Machine learning approach for the prediction of 30-day mortality in patients with sepsis-associated encephalopathy

Area of Science:

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

Background:

• Sepsis-associated encephalopathy (SAE) poses a significant risk of mortality in intensive care units (ICUs).
• Predicting 30-day mortality in SAE patients is crucial for timely intervention and resource allocation.
• Existing predictive tools may require enhancement for improved accuracy and clinical utility.

Purpose of the Study:

• To identify key predictors of 30-day mortality in patients with SAE.
• To develop and validate machine learning (ML) models for predicting SAE mortality risk.
• To compare the performance of various ML models in predicting SAE outcomes.

Main Methods:

• Utilized the Medical Information Mart for Intensive Care (MIMIC)-IV public database for model development and validation.
• Employed recursive feature elimination (RFE) to identify significant predictive variables.
• Evaluated ML models using metrics such as AUC, accuracy, sensitivity, specificity, and the Hosmer-Lemeshow test.

Main Results:

• A cohort of 6994 patients with SAE was analyzed, with a 17.62% 30-day mortality rate.
• Fifteen key predictors were identified, including APSIII, GCS, SOFA, CCI, RDW, BUN, age, respiratory rate, PaO2, temperature, lactate, CRE, malignant cancer, metastatic solid tumor, and PLT.
• Artificial neural network (NNET), logistic regression (LR), and adaptive boosting (Ada) models demonstrated good calibration and high predictive accuracy.

Conclusions:

• Developed ML models effectively predict 30-day mortality risk in SAE patients within the ICU.
• The identified predictors and validated ML models can aid in evaluating patient prognosis.
• An online calculator could enhance the accessibility and application of these predictive models in clinical practice.