A knowledge graph-driven paradigm for holistic symptom management: development and evaluation of a nurse-led KG-QA system in community palliative care.

Abstract

Background: Community nurses play a pivotal role in palliative care but face barriers in managing complex symptoms, such as fragmented knowledge and a lack of community-tailored evidence-based guidance, impairing clinical efficiency. The aim of this study was to develop and evaluate a knowledge graph-based question-answering system for symptom management in community palliative care.

Methods: A three-phase codesign study guided by the Knowledge-to-Action framework was conducted. Phase 1 (Knowledge Creation): A Symptom Management Knowledge Base (Knowledge Product I) was developed through a codesign process involving a multidisciplinary expert panel. This panel adapted a knowledge base created by researchers through systematic evidence synthesis, employing FAME criteria for contextual adaptation. Phase 2 (Action Cycle: Implementation): A semantically structured knowledge graph (Knowledge Product II) was constructed via automated extraction by software developers, followed by manual verification by researchers. Based on this graph, a question-answering system was created and implemented as a WeChat mini-program, resulting in a practical KG-QA system (Knowledge Product III). Phase 3 (Action Cycle: Evaluation): The system's acceptability, usability, and perceived usefulness and ease of use were assessed among experts and community nurses during a two-week evaluation period using the Clinical Nursing Information System Effectiveness Evaluation Scale and the Post-Study System Usability Questionnaire, which is grounded in the Technology Acceptance Model.

Results: The knowledge base comprises 225 evidence items for nine symptoms; the knowledge graph integrates ten entity types, 11 relationship categories, 442 entities and 668 relationships, with the system supporting four query interfaces and three search methods. The evaluations demonstrated high perceived usefulness and ease of use, with strong scores for acceptability (102.25 ± 16.21; 110.56 ± 9.90) and usability (2.47 ± 1.98; 2.23 ± 1.93).

Conclusion: The question-answering system bridges the evidence-practice gap via a nursing-process paradigm, offering a potentially scalable model that aligns with national policies pending further validation. However, these findings are based on a small‑scale, single‑region, short‑term evaluation relying largely on subjective measures. Future research should explore its long-term clinical outcomes and cross-setting scalability.