Using natural language processing to qualitatively assess goals of care conversations for patients with cancer.

person Melissa Greene Duke University School of Medicine, Durham, NC info_outline Melissa Greene, Gloria Broadwater, Donna Niedzwiecki, Thomas William LeBlanc, Jessica Ma, David J. Casarett, Brittany Anne Davidson

Authors person Melissa Greene Duke University School of Medicine, Durham, NC info_outline Melissa Greene, Gloria Broadwater, Donna Niedzwiecki, Thomas William LeBlanc, Jessica Ma, David J. Casarett, Brittany Anne Davidson Organizations Duke University School of Medicine, Durham, NC, Department of Biostatistics & Bioinformatics & CALGB Statistical Center, Duke Cancer Institute, Durham, NC, Duke University, Durham, NC, Division of General Internal Medicine, Department of Medicine, Duke University Medical Center, Durham, NC, Duke Cancer Institute, Durham, NC Abstract Disclosures Research Funding Duke Division of Gynecologic Oncology Background: The importance of goals of care (GOC) discussions during end-of-life (EOL) care is well known. Institutions are increasingly tracking the frequency and timing of GOC documentation, but qualitative assessments of this content on a large scale have been limited given the person-power required and constraints of electronic health systems. Our aim is to use natural language processing to qualitatively assess the content of GOC documentation and associations with EOL care in patients who died with cancer. Methods: This is a retrospective review of patients with solid or hematologic malignancies who died between 2018-2022 with documented GOC notes in the last 12 months of life at a single academic National Cancer Institute-designated cancer center. Eight GOC components were identified based on prior literature: current understanding of patient’s illness, information preferences, disclosure of prognosis, goals, fears, acceptable function, trade-offs, and family involvement. GOC notes were extracted from the electronic health record and text annotation software Clinical Regex was utilized to search for the aggregate presence of these GOC components. Associations between these 8 components and receipt of aggressive EOL care (chemotherapy within 14 days of death, no hospice care, or hospice admission ≤3 days of death) were evaluated. Results: 2,031 patients met inclusion criteria. The most common disease sites were gynecologic (22.6%), gastrointestinal (20.3%), and thoracic (16.7%). The most common GOC components addressed were family involvement (75.0%), patient goals (72.0%), acceptable function, and decision making & information preferences (both 65.8%). Only 5.4% had all 8 components addressed in documentation the last 12 months of life. More comprehensive GOC notes were associated with lower rates of aggressive EOL care; 73.2% received aggressive EOL when 0/8 components were documented, compared to 56.8% and 50.3% when 6 or 7 components were discussed, respectively. In multivariate logistic regression, GOC components documented (≤6 vs ≥7), primary tumor site, and inpatient palliative care referral were independent predictors of aggressive EOL care (p-values <0.0001). Conclusions: GOC conversation documentation is largely not comprehensive. Increasingly comprehensive documentation is associated with a lower likelihood of receiving aggressive EOL care, suggesting opportunities to improve both the quality and documentation of GOC conversations may impact the EOL care for patients with cancer. Opportunities for increasing outpatient palliative care referrals may also improve the quality of the EOL experience for patients. # of GOC components discussed and percentage that received aggressive EOL care. Number of GOC Components Addressed Received Aggressive EOL Care 0 73.2% ≥ 1 60.7% ≥ 2 60.0% ≥ 3 60.6% ≥ 4 61.1% ≥ 5 59.0% ≥ 6 56.8% ≥ 7 50.3% All 8 53.6%