To develop an embedding model that integrates tabular and textual data for comprehensive patient representation in sepsis research, addressing the limitations of existing models.
Key Findings:
SepsisDRM achieved AUC scores of 0.92, 0.94, and 0.78 for retrospective, prospective, and external datasets, respectively, indicating strong predictive performance.
The model effectively stratified patients into four clinically interpretable phenotypes.
SepsisDRM demonstrates strong generalization across diverse sepsis-related tasks without the need for task-specific tuning.
Interpretation:
SepsisDRM represents a significant advancement in sepsis research by integrating diverse data types, potentially enhancing patient stratification and outcome prediction, which could lead to improved clinical decision-making.
Limitations:
The datasets used for training and testing SepsisDRM are not publicly available due to their potentially identifiable nature, which may limit reproducibility.
The model's performance may vary with different populations or settings not represented in the training data, introducing potential biases.
Conclusion:
SepsisDRM establishes a novel paradigm for sepsis research and offers a promising approach for integrating tabular and textual data in other medical fields, paving the way for future research advancements.
