Automated Annotation of Pain Chronicity in Patients With Back Pain by Using Electronic Health Records: Retrospective...

Area of Science:

• Biomedical Informatics
• Clinical Informatics
• Data Science in Healthcare

Background:

• Chronic back pain presents diagnostic challenges due to complex biopsychosocial factors.
• Manual annotation of pain chronicity in electronic health records (EHRs) is time-consuming and prone to variability.
• Standardized criteria for pain chronicity assessment are lacking in clinical practice.

Purpose of the Study:

• To investigate the association between expert-annotated pain chronicity and social determinants within EHR data.
• To assess the feasibility of extracting pain chronicity directly from EHRs without manual expert input.
• To develop and evaluate machine learning models for automated pain chronicity prediction.

Main Methods:

• Utilized univariate regression to analyze associations between structured EHR variables and clinician-annotated pain chronicity.
• Trained a random forest model incorporating structured EHR data and unstructured clinical notes processed by a natural language processing (NLP) tool.
• Extracted features included clinical keywords, pain duration, and International Classification of Diseases, Tenth Revision (ICD-10) codes.

Main Results:

• Univariate analysis revealed significant associations between pain chronicity and variables such as pain severity, imaging orders, neurology visits, and Medi-Cal insurance.
• A random forest model using structured data achieved a high correlation (0.887) with a mean absolute error (MAE) of 18.45.
• An enhanced model incorporating NLP-extracted information from unstructured notes demonstrated superior performance with a correlation of 0.968 and an MAE of 10.87.

Conclusions:

• Automated prediction of pain chronicity from EHR data is feasible using machine learning and NLP techniques.
• This automated approach can facilitate the study of chronic back pain on larger datasets, bypassing the need for manual annotation.
• The findings support the integration of NLP tools for efficient and reliable pain chronicity assessment in clinical research.