Optimize individualized energy delivery for septic patients using predictive deep learning models

Area of Science:

• Critical Care Medicine
• Nutritional Support
• Artificial Intelligence in Healthcare

Background:

• Sepsis management requires precise nutritional support.
• Individualized energy delivery is critical for septic patients in intensive care units (ICUs).
• Current energy delivery strategies may not account for dynamic metabolic changes during sepsis.

Purpose of the Study:

• To develop and validate deep learning models for optimizing individualized energy delivery in adult septic patients.
• To identify optimal energy targets across different metabolic phases of sepsis.

Main Methods:

• A retrospective study of 179 adult septic patients in the ICU, collecting 47 indicators over 14 days.
• Data divided into three metabolic phases: acute early, acute late, and rehabilitation.
• Deep learning models established for optimal energy targets in each phase, with external validation.

Main Results:

• Optimal energy targets identified as 900 kcal/d (early acute), 2300 kcal/d (late acute), and 2000 kcal/d (rehabilitation).
• Excessive energy in the early acute phase and insufficient energy in the late acute phase increased mortality.
• Both excessive and insufficient energy delivery in the rehabilitation phase were associated with higher mortality risk.

Conclusions:

• Time-series deep learning models can optimize energy delivery for septic patients in the ICU.
• Permissive underfeeding is advised only in the early acute phase.
• Increased energy intake in later phases may enhance survival and address energy deficits.