On classifying sepsis heterogeneity in the ICU: insight using machine learning
Problem
Current ML models predicting sepsis from electronic health records ignore the condition's heterogeneity, limiting risk recognition in the ICU.
Why it matters
Sepsis heterogeneity matters for prognosis and treatment of ICU patients.
What the AI did
Identified clinically significant sepsis subpopulations with distinct organ dysfunction patterns and used them as background knowledge for feature selection in random forest, gradient boosting, and SVM classifiers.
Results
Subpopulation-based features yielded superior performance distinguishing sepsis patients on 13,728 ICU records; exact metrics truncated
Impact
Improved EHR-based recognition of patients at risk of sepsis in classification experiments.
Record
| Venue | Journal of the American Medical Informatics Association |
|---|---|
| Published | 2019 |
| Work type | article |
| Authors | Zina M. Ibrahim, Honghan Wu, Ahmed Hamoud, Lukas Stappen, Richard Dobson, Andrea Agarossi |
| Citations | 73 (OpenAlex, at import time) |
| References | 45 |
| Model or method | random forest, gradient boost trees, support vector machines |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1093/jamia/ocz211 |
| Open access | hybrid |
| Keywords | sepsis, ICU, machine learning, EHR, heterogeneity, classification |
Importance rating
significant — thousands to millions affected
Abstract states sepsis heterogeneity has emerging importance in prognosis and treatment of ICU patients; improved risk recognition is significant, but no scale quantified.