Development and Validation of Machine Learning Model to Predict Refractory Septic Shock

Area of Science:

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

Background:

• Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection.
• Refractory septic shock (RSS) is associated with high mortality rates and presents a significant clinical challenge.
• Accurate and early prediction of RSS is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting refractory septic shock (RSS) in adult intensive care unit (ICU) patients.
• To identify key clinical and laboratory features predictive of RSS.
• To assess the model's performance in a prospective validation cohort.

Main Methods:

• An ambispective study design was employed, utilizing data from 1,008 patients for model development and 102 patients for prospective validation.
• A three-tiered feature selection process was used to identify significant variables.
• Random Forest classifiers were trained and optimized using selected features.

Main Results:

• The best-performing ML model incorporated 27 clinical and laboratory features.
• The model achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.877 in the training cohort and 0.839 in prospective validation.
• The model demonstrated high accuracy, precision, and recall in predicting RSS.

Conclusions:

• The validated ML model shows strong predictive ability and interpretability for identifying patients at risk of refractory septic shock.
• Early identification of high-risk patients can facilitate timely interventions and potentially improve outcomes.
• Further multicenter studies are warranted to confirm the generalizability of the model for widespread clinical implementation.