Deep-CSA: Deep Contrastive Learning for Dynamic Survival Analysis With Competing Risks.

Area of Science:

• Biostatistics
• Machine Learning
• Clinical Informatics

Background:

• Survival analysis (SA) is crucial for time-to-event data.
• Conventional SA models struggle with dynamic covariate relationships and competing risks.
• Modeling multiple, simultaneous health events in disease progression remains a challenge.

Purpose of the Study:

• To propose a novel deep contrastive learning model for disease progression analysis.
• To effectively model dynamic correlations between covariates and competing risks.
• To enhance understanding of complex disease trajectories using longitudinal data.

Main Methods:

• Developed a self-supervised deep contrastive learning framework.
• Learned dynamic subject representations from longitudinal observational data.
• Captured time-varying relationships between covariates and specific competing risks.

Main Results:

• The proposed model demonstrated effectiveness on MIMIC-III and EICU datasets.
• Achieved significant improvements compared to existing state-of-the-art SA models.
• Successfully captured dynamic covariate-risk relationships in disease progression.

Conclusions:

• The novel deep contrastive learning model offers a powerful approach for competing risks SA.
• This method provides deeper insights into disease progression from longitudinal data.
• The model shows promise for improving clinical outcome prediction and understanding.