Logo image
Back
Identifying Surrogate Decision Maker and Code Status Discussions in Advance Care Planning Notes Using Natural Language Processing (GP701)
Abstract   Open access   Peer reviewed

Identifying Surrogate Decision Maker and Code Status Discussions in Advance Care Planning Notes Using Natural Language Processing (GP701)

Alaa Albashayreh and Stephanie Gilbertson-White
Journal of pain and symptom management, Vol.63(6), p.1112
06/01/2022
DOI: 10.1016/j.jpainsymman.2022.04.092
url
https://doi.org/10.1016/j.jpainsymman.2022.04.092View
Published (Version of record) Open Access

Abstract

Outcomes. 1. Discuss the process of extracting advance care planning information from electronic health records notes 2. Describe the performance of NimbleMiner, a natural language processing tool based on word embeddings, in identifying advance care planning information as compared to chart review and structured administrative data Importance. Electronic health record (EHR) notes are a rich source for assessing elements of palliative care such as discussing a surrogate decision maker and a code status. However, complex language and unstructured text are common barriers to extracting such palliative care elements. Natural language processing (NLP) is an efficient tool for automating information extraction from EHR notes. Objective(s). To describe the use of NLP to identify documentation of surrogate decision maker and code status discussions in the advance care planning notes of patients with cancer. Method(s). We identified patients with a diagnosis of cancer who had palliative care consultations at a large teaching hospital in Iowa. We extracted administrative data and notes, including advance care planning notes. A dictionary of terms for surrogate decision maker and code status discussions was created. We used NimbleMiner, an NLP tool based on word embeddings, to process advance care planning notes and identify surrogate decision maker and code status information. Results. We included a sample of 1,563 advance care planning notes for 1,563 unique patients (mean age = 70, standard deviation = 14.1). Of the total notes, NLP determined 91.8% (n = 1,435) of notes had documentation of a surrogate decision maker, and 89.4% (n = 1,335) of notes had documentation of code status. Compared to chart review and structured data, NLP achieved F1 scores of 0.88 and 0.93 for identifying the documentation of a surrogate decision maker and code status, respectively. Conclusion(s). NLP is an efficient tool to automate the extraction of advance care planning information from EHR notes. Impact. The use of real-world EHR notes and NLP has the potential to improve evaluating palliative care metrics and advance care planning.
 Automation Electronic Health Records Medical Diagnosis Palliative Care Advance directives Cancer Care plans Chart reviews Computerized medical records Documentation Health records Natural language processing Teaching

Details

Metrics

9 Record Views
Logo image