Abstract 416P
Background
Non-small cell lung cancer (NSCLC) is a leading cause of cancer mortality, with brain metastases affecting 30–60% of patients and significantly worsening prognosis. Despite advances in systemic therapies, these patients are often excluded from clinical trials, limiting insights into effective treatments. This study aimed to develop a machine-learning model to predict survival in NSCLC patients with brain metastases and evaluate the impact of local treatments.
Methods
A retrospective cohort of 97 NSCLC patients with brain metastases treated at a single center was augmented with synthetic data (n=470) using the Synthetic Data Vault library. Predictive models, including Decision Trees, Random Forest (50 and 100 trees), and Deep Neural Networks (Deepnet), were trained using 38 clinical and molecular variables. Performance metrics included accuracy, sensitivity, F1-score, and AUC-ROC.
Results
The Random Forest model with 100 trees showed the best predictive performance (AUC-ROC: 0.7601, F1-score: 0.8188), out-performing Decision Trees and Deepnet. Among predictors, clinical variables such as ECOG performance status and molecular alterations, particularly EGFR mutations, contributed significantly to the models’ accuracy. Local treatments for brain metastases were associated with improved survival outcomes.
Conclusions
Machine learning models, particularly Random Forest, can effectively predict survival in NSCLC patients with brain metastases and support clinical decision-making. While results are promising, limitations such as small sample size and reliance on synthetic data highlight the need for further validation in larger, real-world cohorts.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.