Using Natural Language Processing to Assess Goals-of-Care Conversations for Patients With Cancer.

Area of Science:

• Oncology
• Palliative Care
• Health Informatics

Background:

• Goals-of-care (GOC) discussions are crucial for advanced serious illness and end-of-life (EOL) care.
• Institutions increasingly track GOC documentation frequency and timing.
• Large-scale assessments of GOC documentation content quality have been limited.

Purpose of the Study:

• To use natural language processing (NLP) to assess GOC documentation quality in cancer patients.
• To evaluate the association between GOC documentation quality and end-of-life (EOL) care received by cancer patients.

Main Methods:

• Retrospective review of cancer patients who died between 2018-2022 with documented GOC notes in the last year of life.
• Identified eight key GOC components (illness understanding, information preferences, prognostic disclosure, goals, fears, acceptable function, trade-offs, family involvement).
• Utilized NLP to search for the presence of these components and evaluated associations with aggressive EOL care.

Main Results:

• 2,031 patients met inclusion criteria; only 5.4% had all eight GOC components documented.
• More comprehensive GOC notes correlated with lower rates of aggressive EOL care (e.g., 50.3% with 6-7 components vs. 73.2% with 0 components).
• GOC component documentation (≤6 vs. ≥7 components) and primary tumor site were independent predictors of aggressive EOL care.

Conclusions:

• Higher quality and more comprehensive GOC documentation are associated with reduced likelihood of aggressive EOL care for cancer patients.
• Improving the quality and documentation of GOC conversations presents an opportunity to positively influence EOL care.
• NLP offers a scalable method for assessing GOC documentation quality.