708 posters, 8 topics, 16 sessions, 1,190 authors, 600 institutions
ePostersLive by SciGen Technologies S.A. All rights reserved.
ISQua's 39th International Conference 2023
August 27 - 30, 2023 | Seoul, Republic of Korea
P503
ARTIFICIAL INTELLIGENCE-BASED HEALTHCARE-ASSOCIATED BLOODSTREAM INFECTION SURVEILLANCE AND CLASSIFICATION SYSTEM
Poster Presenter
Part of Topic
Digital Healthcare and Innovation
Artificial intelligence-based healthcare-associated bloodstream infection surveillance and classification system
Introduction
•Central line-associated bloodstream infections (CLABSI) are a significant healthcare concern, with the potential to cause prolonged hospital stays, increased costs, and increased risk of mortality.
•However, identifying CLABSIs is a labor-intensive process, and performance depends on the expertise of infection control personnel (ICP).
•This study aims to develop a web-based system integrated with AI to detect and classify CLABSIs automatically to improve efficiency and accuracy.
Methods
•The Research Ethics Committee of Chung Shan Medical University Hospital approved the study (No. CS2-23009).
•A rule-based AI was developed to detect CLABSIs from structured data, such as vital parameters, and processed data based on the surveillance definitions defined by the Taiwan Centers for Disease Control in 2018.
•The initial detection performance was evaluated by comparing it to ICP-based surveillance results, and any inconsistencies were peer-reviewed to verify the accuracy and confirm the detection performance.
Results
•First, an additional module was developed to standardize the data (identifying the name, placement, indication, and access of central lines from the catheter records).
•Next, natural language processing was applied to identify the pathogens from laboratory reports and signs or symptoms from nursing records (standardizing the diverse symptom description in Chinese or English and typos).
•Finally, a rule-based AI was developed.
•A total of 22,961 samples from 9,173 inpatients were positive blood specimens, triggering the surveillance and classification procedure. According to the ICP-based results, 299 samples (1.3%) from 262 inpatients were identified as CLABSI.
•After peer review, we corrected 73 false positive (FP) as CLABSI, indicating the TP increased from 291 to 364. The sensitivity, specificity, positive predictive value, and negative predictive value (NPV) of the final AI system's performance were 97.85%, 99.11%, 64.65%, and 99.96%.
Conclusion
Using an AI-based model with high negative predictive value could significantly reduce the labor-intensive of infection control personnel and improve the surveillance quality of central line-associated bloodstream infections (CLABSI). Speedy detection of CLABSI can aid in improving healthcare quality and cost-effectiveness.