A web-based prediction model for brain metastasis in non-small cell lung cancer patients.

BACKGROUND

Brain metastasis (BM) stands as a significant contributor to mortality among cancer patients. The factors contributing to BM remain incompletely elucidated. Accurate prediction of BM occurrence is essential for effective disease prevention and control during patient management.

METHODS

Clinical characteristics from a cohort of 39,930 non-small cell lung cancer (NSCLC) patients were meticulously collected. 12 candidate variables were screened out through Least absolute shrinkage and selection operator (LASSO) regression analysis. This dataset was divided into training, testing, and validation sets in a ratio of 6:2:2. Additionally, 276 NSCLC patient records from Shanghai Pulmonary Hospital (SPH) were used as an external cohort. Subsequently, seven machine learning models were constructed employing diverse algorithms, namely Logistic Regression (LR), Classification and Regression Tree (CART), Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Gradient Boosting Machine (GBM), and eXtreme Gradient Boosting (XGBOOST). SHapley Additive exPlanations (SHAP) analysis applied to visualize the importance of each feature. The evaluation of these models was based on the area under the receiver-operating characteristic curve (ROC), facilitating a comprehensive comparison of their predictive performances. The GBM model, demonstrating superior accuracy, was selected for brain metastasis (BM) prediction. A dedicated website was developed to visualize and communicate the predictive model's outcomes.

RESULTS

Through a comprehensive evaluation incorporating metrics such as the area under the curve (AUC), accuracy, sensitivity, specificity, Kappa value, and calibration curve, the model generated by the Gradient Boosting Machine (GBM) algorithm demonstrated exceptional and near-perfect performance in both the validation set (AUC: 0.8276) and the test set (AUC: 0.8301), however, performance was lower in the SPH cohort (AUC: 0.6100). To enhance interpretability, the relative importance of the 12 candidate variables was ranked, providing clinicians with a clear understanding of how each factor contributes to the prediction of BM risk.

CONCLUSION

Successfully developed was a machine learning model, characterized by high accuracy and precision, along with a web-based predictor tailored for NSCLC patients. This advancement enables the accurate prediction of BM risk, facilitating the implementation of targeted preventive measures for individuals deemed at high risk.