Development and Preliminary Evaluation of a Visual Annotation Tool to Rapidly Collect Expert-Annotated Weight Errors...

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Patient Safety

Background:

• Incorrect patient weights entered into electronic health records (EHR) can lead to significant patient harm, particularly in pediatrics due to weight-based dosing.
• Manual chart reviews for weight errors are impractical in clinical settings, and existing EHR alerts are insufficient.

Purpose of the Study:

• To develop an advanced algorithm for detecting patient weight errors in EHR systems using supervised machine learning.
• To design and preliminarily evaluate a visual annotation tool for rapid collection of expert-annotated weight errors.

Main Methods:

• Agile software development methodology was employed to design the visual annotation tool.
• The tool's design leveraged the observation that medical experts can readily identify infrequent weight errors.

Main Results:

• Preliminary evaluation yielded positive user feedback on the visual annotation tool.
• The tool successfully facilitated the initial collection of expert-annotated weight errors.

Conclusions:

• The developed visual annotation tool is a promising approach for efficiently gathering labeled data for training machine learning algorithms to detect EHR weight errors.
• Positive user feedback indicates readiness for a formal user-centered evaluation, paving the way for improved patient safety through accurate weight data.