Research Square · Aug 20, 2026
Machine Learning-Based Prediction of Lung Cancer Risk in COPD Patients Using Clinical Severity Indicators and Comorbidity Profile
By Alshraideh A, Alshraideh M, Alkayed Y et al.
Lung cancer remains a major cause of cancer-related mortality worldwide. It is clinically relevant in patients with chronic obstructive pulmonary disease (COPD), in whom shared risk factors and biological mechanisms may increase susceptibility. Early identification of COPD patients with elevated lung-cancer risk status could support screening prioritization and timely diagnostic evaluation. This study developed and internally evaluated a machine-learning framework for lung-cancer risk stratification among COPD patients, using demographic, physiological, disease-severity, and comorbidity variables. The analytic dataset comprised 2300 COPD patient records, with 32% labeled as lung-cancer cases and 68% as non-lung-cancer cases. Data preprocessing was implemented in leakage-controlled, training-only workflows that included imputation, categorical encoding, standardization, feature selection, class imbalance handling with SMOTE applied only to the training data, and stratified validation. Because temporal follow-up information was not available in the manuscript, the modeling task was framed as lung cancer risk-status classification rather than as definitive incident cancer prediction. Six supervised models were compared: Logistic Regression, Support Vector Machine, Decision Tree, K-Nearest Neighbors, Random Forest, and Artificial Neural Network (ANN). In the reported internal validation, the ANN achieved the highest performance, with an accuracy of 0.90, a recall of 0.92, and an AUC-ROC of 0.94; Random Forest showed comparable discrimination with an AUC-ROC of 0.93. McNemar testing suggested that ANN and Random Forest performed similarly, whereas ANN outperformed the linear baseline. Calibration analysis using the Brier Score showed the lowest value for ANN (0.097), followed by Random Forest (0.109). Smoking exposure, COPD severity, FEV1, and age were the most influential predictors in Random Forest feature-importance analysis. These findings suggest that nonlinear machine-learning models may support stratification of COPD-related lung cancer risk status. However, the results are internally validated and hypothesis-generating, and they require external validation, transparent outcome ascertainment, uncertainty estimation, and additional explainability analyses before clinical implementation.