Dynamic Sepsis Prediction for Intensive Care Unit Patients Using XGBoost-Based Model With Novel Time-Dependent...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Critical Care Medicine

Background:

• Sepsis, a life-threatening systemic inflammatory response, presents diagnostic challenges due to variable clinical manifestations and low specificity of traditional criteria.
• High incidence and mortality rates of sepsis pose significant threats to patients and healthcare systems, particularly in Intensive Care Units (ICUs).

Purpose of the Study:

• To develop and validate a novel Artificial Intelligence (AI) model for dynamic sepsis prediction and risk assessment.
• To improve the accuracy and efficiency of sepsis diagnosis compared to existing methods.

Main Methods:

• Utilized the XGBoost machine learning framework with demographic, vital signs, laboratory, and medical intervention data.
• Implemented time-dependent feature construction for time-series data and count-dependent feature construction for clinical intervention data.
• Proposed a novel objective function with first-order and second-order gradients for optimized model training.

Main Results:

• The proposed AI model demonstrated superior performance in sepsis prediction compared to state-of-the-art methods.
• Achieved significant improvements in Area Under the Receiver Operating Characteristic Curve (AUROC): 5.4% on the MIMIC-III dataset and 2.1% on the PhysioNet Challenge 2019 dataset.
• Validated the model's effectiveness using established clinical datasets.

Conclusions:

• The AI-driven approach offers a promising advancement for accurate and dynamic sepsis prediction and risk assessment.
• The model's data processing and training methodologies are adaptable to various electronic health record systems, indicating broad clinical applicability.
• This AI model has the potential to enhance early sepsis detection, leading to improved patient management and outcomes in critical care settings.