Artificial Intelligence May Predict Early Sepsis After Liver Transplantation
Rishikesan Kamaleswaran1,2, Sanjaya K Sataphaty3, Valeria R Mas4
1Emory University School of Medicine, Atlanta, GA, United States.
Frontiers in Physiology
|September 23, 2021
Summary
This study developed an AI model using continuous physiologic data to predict post-liver transplant sepsis. The model achieved high accuracy in identifying sepsis 12 hours in advance, improving patient outcomes.
Area of Science:
- Critical Care Medicine
- Medical Informatics
- Artificial Intelligence in Healthcare
Background:
- Sepsis following liver transplantation presents a significant clinical challenge, impacting patient prognosis.
- Early prediction of post-operative sepsis is crucial for timely intervention and improved outcomes.
Purpose of the Study:
- To develop and validate an artificial intelligence (AI) method for early prediction of post-operative sepsis in liver transplant recipients.
- To identify key physiological markers ('physiomarkers') from continuous monitoring data predictive of sepsis onset.
Main Methods:
- A cohort of 5,748 intensive care unit (ICU) patients, including 92 post-liver transplant sepsis cases, was analyzed.
- Continuous minute-by-minute physiologic data (heart rate, respiratory rate, SpO2, blood pressure) were processed to generate 155 features.
- An eXtreme Gradient Boost (XGB) classifier was trained and validated using 5-fold cross-validation.
Main Results:
- The AI model demonstrated high performance, with an average sensitivity of 0.94, specificity of 0.90, and AUC of 0.97.
- The model successfully predicted sepsis up to 12 hours before meeting Sepsis-3 criteria.
- Key features identified included various blood pressure parameters.
Conclusions:
- Machine learning and deep learning models can effectively utilize continuous streaming data for sepsis prediction in transplant ICUs.
- This AI approach shows promise for enhanced patient monitoring and early sepsis detection in critical care settings.
