Procedure and Materials

All interviews were conducted digitally via video calls, with screen sharing used to provide visual support for the interview questions. The interviews were conducted between May 2022 and February 2023, with a mean duration of 79 (range 58-119) minutes. Of the 24 interviews, 17 (71%) were conducted by pairs of researchers, and 7 (29%) were conducted by a single researcher. A multidisciplinary group of researchers (DRML, NES, SIA, HHN, WPCB, and AP) developed the interview protocol. Minor adaptations were made to the protocol after pilot testing with the first 2 participants. The interviews were conducted in Dutch.

The interview protocol (Multimedia Appendix 1) was structured as follows. In the first part of the interviews, a general introduction was given about the AI-DSS developed in the HAAL project. This concerns a dashboard that acquires, presents, and uses data generated by various digital care and well-being technologies that can be deployed in the homes of older adults. When used, these technologies collect data on the physical activity, eating and sleeping patterns, cognitive functioning, mood, social contact, and medication intake of older adults. All technologies were explained and shown to participants using a visual illustration, and questions were asked about the perceived relevance of and the familiarity of participants with the various technologies and data.

In the second part, participants were invited to reflect on the opportunities of AI-assisted decision-making in LTC. A description and visual illustration were provided, and questions related to the 2 developed imaginary scenarios were asked. The first scenario describes a dashboard with descriptive analytical functions only. The dashboard provides an overview of the data collected over time via a tailored selection of digital care and well-being technologies. In the dashboard, specific collected data are marked by a color (red, orange, or green) to signify varying levels of risk or urgency associated with them. Apart from the application of this coloring scheme, the data are not interpreted by algorithms. The primary goal of this dashboard is to make the data generated by various technologies available to caregivers in one place to prevent them from looking at separate overviews and apps.

The second scenario describes a more advanced dashboard with descriptive, predictive, and prescriptive functions. In this scenario, the data generated by the selected care technologies are not only integrated into one system and color marked to signify risk levels but also automatically processed into actionable insights by algorithms. Actionable insights could entail predictions of the risk for future emergency situations, such as a fall, and recommendations about possible follow-up actions, such as stimulating the physical activity of a client if the data indicate a relatively inactive period.

Both scenarios left room for the interview participants to indicate whether, and for which types of caregivers and other stakeholders in LTC, the respective dashboard might be relevant and why. After questions in this regard, participants were asked which of the 2 dashboards they would prefer and why. In addition, a short explanation was provided about the term AI, including everyday examples, after which the participants were asked what role they hope AI will play in the future of LTC.

In the third part, the participants were asked about the risks related to the use of AI-DSSs in LTC, as well as mitigation strategies. Participants were first invited to openly discuss any risks or concerns linked to both scenarios and consider whether they perceived any explicit differences in the risks associated with more advanced AI-DSSs compared to low-complexity AI-DSSs. Subsequently, targeted questions about risks were asked by using the 6 probing themes based on the responsible AI principles from the WHO [43]. After a brief explanation of each principle, participants were asked about their views on the respective principle in the context of AI-assisted decision-making in LTC. During discussions of potential risks, participants were encouraged to brainstorm possible strategies and actions to address or mitigate these risks in the design, use, and implementation of AI-DSSs. Finally, the participants were asked whether they had any other suggestions or topics that they wanted to discuss regarding the implications of using AI-DSSs in LTC.