Objective
To develop a generative AI-driven nursing risk assessment model, validate its efficacy in identifying risks (including pressure injury, falls, and malnutrition) among elderly patients with chronic diseases, evaluate its role in enhancing the accuracy of risk assessment and generating personalized early warning and preventive recommendations, and explore its value in clinical decision support.
Methods
A prospective mixed-methods study was conducted. Phase 1: A large language model was fine-tuned on multi-source data from hospital electronic health records (EHRs) to develop a dynamic risk-prediction and personalized recommendation-generation system. Phase 2: A non-randomized controlled trial was implemented to compare outcomes between the intervention group (with AI-generated recommendations integrated into care) and the control group (receiving usual care). The effectiveness and applicability of the system were comprehensively evaluated via quantitative metrics and qualitative analysis of nurse interviews. Additionally, semi-structured interviews were conducted with nurses who used the system, and thematic analysis was employed to explore their user experiences, perceived usefulness, ease of use, and barriers and facilitators to clinical integration.
Results
The generative AI model is expected to yield a higher area under the receiver operating characteristic curve (AUC-ROC) for risk prediction compared to traditional assessment scales. Moreover, the intervention group is anticipated to have a lower incidence of adverse events (e.g., falls, pressure injuries). Qualitative analysis revealed core themes, including "improved assessment efficiency" and "balance between human and machine decision-making".
Conclusion
Generative AI enables precise and prospective assessment of nursing risks in elderly patients with chronic diseases. The personalized intervention recommendations it generates can serve as an efficient decision-support tool, thereby reducing the incidence of nursing-related adverse events.
