Usability and acceptability of an LLM-Enabled collaborative care app for home-based depression support: A multi-stakeholder mixed-methods study

Objective Large language model (LLM)-enabled applications may support emotional expression and care coordination in home-based depression care, but their usability and acceptability across different stakeholders remain insufficiently understood. This study aimed to evaluate the usability and acceptability of CloudEcho from the perspectives of patients, family caregivers, and clinical experts in collaborative home-based depression care, with particular attention to system use, information sharing, role boundaries, and care coordination. Methods A mixed-methods study was conducted with 30 participants in China, comprising 10 patients, 10 family caregivers, and 10 clinical experts. Participants completed the System Usability Scale (SUS) and participated in semi-structured interviews. Quantitative data were analysed descriptively, and interview data were examined using framework-guided thematic analysis. The study focused on stakeholder perceptions of CloudEcho rather than on observed communication or coordination outcomes within matched patient–caregiver dyads. Results CloudEcho achieved an overall mean SUS score of 77.50, with mean scores of 83.25 among patients, 79.25 among family caregivers, and 70.00 among clinical experts. Qualitative findings highlighted the importance of maintaining a bounded chatbot role, preserving selective control over information visibility, and establishing clear safety and responsibility boundaries. Stakeholder priorities differed by role: patients emphasized low-burden expression and emotional comfort, family caregivers focused on care awareness and everyday coordination, and clinical experts placed greater emphasis on privacy, crisis-related safeguards, information reliability, and professional responsibility. Conclusions CloudEcho was generally perceived as usable across the three stakeholder groups, although evaluations reflected distinct care roles and responsibilities. The findings suggest that LLM-enabled tools may be most appropriately positioned as bounded intermediaries that support emotional expression and care coordination rather than as autonomous therapeutic or clinical agents. Clear controls over information visibility, safety boundaries, and human responsibility are therefore important considerations for the design of LLM-enabled support in home-based depression care.