SURVFIT: Doubly sparse rule learning for survival data.

Area of Science:

• Biostatistics
• Machine Learning
• Medical Informatics

Background:

• Survival data analysis is vital for studying disease morbidity and mortality.
• Interpretable models are needed for high-dimensional medical data, but existing rule ensembles are computationally expensive and lack variable sparsity.
• Current methods struggle to balance predictive accuracy with computational efficiency and interpretability in survival analysis.

Purpose of the Study:

• To present SURVFIT, a "doubly sparse" rule extraction formulation for survival data.
• To address the computational expense and lack of variable sparsity in existing rule ensemble methods for survival data.
• To develop an efficient and interpretable model for identifying significant biomarkers from high-dimensional survival data.

Main Methods:

• SURVFIT employs a "doubly sparse" approach, inducing sparsity in both the number of rules and variables within rules.
• Utilizes a quadratic loss function with overlapping group regularization for computational efficiency.
• Incorporates a systematic rule evaluation framework including statistical, decomposition, and sensitivity analyses.

Main Results:

• SURVFIT demonstrates computational efficiency for rule extraction from survival data, considering both rule and variable sparsity.
• The method effectively captures sparse underlying signals while maintaining high predictive accuracy.
• Experiments on synthetic and MIMIC-III sepsis survival data validate the utility of SURVFIT.

Conclusions:

• SURVFIT offers an efficient and interpretable solution for survival data analysis, enhancing biomarker discovery.
• The doubly sparse formulation overcomes limitations of existing methods, enabling practical application to high-dimensional medical data.
• This approach facilitates the identification of significant bio-markers affecting disease morbidity and mortality.