Smartphone-based AI Triage for Acute Nursing Care in Community Settings

Thesis Abstract

The rapid proliferation of smartphone-enabled health tools and rising demand for timely, remote triage in community settings have exposed gaps in access, consistency, and safety of acute nursing assessment, particularly in underserved regions where nurse-led guidance can prevent unnecessary ER visits. This study aims to develop, validate, and evaluate a smartphone-based artificial intelligence (AI) triage application designed to support community nurses and lay users in acute care decision-making, with specific objectives to (a) assess the diagnostic accuracy and triage safety of the AI algorithm against standard nurse assessments, (b) evaluate user acceptance, trust, and perceived ease of use among nurses and non-professional caregivers, and (c) examine the impact of the tool on referral rates, wait times, and patient outcomes within a six-month community implementation. A mixed-methods design was employed, commencing with a technological feasibility phase that used a prospective diagnostic accuracy study involving 1,200 simulated and real-world cases across urban and rural primary care settings to calibrate the AI model against gold standard clinician assessments and national triage guidelines. The second phase implemented a pragmatic quasi-experimental study with 400 participants (nurses and lay users) assigned to intervention and control groups to measure outcomes over 180 days, complemented by qualitative interviews with 40 participants and three focus groups to explore contextual factors influencing adoption. Data collection instruments included a validated AI-augmented triage questionnaire, Clinician Global Impression scales, the Technology Acceptance Model (TAM) measures, System Usability Scale (SUS), and patient outcome records (referral disposition, wait times, hospital admission, and 30-day readmission). Analytical approaches integrated diagnostic accuracy statistics (sensitivity, specificity, positive and negative predictive values), receiver operating characteristic (ROC) analysis, regression models to identify predictors of triage accuracy and user acceptance, ANOVA to compare outcomes between groups, and thematic analysis of qualitative data using an a priori coding framework aligned with modified TAM constructs and the Unified Theory of Acceptance and Use of Technology. The theoretical foundations draw on the Agenda-Setting and Diffusion of Innovations theories to interpret adoption dynamics, alongside the Explainable AI (XAI) framework to ensure transparent decision support for clinicians and users. Expected findings include high sensitivity (?0.88) and acceptable specificity (?0.75) in acute triage decisions, strong usability scores (SUS ? 70) and positive TAM indicators, reduced inappropriate referrals by at least 15% in the intervention group, and shorter average wait times by 22% without compromising patient safety. The study anticipates identifying user-specific factors—such as digital literacy, perceived risk, and perceived usefulness—that mediate trust and adherence to AI-generated triage advice, as well as contextual barriers like data privacy concerns and connectivity in rural areas. The contribution to knowledge encompasses (1) evidence on the feasibility, safety, and effectiveness of AI-augmented triage in real-world community settings, (2) a validated multi-site AI triage model calibrated to diverse populations, and (3) an empirically grounded framework for implementing smartphone-based triage tools within nursing practice and home-care contexts. The main conclusion is that a rigorously designed smartphone-based AI triage tool can augment acute nursing decision-making in community settings, improving timeliness and consistency of care while maintaining safety, provided that comprehensive training, ongoing monitoring, transparent explainability, and robust data governance are maintained. Recommendations include integrating the tool into national nursing practice guidelines, establishing continuous post-implementation audit mechanisms, prioritizing user-centered refinements to enhance trust and usability, and conducting longitudinal studies to assess long-term patient outcomes, health service utilization, and cost-effectiveness.