Rank-Based Transfer Learning for High-Dimensional Survival Data With Application to Sepsis Data

Area of Science:

• Biostatistics
• Machine Learning
• Infectious Disease Epidemiology

Background:

• Sepsis, particularly Methicillin-susceptible Staphylococcus aureus (MSSA) sepsis, presents significant challenges due to high mortality and complex prognosis.
• Studying MSSA sepsis is hindered by limitations in available high-dimensional survival data.
• Existing transfer learning frameworks require adaptation for high-dimensional survival data analysis.

Purpose of the Study:

• To extend transfer learning frameworks for high-dimensional survival data in the context of MSSA sepsis.
• To develop a method for intelligently identifying beneficial source datasets using a C-index based measurement.
• To improve target model performance by transferring knowledge from identified source datasets and applying a debiasing step.

Main Methods:

• Development of a novel measurement index based on the C-index for source dataset selection.
• Implementation of transfer and debiasing steps to leverage information from identified source datasets.
• Rigorous establishment of statistical properties, including $\mathrm{\ell}_1/\mathrm{\ell}_2$-estimation error bounds and detection consistency for transferable source detection.

Main Results:

• The proposed transfer learning algorithm demonstrates improved estimation and prediction accuracy.
• The source detection algorithm exhibits a detection consistency property.
• Analysis of MIMIC-IV sepsis data confirms the practical advantages and significant improvements in survival estimates for MSSA sepsis patients.

Conclusions:

• The developed transfer learning approach effectively addresses data limitations in MSSA sepsis research.
• The method provides enhanced survival estimates, offering practical benefits for clinical applications.
• This work contributes a statistically robust and computationally efficient solution for high-dimensional survival data analysis in critical care settings.