An eXtreme Gradient Boosting machine learning model developed using routine pre-treatment laboratory parameters achieved reliable discrimination between early-stage and advanced-stage non-small cell lung cancer, reaching an area under the receiver operating characteristic curve of 0.875 in a validation cohort.
Non-small cell lung cancer is one of the most prevalent malignancies worldwide, and the clinical stage identified at diagnosis guides treatment strategies and strongly influences clinical outcomes. To investigate whether routine clinical data could assist in this classification, researchers evaluated whether computational models could distinguish early from advanced disease using routinely collected tests.
The retrospective study included 1,240 patients with pathologically confirmed non-small cell lung cancer. Before treatment commenced, sixty-seven laboratory parameters and three clinical variables were collected for each participant.
The patient cohort was randomly divided into a training set comprising 70 percent of the sample and a validation set comprising 30 percent. Investigators constructed two predictive architectures—an eXtreme Gradient Boosting model and a logistic regression model—to differentiate early-stage disease, defined as TNM stages I to II, from advanced-stage disease, defined as TNM stages III to IV.
Model performance was evaluated using receiver operating characteristic curves, bootstrap resampling, and decision curve analysis. In the validation cohort, the eXtreme Gradient Boosting model achieved a significantly higher area under the curve than the logistic regression model, registering 0.875 compared to 0.834 via the DeLong test with a p-value of 0.0061, while also demonstrating higher specificity and accuracy.
Interpretability evaluated through SHapley Additive exPlanations identified carcinoembryonic antigen, sialic acid, and fibrinogen as the most influential predictors in the model. Additionally, decision curve analysis indicated that the gradient boosting framework delivered a net clinical benefit across a range of clinically relevant threshold probabilities.
Subgroup analyses showed strong discriminative performance of the eXtreme Gradient Boosting model in patients with adenocarcinoma, whereas both the gradient boosting and logistic regression models showed limited discrimination in squamous cell carcinoma. The authors concluded that the model based primarily on routine laboratory parameters provides good discrimination for staging and may serve as a noninvasive adjunct for preliminary risk stratification.
Subject of Research: Machine learning classification for non-small cell lung cancer staging using routine laboratory parameters
Article Title: Machine learning-assisted staging classification of non-small cell lung cancer using routine laboratory parameters
Article References: Mu, Y., Zhu, Y., Zhao, C., li, K., li, Y., sun, C., xiao, C., & Li, G. (2026). Machine learning-assisted staging classification of non-small cell lung cancer using routine laboratory parameters. Frontiers in Molecular Biosciences, 13, 1999345. https://doi.org/10.3389/fmolb.2026.1999345
Image Credits: AI Generated
DOI: 10.3389/fmolb.2026.1999345
Keywords: non-small cell lung cancer, XGBoost, routine laboratory parameters, TNM staging, machine learning, carcinoembryonic antigen, sialic acid, fibrinogen, logistic regression, adenocarcinoma, squamous cell carcinoma, SHAP analysis, decision curve analysis, receiver operating characteristic, area under the curve, risk stratification
News Source: Nathaniel Bowman. (November 16, 2026). XGBoost Classifies Non-Small Cell Lung Cancer Staging Using Routine Laboratory Parameters. Scienmag.



