CoxFNN: Interpretable machine learning method for survival analysis.

Area of Science:

• Biostatistics
• Machine Learning in Healthcare
• Survival Data Analysis

Background:

• Survival analysis is crucial for time-to-event data in healthcare, including disease progression and treatment efficacy.
• Traditional survival models present a trade-off between interpretability (linear assumptions) and capturing complex relationships (non-linear, less interpretable).
• Existing methods often struggle to balance analytical sophistication with clear, human-understandable insights.

Purpose of the Study:

• To develop a novel machine learning approach for survival analysis that overcomes limitations of traditional methods.
• To create a model capable of handling non-linear relationships without strict distribution assumptions.
• To enhance interpretability by enabling the learning of human-comprehensible rules from data.

Main Methods:

• An extension of the Cox proportional hazards model was developed using machine learning.
• The model was designed to capture non-linear feature-risk associations.
• Incorporated rule-learning capabilities for enhanced data interpretability.

Main Results:

• The proposed model achieved performance comparable to existing survival analysis techniques.
• Successfully identified significant high-risk factors within the analyzed datasets.
• Demonstrated practical utility by uncovering clinically relevant and understandable rules.

Conclusions:

• The novel machine learning approach offers a promising balance between advanced analysis and interpretability in survival analysis.
• This method has significant potential for real-world healthcare applications, aiding in disease progression and treatment efficacy studies.
• The model's ability to learn interpretable rules enhances its value for clinical decision-making.