Extracting Symptoms Relevant to Geriatric Syndromes from Japanese Pharmaceutical Care Records: A Natural Language Processing Analysis

Abstract

Geriatric syndromes are multifactorial conditions associated with adverse outcomes in older adults, but how community pharmacists capture related symptoms remains unclear. Therefore, a natural language processing (NLP) model was used to analyze pharmaceutical care records. The records of 90651 patients (≥65 y of age) who had prescriptions filled at 58 community pharmacy branches in Japan between January and December 2022 were analyzed. Patients without the home-based patient dispensing fee (hereafter referred to as outpatients) and patients receiving home-based care were included in the study. An existing NLP model (MedNER-CR-JA) was adapted to extract the symptoms from unstructured records. Two pharmacists annotated 500 randomly selected records to evaluate NLP model performance in the target records. The NLP model extracted present symptom tags with F1-scores of 0.75 for outpatients and 0.73 for patients receiving home-based care. A total of 90162 present symptom tags were identified. Of these, “constipation,” “frequent urination,” “falls,” “insomnia,” and “delirium” accounted for a higher proportion of present symptom tags in patients receiving home-based care, whereas “dizziness” and “drowsiness” accounted for a higher proportion of present symptom tags in outpatients. Mentions of “cognitive decline” and “depression” were limited in both patient groups. NLP applied to pharmaceutical care records demonstrated that community pharmacists routinely document symptoms relevant to geriatric syndromes in both outpatient and home-based care settings. These records inherently contain clinically meaningful information on symptoms relevant to geriatric syndromes observed in routine practice and represent valuable real-world data that can support geriatric syndrome-focused evaluation during medication reviews.