Interpretable Machine Learning for Survival Prediction in Small Cell Lung Cancer Patients With Brain Metastases: A Population-Based Study With External Validation

利用可解释机器学习预测小细胞肺癌脑转移患者的生存期:一项基于人群的外部验证研究

阅读:3

Abstract

IntroductionBrain metastases (BM) represent a common and fatal progression in small cell lung cancer (SCLC), yet prognostic tools for this population remain underdeveloped. This study aimed to establish and externally validate a machine learning-based model to predict overall survival (OS) in SCLC patients with BM.MethodsWe extracted clinical data from 2392 SCLC patients with BM from the SEER database to construct prognostic models using Cox regression, AJCC staging, and four machine learning algorithms: Random Survival Forest (RSF), Extreme Gradient Boosting (XGB), Elastic Net (Enet), and Artificial Neural Network (ANN). Key features were selected via Lasso-Cox regression. Model performance was evaluated using time-dependent AUC, calibration curves, Brier scores, precision-recall (PR) curves, and decision curve analysis (DCA). SHAP and partial dependence plots were applied for model interpretability. External validation was conducted using an independent hospital-based cohort of 85 patients, with comparability to the SEER cohort addressed through inverse probability of treatment weighting (IPTW).ResultsAmong all models, the RSF algorithm demonstrated the best overall performance. In the training cohort, it achieved AUCs of 0.738 and 0.809 for 1-year and 2-year OS, respectively. In the internal validation cohort, AUCs were 0.718 and 0.748, and in the external validation cohort, 0.686 and 0.802, respectively. The RSF model also showed favorable calibration and the lowest Brier scores across datasets. SHAP analysis ranked chemotherapy, liver metastasis, N stage, and age as the most influential prognostic features. A web-based calculator was developed to enable real-time individualized risk prediction.ConclusionsThis study presents a robust, interpretable, and externally validated RSF-based model for predicting OS in SCLC patients with BM. The model offers clinically relevant insights and is accessible via an online tool, supporting its potential integration into personalized treatment planning.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。