Enhanced Identification of Care Preference Documentation in Patients' Discharge Summaries Using Pre-Trained Large Language Models

Improving patient-centered care necessitates accurate documentation of care preferences, a crucial aspect often underrepresented in administrative data. Most studies apply care documentation to specific patient populations, rather than the more appropriately broad population of ‘seriously ill’ patients. This study addresses this gap, first by matching decedent patients to non-decedent counterparts by using propensity score matching, accounting for important variables like age, gender, primary diagnoses, and patient co-morbidities, and then by leveraging transformer-based machine learning models, exhibiting an improvement over traditional keyword-based search methods. We trained and fine-tuned Bio_ClinicalBERT and ClinicalLong-former models on a large dataset of patient discharge summaries, enhancing their signal-to-noise ratio by focusing on key textual aspects, thereby capturing contextually nuanced mentions of care preference documentation. These models demonstrated high sensitivity and specificity compared to industry-standard keyword-search methods, proving adept at interpreting complex clinical concepts. This study highlights the potential of transformer-based models specifically trained for clinical domain tasks. Our findings not only contribute to enhancing end-of-life communication and aligning treatment with patients' care objectives but also pave the way for future research in this promising domain, with potential implications for improving patient care quality.