System developed by CLEW allows large-scale AI Model Development….
A fundamental requirement for developing machine learning (ML) predictive models is to accurately identify (i.e., tag) significant events of interest within the data. These tags are critical for training predictive models to identify features that predict them and validate the models’ performance against those tags, which serve as a gold standard against which the models’ predictions are assessed. However, predictive model development usually requires vast amounts of data, making these events’ manual identification and timestamping inefficient, timely, and expensive.
A new study recently published in the Healthcare Informatics Research journal validates an automatic tagging system developed by Clew Medical. This rule-based tagging system was designed to identify within the electronic medical record (EMR) data-specific deterioration events so that predictive models could be developed and validated for those events. More specifically, the system identified and tagged respiratory and hemodynamic deterioration events within a large ICU database. The study validates the performance of the automatic tagging system by comparing it to tagging by expert human reviewers.
The study results demonstrated how the automated tagging system was able to identify respiratory and hemodynamic deterioration events rapidly and efficiently in a large dataset and tag these events close to the accuracy of the human experts; the overall agreement rate between the automated system and the human experts was 89.4% for respiratory failure events, and 87.1% for hemodynamic instability events.
In summary, the study validates the automatic tagging system for respiratory and hemodynamic deterioration events, providing a rapid and accurate tool for mass tagging of a compound clinical database. The system will allow the replacement of human reviewers and considerable time and resource-saving while creating validated, labeled databases for training artificial intelligence algorithms.
Jeddah D, Chen O, Lipsky A, Forgacs A, Celniker G, Lilly C, Pessach I. Validation of an Automatic Tagging System for Identifying Respiratory and Hemodynamic Deterioration Events in the Intensive Care Unit. Healthcare Informatics Research, 2021 July; 27(3): 241-248.