Machine Learning Predictive Model for Septic Shock in Acute Pancreatitis with Sepsis

Area of Science:

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

Background:

• Acute pancreatitis (AP) can progress to fatal septic shock.
• Early identification of high-risk patients is crucial for timely intervention.
• Machine learning (ML) offers potential for improved early prediction models.

Purpose of the Study:

• To develop and validate an ML model for early prediction of septic shock in AP patients with sepsis within 28 days of admission.
• To guide emergency physicians in clinical decision-making and resource allocation.

Main Methods:

• Retrospective cohort study of 604 AP patients with sepsis from three tertiary hospitals.
• Data split into training (70%) and testing (30%) sets.
• Ten ML classifiers evaluated, with an auto-encoder (AE) model optimized using mean normalization, Pearson correlation coefficient (PCC), and recursive feature elimination (RFE).

Main Results:

• The optimized AE model achieved high performance: AUC 0.900 (accuracy 0.868) on validation and AUC 0.879 (accuracy 0.790) on testing datasets.
• AE model demonstrated superior predictive performance compared to traditional scoring systems (SOFA, qSOFA, APACHE II, SAPS).

Conclusions:

• The AE model significantly outperforms traditional scoring systems in predicting septic shock in AP patients with sepsis.
• This ML-driven approach facilitates early identification of high-risk patients, enabling prompt medical interventions.