Integrating Machine Learning for Early COPD Prediction in Lung Cancer Patients: A Focus on Systemic Coagulation-Inflammation Index.

BACKGROUND: Chronic obstructive pulmonary disease (COPD) often coexists with lung cancer, worsening clinical outcomes. This study aimed to develop machine learning models for early COPD screening in lung cancer patients using clinical variables and a novel systemic coagulation-inflammation index (SCI).

METHODS: We retrospectively enrolled 1016 patients, extracting demographic, smoking, vital, and laboratory data. After feature selection with Boruta and least absolute shrinkage and selection operator (LASSO), six models-logistic regression, decision tree (DT), multilayer perceptron (MLP), support vector machine (SVM), gradient boosting decision tree (GBDT), and extreme gradient boosting (XGBoost)-were trained on 70% of the data and tested on 30%. Model performance was assessed with area under the curve (AUC), accuracy, sensitivity, specificity, precision, F1 score, and calibration, while SHapley Additive exPlanations (SHAP) and contour plots helped interpret the best model and explore predictor interactions.

RESULTS: Among the 1,016 patients, 182 (17.9%) had concomitant COPD. Boruta and LASSO identified 8 key predictors: age, historical smoking index (HSI), SCI, eosinophils (EOS), bicarbonate (HCO3 -), sex, lymphocytes (LYM), and hemoglobin (HGB). The GBDT model showed the best performance, with an area under the curve (AUC) of 0.74 (95% CI: 0.69-0.80) and moderate sensitivity (0.68) and specificity (0.66). SHAP analysis revealed age, HSI, and SCI as major risk factors, with contour plots indicating a synergistic effect between SCI, age, and HSI on COPD risk.

CONCLUSION: A GBDT-based model built from routine clinical variables and SCI showed moderate discrimination for identifying patients with lung cancer at high risk of concomitant COPD.