Deciphering clinical abbreviations with a privacy protecting machine learning system.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Computational Linguistics

Background:

• Physicians frequently use abbreviations and shorthand in clinical notes, leading to ambiguity and comprehension challenges.
• Abbreviations can include clinical jargon, ambiguous terms, or domain-specific vernacular, requiring expert knowledge for accurate interpretation.
• Existing methods for deciphering abbreviations may be limited in scope or require privacy-compromising data.

Purpose of the Study:

• To develop and evaluate a machine learning model for the automated detection and expansion of abbreviations in clinical text.
• To create a generalizable method for deciphering medical shorthand without using sensitive patient data.
• To assess the model's performance against human expert interpretation.

Main Methods:

• Training machine learning models on publicly available web data to recognize and translate abbreviations.
• Developing a single translation model capable of simultaneously detecting and expanding thousands of abbreviations.
• Testing the model on multiple external datasets of real clinical notes.

Main Results:

• The translation model achieved high accuracies ranging from 92.1% to 97.1% on diverse test datasets.
• The model's performance in deciphering abbreviations met or surpassed that of board-certified physicians (97.6% vs. 88.7% total accuracy).
• The developed method effectively decodes abbreviations and shorthand in contextually relevant ways.

Conclusions:

• A generalizable machine learning approach can accurately decipher medical abbreviations and shorthand from clinical notes.
• This privacy-preserving method enhances the clarity and accessibility of clinical documentation.
• The model offers a valuable tool for improving medical record comprehension and potentially clinical decision-making.