Survival Prediction After Neurosurgical Resection of Brain Metastases: A Machine Learning Approach

Area of Science:

• Neurosurgery
• Oncology
• Machine Learning
• Biostatistics

Background:

• Current prognostic models for brain metastases (BM) are primarily based on radiotherapy patients, limiting their applicability to surgical cases.
• There is a need for reliable prognostic tools specifically for patients undergoing surgical resection of brain metastases.

Purpose of the Study:

• To develop and validate a machine learning model for predicting 6-month survival in patients after brain metastases resection.
• To compare the performance of various machine learning algorithms against established prognostic models.

Main Methods:

• Utilized a dataset of 1062 patients who underwent resection for brain metastases, split into training and testing sets.
• Trained and evaluated seven machine learning algorithms, including logistic regression, and compared them with the diagnosis-specific graded prognostic assessment.
• Assessed model performance using area under the curve (AUC) and calibration metrics.

Main Results:

• Logistic regression demonstrated the best performance with an AUC of 0.71 and strong calibration.
• The developed model effectively stratified patients into distinct risk groups (regular, high, very high) for 6-month mortality.
• The model's predictions significantly correlated with both 6-month and longitudinal overall survival (P < .0005).

Conclusions:

• A validated prediction model for 6-month survival post-neurosurgical resection of brain metastases has been developed.
• The model provides accurate risk stratification, aiding clinical decision-making for neurosurgical oncology patients.
• External validation is recommended for future research to confirm the model's generalizability.