Harnessing Natural Language Processing and High-Dimensional Clinical Notes to Detect Goals-of-Care and Surrogate-Designation Conversations | JoVE Visualize

Area of Science:

Medical Informatics
Clinical Documentation Analysis
Natural Language Processing in Healthcare

Background:

Advance care planning (ACP) is vital for patient-centered care, encompassing goals-of-care and surrogate designation.
Optimal timing and participants for ACP discussions remain challenging to determine in clinical practice.
Accurate documentation of ACP elements in clinical notes is essential for continuity of care.

Purpose of the Study:

To explore the frequency, timing, and predictors of documenting goals-of-care and surrogate-designation conversations in clinical notes.
To evaluate the efficacy of natural language processing (NLP) models in detecting ACP conversations within electronic health records.
To identify temporal patterns and patient characteristics associated with the documentation of advance care planning.

Main Methods:

Retrospective observational study analyzing clinical notes from a Midwestern United States hospital.
Leveraged high-dimensional data and NLP, including a transformer-based model (BERT), to detect ACP conversations.
Manual labeling of a gold-standard dataset (n=913 notes) for model training and validation, achieving 90.5% inter-annotator agreement.

Main Results:

The BERT model achieved high accuracy (F1 score of 93.6) in detecting goals-of-care and surrogate-designation conversations.
ACP conversations were detected in 85% (goals-of-care) and 60% (surrogate-designation) of 4,341 patients' records.
Goals-of-care and surrogate-designation conversations were documented at median 28 and 8 days before death, respectively; patient characteristics and palliative care referral were significant predictors.

Conclusions:

NLP, particularly BERT, can accurately detect advance care planning conversations in clinical narratives.
Significant temporal patterns, including late documentation, were identified, underscoring the need for earlier interventions.
Understanding patient characteristics associated with ACP documentation can inform strategies to improve patient-centered care and timely discussions.