A Machine Learning Model for Early Prediction and Detection of Sepsis in Intensive Care Unit Patients

Area of Science:

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

Background:

• Sepsis presents a major healthcare challenge with high mortality rates in intensive care units (ICUs).
• Delayed diagnosis and treatment of sepsis lead to advanced stages, increased fatalities, and higher healthcare costs.
• Current sepsis detection methods, relying on manual scores or limited features, lack automation and optimal accuracy.

Purpose of the Study:

• To develop and evaluate a machine learning model for early detection and prediction of sepsis in ICU patients.
• To compare the performance of various machine learning algorithms for sepsis prediction.
• To introduce an advanced ensemble model for enhanced sepsis detection accuracy.

Main Methods:

• Utilized clinical laboratory values and vital signs from an ICU patient database.
• Simulated and evaluated multiple machine learning models: Random Forest (RF), Linear Regression (LR), Support Vector Machine (SVM), Naive Bayes (NB), an initial ensemble, XGBoost, and a proposed comprehensive ensemble model.
• Performance was assessed using balanced accuracy metrics.

Main Results:

• The proposed ensemble model, combining SVM, RF, NB, LR, and XGBoost, achieved the highest balanced accuracy of 0.96.
• Individual models showed varying performance: RF (0.90), LR (0.73), SVM (0.93), NB (0.74), initial ensemble (0.94), and XGBoost (0.95).
• The proposed ensemble model outperformed all other evaluated models in sepsis prediction accuracy.

Conclusions:

• The developed machine learning ensemble model demonstrates superior performance for early sepsis detection in ICUs.
• Accurate and timely sepsis prediction using this model can potentially reduce mortality rates and optimize resource allocation.
• This automated approach offers a significant advancement over existing sepsis detection methods.