A Real-Time Dynamic Warning Method for MODS in Trauma Sepsis Patients Based on a Pre-Trained Transfer Learning...

Area of Science:

• Critical Care Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Multiple organ dysfunction syndrome (MODS) is a significant complication in trauma sepsis with poor prognosis.
• Early and individualized risk assessment is crucial for timely intervention in trauma sepsis patients.

Purpose of the Study:

• To develop an interpretable, multicenter-validated prediction model for early risk assessment of MODS in trauma sepsis.
• To enable individualized risk stratification and guide timely clinical care.

Main Methods:

• Utilized MIMIC-IV and eICU datasets for model development.
• Employed a pre-trained transfer-learning model with a separation processing strategy.
• Assessed model interpretability using SHAP (SHapley Additive exPlanations).

Main Results:

• Internal validation (700 patients) and external validation (110 patients) demonstrated robust performance.
• The best pre-trained model achieved an average AUC of 0.906 across 6-, 12-, and 24-hour prediction windows.
• Fine-tuning on limited trauma sepsis data (100 cases) yielded an AUC of 0.846, outperforming non-pre-trained models.

Conclusions:

• The pre-trained MODS prediction model exhibits strong discrimination, generalizability, and interpretability.
• The model's portability supports its clinical potential for early identification of high-risk trauma sepsis patients.
• Platelet count emerged as a key predictor of MODS in trauma sepsis.