Abstract
BACKGROUND: Brain metastasis, the most prevalent site of lung cancer metastasis, implies a grim prognosis. Adopting the best treatment approach is crucial for improving the survival of these patients. Therefore, this study aimed to develop a personalized prognostic model for brain-metastasized non-small cell lung cancer (BM-NSCLC) patients to aid in clinical decision-making. METHODS: The study enrolled BM-NSCLC patients who were single-primary and had not undergone radical surgery from 2010 to 2021. The Kaplan-Meier method analysis was utilized to assess overall survival (OS) and cancer-specific survival (CSS) under different treatments. Univariable and multivariable Cox regression analyses were conducted to ascertain independent prognostic factors. The dataset was partitioned into training (70%) and validation (30%) cohorts for the development and assessment of random forest (RF), logistic regression (LR), support vector machine (SVM), and K-nearest neighbor (KNN) models. The efficacy of the models was evaluated through the calculation of area under the curve (AUC) of the receiver operating characteristic (ROC) curve and decision curve analysis (DCA). A user-friendly web app was developed via shinyapps.io to increase the accessibility for clinicians. RESULTS: A total of 3,171 eligible samples were ultimately included in the study. Survival analysis indicated that patients who underwent metastasis site surgery combined with radiotherapy based on chemotherapy exhibited a more favorable prognosis compared to alternative treatment modalities within the scope of this study. The RF model demonstrated superior predictive accuracy for 1-year-OS, with an AUC of 0.89 in validation cohorts (n=951), and a more refined DCA profile. CONCLUSIONS: In the case of patients with BM-NSCLC, the integration of radiation therapy with surgery for metastasis site based on systematic treatment yielded the most significant benefits. The importance of a comprehensive treatment strategy that integrates chemotherapy, surgery, and radiotherapy for these patients was emphasized. Additionally, a clinical decision-support tool constructed from this dataset, demonstrated robust discrimination, excellent calibration, and notable clinical utility. This tool will effectively assist clinical practitioners in making more personalized clinical decisions for patients.