AI decision support for symptom management in hursing homes: Usage framework and requirements for AI-supported non-urgent symptom management

Abstract

Norwegian nursing homes face an increased work burden due to population ageing and insufficient staffing. Frontline staff manage frequent non-urgent symptom changes and decisions about assessment, follow-up, handover, and escalation. The thesis examined whether and how an AI system could support the non-urgent symptom change management and escalation. A scoping review and qualitative study design were employed. For the qualitative study, data were collected in Stavanger, Norway, at three nursing homes using individual interviews (n = 15) and one focus group that used a case-based discussion and a storyboard/prototype walkthrough and were analysed using thematic analysis. Results show the decision-making experience was described as a process rather than a single decision, with varying support needs across different roles. Concise assessment support and easily communicable guidance, as well as criteria for escalation, were important for healthcare professionals, whereas prioritisation, concise summaries, and next steps in accordance with clinical roles were important for nurses. The physician perspective highlighted the need for high-quality information from the bedside healthcare professionals and trend summaries, recognising that there is rarely a single generic answer for complex residents. The study concludes that for an AI CDSS to be safe and usable, it must provide "actionable explainability", fit into the workflow, be fast, integrated with the Electronic Patient Journal (EPJ), provide role-specific outputs, and consider residents' baselines. Finally, it requires clear governance to ensure AI serves as a "second opinion" under human clinical accountability.