Predicting In-Hospital Mortality in Intensive Care Unit Patients Using Causal SurvivalNet With Serum Chloride and...

Area of Science:

• Critical Care Medicine
• Clinical Chemistry
• Biostatistics
• Data Science in Healthcare

Background:

• Serum chloride levels are potential prognostic indicators in intensive care units (ICUs).
• Utilizing chloride levels can refine risk stratification and treatment, improving patient outcomes and resource allocation.

Purpose of the Study:

• To quantitatively analyze the association between serum chloride levels upon intensive care unit (ICU) admission and in-hospital mortality.
• To develop a personalized survival curve prediction deep learning model for enhanced risk stratification and clinical decision-making.

Main Methods:

• A large-scale, multicohort study involving 189,462 ICU patients from US and Chinese databases (MIMIC-IV, eICU-CRD, Yantai Yuhuangding Hospital, Zigong Fourth People's Hospital).
• Causal graph analysis, restricted cubic splines, Cox proportional hazards models, Cox frailty model, Kaplan-Meier curves, and sensitivity analyses were used to evaluate associations.
• A deep learning model (Causal SurvivalNet) was developed for individualized survival predictions.

Main Results:

• Admission serum chloride levels significantly correlate with 28-day mortality, with critical thresholds identified at 103 mEq/L and 115 mEq/L.
• Patients with chloride levels ≤103 mEq/L (HR 1.36) and >115 mEq/L (HR 1.27) exhibited higher mortality risks.
• Cross-cohort validations confirmed these ranges, and the Causal SurvivalNet model demonstrated accurate individual survival predictions (Brier scores 0.09-0.15).

Conclusions:

• Initial serum chloride levels are valuable for enhancing prognostic accuracy in ICU patients.
• These findings support the development of tailored treatment plans for critical care settings based on chloride levels.