Authors:
Thomas Hartvigsen
1
;
Cansu Sen
1
;
Sarah Brownell
2
;
Erin Teeple
1
;
Xiangnan Kong
1
and
Elke Rundensteiner
1
Affiliations:
1
Worcester Polytechnic Institute, United States
;
2
Simmons College, United States
Keyword(s):
MRSA, Healthcare-associated Infections, Risk Stratification, Machine Learning, Electronic Health Records.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Biomedical Engineering
;
Data Mining
;
Databases and Information Systems Integration
;
Electronic Health Records and Standards
;
Enterprise Information Systems
;
Health Information Systems
;
Pattern Recognition and Machine Learning
;
Physiological Modeling
;
Sensor Networks
;
Signal Processing
;
Soft Computing
Abstract:
Despite eradication efforts, Methicillin-resistant Staphylococcus aureus (MRSA) remains a common cause
of serious hospital-acquired infections (HAI) in the United States. Electronic Health Record (EHR) systems
capture MRSA infection events along with detailed patient information preceding diagnosis. In this work, we
design and apply machine learning methods to support early recognition of MRSA infection by estimating risk
at several time points during hospitalization. We use EHR data including on-admission and throughout-stay
patient information. On-admission features capture clinical and non-clinical information while throughout-stay
features include vital signs, medications, laboratory studies, and other clinical assessments. We evaluate
prediction accuracy achieved by core Machine Learning methods, namely Logistic Regression, Support Vector
Machine, and Random Forest classifiers, when mining these different types of EHR features to detect patterns
predictive of MRSA infec
tion. We evaluate classification performance using MIMIC III – a critical care
data set comprised of 12 years of patient records from the Beth Israel Deaconess Medical Center Intensive
Care Unit in Boston, MA. Our methods can achieve near-perfect MRSA prediction accuracies one day before
documented clinical diagnosis. Also, they perform well for early MRSA prediction many days in advance of
diagnosis. These findings underscore the potential clinical applicability of machine learning techniques.
(More)