A newly developed digital health education platform combining an evidence-based knowledge graph with an intelligent question-answering system demonstrated favorable usability and preliminary functional performance among healthcare staff and hospitalized individuals. Chronic rhinosinusitis affects between 4.8 percent and 12.0 percent of the global population, significantly impairing patient quality of life. While patient education is recognized as an important component of managing the condition, traditional educational approaches remain constrained by limited time among medical staff, and existing online health information often suffers from uneven quality.
To address these challenges, researchers designed a structured, evidence-based chronic rhinosinusitis health education knowledge base. The development process involved comprehensive literature retrieval, methodological quality appraisal, evidence synthesis, and structured expert consultation. Following the synthesis of evidence, the research team manually annotated the finalized educational content and converted it into machine-readable semantic triples. To support user interaction and conceptual clarity, the resulting knowledge graph was visually rendered using the NRD Studio platform, establishing a network comprising 287 distinct nodes, 42 relationship types, and 419 inter-node links.
The system’s conversational component was built using the Coze platform, which integrated the structured knowledge base to execute retrieval-augmented response generation. To ensure practical mobile accessibility, the entire architecture was deployed through a dedicated WeChat official account. This mobile channel served as a unified access point where end users could directly engage with the intelligent question-answering interface while also browsing the visual representations of the underlying knowledge graph.
Prior to clinical user testing, the system underwent functional evaluation on a development set of 30 standardized questions across six core domains. During initial trials, seven of the 30 questions revealed issues concerning incomplete information or inappropriate wording. The investigative team refined the knowledge-base entries and system prompts to correct these deficiencies, subsequently retesting the complete set of questions two days later under independent evaluation by clinical nurses.
During the retest evaluation, two independent clinical evaluators assessed the responses for content consistency, a study-specific safety criterion, and the appropriate refusal of out-of-scope questions. Both evaluators assigned scores of 1 for content consistency and study-specific safety across all 25 in-scope questions, while all five out-of-scope inquiries were confirmed to be appropriately refused. Inter-rater agreement reached 100 percent, yielding a prevalence-adjusted and bias-adjusted kappa value of 1.00, though Cohen’s kappa could not be statistically estimated due to identical rating outcomes across both evaluators.
Real-world usability was subsequently assessed by recruiting a convenience sample of 15 clinical nurses and 15 hospitalized patients diagnosed with chronic rhinosinusitis, all of whom interacted with the WeChat-based system over a one-week period. Standardized evaluation via the System Usability Scale yielded a mean usability score of 90.33 ± 7.31 with a 95 percent confidence interval of 86.28 to 94.38 among participating nurses. Hospitalized patients recorded a mean System Usability Scale score of 85.00 ± 4.91 with a 95 percent confidence interval of 82.29 to 87.72. In addition, patients completed a health education satisfaction questionnaire, reporting a mean satisfaction score of 68.53 ± 2.42 with a 95 percent confidence interval of 67.19 to 69.87 out of a maximum possible score of 72.
The authors emphasized that while the knowledge graph–supported question-answering tool demonstrated strong usability and favorable preliminary performance, the functional testing was conducted on a developmental question set rather than an independent validation cohort. Consequently, these initial scores do not establish independent clinical safety, validate subsequently revised reference answers, or define generalizable accuracy rates. The platform may serve as a supportive digital supplement to hospital-based education, but comprehensive validation across independent test sets remains essential before making broader determinations regarding its clinical safety and instructional efficacy.
The research was conducted through an academic collaboration involving investigators from Chengde Medical University and its affiliated hospital, The Second Affiliated Hospital of Xuzhou Medical University, and Shijiazhuang People’s Medical College in China, alongside the University of Manchester in the United Kingdom. Published in Frontiers in Public Health under its digital public health section, the initiative was developed to explore whether combining structured knowledge graphs with conversational artificial intelligence could alleviate the educational workload placed on clinical personnel while mitigating the risks posed by uneven online health advice.
Methodologically, the authors structured the project around a formative development workflow rather than a summative trial. Because the second round of functional testing took place on the exact same 30-question development set only two days after addressing the seven identified wording and completeness deficiencies, the resulting ratings reflect the immediate effectiveness of targeted prompt and knowledge-base adjustments. The investigators explicitly cautioned that these development-set outcomes should not be interpreted as establishing a generalizable rate of accuracy or confirming clinical safety across uncurated inquiries.
Furthermore, the findings emphasize the system’s intended role as an auxiliary educational instrument rather than a standalone clinical resource. Although the mobile deployment over WeChat achieved favorable usability ratings from the convenience sample of hospitalized patients and clinical nurses during the one-week evaluation, broader adoption hinges on future prospective validation. The researchers reiterated that establishing genuine clinical safety and instructional dependability will require rigorous testing on diverse, independent test sets outside the original developmental parameters.
Subject of Research: Usability and preliminary functional evaluation of an evidence-based knowledge graph question-answering system for chronic rhinosinusitis health education
Article Title: Development and preliminary usability evaluation of an evidence-based knowledge graph–supported intelligent question-answering system for chronic rhinosinusitis health education
Article References: Wang, W., Zhang, Z., Li, X., Zhang, Y., Wang, N., Shi, M., & Zhang, X. (2026). Development and preliminary usability evaluation of an evidence-based knowledge graph–supported intelligent question-answering system for chronic rhinosinusitis health education. Frontiers in Public Health, 14, Article 1988483. https://doi.org/10.3389/fpubh.2026.1988483
Image Credits: AI Generated
DOI: 10.3389/fpubh.2026.1988483
Keywords: Chronic rhinosinusitis, Knowledge graph, Intelligent question-answering, Health education, Digital health, Retrieval-augmented generation, System Usability Scale, Patient satisfaction, Clinical nursing, Hospitalized patients, WeChat Official Account, NRD Studio, Coze platform, Content consistency, Evidence synthesis, Development set testing
News Source: Courtney Benton. (November 12, 2026). Knowledge Graph AI System Shows High Usability for Chronic Rhinosinusitis Education. Scienmag.
