Predicting breast cancer 5-year survival using machine learning: A systematic review

Area of Science:

• Oncology
• Biostatistics
• Computer Science

Background:

• Accurate prediction of 5-year survival rates for breast cancer patients is a critical challenge in oncology research.
• Machine learning (ML) offers potential for improved prediction accuracy but its effectiveness and methodologies remain debated.
• This systematic review critically appraises current research on ML applications for breast cancer survival prediction.

Purpose of the Study:

• To systematically identify and evaluate studies applying machine learning (ML) to predict the 5-year survival rate of breast cancer patients.
• To critically appraise the methodologies and performance of ML models used in breast cancer survival prediction.
• To identify limitations and areas for future research in this domain.

Main Methods:

• A systematic literature search was conducted across PubMed, Embase, and Web of Science databases up to November 30, 2020, adhering to PRISMA guidelines.
• Studies were included if they utilized ML for breast cancer survival prediction models with measurable performance metrics.
• Data extraction focused on literature, database, data preparation, modeling, performance evaluation, and predictor information.

Main Results:

• Thirty-one studies were included, with most published after 2013. Common ML methods included decision trees, artificial neural networks, support vector machines, and ensemble learning.
• Reported performance metrics varied widely: accuracy (0.510-0.971), sensitivity (0.037-1), specificity (0.008-0.993), and AUC (0.500-0.972).
• All models underwent internal validation, with only one external validation, highlighting a significant gap in robust model assessment.

Conclusions:

• Machine learning models do not consistently demonstrate superior performance compared to traditional statistical methods for breast cancer survival prediction.
• Key limitations include insufficient data preprocessing, diverse feature selection, and inadequate validation strategies.
• Future research necessitates greater standardization in methodology and rigorous external validation to enhance the reliability and generalizability of ML models.