1.00 0.042 Grade 2 3.05 (1.06–8.76) 2.20 (0.74–6.55) Grade 3 6.08 (2.117.4) 3.62 (1.15–11.39) Vascular invasion Yes 1.00 < 0.001 1.00 0.021 No 0.38 (0.24–0.60) 0.49 (0.26–0.91) Lymphatic invasion Yes 2.03 (1.31–3.14) 0.001 No 1.00 Wedge resection Performed 1.00 0.746 Not performed 0.91 (0.52–1.58) Mutation status Negative 1.00 0.560 Positive 0.78 (0.34–1.78) Adjuvant chemotherapy Yes 1.30 (0.81–2.10) 0.268 No 1.00 Number of cycles 1.01 (0.90–1.13) 0.841 Adjuvant radiotherapy Yes 1.00 < 0.001 1.00 0.012 No 0.40 (0.23–0.69) 0.41 (0.22–0.78) Adjuvant chemoradiotherapy Yes 1.00 0.865 No 1.09 (0.39–2.98) The classification performance of multiple ML algorithms was evaluated using the full dataset consisting of 17 variables (sex, age at diagnosis, tumor size, pT stage, pN stage, visceral pleural involvement, surgical margin status, tumor grade, vascular invasion, wedge resection (performed/not performed), lymphatic invasion, histological subtype, ECOG performance status, adjuvant chemotherapy regimen, number of chemotherapy cycles, receipt of adjuvant radiotherapy and receipt of chemoradiotherapy ), as well as reduced feature subsets obtained through ANOVA, Chi-square, and Kruskal–Wallis feature selection methods. Model performance was compared using accuracy as the primary evaluation metric (Table 3 ). Table 3 Comparison of recurrence prediction accuracy across machine learning methods Model All Features 13 Features by ANOVA 13 Features by Chi2 14 Features by Kruskal-Wallis AdaBoost 0.7925 0.7736 0.7547 0.7736 Support Vector Classifier RBF 0.7736 0.7170 0.7736 0.7170 MLPClassifier 0.7736 0.6604 0.6792 0.6038 BernoulliNB 0.7736 0.7358 0.7358 0.7358 Support Vector Classifier Poly 0.7547 0.7170 0.7925 0.7358 GradientBoosting 0.7547 0.7358 0.7170 0.6604 RidgeClassifier 0.7358 0.7547 0.7358 0.7170 BaggingClassifier 0.7358 0.6415 0.6981 0.6604 QuadraticDiscriminant 0.7358 0.7547 0.7736 0.7736 ExtraTrees 0.7358 0.7170 0.7547 0.6981 RandomForest 0.7358 0.7547 0.7358 0.7358 Calibrated SVM 0.7170 0.7358 0.7358 0.7170 LinearDiscriminant 0.6981 0.7547 0.7170 0.7358 Nu-Support Vector Classifier 0.6981 0.7547 0.7358 0.7358 KNeighbors 0.6981 0.7170 0.7358 0.6038 Linear Support Vector Classifier 0.6792 0.7736 0.6981 0.7170 MLPClassifier_Deep 0.6792 0.6415 0.6792 0.6415 SGDClassifier 0.6604 0.6792 0.6038 0.6604 Logistic Regression 0.6604 0.7547 0.7358 0.7170 Machine learning accuracy results showed that AdaBoost achieved the highest performance using the full feature set (accuracy ≈ 0.79). SVC-RBF and MLP-based models also demonstrated strong performance, with accuracies around 0.77. Model performance remained broadly comparable across feature-selection strategies, and no single selection method consistently outperformed the others. Ensemble- and kernel-based classifiers generally showed more stable performance, whereas simpler linear models tended to yield lower accuracy. Overall, predictive accuracy across models ranged approximately between 0.66 and 0.79 (Table 3 ). Based on overall accuracy, confusion matrices were generated for the two best-performing models, AdaBoost and the Support Vector Classifier with an RBF kernel (SVC-RBF). For both models, recurrence predictions were classified as progression present or absent. The confusion matrices showed that AdaBoost correctly classified 36 progression cases and 6 non-progression cases, with 1 false-negative and 10 false-positive predictions. SVC-RBF correctly identified 32 progression cases and 9 non-progression cases, with 5 false-negative and 7 false-positive predictions (Fig. 1 ). The area under the receiver operating characteristic curve (AUC) was calculated to further evaluate the discriminatory performance of the ML models in predicting disease recurrence. Among the evaluated algorithms, the Support Vector Classifier with RBF kernel (SVC-RBF) achieved the highest AUC value (AUC = 0.814), indicating superior discriminative ability. This was followed by the Support Vector Classifier with polynomial kernel (AUC = 0.779) and the Calibrated Support Vector Machine (AUC = 0.780) (Fig. 2 ). The F1 score was used to evaluate the balance between precision and recall for all ML models in predicting disease recurrence. Among the evaluated classifiers, AdaBoost achieved the highest F1 score (F1 = 0.867), indicating the best overall balance between sensitivity and precision. This was followed by Bernoulli Naive Bayes (F1 = 0.853) and MLPClassifier (F1 = 0.850), both of which demonstrated strong classification performance (Fig. 3 ). Feature importance analysis across models showed that histological subtypes, including adenocarcinoma, squamous cell carcinoma, and pleomorphic carcinoma, were the most frequently selected features. Tumor grade, tumor size, and ECOG performance status were also commonly retained across models. Less frequent but consistent contributions were observed for large cell carcinoma, histology not otherwise specified, age at diagnosis, and pN stage (Fig. 4 ). SHAP analysis for the AdaBoost model, one of the best-performing classifiers, indicated that ECOG-PS and tumor size had the greatest contributions to disease recurrence predictions. The number of chemotherapy cycles and vascular invasion status showed moderate importance, whereas age at diagnosis, lymphatic invasion, adjuvant radiotherapy, tumor grade, and pathological substage had comparatively lower contributions to the model predictions (Fig. 5 ). Discussion This study demonstrates that recurrence risk after curative resection in NSCLC can be predicted with high accuracy using multiple ML algorithms. In particular, AdaBoost and SVC-RBF models demonstrated robust and consistent predictive performance across multiple feature selection strategies, indicating their stability and reliability in predicting recurrence risk. Lastly, the variables highlighted by the ML models largely overlapped with those identified as independent prognostic factors in Cox regression analysis, supporting the clinical relevance and interpretability of the ML–based predictions. Despite applying different feature selection strategies (ANOVA, Chi-square, and Kruskal–Wallis), the overall model performance remained comparable across most classifiers. Several models, including AdaBoost, Support Vector Classifiers (RBF and polynomial kernels), Quadratic Discriminant Analysis, and Random Forest, demonstrated relatively stable performance when trained on either the full feature set or reduced feature subsets. While certain algorithms, such as Logistic Regression, Linear SVC, and Ridge Classifier, showed modest performance improvements after feature selection, others, particularly MLP-based models and k-nearest neighbors, experienced a decline in accuracy when the number of input variables was reduced. Importantly, no single feature selection method consistently outperformed the others across all models. Collectively, these findings suggest that predictive performance was primarily driven by a core group of robust variables, with the choice of feature selection method exerting only a secondary influence on model accuracy. Taken together, the performance metrics highlight complementary strengths among the evaluated models in predicting disease recurrence. Confusion matrix analysis showed that AdaBoost was more effective in minimizing false-negative predictions, thereby improving identification of recurrent cases, whereas SVC-RBF was more conservative, yielding fewer false-positive classifications among non-recurrent patients. Although overall accuracy was comparable, the distribution of misclassifications differed meaningfully between models. Discriminative performance analysis further demonstrated that SVC-RBF achieved the highest AUC, indicating superior ability to separate recurrent from non-recurrent cases across varying decision thresholds, likely due to its capacity to model complex non-linear relationships. In contrast, AdaBoost achieved the highest F1 score, reflecting a more favorable balance between sensitivity and precision and underscoring its robustness in clinically balanced classification settings. Overall, these findings suggest that while kernel-based methods such as SVC-RBF may be optimal for maximizing discriminative power, ensemble-based approaches like AdaBoost may be better suited for scenarios in which balanced detection of disease recurrence and minimization of missed recurrences are prioritized. Feature importance analysis across multiple ML models demonstrated that histological subtypes were consistently retained, indicating that tumor histology represents a stable and informative variable across different modeling approaches. The frequent selection of tumor grade, tumor size, and ECOG performance status further supports the relevance of both pathological characteristics and baseline clinical condition in recurrence prediction. In contrast, SHAP analysis of the AdaBoost model, one of the best-performing classifiers, highlighted ECOG performance status and tumor size as the dominant contributors to individual predictions, suggesting that these factors exert the most direct influence on the model’s decision-making process. Variables such as chemotherapy cycle number and vascular invasion provided additional, moderate contributions, while other features, including age at diagnosis, lymphatic invasion, adjuvant radiotherapy, tumor grade, and pathological substage, played a more limited role in shaping predictions within this model. Taken together, these results indicate that while histological features are important for model stability and selection across algorithms, clinical status and tumor burden have a more prominent impact on the predictive behavior of high-performing models. This distinction underscores the value of integrating both model-agnostic feature selection and model-specific interpretability approaches to better understand the relative roles of clinical and pathological variables in disease recurrence. Given that the primary objective of this study was to predict recurrence, the findings of the Cox regression and SHAP analyses can be interpreted as complementary but functionally distinct. In the multivariate Cox model, ECOG performance status, advanced T stage, tumor grade, vascular invasion, and adjuvant radiotherapy status emerged as independent predictors of recurrence-free survival, indicating variables with a consistent association with recurrence risk at the population level. In contrast, SHAP analysis of the AdaBoost model identified ECOG performance status and tumor size as the main drivers of individual recurrence predictions, while chemotherapy cycle number and vascular invasion contributed more modestly. Other variables identified as prognostic in the Cox analysis contributed less to the machine-learning model’s predictive performance. This difference likely reflects the distinct purposes of the two approaches: Cox regression estimates the overall effect of each variable on recurrence risk across the population over time, whereas SHAP explains which variables most strongly influence predictions for individual patients within a non-linear model. Consequently, factors that show consistent prognostic associations at the population level may exert a smaller impact on individual-level predictions. Machine-learning models may therefore provide additional value for individualized recurrence risk estimation. Similar machine learning–based approaches combining feature selection with kernel-based support vector methods have demonstrated strong prognostic performance in other solid tumors. In hepatocellular carcinoma, ML models integrating clinical and pathological variables achieved accuracy rates exceeding 85–90% for survival prediction across different disease stages [ 20 ]. Likewise, in advanced pancreatic cancer, support vector machine–based models combined with feature selection methods yielded survival prediction accuracies of approximately 88%, highlighting the robustness and generalizability of such approaches across distinct oncologic settings [ 21 ]. Extending this prognostic framework from survival outcomes to disease recurrence, several studies have investigated the use of machine learning models specifically for predicting postoperative recurrence risk in patients with resected NSCLC. In a study including 1,387 patients with early-stage resected NSCLC, ML models achieved a best accuracy of 76% and an AUC of 0.81; notably, F1 scores were not reported [ 22 ]. In another study, a multimodal deep learning–based ensemble model integrating clinical data with handcrafted and deep learning–based radiomic features achieved an accuracy of approximately 73% and an F1 score of 0.78, with the ensemble approach yielding the highest ROC–AUC among the evaluated models [ 23 ]. In a recent study conducted in a cohort of 309 surgically resected NSCLC patients, ML models and Cox proportional hazards regression demonstrated comparable performance for recurrence prediction, with reported AUC values ranging between approximately 0.74 and 0.77 for 2-year and 5-year postoperative recurrence events, while accuracy- and F1-based metrics were not provided [ 24 ]. Compared to these studies, our study demonstrated comparable predictive performance for postoperative recurrence. Specifically, the SVC-RBF model achieved an AUC of 0.814, an accuracy of approximately 77%, and an F1 score of 0.842, while AdaBoost yielded the highest accuracy (≈ 79%) and F1 score (0.867), albeit with a lower AUC of 0.680. The performance of our models was consistent with that reported in the aforementioned studies. Moreover, our study uniquely incorporated confusion matrix–based evaluation alongside feature importance and SHAP analyses, thereby enhancing model interpretability and reinforcing the clinical relevance of the findings. This study has several limitations. First, genomic and molecular data were not included in the analysis, as such information remains largely unavailable and insufficiently characterized in routine clinical practice for this patient population. Second, the lack of temporal modeling represents a limitation of the present study, since longitudinal changes in clinical and pathological variables could not be captured. Finally, the single-center design may introduce institutional bias and limit the generalizability of the findings to broader populations. Conclusions This study demonstrates that ML–based models can reliably predict recurrence risk after curative resection in NSCLC using routinely available clinicopathologic variables. The complementary strengths of kernel-based and ensemble approaches highlight their potential utility for individualized recurrence risk assessment. These findings support the integration of ML models as decision-support tools in postoperative risk stratification, while emphasizing the need for future studies incorporating genomic data to further enhance predictive performance and generalizability. Declarations Data availability Because of ethical and privacy considerations, the raw data are not publicly accessible. However, the datasets generated and analyzed during this study can be obtained from the corresponding author upon reasonable request. Acknowledgments The authors thank Furkan Aydos and Hatice Rüveyda Akça for their support in the preparation of this study. Author contributions Conceptualization: Ugur Ozberk, Selin Akturk Esen; Methodology: Ugur Ozberk, Selin Akturk Esen; Formal analysis and investigation: Ugur Ozberk, Serkan Keskin, Hilal Arslan, Melike Cobankaya; Data curation: Ugur Ozberk, Serkan Keskin, Oznur Bal, Efnan Algın; Writing - original draft preparation: Ugur Ozberk; Writing - review and editing: Selin Akturk Esen, Burak Bilgin, Mehmet Ali Nahit Sendur, Dogan Uncu; Resources: Oznur Bal, Efnan Algın, Burak Bilgin; Supervision: Mehmet Ali Nahit Sendur, Dogan Uncu Declaration of Conflict of Interests The authors report no financial or personal conflicts of interest associated with this study. Funding This study was conducted without any external financial support. Institutional Review Board Statement The study was approved by the Clinical Research Ethics Committee of Ankara City Hospital (Decision No: TABED 1/1772/2025, Date: 22/10/2025) and was conducted in accordance with the principles of the Declaration of Helsinki. Informed Consent Statement Informed consent was waived by the Clinical Research Ethics Committee of Ankara City Hospital due to the retrospective design of the study and the use of anonymized data obtained from medical records. References Thai, A. A., Solomon, B. J., Sequist, L. V., Gainor, J. F. & Heist, R. S. Lung cancer. Lancet 398 , 535–554. 10.1016/S0140-6736(21)00312-3 (2021). Su, S. et al. Patterns of survival and recurrence after surgical treatment of early stage non-small cell lung carcinoma in the ACOSOG Z0030 (ALLIANCE) trial. J. Thorac. Cardiovasc. Surg. 147 , 747–752. 10.1016/j.jtcvs.2013.10.001 (2014). Discussion 752 – 743. Uramoto, H. & Tanaka, F. Recurrence after surgery in patients with NSCLC. Transl Lung Cancer Res. 3 , 242–249. 10.3978/j.issn.2218-6751.2013.12.05 (2014). Farina, E., Nabhen, J. J., Dacoregio, M. I., Batalini, F. & Moraes, F. Y. An overview of artificial intelligence in oncology. Future Sci. OA . 8 , FSO787. 10.2144/fsoa-2021-0074 (2022). Gould, M. K., Huang, B. Z., Tammemagi, M. C., Kinar, Y. & Shiff, R. Machine Learning for Early Lung Cancer Identification Using Routine Clinical and Laboratory Data. Am. J. Respir Crit. Care Med. 204 , 445–453. 10.1164/rccm.202007-2791OC (2021). Lynch, C. M. et al. Prediction of lung cancer patient survival via supervised machine learning classification techniques. Int. J. Med. Inf. 108 , 1–8. 10.1016/j.ijmedinf.2017.09.013 (2017). Detterbeck, F. C. et al. The Proposed Ninth Edition TNM Classification of Lung Cancer. Chest 166 , 882–895. 10.1016/j.chest.2024.05.026 (2024). Liu, H. & Setiono, R. in Proceedings of 7th IEEE international conference on tools with artificial intelligence. 388–391 (Ieee). Guyon, I. & Elisseeff, A. An introduction to variable and feature selection. J. Mach. Learn. Res. 3 , 1157–1182 (2003). Breiman, L., Friedman, J., Olshen, R. A. & Stone, C. J. Classification and regression trees (Chapman and Hall/CRC, 2017). Hosmer, D. W. Jr, Lemeshow, S. & Sturdivant, R. X. Applied logistic regression (Wiley, 2013). Cortes, C. & Vapnik, V. Support-vector networks. Mach. Learn. 20 , 273–297 (1995). Zhang, H. The optimality of naive Bayes. Aa 1 , 3 (2004). Ripley, B. D. Neural networks and related methods for classification. J. Roy. Stat. Soc.: Ser. B (Methodol.) . 56 , 409–437 (1994). Hofmann, T., Schölkopf, B. & Smola, A. J. Kernel methods in machine learning. (2008). Zhang, Z. Introduction to machine learning: k-nearest neighbors. Annals translational Med. 4 , 218 (2016). Mienye, I. D. & Sun, Y. A survey of ensemble learning: Concepts, algorithms, applications, and prospects. Ieee Access. 10 , 99129–99149 (2022). Chen, T. & XGBoost A Scalable Tree Boosting System. Cornell University (2016). Ke, G. et al. Lightgbm: A highly efficient gradient boosting decision tree. Advances neural Inform. Process. systems 30 (2017). Seven, I. et al. Predicting hepatocellular carcinoma survival with artificial intelligence. Sci. Rep. 15 , 6226. 10.1038/s41598-025-90884-6 (2025). Seven, I. et al. Predicting survival outcomes in advanced pancreatic cancer using machine learning methods. Med. (Baltim). 104 , e43904. 10.1097/MD.0000000000043904 (2025). Janik, A. et al. Machine Learning-Assisted Recurrence Prediction for Patients With Early-Stage Non-Small-Cell Lung Cancer. JCO Clin. Cancer Inf. 7 , e2200062. 10.1200/CCI.22.00062 (2023). Kim, G., Moon, S. & Choi, J. H. Deep Learning with Multimodal Integration for Predicting Recurrence in Patients with Non-Small Cell Lung Cancer. Sens. (Basel) . 22 10.3390/s22176594 (2022). Pu, L., Dhupar, R. & Meng, X. Predicting Postoperative Lung Cancer Recurrence and Survival Using Cox Proportional Hazards Regression and Machine Learning. Cancers (Basel) . 17 10.3390/cancers17010033 (2024). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 09 Apr, 2026 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 27 Mar, 2026 Reviews received at journal 19 Mar, 2026 Reviewers agreed at journal 19 Mar, 2026 Reviews received at journal 18 Mar, 2026 Reviewers agreed at journal 18 Mar, 2026 Reviewers agreed at journal 18 Mar, 2026 Reviewers invited by journal 17 Mar, 2026 Editor assigned by journal 17 Mar, 2026 Editor invited by journal 24 Feb, 2026 Submission checks completed at journal 20 Feb, 2026 First submitted to journal 20 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8912361","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":609100904,"identity":"206684eb-fb0b-4816-99a5-e9b34bda7754","order_by":0,"name":"Ugur Ozberk","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6klEQVRIiWNgGAWjYFCCBAhlwMDAzPAByGBjJ0UL4wyQFmZStDDzgFiEtOi25xi/5t1jY2/O3vvY2ObXNnk+ZgbGDx9zcGsxO/PGzJrnWRqzZc9x4+TcvtuGbcwMzJIzt+HRciPHzJjnwGE2gxtpzIdze24zArWwMfMSoYXH4P4z5sOWPbftidFi/BioRcLgBhtzMsOP24mEtZx5VsY450CagWVPGrNhb8Pt5DZmxmb8fjmevPnDmwOgEDvGLPHjz23b+e3NBz98xKMFCNgk4EzGNjDZgFc9EDB/QLD/EFI8CkbBKBgFIxEAAMjOTmEd4hILAAAAAElFTkSuQmCC","orcid":"","institution":"Ankara City Hospital","correspondingAuthor":true,"prefix":"","firstName":"Ugur","middleName":"","lastName":"Ozberk","suffix":""},{"id":609100913,"identity":"15a9dc0a-b23a-41e2-a07b-7a7c8f8f4d04","order_by":1,"name":"Selin Akturk Esen","email":"","orcid":"","institution":"Bursa Nilüfer Doruk Hospital","correspondingAuthor":false,"prefix":"","firstName":"Selin","middleName":"Akturk","lastName":"Esen","suffix":""},{"id":609100914,"identity":"c14e2da5-4265-40be-a85f-1bc50419ca2a","order_by":2,"name":"Hilal Arslan","email":"","orcid":"","institution":"Ankara Yıldırım Beyazıt University","correspondingAuthor":false,"prefix":"","firstName":"Hilal","middleName":"","lastName":"Arslan","suffix":""},{"id":609100915,"identity":"7f35d8c8-3e4d-4f7a-8ceb-8d02f98c5688","order_by":3,"name":"Oznur Bal","email":"","orcid":"","institution":"Ankara City Hospital","correspondingAuthor":false,"prefix":"","firstName":"Oznur","middleName":"","lastName":"Bal","suffix":""},{"id":609100916,"identity":"2865efbf-2225-496f-8649-8c3c9a2fd45e","order_by":4,"name":"Efnan Algın","email":"","orcid":"","institution":"Ankara City Hospital","correspondingAuthor":false,"prefix":"","firstName":"Efnan","middleName":"","lastName":"Algın","suffix":""},{"id":609100917,"identity":"1b7f7604-f2d6-4d52-a77a-ec64274d8a32","order_by":5,"name":"Serkan Keskin","email":"","orcid":"","institution":"Ankara City Hospital","correspondingAuthor":false,"prefix":"","firstName":"Serkan","middleName":"","lastName":"Keskin","suffix":""},{"id":609100921,"identity":"9a41b799-812d-41c3-baa1-86ee9213f052","order_by":6,"name":"Burak Bilgin","email":"","orcid":"","institution":"Ankara City Hospital","correspondingAuthor":false,"prefix":"","firstName":"Burak","middleName":"","lastName":"Bilgin","suffix":""},{"id":609100923,"identity":"124fa413-65d0-481a-bca4-09bbbeddb0ed","order_by":7,"name":"Melike Cobankaya","email":"","orcid":"","institution":"Ankara Yıldırım Beyazıt University","correspondingAuthor":false,"prefix":"","firstName":"Melike","middleName":"","lastName":"Cobankaya","suffix":""},{"id":609100925,"identity":"952f1bd0-320b-4cbf-9a44-c6a64ceebde4","order_by":8,"name":"Mehmet Ali Nahit Sendur","email":"","orcid":"","institution":"Ankara City Hospital","correspondingAuthor":false,"prefix":"","firstName":"Mehmet","middleName":"Ali Nahit","lastName":"Sendur","suffix":""},{"id":609100927,"identity":"58c5ff3f-48da-451f-963c-8da095533817","order_by":9,"name":"Dogan Uncu","email":"","orcid":"","institution":"Ankara City Hospital","correspondingAuthor":false,"prefix":"","firstName":"Dogan","middleName":"","lastName":"Uncu","suffix":""}],"badges":[],"createdAt":"2026-02-18 21:08:47","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8912361/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8912361/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-026-47862-3","type":"published","date":"2026-04-09T15:57:47+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":105149093,"identity":"a47a8f5d-4d4d-46da-a040-18cfe0690474","added_by":"auto","created_at":"2026-03-22 14:52:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":59579,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion matrices of the best-performing models (AdaBoost and SVC-RBF) based on accuracy\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8912361/v1/cfab47938956116372e8d085.png"},{"id":105149092,"identity":"3880bd0a-1d67-4642-85de-4700a6f28b1e","added_by":"auto","created_at":"2026-03-22 14:52:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":250595,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparison of AUC scores across machine learning models for recurrence prediction\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8912361/v1/e6fccf823b59af749c14c323.png"},{"id":105903764,"identity":"0066e336-1454-41d9-a0c9-8f8459b35563","added_by":"auto","created_at":"2026-04-01 09:52:39","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":224445,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparison of F1 scores across machine learning models for recurrence prediction\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8912361/v1/6530f99a0d86725c2307838d.png"},{"id":105149090,"identity":"7e3817da-c8f5-4cf5-8c7b-de121c5922b9","added_by":"auto","created_at":"2026-03-22 14:52:46","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":63793,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDistribution of key predictive features across machine learning models\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8912361/v1/675ad255f1204d7bac1da9e8.png"},{"id":105149089,"identity":"de455f7a-aaab-40b9-b1e3-abd4f4c1e01c","added_by":"auto","created_at":"2026-03-22 14:52:46","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":59935,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFeature importance based on SHAP analysis for recurrence prediction\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8912361/v1/2b47bb01bcd60ef6dafaa1b0.png"},{"id":106810665,"identity":"c0ab9777-8040-4016-a77b-0edafcb58c01","added_by":"auto","created_at":"2026-04-13 16:16:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2232500,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8912361/v1/85751fba-ae4b-4a53-a025-c6e3613f434b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning–Based Prediction of Recurrence After Curative Resection in Non–Small Cell Lung Cancer","fulltext":[{"header":"Introduction","content":"\u003cp\u003eLung cancer is among the most commonly diagnosed cancers and remains a leading cause of cancer-related mortality. Each year, an estimated two million new cases and 1.76\u0026nbsp;million deaths occur worldwide [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Despite these concerning statistics, survival rates are favorable in early-stage disease [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. However, disease recurrence remains a major clinical challenge. Despite curative resection, approximately 30\u0026ndash;55% of patients with non\u0026ndash;small cell lung cancer (NSCLC) develop recurrence and ultimately die from their disease, with the recurrence risk varying by pathologic stage, nodal status, extent of resection, and histologic subtype [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eArtificial intelligence (AI) has advanced oncologic care, with benefits spanning diagnosis through treatment, supported by techniques such as machine learning (ML) and deep learning [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. In support of this concept, a large population-based study demonstrated the potential of machine learning to predict lung cancer risk prior to clinical diagnosis [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Using routinely collected clinical and laboratory data from more than 6,500 patients with non\u0026ndash;small cell lung cancer and nearly 190,000 control subjects, the authors developed a machine learning model that outperformed established risk-prediction tools, including the mPLCOm2012 model, in identifying individuals at high risk up to 9\u0026ndash;12 months before diagnosis. The machine learning approach achieved superior discriminative performance and sensitivity at high specificity, highlighting the added value of data-driven models for early risk stratification in lung cancer. In another study, supervised machine learning techniques applied to the SEER database demonstrated that lung cancer survival time could be predicted using routinely collected clinical variables, with gradient boosting and ensemble models achieving performance comparable to classical Cox proportional hazards models [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAI may also play an important role in predicting lung cancer recurrence. AI-enabled estimation of lung cancer recurrence risk can optimize remission-phase management and facilitate earlier detection, allowing clinicians to concentrate resources on higher-risk patients and thereby improving overall care. This study aimed to resolve persistent limitations in postoperative recurrence prediction for NSCLC by employing ML approaches. The specific aims were: (i) to evaluate the ability of diverse ML algorithms to predict postoperative recurrence using a comprehensive set of clinicopathological and treatment-related variables; (ii) to identify key clinicopathologic, perioperative, and laboratory determinants of recurrence using feature-selection procedures (χ\u0026sup2; tests, ANOVA, Kruskal\u0026ndash;Wallis); and (iii) to assess the concordance between ML\u0026ndash;based predictions and prognostic factors identified by Cox proportional hazards analysis, while improving model interpretability through feature importance and SHAP analyses to elucidate the drivers of individual-level predictions. Through this study, identifying patients at high risk of recurrence in non\u0026ndash;small cell lung cancer, a disease with high relapse rates, may enable more individualized postoperative management and improved treatment and surveillance strategies.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eThis retrospective study was conducted at the Department of Medical Oncology, Ankara Bilkent City Hospital. Ethical approval was obtained from the Institutional Review Board of Ankara Bilkent City Hospital (Decision No: TABED 1/1772/2025, Date: 22/10/2025), and the study was performed in accordance with the principles of the Declaration of Helsinki. Owing to the retrospective nature of the study, the requirement for written informed consent was waived.\u003c/p\u003e \u003cp\u003ePatients with histopathologically confirmed NSCLC who underwent curative-intent surgical resection and were subsequently followed for disease recurrence between January 2020 and June 2025 were included in the study. Patients with incomplete clinical or pathological data, missing follow-up information, or non-curative surgical intent were excluded.\u003c/p\u003e \u003cp\u003eClinical, pathological, and treatment-related data were collected retrospectively from the hospital\u0026rsquo;s electronic medical record system and archival files. Recorded variables included age at diagnosis, sex, ECOG performance status, histological subtype, tumor size, pathological T and N stage classified according to the New Ninth Edition TNM Classification for Lung Cancer [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], tumor grade, visceral pleural invasion, surgical margin status, vascular and lymphatic invasion, type of surgical resection, adjuvant chemotherapy regimen and number of cycles, receipt of adjuvant radiotherapy or chemoradiotherapy, date of diagnosis, and date of recurrence.\u003c/p\u003e \u003cp\u003eRecurrence was defined as radiological or histopathological evidence of locoregional or distant disease relapse following curative resection. Patients were followed until documented recurrence, death, or last clinical follow-up.\u003c/p\u003e\n\u003ch3\u003e1. SPSS Software Version 25.0 Analysis\u003c/h3\u003e\n\u003cp\u003eData analysis was conducted using SPSS software version 25.0 (IBM Corp., Armonk, NY, USA). Continuous variables were summarized as median values (minimum and maximum ranges), while categorical variables were presented as frequencies and percentages. Recurrence-free survival (RFS) was defined as the time from curative resection to the date of first documented disease recurrence or death, whichever occurred first. Factors associated with RFS were initially evaluated using univariate Cox proportional hazards regression analysis and multivariate Cox regression model to identify independent prognostic factors. A p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e\n\u003ch3\u003e2. Feature Selection Process\u003c/h3\u003e\n\u003cp\u003eFeature importance for predicting patient outcomes was assessed using three complementary statistical methods: Chi-square (Chi\u0026sup2;), Analysis of Variance (ANOVA), and the Kruskal\u0026ndash;Wallis test. These techniques were selected to capture different statistical properties of categorical and continuous variables and to improve model interpretability and performance. Features with statistically significant results (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) were considered informative and retained.\u003c/p\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Chi-square test\u003c/h2\u003e \u003cp\u003eThe Chi-square test was applied to evaluate the association between categorical features and the target variable by comparing observed and expected frequencies under the assumption of independence [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Analysis of variance test\u003c/h2\u003e \u003cp\u003eAnalysis of variance test was used to identify continuous variables showing significant mean differences across outcome groups [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Features with p-values below 0.05 were interpreted as having discriminative power and were included in the reduced feature set.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Kruskal\u0026ndash;Wallis test\u003c/h2\u003e \u003cp\u003eFor variables that violated normality assumptions, the Kruskal\u0026ndash;Wallis test, a non-parametric alternative to ANOVA, was employed [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Features with significant median differences across groups were selected, enabling a more robust and assumption-free feature selection strategy.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e3. Machine Learning Methods\u003c/h3\u003e\n\u003cp\u003eA comprehensive comparative analysis of multiple supervised ML algorithms was conducted to evaluate classification performance on the study dataset.\u003c/p\u003e \u003cp\u003eModel performance was evaluated using standard classification metrics derived from the confusion matrix, which consists of true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN). Precision was defined as TP/(TP\u0026thinsp;+\u0026thinsp;FP), representing the proportion of predicted positive cases that were correctly identified. Recall (sensitivity) was calculated as TP/(TP\u0026thinsp;+\u0026thinsp;FN) to quantify the model\u0026rsquo;s ability to detect actual positive cases. The F1 score, computed as the harmonic mean of precision and recall, provided a balanced measure of predictive performance. Overall accuracy was determined using (TP\u0026thinsp;+\u0026thinsp;TN)/(TP\u0026thinsp;+\u0026thinsp;TN\u0026thinsp;+\u0026thinsp;FP\u0026thinsp;+\u0026thinsp;FN), reflecting the total proportion of correct predictions. Additionally, the area under the receiver operating characteristic curve (AUC) was calculated using the trapezoidal rule to assess the threshold-independent discriminative capacity of each model. All metrics ranged from 0 to 1, with higher values indicating superior classification performance.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1. Decision Tree Models\u003c/h2\u003e \u003cp\u003eDecision Trees were implemented to derive interpretable rules and partition the dataset into homogeneous subgroups [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. The study evaluated multiple tree-based variants including Fine, Medium, Coarse Trees, Boosted Trees, Bagged Trees, and RUSBoosted Trees. A grid search was conducted to optimize key hyperparameters such as minimum leaf size (tested at 1, 5, 20, and 50) and the Gini index as the splitting criterion. The best performance was achieved with a minimum leaf size of 1, indicating that finer granularity provided superior classification capability.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.2. Regression-Based Models\u003c/h2\u003e \u003cp\u003eRegression models, particularly Binary GLM Logistic Regression and Efficient Logistic Regression, were adopted to capture linear relationships between predictors and recurrence outcomes [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. These models estimate the probability of class membership through logistic functions, enabling clear interpretability and strong baseline predictive performance. Their inclusion ensured that both parametric and non-parametric modeling perspectives were represented within the study.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Support Vector Machines\u003c/h2\u003e \u003cp\u003eSupport Vector Machines (SVM) were used as robust classifiers capable of handling high-dimensional and non-linear feature interactions [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Seven SVM configurations were examined, including Linear, Efficient Linear, Quadratic, Cubic, and Gaussian (Fine, Medium, Coarse) kernels. Hyperparameter optimization was conducted using grid search across values of the regularization constant C (0.1, 1, 10, 100, 1000). The optimal performance was consistently obtained at C\u0026thinsp;=\u0026thinsp;1, balancing margin maximization and misclassification tolerance effectively.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.4. Naive Bayes Models\u003c/h2\u003e \u003cp\u003eNaive Bayes classifiers, including Gaussian and Kernel-based variants, were utilized owing to their computational efficiency and probabilistic interpretability [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. These models assume conditional independence among features, enabling fast and effective classification, especially for high-dimensional datasets. Their inclusion provided a lightweight yet powerful comparative baseline against more complex methods.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.5. Neural Network Models\u003c/h2\u003e \u003cp\u003eArtificial Neural Networks were applied to model complex, non-linear relationships within the dataset [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Five architectures were employed: Trilayered Neural Network, Narrow Neural Network, Wide Neural Network, Bilayered Neural Network, and Medium Neural Network. Hidden layers used sigmoid activation functions, while output layers used softmax activation. Network structures varied from shallow (5 neurons in a single layer) to deep (three layers with 10 neurons each), enabling evaluation of how architectural complexity influences predictive performance.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.6. Kernel-Based Methods\u003c/h2\u003e \u003cp\u003eKernel transformation techniques were incorporated to address non-linear classification challenges [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Two kernelized models were evaluated: SVM Kernel and Logistic Regression Kernel. These methods map input features into higher-dimensional spaces where linear separation is more feasible, improving classification power for datasets with inherently non-linear decision boundaries.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e3.7. K-Nearest Neighbor (KNN)\u003c/h2\u003e \u003cp\u003eK-Nearest Neighbor was employed as a distance-based, instance-learning algorithm [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Several variants including Weighted, Cubic, Medium, Cosine, Coarse, and Fine KNN were tested. Hyperparameters evaluated via grid search included the number of neighbors (k ranging from 1 to 30) and distance metrics (Euclidean, Minkowski, Chebyshev). Optimal performance occurred at k\u0026thinsp;=\u0026thinsp;1 with Euclidean distance, indicating that local neighborhood similarity was highly informative for classification in this dataset.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e3.8. Ensemble Classifiers\u003c/h2\u003e \u003cp\u003eEnsemble classifiers combine the predictive capabilities of multiple base learners to improve model performance and robustness beyond that of individual algorithms [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. In this study, several ensemble methods were implemented, including Bagged Trees, Subspace KNN, Boosted Trees, RUSBoosted Trees, and Subspace Discriminant models. Additionally, two advanced gradient boosting algorithms, Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM), were incorporated to further enhance predictive accuracy.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e3.9. XGBoost and LightGBM\u003c/h2\u003e \u003cp\u003eXGBoost employs decision trees as base learners and builds them sequentially in a boosting framework, where each new tree focuses on correcting the errors of preceding trees [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. This iterative refinement process increases accuracy and robustness, making XGBoost one of the most powerful boosting algorithms available.\u003c/p\u003e \u003cp\u003eLightGBM, an optimized gradient boosting framework, utilizes a leaf-wise tree growth strategy, enabling faster processing and reduced memory usage compared to traditional boosting methods [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Its ability to handle large datasets efficiently contributes to its growing popularity in ML applications.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eA total of 265 patients were included in the study. The majority were male (n\u0026thinsp;=\u0026thinsp;220, 83%), while 45 (17%) were female. The median age at diagnosis was 63 (39\u0026ndash;85) years. Most patients had an ECOG performance status of 1 (n\u0026thinsp;=\u0026thinsp;208, 78.5%), followed by ECOG 2 (n\u0026thinsp;=\u0026thinsp;35, 13.2%) and ECOG 0 (n\u0026thinsp;=\u0026thinsp;22, 8.3%). The most frequent histological subtype was adenocarcinoma (n\u0026thinsp;=\u0026thinsp;144, 54.3%), followed by squamous cell carcinoma (n\u0026thinsp;=\u0026thinsp;105, 39.6%). T-stage distribution was as follows: T1a 5 (1.9%), T1b 51 (19.2%), T1c 32 (12.1%), T2a 72 (27.2%), T2b 45 (17%), T3 42 (15.8%), and T4 18 (6.8%). Nodal status showed N0 in 176 (66.4%), N1 in 54 (20.4%), and N2 disease in 35 (13.2%). Visceral pleural invasion was present in 104 (39.2%), while surgical margins were negative in 244 (92.1%). Tumor grade distribution included Grade 1: 37 (14%), Grade 2: 93 (35.1%), and Grade 3: 63 (23.8%). Vascular invasion was detected in 109 (41.1%). Wedge resection was performed in 48 (18.1%), and lymphatic invasion was present in 89 (33.6%). Mutation testing revealed EGFR positivity in 11 (4.2%), ALK in 7 (2.6%), ROS1 in 1 (0.4%), and BRAF in 3 (1.1%), while 69 (26%) were mutation-negative. Adjuvant chemotherapy was administered to 177 (66.8%), adjuvant radiotherapy to 31 (11.7%), and adjuvant chemoradiotherapy to 12 (4.5%). Recurrence occurred in 82 patients (30.9%) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBaseline clinicopathologic and treatment characteristics of the patients\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMedian (Min-Max)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eN (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge at diagnosis (years)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63 (39\u0026ndash;85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor size (cm)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3.2 (0.8\u0026ndash;16)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eNumber of cycles\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (0\u0026ndash;6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex (Male)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e220 (83)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eECOG performance score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECOG score 0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22 (8.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECOG score 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e208 (78.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECOG score 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e35 (13.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHistologic subtype\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdenocarcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e144 (54.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSquamous cell carcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e109 (41.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLarge-cell carcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6 (2.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePleomorphic carcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (1.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNOS (Not otherwise specified)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (0.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eT stage\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1a\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5 (1.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1b\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51 (19.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1c\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32 (12.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT2a\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e72 (27.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT2b\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e45 (17)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e42 (15.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18 (6.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eN stage\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e176 (66.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e54 (20.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e35 (13.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eVisceral pleural invasion\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e104 (39.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e158 (59.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSurgical margin\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e244 (92.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17 (6.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (1.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor grade\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e37 (14)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e93 (35.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63 (23.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e72 (27.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eVascular invasion\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e109 (41.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e153 (57.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLymphatic invasion\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e89 (33.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e174 (65.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (0.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eWedge resection\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerformed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e48 (18.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot performed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e215 (81.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (0.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMutation status\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e69 (26)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEGFR-mutant\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11 (4.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALK-mutant\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (2.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eROS1-mutant\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBRAF-mutant\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (1.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e174 (65.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdjuvant chemotherapy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e88 (33.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCisplatin\u0026ndash;vinorelbine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e125 (47.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCisplatin\u0026ndash;gemcitabine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20 (7.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCarboplatin\u0026ndash;paclitaxel\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15 (5.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCisplatin\u0026ndash;pemetrexed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e9 (3.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCarboplatin\u0026ndash;vinorelbine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e5 (1.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCarboplatin\u0026ndash;gemcitabine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (0.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCarboplatin\u0026ndash;pemetrexed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (0.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdjuvant radiotherapy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e31 (11.7)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e234 (88.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdjuvant chemoradiotherapy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12 (4.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e253 (95.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRecurrence\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e82 (30.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e183 (69.1)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn the univariate Cox regression analysis, several variables were significantly associated with recurrence-free survival (RFS). Higher ECOG performance status, larger tumor size, advanced T stage, higher tumor grade, presence of vascular invasion, lymphatic invasion, visceral pleural invasion, and receipt of adjuvant radiotherapy were all associated with poorer RFS. Female sex was associated with better RFS compared with male sex. Other variables were not significantly associated with RFS (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn the multivariate Cox regression analysis, ECOG performance status remained an independent predictor of RFS (p\u0026thinsp;=\u0026thinsp;0.012), with ECOG 2 showing a markedly increased risk compared with ECOG 0 (HR 3.84, 95% CI 1.28\u0026ndash;11.52). Advanced T stage also retained significance (p\u0026thinsp;=\u0026thinsp;0.012), with T4 disease associated with a substantially elevated recurrence risk (HR 3.24, 95% CI 1.51\u0026ndash;6.93). Higher tumor grade remained independently associated with poorer RFS (p\u0026thinsp;=\u0026thinsp;0.042), particularly for grade 3 tumors (HR 3.62, 95% CI 1.15\u0026ndash;11.39). Absence of vascular invasion continued to confer a protective effect (HR 0.49, 95% CI 0.26\u0026ndash;0.91, p\u0026thinsp;=\u0026thinsp;0.021). Additionally, patients who did not receive adjuvant radiotherapy demonstrated better RFS than those who did (HR 0.41, 95% CI 0.22\u0026ndash;0.78, p\u0026thinsp;=\u0026thinsp;0.012). Other variables were not independently associated with RFS in the multivariate model (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eUnivariate and Multivariate Cox Regression Analysis for Recurrence-Free Survival\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHR (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eHR (95% CI)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ep-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge at diagnosis (years)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.02 (0.99\u0026ndash;1.05)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.080\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.047\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.49 (0.24\u0026ndash;0.99)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eECOG performance score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECOG score 0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.003\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.012\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECOG score 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.64 (0.59\u0026ndash;4.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.09 (0.73\u0026ndash;6.02)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eECOG score 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.54(1.22\u0026ndash;10.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.84 (1.28\u0026ndash;11.52)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHistologic subtype\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdenocarcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.659\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSquamous cell carcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.99 (0.63\u0026ndash;1.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.47 (0.62\u0026ndash;3.46)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor size (cm)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.16 (1.07\u0026ndash;1.26)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eT stage\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.012\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.17 (0.68\u0026ndash;2.01)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.08 (0.62\u0026ndash;1.88)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.45 (0.73\u0026ndash;2.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.15 (0.57\u0026ndash;2.32)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eT4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e5.14 (2.50-10.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.24 (1.51\u0026ndash;6.93)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eN stage\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.568\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.86 (0.52\u0026ndash;1.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eN2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.08 (0.61\u0026ndash;1.92)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eVisceral pleural invasion\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.019\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.59 (0.38\u0026ndash;0.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSurgical margin\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.052\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.99 (0.99\u0026ndash;3.98)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor grade\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.042\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.05 (1.06\u0026ndash;8.76)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.20 (0.74\u0026ndash;6.55)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6.08 (2.117.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e3.62 (1.15\u0026ndash;11.39)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eVascular invasion\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.021\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.38 (0.24\u0026ndash;0.60)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.49 (0.26\u0026ndash;0.91)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLymphatic invasion\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.03 (1.31\u0026ndash;3.14)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eWedge resection\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerformed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.746\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot performed\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.91 (0.52\u0026ndash;1.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMutation status\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNegative\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.560\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePositive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.78 (0.34\u0026ndash;1.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdjuvant chemotherapy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.30 (0.81\u0026ndash;2.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eNumber of cycles\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.01 (0.90\u0026ndash;1.13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.841\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdjuvant radiotherapy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e\u0026lt;\u0026thinsp;0.001\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.012\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.40 (0.23\u0026ndash;0.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.41 (0.22\u0026ndash;0.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdjuvant chemoradiotherapy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.00\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.865\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.09 (0.39\u0026ndash;2.98)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe classification performance of multiple ML algorithms was evaluated using the full dataset consisting of 17 variables (sex, age at diagnosis, tumor size, pT stage, pN stage, visceral pleural involvement, surgical margin status, tumor grade, vascular invasion, wedge resection (performed/not performed), lymphatic invasion, histological subtype, ECOG performance status, adjuvant chemotherapy regimen, number of chemotherapy cycles, receipt of adjuvant radiotherapy and receipt of chemoradiotherapy ), as well as reduced feature subsets obtained through ANOVA, Chi-square, and Kruskal\u0026ndash;Wallis feature selection methods. Model performance was compared using accuracy as the primary evaluation metric (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of recurrence prediction accuracy across machine learning methods\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAll Features\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13 Features by ANOVA\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13 Features by Chi2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e14 Features by Kruskal-Wallis\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdaBoost\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7925\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSupport Vector Classifier RBF\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMLPClassifier\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.6604\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6792\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6038\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBernoulliNB\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSupport Vector Classifier Poly\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7925\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGradientBoosting\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6604\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRidgeClassifier\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eBaggingClassifier\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.6415\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6981\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6604\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eQuadraticDiscriminant\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExtraTrees\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6981\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRandomForest\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCalibrated SVM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLinearDiscriminant\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.6981\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eNu-Support Vector Classifier\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.6981\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eKNeighbors\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.6981\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6038\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLinear Support Vector Classifier\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.6792\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7736\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6981\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMLPClassifier_Deep\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.6792\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.6415\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6792\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6415\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSGDClassifier\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.6604\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.6792\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.6038\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.6604\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLogistic Regression\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.6604\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7170\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eMachine learning accuracy results showed that AdaBoost achieved the highest performance using the full feature set (accuracy\u0026thinsp;\u0026asymp;\u0026thinsp;0.79). SVC-RBF and MLP-based models also demonstrated strong performance, with accuracies around 0.77. Model performance remained broadly comparable across feature-selection strategies, and no single selection method consistently outperformed the others. Ensemble- and kernel-based classifiers generally showed more stable performance, whereas simpler linear models tended to yield lower accuracy. Overall, predictive accuracy across models ranged approximately between 0.66 and 0.79 (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBased on overall accuracy, confusion matrices were generated for the two best-performing models, AdaBoost and the Support Vector Classifier with an RBF kernel (SVC-RBF). For both models, recurrence predictions were classified as progression present or absent. The confusion matrices showed that AdaBoost correctly classified 36 progression cases and 6 non-progression cases, with 1 false-negative and 10 false-positive predictions. SVC-RBF correctly identified 32 progression cases and 9 non-progression cases, with 5 false-negative and 7 false-positive predictions (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe area under the receiver operating characteristic curve (AUC) was calculated to further evaluate the discriminatory performance of the ML models in predicting disease recurrence. Among the evaluated algorithms, the Support Vector Classifier with RBF kernel (SVC-RBF) achieved the highest AUC value (AUC\u0026thinsp;=\u0026thinsp;0.814), indicating superior discriminative ability. This was followed by the Support Vector Classifier with polynomial kernel (AUC\u0026thinsp;=\u0026thinsp;0.779) and the Calibrated Support Vector Machine (AUC\u0026thinsp;=\u0026thinsp;0.780) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe F1 score was used to evaluate the balance between precision and recall for all ML models in predicting disease recurrence. Among the evaluated classifiers, AdaBoost achieved the highest F1 score (F1\u0026thinsp;=\u0026thinsp;0.867), indicating the best overall balance between sensitivity and precision. This was followed by Bernoulli Naive Bayes (F1\u0026thinsp;=\u0026thinsp;0.853) and MLPClassifier (F1\u0026thinsp;=\u0026thinsp;0.850), both of which demonstrated strong classification performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFeature importance analysis across models showed that histological subtypes, including adenocarcinoma, squamous cell carcinoma, and pleomorphic carcinoma, were the most frequently selected features. Tumor grade, tumor size, and ECOG performance status were also commonly retained across models. Less frequent but consistent contributions were observed for large cell carcinoma, histology not otherwise specified, age at diagnosis, and pN stage (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSHAP analysis for the AdaBoost model, one of the best-performing classifiers, indicated that ECOG-PS and tumor size had the greatest contributions to disease recurrence predictions. The number of chemotherapy cycles and vascular invasion status showed moderate importance, whereas age at diagnosis, lymphatic invasion, adjuvant radiotherapy, tumor grade, and pathological substage had comparatively lower contributions to the model predictions (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study demonstrates that recurrence risk after curative resection in NSCLC can be predicted with high accuracy using multiple ML algorithms. In particular, AdaBoost and SVC-RBF models demonstrated robust and consistent predictive performance across multiple feature selection strategies, indicating their stability and reliability in predicting recurrence risk. Lastly, the variables highlighted by the ML models largely overlapped with those identified as independent prognostic factors in Cox regression analysis, supporting the clinical relevance and interpretability of the ML\u0026ndash;based predictions.\u003c/p\u003e \u003cp\u003eDespite applying different feature selection strategies (ANOVA, Chi-square, and Kruskal\u0026ndash;Wallis), the overall model performance remained comparable across most classifiers. Several models, including AdaBoost, Support Vector Classifiers (RBF and polynomial kernels), Quadratic Discriminant Analysis, and Random Forest, demonstrated relatively stable performance when trained on either the full feature set or reduced feature subsets. While certain algorithms, such as Logistic Regression, Linear SVC, and Ridge Classifier, showed modest performance improvements after feature selection, others, particularly MLP-based models and k-nearest neighbors, experienced a decline in accuracy when the number of input variables was reduced. Importantly, no single feature selection method consistently outperformed the others across all models. Collectively, these findings suggest that predictive performance was primarily driven by a core group of robust variables, with the choice of feature selection method exerting only a secondary influence on model accuracy.\u003c/p\u003e \u003cp\u003eTaken together, the performance metrics highlight complementary strengths among the evaluated models in predicting disease recurrence. Confusion matrix analysis showed that AdaBoost was more effective in minimizing false-negative predictions, thereby improving identification of recurrent cases, whereas SVC-RBF was more conservative, yielding fewer false-positive classifications among non-recurrent patients. Although overall accuracy was comparable, the distribution of misclassifications differed meaningfully between models.\u003c/p\u003e \u003cp\u003eDiscriminative performance analysis further demonstrated that SVC-RBF achieved the highest AUC, indicating superior ability to separate recurrent from non-recurrent cases across varying decision thresholds, likely due to its capacity to model complex non-linear relationships. In contrast, AdaBoost achieved the highest F1 score, reflecting a more favorable balance between sensitivity and precision and underscoring its robustness in clinically balanced classification settings. Overall, these findings suggest that while kernel-based methods such as SVC-RBF may be optimal for maximizing discriminative power, ensemble-based approaches like AdaBoost may be better suited for scenarios in which balanced detection of disease recurrence and minimization of missed recurrences are prioritized.\u003c/p\u003e \u003cp\u003eFeature importance analysis across multiple ML models demonstrated that histological subtypes were consistently retained, indicating that tumor histology represents a stable and informative variable across different modeling approaches. The frequent selection of tumor grade, tumor size, and ECOG performance status further supports the relevance of both pathological characteristics and baseline clinical condition in recurrence prediction. In contrast, SHAP analysis of the AdaBoost model, one of the best-performing classifiers, highlighted ECOG performance status and tumor size as the dominant contributors to individual predictions, suggesting that these factors exert the most direct influence on the model\u0026rsquo;s decision-making process. Variables such as chemotherapy cycle number and vascular invasion provided additional, moderate contributions, while other features, including age at diagnosis, lymphatic invasion, adjuvant radiotherapy, tumor grade, and pathological substage, played a more limited role in shaping predictions within this model. Taken together, these results indicate that while histological features are important for model stability and selection across algorithms, clinical status and tumor burden have a more prominent impact on the predictive behavior of high-performing models. This distinction underscores the value of integrating both model-agnostic feature selection and model-specific interpretability approaches to better understand the relative roles of clinical and pathological variables in disease recurrence.\u003c/p\u003e \u003cp\u003eGiven that the primary objective of this study was to predict recurrence, the findings of the Cox regression and SHAP analyses can be interpreted as complementary but functionally distinct. In the multivariate Cox model, ECOG performance status, advanced T stage, tumor grade, vascular invasion, and adjuvant radiotherapy status emerged as independent predictors of recurrence-free survival, indicating variables with a consistent association with recurrence risk at the population level. In contrast, SHAP analysis of the AdaBoost model identified ECOG performance status and tumor size as the main drivers of individual recurrence predictions, while chemotherapy cycle number and vascular invasion contributed more modestly. Other variables identified as prognostic in the Cox analysis contributed less to the machine-learning model\u0026rsquo;s predictive performance. This difference likely reflects the distinct purposes of the two approaches: Cox regression estimates the overall effect of each variable on recurrence risk across the population over time, whereas SHAP explains which variables most strongly influence predictions for individual patients within a non-linear model. Consequently, factors that show consistent prognostic associations at the population level may exert a smaller impact on individual-level predictions. Machine-learning models may therefore provide additional value for individualized recurrence risk estimation.\u003c/p\u003e \u003cp\u003eSimilar machine learning\u0026ndash;based approaches combining feature selection with kernel-based support vector methods have demonstrated strong prognostic performance in other solid tumors. In hepatocellular carcinoma, ML models integrating clinical and pathological variables achieved accuracy rates exceeding 85\u0026ndash;90% for survival prediction across different disease stages [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Likewise, in advanced pancreatic cancer, support vector machine\u0026ndash;based models combined with feature selection methods yielded survival prediction accuracies of approximately 88%, highlighting the robustness and generalizability of such approaches across distinct oncologic settings [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eExtending this prognostic framework from survival outcomes to disease recurrence, several studies have investigated the use of machine learning models specifically for predicting postoperative recurrence risk in patients with resected NSCLC. In a study including 1,387 patients with early-stage resected NSCLC, ML models achieved a best accuracy of 76% and an AUC of 0.81; notably, F1 scores were not reported [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. In another study, a multimodal deep learning\u0026ndash;based ensemble model integrating clinical data with handcrafted and deep learning\u0026ndash;based radiomic features achieved an accuracy of approximately 73% and an F1 score of 0.78, with the ensemble approach yielding the highest ROC\u0026ndash;AUC among the evaluated models [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. In a recent study conducted in a cohort of 309 surgically resected NSCLC patients, ML models and Cox proportional hazards regression demonstrated comparable performance for recurrence prediction, with reported AUC values ranging between approximately 0.74 and 0.77 for 2-year and 5-year postoperative recurrence events, while accuracy- and F1-based metrics were not provided [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eCompared to these studies, our study demonstrated comparable predictive performance for postoperative recurrence. Specifically, the SVC-RBF model achieved an AUC of 0.814, an accuracy of approximately 77%, and an F1 score of 0.842, while AdaBoost yielded the highest accuracy (\u0026asymp;\u0026thinsp;79%) and F1 score (0.867), albeit with a lower AUC of 0.680. The performance of our models was consistent with that reported in the aforementioned studies. Moreover, our study uniquely incorporated confusion matrix\u0026ndash;based evaluation alongside feature importance and SHAP analyses, thereby enhancing model interpretability and reinforcing the clinical relevance of the findings.\u003c/p\u003e \u003cp\u003eThis study has several limitations. First, genomic and molecular data were not included in the analysis, as such information remains largely unavailable and insufficiently characterized in routine clinical practice for this patient population. Second, the lack of temporal modeling represents a limitation of the present study, since longitudinal changes in clinical and pathological variables could not be captured. Finally, the single-center design may introduce institutional bias and limit the generalizability of the findings to broader populations.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study demonstrates that ML\u0026ndash;based models can reliably predict recurrence risk after curative resection in NSCLC using routinely available clinicopathologic variables. The complementary strengths of kernel-based and ensemble approaches highlight their potential utility for individualized recurrence risk assessment. These findings support the integration of ML models as decision-support tools in postoperative risk stratification, while emphasizing the need for future studies incorporating genomic data to further enhance predictive performance and generalizability.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBecause of ethical and privacy considerations, the raw data are not publicly accessible. However, the datasets generated and analyzed during this study can be obtained from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank Furkan Aydos and Hatice Rüveyda Akça for their support in the preparation of this study.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization: Ugur Ozberk, Selin Akturk Esen; Methodology: Ugur Ozberk, Selin Akturk Esen; Formal analysis and investigation: Ugur Ozberk, Serkan Keskin, Hilal Arslan, Melike Cobankaya; Data curation: Ugur Ozberk, Serkan Keskin, Oznur Bal, Efnan Algın; Writing - original draft preparation: Ugur Ozberk; Writing - review and editing: Selin Akturk Esen, Burak Bilgin, Mehmet Ali Nahit Sendur, Dogan Uncu; Resources: Oznur Bal, Efnan Algın, Burak Bilgin; Supervision: Mehmet Ali Nahit Sendur, Dogan Uncu\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of Conflict of Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors report no financial or personal conflicts of interest associated with this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted without any external financial support.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInstitutional Review Board Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Clinical Research Ethics Committee of Ankara City Hospital (Decision No: TABED 1/1772/2025, Date: 22/10/2025) and was conducted in accordance with the principles of the Declaration of Helsinki.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eInformed Consent Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInformed consent was waived by the Clinical Research Ethics Committee of Ankara City Hospital due to the retrospective design of the study and the use of anonymized data obtained from medical records.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eThai, A. A., Solomon, B. J., Sequist, L. V., Gainor, J. F. \u0026amp; Heist, R. S. Lung cancer. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e398\u003c/b\u003e, 535\u0026ndash;554. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0140-6736(21)00312-3\u003c/span\u003e\u003cspan address=\"10.1016/S0140-6736(21)00312-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSu, S. et al. Patterns of survival and recurrence after surgical treatment of early stage non-small cell lung carcinoma in the ACOSOG Z0030 (ALLIANCE) trial. \u003cem\u003eJ. Thorac. Cardiovasc. Surg.\u003c/em\u003e \u003cb\u003e147\u003c/b\u003e, 747\u0026ndash;752. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jtcvs.2013.10.001\u003c/span\u003e\u003cspan address=\"10.1016/j.jtcvs.2013.10.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014). Discussion 752\u0026thinsp;\u0026ndash;\u0026thinsp;743.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUramoto, H. \u0026amp; Tanaka, F. Recurrence after surgery in patients with NSCLC. \u003cem\u003eTransl Lung Cancer Res.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 242\u0026ndash;249. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3978/j.issn.2218-6751.2013.12.05\u003c/span\u003e\u003cspan address=\"10.3978/j.issn.2218-6751.2013.12.05\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFarina, E., Nabhen, J. J., Dacoregio, M. I., Batalini, F. \u0026amp; Moraes, F. Y. An overview of artificial intelligence in oncology. \u003cem\u003eFuture Sci. OA\u003c/em\u003e. \u003cb\u003e8\u003c/b\u003e, FSO787. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.2144/fsoa-2021-0074\u003c/span\u003e\u003cspan address=\"10.2144/fsoa-2021-0074\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGould, M. K., Huang, B. Z., Tammemagi, M. C., Kinar, Y. \u0026amp; Shiff, R. Machine Learning for Early Lung Cancer Identification Using Routine Clinical and Laboratory Data. \u003cem\u003eAm. J. Respir Crit. Care Med.\u003c/em\u003e \u003cb\u003e204\u003c/b\u003e, 445\u0026ndash;453. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1164/rccm.202007-2791OC\u003c/span\u003e\u003cspan address=\"10.1164/rccm.202007-2791OC\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLynch, C. M. et al. Prediction of lung cancer patient survival via supervised machine learning classification techniques. \u003cem\u003eInt. J. Med. Inf.\u003c/em\u003e \u003cb\u003e108\u003c/b\u003e, 1\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijmedinf.2017.09.013\u003c/span\u003e\u003cspan address=\"10.1016/j.ijmedinf.2017.09.013\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDetterbeck, F. C. et al. The Proposed Ninth Edition TNM Classification of Lung Cancer. \u003cem\u003eChest\u003c/em\u003e \u003cb\u003e166\u003c/b\u003e, 882\u0026ndash;895. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.chest.2024.05.026\u003c/span\u003e\u003cspan address=\"10.1016/j.chest.2024.05.026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, H. \u0026amp; Setiono, R. in \u003cem\u003eProceedings of 7th IEEE international conference on tools with artificial intelligence.\u003c/em\u003e 388\u0026ndash;391 (Ieee).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuyon, I. \u0026amp; Elisseeff, A. An introduction to variable and feature selection. \u003cem\u003eJ. Mach. Learn. Res.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 1157\u0026ndash;1182 (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBreiman, L., Friedman, J., Olshen, R. A. \u0026amp; Stone, C. J. \u003cem\u003eClassification and regression trees\u003c/em\u003e (Chapman and Hall/CRC, 2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHosmer, D. W. Jr, Lemeshow, S. \u0026amp; Sturdivant, R. X. \u003cem\u003eApplied logistic regression\u003c/em\u003e (Wiley, 2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCortes, C. \u0026amp; Vapnik, V. Support-vector networks. \u003cem\u003eMach. Learn.\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e, 273\u0026ndash;297 (1995).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, H. The optimality of naive Bayes. \u003cem\u003eAa\u003c/em\u003e \u003cb\u003e1\u003c/b\u003e, 3 (2004).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRipley, B. D. Neural networks and related methods for classification. \u003cem\u003eJ. Roy. Stat. Soc.: Ser. B (Methodol.)\u003c/em\u003e. \u003cb\u003e56\u003c/b\u003e, 409\u0026ndash;437 (1994).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHofmann, T., Sch\u0026ouml;lkopf, B. \u0026amp; Smola, A. J. Kernel methods in machine learning. (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Z. Introduction to machine learning: k-nearest neighbors. \u003cem\u003eAnnals translational Med.\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, 218 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMienye, I. D. \u0026amp; Sun, Y. A survey of ensemble learning: Concepts, algorithms, applications, and prospects. \u003cem\u003eIeee Access.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 99129\u0026ndash;99149 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, T. \u0026amp; XGBoost A Scalable Tree Boosting System. \u003cem\u003eCornell University\u003c/em\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKe, G. et al. Lightgbm: A highly efficient gradient boosting decision tree. \u003cem\u003eAdvances neural Inform. Process. systems\u003c/em\u003e 30 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeven, I. et al. Predicting hepatocellular carcinoma survival with artificial intelligence. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, 6226. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-025-90884-6\u003c/span\u003e\u003cspan address=\"10.1038/s41598-025-90884-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeven, I. et al. Predicting survival outcomes in advanced pancreatic cancer using machine learning methods. \u003cem\u003eMed. (Baltim).\u003c/em\u003e \u003cb\u003e104\u003c/b\u003e, e43904. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/MD.0000000000043904\u003c/span\u003e\u003cspan address=\"10.1097/MD.0000000000043904\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJanik, A. et al. Machine Learning-Assisted Recurrence Prediction for Patients With Early-Stage Non-Small-Cell Lung Cancer. \u003cem\u003eJCO Clin. Cancer Inf.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, e2200062. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1200/CCI.22.00062\u003c/span\u003e\u003cspan address=\"10.1200/CCI.22.00062\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, G., Moon, S. \u0026amp; Choi, J. H. Deep Learning with Multimodal Integration for Predicting Recurrence in Patients with Non-Small Cell Lung Cancer. \u003cem\u003eSens. (Basel)\u003c/em\u003e. \u003cb\u003e22\u003c/b\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/s22176594\u003c/span\u003e\u003cspan address=\"10.3390/s22176594\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePu, L., Dhupar, R. \u0026amp; Meng, X. Predicting Postoperative Lung Cancer Recurrence and Survival Using Cox Proportional Hazards Regression and Machine Learning. \u003cem\u003eCancers (Basel)\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/cancers17010033\u003c/span\u003e\u003cspan address=\"10.3390/cancers17010033\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Non–small cell lung cancer, recurrence, machine learning","lastPublishedDoi":"10.21203/rs.3.rs-8912361/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8912361/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRecurrence after curative resection remains a major clinical challenge in non\u0026ndash;small cell lung cancer (NSCLC), and improved postoperative risk stratification is needed. Machine learning (ML) approaches may enhance recurrence prediction using routinely available clinicopathologic data. We analyzed 265 patients who underwent curative lung cancer surgery. Recurrence was the primary endpoint. Seventeen clinical, pathological, and treatment-related variables were evaluated. Multiple supervised ML classifiers were trained using the full dataset and reduced feature sets generated by ANOVA, chi-square, and Kruskal\u0026ndash;Wallis methods. Model performance was assessed using accuracy, area under the curve (AUC), and F1 score. Prognostic factors were examined with Cox regression, and model interpretability was explored through feature importance and SHAP analysis. Recurrence occurred in 82 patients (30.9%). AdaBoost achieved the highest accuracy (0.79) and F1 score (0.87), whereas SVC-RBF showed the highest AUC (0.81). Performance remained stable across feature-selection strategies. Histologic subtype, tumor size, tumor grade, and ECOG performance status were consistently influential variables, with ECOG status and tumor size dominating SHAP-based predictions. These findings indicate that ML models using routine clinicopathologic variables can reliably predict recurrence after NSCLC surgery and support individualized postoperative risk assessment.\u003c/p\u003e","manuscriptTitle":"Machine Learning–Based Prediction of Recurrence After Curative Resection in Non–Small Cell Lung Cancer","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-22 14:52:31","doi":"10.21203/rs.3.rs-8912361/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-27T08:15:49+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-19T11:16:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"99425727830275897846408206007736567873","date":"2026-03-19T10:44:34+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-18T08:59:10+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"120778307986192398680591792216308379156","date":"2026-03-18T07:26:09+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"335484176178234914440500525739776960477","date":"2026-03-18T04:25:23+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-17T16:26:26+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-17T16:22:06+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-25T04:47:48+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-20T18:13:15+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2026-02-20T18:08:25+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"50f354b0-b575-43e7-b34d-40be071d4c2a","owner":[],"postedDate":"March 22nd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":64814463,"name":"Biological sciences/Cancer"},{"id":64814464,"name":"Health sciences/Oncology"}],"tags":[],"updatedAt":"2026-04-13T16:14:50+00:00","versionOfRecord":{"articleIdentity":"rs-8912361","link":"https://doi.org/10.1038/s41598-026-47862-3","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2026-04-09 15:57:47","publishedOnDateReadable":"April 9th, 2026"},"versionCreatedAt":"2026-03-22 14:52:31","video":"","vorDoi":"10.1038/s41598-026-47862-3","vorDoiUrl":"https://doi.org/10.1038/s41598-026-47862-3","workflowStages":[]},"version":"v1","identity":"rs-8912361","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8912361","identity":"rs-8912361","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}