Evolving prognostic paradigms in lung adenocarcinoma with brain metastases: a web-based predictive model enhanced by machine learning.

Abstract

Introduction: Patients with lung adenocarcinoma (LUAD) who develop brain metastases (BM) face significantly poor prognoses. A well-crafted prognostic model could greatly assist clinicians in patient counseling and in devising tailored therapeutic strategies.

Methods: The study cohort comprised LUAD patients with BM identified from the surveillance, epidemiology, and end results database between 2000 and 2018. We pinpointed independent prognostic features for overall survival (OS) using Lasso regression analyses. Predictive models were built using Random Forest, XGBoost, Decision Trees, and Artificial Neural Networks, with their performance evaluated via metrics including the area under the receiver operating characteristic curve (AUC), calibration plots, brier score, and decision curve analysis (DCA).

Results: We extracted a total of 9121 eligible patients from the database, identifying eleven clinical parameters that significantly influenced OS prognostication. The XGBoost model exhibited superior discriminative power, achieving AUC values of 0.829 and 0.827 for 1- and 2-year survival, respectively, in the training cohort, and 0.816 and 0.809 in the validation cohort. In comparison to other models, the XGBoost model excelled in both training and validation phases, as demonstrated by substantial differences in AUC, DCA, calibration, and Brier score. This model has been made accessible via a web-based platform.

Conclusions: This study has developed an XGBoost-based machine learning model with an accompanying web-based application, providing a novel resource for clinicians to support personalized decision-making and enhance treatment outcomes for LUAD patients with BM.