Answers are grounded in this paper and cite their sources.
Background Non-small cell lung cancer (NSCLC) is one of the most prevalent malignancies worldwide. The stage at diagnosis guides treatment strategies and strongly influences clinical outcomes. This study aimed to develop a machine learning-based model using routine laboratory parameters to assist in the staging classification of NSCLC. Methods A total of 1,240 patients with pathologically confirmed NSCLC were retrospectively included. Sixty-seven laboratory parameters and three clinical variables were collected prior to treatment. Patients were randomly assigned to a training set (70%) and a validation set (30%). eXtreme Gradient Boosting (XGBoost) and logistic regression (LR) models were developed to distinguish early-stage disease (TNM stages I-II) from advanced-stage disease (TNM stages III-IV). Model performance was evaluated using receiver operating characteristic (ROC) curves, bootstrap resampling, and decision curve analysis (DCA). Interpretability was assessed using SHapley Additive exPlanations (SHAP). Subgroup analyses were conducted for patients with adenocarcinoma (ADC) and squamous cell carcinoma (SCC). Results In the validation set, the XGBoost model achieved a higher area under the curve (AUC) than the LR model (0.875 vs. 0.834, DeLong test, p = 0.0061), with higher specificity and accuracy. SHAP analysis identified carcinoembryonic antigen (CEA), sialic acid, and fibrinogen as the most influential predictors. Subgroup analysis showed strong discriminative performance of the XGBoost model in patients with ADC, whereas both models showed limited discrimination in SCC. DCA indicated that the XGBoost model provided a net benefit across a range of clinically relevant threshold probabilities. Conclusion The XGBoost model based primarily on routine laboratory parameters showed good discrimination for NSCLC staging and may serve as a noninvasive adjunct for preliminary risk stratification.
No takes yet. Share an insight, caveat, or question.
Mu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: