Machine Learning for Predicting Oxygen Delivery in ICU Patients
Author Information
Author(s): Holman Heather BS, Baronov Dimitar PhD, McMurray Jeff MD, Kilic Arman MD, Katz Marc MD, Zeigler Sanford MD
Primary Institution: Medical University of South Carolina
Hypothesis
An ML algorithm could predict impaired delivery of oxygen (IDO2) with comparable discrimination to invasive mixed venous oxygen saturation (SvO2) measurement.
Conclusion
The IDO2 index is capable of detecting SvO2 ≤50% with good discriminatory function in non-MCS CVICU patients in a variety of monitoring situations.
Supporting Evidence
- The IDO2 index showed an AUC of 0.89 for detecting SvO2 ≤50%.
- The model maintained predictive capability even without PAC or ScvO2 data.
- Patients requiring mechanical circulatory support were excluded from the study.
Takeaway
This study shows that a computer program can help doctors know if patients are getting enough oxygen without needing to use complicated equipment.
Methodology
The study involved 230 patients managed with a pulmonary artery catheter, using a machine learning algorithm to predict oxygen delivery based on physiological data.
Potential Biases
The model relies on assumptions of normal physiology and may not account for variations in patient conditions.
Limitations
The study is limited to a CVICU population, and results may not be applicable to other settings; patients requiring mechanical circulatory support were excluded.
Participant Demographics
{"male_percentage":69,"postoperative_percentage":78,"age_median":64.6,"age_range":"54.4-73.3"}
Statistical Information
Confidence Interval
0.87-0.91
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website