Detect & Reduce Excessive Antibiotic Medication (DREAM)
Machine learning guided antibiotic therapy discontinuation to reduce overuse in hospitals
OPEN ACCESS
Author / Producer
Date
2023
Publication Type
Master Thesis
ETH Bibliography
yes
Citations
Altmetric
OPEN ACCESS
Data
Rights / License
Abstract
This thesis addresses the critical problem of antibiotic overuse in Intensive Care Units (ICUs), a prevalent issue that leads to numerous risks such as increased antimicrobial resistance, severe side effects, and elevated healthcare costs. Despite the growing awareness and countermovements advocating for judicious antibiotic use, a not inconsiderable proportion of unnecessary antibiotic treatments are stopped too late. Bridging this gap, our study presents a novel flexible Machine Learning (ML) pipeline, comprising two innovative models that employ a direct and an indirect method to assess the necessity of antibiotic treatments during the initial critical days of patient care and aiming to provide clinicians with a reliable tool to discontinue unnecessary antibiotic treatments safely. By comparing the outcomes of these models using a custom performance metric, are we able to compare different approaches with each other. Additionally, the models’ decision-making processes are elucidated through SHAP (SHapley Additive exPlanations) values, offering deep insights into the factors influencing their recommendations. A significant part of our research is dedicated to validating the models’ transferability across unseen datasets, ensuring their applicability in various clinical settings. This work not only contributes to the enhancement of clinical decision-making but also stands as a significant step towards reducing the adverse effects associated with antibiotic overuse in ICUs.
Permanent link
Publication status
published
External links
Editor
Contributors
Examiner : Vogt, Julia
Examiner : Sutter, Thomas
Examiner : Vandenhirtz, Moritz
Book title
Journal / series
Volume
Pages / Article No.
Publisher
ETH Zurich
Event
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Organisational unit
09670 - Vogt, Julia / Vogt, Julia