Identification of Prediabetes Discussions in Unstructured Clinical Documentation: Validation of a Natural Language...

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Prediabetes impacts one in three US adults, yet many do not receive evidence-based interventions.
• Understanding provider-patient discussions about prediabetes is crucial for enhancing care quality.

Purpose of the Study:

• To develop a natural language processing (NLP) algorithm utilizing machine learning to detect prediabetes conversations within narrative clinical documentation.
• To improve the identification of prediabetes management discussions in electronic health records.

Main Methods:

• A keyword search strategy was employed to locate potential prediabetes discussions in clinical notes.
• Manual review of identified notes was conducted to confirm actual prediabetes discussions.
• Seven distinct machine learning models were applied and evaluated against manually annotated data.

Main Results:

• Machine learning classifiers achieved high performance, nearing human accuracy in identifying prediabetes discussions.
• The developed NLP algorithm demonstrated up to 98% precision and recall in classifying prediabetes conversations.

Conclusions:

• Prediabetes discussions within clinical documentation can be accurately identified using a developed NLP algorithm.
• This approach offers a scalable method to analyze prediabetes management practices in primary care.
• Findings can inform the development of targeted interventions to improve guideline-concordant prediabetes care.