Advanced Care Planning Content Encoding with Natural Language Processing.

Area of Science:

• Clinical Informatics
• Natural Language Processing
• Health Services Research

Background:

• Advanced care planning (ACP) is crucial for patient-centered care but suffers from poor adoption and variable documentation completeness.
• Gaps in ACP documentation at the point of care hinder effective patient care coordination and decision-making.

Purpose of the Study:

• To evaluate the utility of Natural Language Processing (NLP) for identifying Advanced Care Planning (ACP) themes within electronic health records.
• To assess the variability in ACP documentation across a large corpus of clinical notes using NLP.

Main Methods:

• ACP themes were annotated by four experts on palliative care notes, achieving high inter-annotator agreement (Fleiss kappa = 0.753).
• Supervised machine learning models (bert-base-uncased, Bio_ClinicalBERT) were trained using 5-fold cross-validation, demonstrating strong performance (F1=0.8, precision=0.75, recall=0.86).
• The trained NLP models were applied to a corpus of 12,711 clinical notes to analyze ACP documentation patterns.

Main Results:

• The NLP models successfully identified ACP themes with high accuracy, indicating their potential for automated analysis.
• Application of the models to the full note corpus revealed significant variability in the completeness and presence of ACP information.
• The study highlights NLP's capability to systematically assess documentation quality in large-scale clinical data.

Conclusions:

• Natural Language Processing offers a promising informatics-based approach to support and improve Advanced Care Planning (ACP).
• NLP tools can help identify documentation gaps and support patient-centered care by ensuring comprehensive ACP information.
• Further development and implementation of NLP can enhance the consistency and quality of ACP documentation in clinical practice.