Personalized and real time hemodynamic management in critical care using Dynamic Cohort Ensemble Learning (DynaCEL)

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Clinical Data Science

Background:

• Effective hemodynamic management in intensive care units (ICUs) necessitates adaptive, individualized targets.
• Current management often relies on static or generalized hemodynamic targets.
• Dynamic clinical conditions require real-time adjustments to patient care.

Purpose of the Study:

• To develop and validate Dynamic Cohort Ensemble Learning (DynaCEL), a real-time framework for personalized hemodynamic targets.
• To assess the association between DynaCEL-guided hemodynamic targets and patient mortality.
• To explore the temporal and inter-individual variability in optimal hemodynamic targets.

Main Methods:

• Developed DynaCEL, a real-time framework modeling ICU admission time points as distinct temporal cohorts.
• Trained DynaCEL on the eICU database and validated on MIMIC-IV and Indiana University Health datasets.
• Utilized propensity score matching for comparative analysis of DynaCEL-predicted targets versus fixed targets regarding 24-hour mortality.

Main Results:

• DynaCEL demonstrated robust predictive performance with AUCs ranging from 0.83 to 0.91 across datasets.
• Proximity to DynaCEL-predicted targets was associated with significantly lower 24-hour mortality in the MIMIC-IV cohort.
• Greater deviations from personalized targets correlated with increased mortality, indicating a dose-response relationship.

Conclusions:

• DynaCEL provides an interpretable and scalable approach to precision hemodynamic management in ICUs.
• Personalized hemodynamic targets show potential for improving patient outcomes.
• Prospective clinical trials are necessary to establish the definitive clinical utility of DynaCEL.