Model development and validation for predicting small-cell lung cancer bone metastasis utilizing diverse machine...

Area of Science:

• Oncology
• Machine Learning
• Biostatistics

Background:

• Small cell lung cancer (SCLC) poses significant challenges in predicting bone metastasis (BM).
• Accurate prediction of BM is crucial for timely intervention and improved patient outcomes in SCLC.
• Existing predictive models may lack the precision required for effective clinical application.

Purpose of the Study:

• To develop a high-performance machine learning algorithm for predicting bone metastasis (BM) in small cell lung cancer (SCLC).
• To create a user-friendly, web-based predictor tool for clinical decision support.
• To identify key demographic and clinicopathological factors associated with BM risk in SCLC patients.

Main Methods:

• Utilized a large dataset (89,366 patients) from the Surveillance, Epidemiology, and End Results database (2010-2018).
• Developed and compared 12 machine learning models, including XGBoost, SVM, and Random Forest.
• Evaluated model performance using metrics such as AUC, accuracy, precision, recall, F1-score, and Brier score.

Main Results:

• The XGBoost algorithm demonstrated superior predictive performance with AUC scores of 0.965 (training), 0.962 (validation), and 0.961 (testing).
• Identified age, T stage, N stage, liver/lung/brain metastasis, marital status, income, M stage, and AJCC stage as independent risk factors for BM in SCLC.
• Successfully developed a web-based predictor tool based on the validated XGBoost model.

Conclusions:

• The XGBoost-based machine learning model offers a highly accurate method for predicting bone metastasis in SCLC.
• The developed web predictor can serve as a valuable tool for clinicians in managing SCLC patients.
• Understanding identified risk factors can guide personalized risk assessment and treatment strategies for SCLC patients at risk of BM.