Applying Machine Learning for Prescriptive Support: A Use Case with Unfractionated Heparin in Intensive Care Units
Boris Delange1,2, Guillaume Bouzille1, Isabelle Gouin3
1CHU Rennes, INSERM, LTSI-UMR 1099, Univ Rennes, 35000 Rennes, France.
Studies in Health Technology and Informatics
|August 23, 2024
Summary
Machine learning models can predict unfractionated heparin dosing errors using anti-Xa levels. These tools, based on random forest and XGB algorithms, aim to improve patient safety in intensive care units.
Area of Science:
- Pharmacokinetics
- Machine Learning
- Clinical Decision Support
Background:
- Continuous unfractionated heparin is crucial in intensive care but challenging to dose accurately due to complex pharmacokinetics.
- Over- and under-dosing of heparin can lead to adverse patient outcomes.
Purpose of the Study:
- To develop and evaluate machine learning models for predicting unfractionated heparin over- and under-dosing.
- To enhance model interpretability for clinical acceptance.
Main Methods:
- Retrospective analysis of a monocentric dataset.
- Development of random forest and XGBoost machine learning models using anti-Xa results.
- Assessment of model performance using Area Under the Receiver Operating Characteristic curve (AUROC).
Main Results:
- Both random forest and XGBoost models achieved a mean AUROC of 0.80.
- Feature importance analysis was used to improve model interpretability.
Conclusions:
- Machine learning models show promise in predicting heparin dosing errors.
- Prospective validation and integration into computerized physician order entry systems could establish these models as clinical decision support tools.
