Machine learning-based prognostic models and factors influencing the benefit of surgery on primary lesion for patients with lung cancer brain metastases.

Abstract

Brain metastasis is very common in lung cancer and it's a fatal disease with extremely poor prognosis. Until now, there has been a lack of accurate and efficient prognostic models for patients with lung cancer brain metastases (LCBM), and the factors influencing the effectiveness of the surgery on primary lesion for these patients remain unclear. We used 7 machine learning algorithms to create prognostic models to predict the overall survival (OS) of LCBM based on the data from the Surveillance Epidemiology and End Results. Then, a series of validation methods, including area under the curve values, receiver operating characteristic curve analysis, calibration curves, decision curve analysis and external data validation were used to confirm the high discrimination, accuracy, and clinical applicability of the XGBoost models. Propensity score matching adjusted analysis was conducted for further stratified analysis to find factors influencing the benefit of surgery on primary lesion for LCBM. Models using XGBoost algorithm performed best. Surgery on primary lesion was a favorable independent prognostic factor for LCBM. Age > 70 years old, blacks, grade IV, stage T4, N3, other distant organ metastases, squamous cell carcinoma, large cell carcinoma and no radiation were all unfavorable factors of primary lung tumor surgery for the prognosis of LCBM. Our study is the first one to create highly accurate AI models to predict the OS of LCBM. Our in-depth stratified analysis found some influence factors of surgery on primary lesion for the prognosis of LCBM.