21 of 58 invited experts took part in the online workshop, 53 experts completed the online survey and the perspectives of 14 experts were included in the interviews. Table 1 shows the number of participants per stakeholder group for the online workshop and expert interviews. Most participants of the online survey (n = 28, 52.8%) belonged to a nursing care related area of expertise such as nursing practice, management or education. 16 participants (34.0%) reported belonging to another area of expertise. The majority of participants of the online survey (n = 39, 73.6%) had not previously taken part in the online workshop and 25 participants (47.2%) reported experience in research projects on AI in nursing practice.
Table 1
Number of participants of the online workshop and expert interview per stakeholder group.
| Number of participants |
Status group | Online workshop | Expert interviews |
Management/DoNa/Provider of Home Care | 2 | |
Management/DoN/Provider of Nursing Home Care | 1 |
Management/DoN/Provider of Daycare | 1 |
Management/DoN/Provider of Hospital Care | 2 |
Informal Caregiver | 2 |
Digitization Officer/ Digitization in Nursing Care | 3 |
Registered Nurseb | 1 |
Nursing education and Nursing science | 4 | 5 |
AI research and development | 5 | 6 |
Ethics and Legal research | -- | 3 |
Total | 21 | 14 |
a) DoN = Director of Nursing. b) Five nurses predominantly working in direct patient care where invited to the workshop of which one participated. The overall sample contains more than one registered nurse as individuals in the groups Management/DoN/Provider and Nursing education and Nursing science also held a nursing degree. |
Needs and application scenarios for AI in nursing care
Participants of the online workshop predominantly brought up need and application scenarios for AI targeting the micro level of nursing care. Recurring considerations entailed specific actions and decisions to be executed by nurses during the care process, such as assessment of care needs, selection of suitable and effective interventions, and monitoring and evaluation of health status and outcomes. Informal caregivers emphasized needs and application scenarios in counselling, quality of life and experience and prevention of stressors. Participants from care organizations expressed the need for intelligent staff mix models. Further, application scenarios for planning of care services and route planning in home care were highlighted. Table 2 lists topics that were highlighted as needs as well as promising application scenarios for AI in nursing care and anchor-citations.
Table 2
Needs and promising application scenarios for AI in nursing practice emphasized during the online workshop.
Need and promising application scenario | Anchor-citation |
Care needs assessment and risk assessment | "The fact that I continuously use a certain system that records my everyday life, that then registers when there are deviations from this everyday life. Or when I think of movement sensors in the apartment, when the system suddenly notices that I get up more often at night to go to the toilet than I used to, that the system then realizes that this could be a sign of incontinence or a disturbed day-night rhythm […]."[Management/DoN/Provider of Home Care] |
Care planning in complex care situations | "That AI simply supports me in finding adequate interventions, especially for very complex care cases or very multimorbid people. There are also interactions sometimes […]”[Management/DoN/Provider of Hospital Care] “Assessment or nursing assessment and transfer into nursing care planning. Where [...] individual complex situations [...] are covered. Not just pure if-then functions.” [Management/DoN/Provider of Hospital Care] |
Evidence-based decision support | "Keyword evidence-based decision support [...]. I want to emphasize this again because it is something very concrete that could be implemented directly. The databases are there, the preparation of the data, the knowledge access, that is missing at this point. Quick decision-making tools to really do better." [Nursing Education and Science] "The most ideal [...] would of course be an AI on site at the bedside. That I can also use exactly this knowledge on site. [...] The variant I would have a voice control and interact verbally with the AI.” [Management/DoN/Provider of Hospital Care] |
Medication management | “The [...] medication management. How much are people taking of what, are they actually doing it, what is being taken [...] that would be a nice topic for automation." [AI research and development] |
Physical and cognitive activation | "Activation. Both physical and cognitive I can imagine very well. [Example Tovertafel]. It would be great to have such tools also generated at home and also everything that has to do with training at home, with solidification of skills or improvements.” [Informal Caregiver] |
Route planning | “That both traffic reports, the supply wishes of the customers, the available employees and their qualifications, etc. are all taken into account and that a finished tour is proposed […]. I think this is very important to support the care managers [...] and [...] to free up more time for other activities. [...] If you do more research in that area and develop that further, I think that can be a very useful support." [Management/DoN/Provider of Home Care] |
Nurse rostering, staff mix | „How do we get service planning on the one hand and the planning of concrete nursing care with staff deployment – and not so much with the question of how many personnel are to be deployed, because there are corresponding [legal] specifications – but with the question of how this is to be distributed qualitatively and quantitatively. [...] In a correlation of content, between the staff deployment, with all its facets, qualification and such things, and the concrete activities. So also, for example, such a question: Doesn't the care of Mrs. Klarenbach in a cognitively restricted scenario actually have to be allowed to last longer than, so to speak, usually?” [AI research and development] |
Counselling of informal care givers and patients | "I can think of care counselling, which would also be suitable, at least initially, for AI use. [...] AI could first provide support upstream to give orientation and sounding out and then guide you directly to the help that you need and that will then help you further." [Management/DoN/Provider of Home Care] |
Quality of life and stressors of informal caregivers | "Also, to collect specific needs of the relatives. What are the needs of those standing at the bedside? To call up and recognize the quality of life and well-being based on certain parameters, whether it's through digitalized image processing or observation modules [...] also for people with cognitive impairments.” [Informal Caregiver[ |
Education, mediation of knowledge and competencies (nurses, other health professionals, informal care givers) | “For example, how to qualify nursing assistants to take on more tasks at a distance, even if they don't have the expertise. Here, AI could provide support, for example, with the help of [smart glasses] that then impart knowledge, transmit what the nursing assistant sees and can also evaluate it directly and then give her a recommendation for action. If the nursing assistant looks at a wound, that the AI can then convert what it sees and say I now recommend this and that care.” [Management/DoN/Provider of Home Care] "The concrete ideas […] I can think of are education and communication. [...] Also information sharing. What is where? That's what's wrong in some cases, that people on the ground don't know who is where, who can provide what support. Education [...] I could very well imagine that there are certain care topics that perhaps generate automatically, so that [caregiving relatives notice] oh there's a problem right now. [...] big issues, what about even if...for example, food refusal, for example, pain conditions." [Informal Caregiver] |
DoN = Director of Nursing |
Table 2 Needs and promising application scenarios for AI in nursing practice emphasized during the online workshop.
Some of the nursing scientists taking part in the expert interviews regarded AI as more suitable to support less complex decisions in care. They rated the nature of the overall nursing process as too complex and multidimensional to be managed by AI. However, the potential of AI to compile and provide information and thus ultimately initiate a decision was repeatedly emphasized. The use of AI in home care to support independent living for people in need of care and to detect risks or emergency situations was also mentioned. Staff shortages in long-term care and hospital settings and the respective challenges arising from a more heterogeneous staff and skill mix were named as reasons for needs. The transfer of clinical guidelines into AI applications for monitoring care provision and outcomes could support staff and skill mix as well as interdisciplinary and intersectoral collaboration. In this context, a high potential for the use of AI was described by linking different data sources while taking into account the expertise of the professional groups involved in healthcare.
“How can we support the different competencies that we have in nursing, that we have in the qualification mix, how can we support the qualification mix, so to speak, with assistive technologies in order to provide the right information at the right time in the right place?” (Nursing Scientist)
Experts from AI R&D rated the daily interactions occurring in nursing care with a few restrictions as too complex to be supported by an omni-purpose AI. However, they pointed out that complexity must be considered depending on the respective situation and the specific research question. For the detection of situations in which no model or theory is yet available, for example in the case of multimorbidity, they named deep-learning systems as suitable for generating knowledge and understanding the complexity of these situations and, in the course of this, for developing rule-based systems based on this knowledge. Further, the amount of data available influences the degree of complexity and ultimately the methodological decision for or against the use of ML.
"These deep learning systems [...], you can use them as tools to try to understand complex issues for which you don't yet have a model, for which you don't yet have a theory. […] This could be something like these multimorbidity situations. Complex situations where you have to make decisions under complex circumstances that can be very different. […]" (AI research and development)
Experts from AI R&D commented on what they considered to be important conditions or information needed for needs-driven development of AI applications. They pointed out specific challenges, for instance when patients may behave differently because of unconsidered medication effects or participate less frequently in research than would be necessary for an algorithm to be trained. In order to avoid mis- or underrepresentation of individual socio-demographic characteristics in the data, they pointed out the need to include as heterogeneous groups of people as possible to obtain a suitable data base. In this context, they also mentioned data protection as a hurdle that would come into play in particular if individual care situations were to be mapped. Further, a lack of standardized comparable data for nursing care complicates the further development of learning systems.
Experts from ethics and legal research pointed towards general challenges associated with the automation of decisions. For example, the process of automation through the use of AI creates dangers, regardless of the type of AI system. For example, discrimination would not be recognized, the technology would be considered neutral, and responsibility would be handed over to technology. This way of dealing with technology favors intransparency, people are deprived of the possibility to object, and the evaluation of complex amounts of data may pave the way for new kinds of abuse of data or information. These dangers may be particularly pronounced in nursing, but they do not represent nursing-specific characteristics. Depending on the context in which an AI system is used in nursing and the degree of human involvement, different risks arise and different solutions are needed to guarantee an ethical use of AI. If humans represent an intermediate instance in the system or if the system is in direct contact with the person in need of care and interacts with them, a differentiated discussion and decision on the ethical design is required in each case. Depending on the application situation, a human intermediary between the AI system and the action becomes more important, as does the qualification of the nursing staff, who represent such an intermediary.
"I think what's important here is that the risks are different depending on whether it's a system where a human being acts as an intermediary. So if it's a system that, for example, makes a prediction or a diagnosis that is then used by a nurse, a caregiver, a doctor, then it's a different situation than a system that interacts directly with the person concerned. And the approaches we need to guarantee ethical use are different depending on whether there is a human intermediary or whether it is a system that is in direct contact with the people concerned. In any case, a differentiated discussion is needed regarding the ethical design, because topics such as traceability, complaints, and responsibility must be organized and implemented in a completely different way in these two cases.” [Ethics and legal research]
Priorities and rating of scenarios
Participants of the online survey were also asked to rank the ten needs and application scenarios derived from the online workshop, as they thought they should be prioritized in research funding. Table 3 shows the results of said ranking with rank 1 representing the highest priority and rank 10 representing the lowest priority. The topic of care needs assessment and risk assessment was ranked particularly high, with a median rank of 3 and an average rank of 3.6. The topics evidence-based decision support, care planning in complex care situations and nurse rostering were rated in the median 4th place while physical and cognitive activation and the quality of life and stressor of informal caregivers were ranked at the bottom of the list.
Table 3
Ranking of topics for research funding (rank 1 (highest) to rank 10 (lowest)), online survey.
Topic | Mean rank | Median rank |
Care needs assessment and risk assessment | 3.60 | 3 |
Evidence-based decision support (e.g. selection of interventions, consultation of other health professionals) | 4.26 | 4 |
Care planning in complex care situations | 4.28 | 4 |
Nurse rostering, staff mix | 4.42 | 4 |
Education, mediation of knowledge and competencies | 5.34 | 5 |
Route planning (in home care and inpatient care) | 5.51 | 7 |
Medication management | 5.57 | 6 |
Counselling of informal care givers and patients | 5.79 | 6 |
Physical and cognitive activation | 6.25 | 7 |
Quality of life and stressors of informal caregivers | 6.26 | 7 |
Participants were asked to evaluate hypothetical AI systems (without further specification of the system design) for the ten needs and application scenarios with regard to expected benefit for nursing professionals and nursing assistants, for care patients, for informal caregivers and the technical feasibility at the time of the study. The assessment was made on a scale from 0 (no expected benefit or impossible feasibility) to 10 (very high benefit or very good feasibility). For Fig. 3, the three assessments of the expected benefits were combined into a mean value and plotted against the assessment of technical feasibility. Participants expected AI systems to have a medium to high benefit in all ten needs and application scenarios, as well as medium to good feasibility. Route planning can be considered the application scenario with the highest average benefit and at the same time the best technical feasibility. AI Systems aiming at the quality of life of informal caregivers received the lowest ratings in both aspects.
Participants were asked to indicate for each application scenario whether AI systems were relevant for one or more settings or equally relevant for all settings (Table 4). AI solutions for the application scenarios of medication management, care needs assessment and risk assessment and education and mediation of knowledge and competencies were considered relevant for all settings by the majority of respondents. Solutions related to counselling, care planning in complex care situations, physical and cognitive activation, and evidence-based decision support were also named as relevant for all settings by a large proportion of participants, but also received differentiated assessments. For example, solutions for care planning in complex care situations are primarily named as relevant in settings with the participation of professional caregivers, while for counselling, a particular relevance in home care by relatives is emphasized.
Table 4
Particularly relevant nursing settings for key applications of AI: Proportion of people who find the respective setting relevant for the focus of application (multiple answers possible).
| Setting |
Key applications | All | Home care | Informal care giving | Inpatient long-term care | Daycare | No answer |
Education, mediation of knowledge and competencies | 69.81% | 5.66% | 16.98% | 7.55% | 5.66% | 1.89% |
Medication management | 64.15% | 18.87% | 9.43% | 22.64% | 5.66% | 1.89% |
Care needs assessment and risk assessment | 62.26% | 26.42% | 7.55% | 26.42% | 13.21% | 1.89% |
Physical and cognitive activation | 49.06% | 7.55% | 18.87% | 16.98% | 16.98% | 11.32% |
Evidence-based decision support | 47.17% | 16.98% | 18.87% | 13.21% | 3.77% | 3.77% |
Counselling | 45.28% | 15.09% | 37.74% | 7.55% | 7.55% | 3.77% |
Care planning in complex care situations | 45.28% | 28.30% | 11.32% | 37.74% | 20.75% | 3.77% |
Nurse rostering, staff mix | 30.19% | 52.83% | 1.89% | 50.94% | 35.85% | 0.00% |
Quality of life and stressors of informal caregivers | 18.87% | 18.87% | 69.81% | 0.00% | 15.09% | 5.66% |
Route planning | 9.43% | 81.13% | 0.00% | 16.98% | 18.87% | 1.89% |
Requirements and characteristics of successful research projects
Figure 4 shows the categories that summarize requirements and characteristics of successful research projects from the perspective of the stakeholders included in the study. Next to regulatory, processual and technological requirements, success criteria include ethical and legal aspects and supportive communities and eco-systems as well as the inclusion or reflection of existing frameworks and instruments for the development and implementation of technologies and interventions in nursing and health care, such as the NASSS (21) or RE-AIM (32, 33) framework.
Regulatory requirements were concerned with the analysis of the data and models of data sharing in compliance with data protection regulations. As R&D projects often include personal data for which informed consent is mandatory, possibilities to guarantee a fundamental use of the data should already be assessed during the planning phase of a project. This includes the consideration of technical measures to comply with data protection (e.g., anonymization, pseudonymization, or synthetization) and particular challenges arising when applying ML components that learn from personal data, such as clarification on how to proceed when data subjects withdraw consent to data processing. The experts additionally recommended to always consult an ethics committee and obtain an ethical clearance if this is not already mandatory by other regulations. If applicable, the study should then be registered in a study registry.
Processual and translational requirements comprise of aspects concerning project planning, execution and management. While R&D projects were considered as mostly being driven by AI researchers, the role of application partners and care facilities was of high relevance. If the application or evaluation of an AI solution is to be tested in practice, care facilities or institutions should be planned as permanent consortium members. The inclusion of several nursing homes or home care services, for which personnel resources to accompany the project should be available, was deemed as desirable to prevent the threat of temporary delays or failure of the project implementation due to shortcomings and staff shortages. Close involvement of the application partners from the very beginning of the project planning supports a needs- or practice-driven approach to R&D that focuses on the needs of the target group and the expected benefits for nursing practice. Further, the involvement of nursing scientists could – next to providing scientific support – also act as a mediator between nursing practice and developers. Choosing and rigorously implementing a participatory design approach supports to focus on specific care scenarios, the emergence and significance of available data, the definition of expected benefits and target groups and the incorporation of underlying conditions and mechanisms of the health care system, for which usual care pathways and interfaces to other institutions and sectors should also be taken into account.
While requirements for the technical infrastructure and the discussion of technical implementation details are part of every R&D project, the experts pointed out that specifications should be made in advance and the existing infrastructure in participating care facilities should be evaluated during the projects’ planning phase. If data from electronic nursing or health records or other documentation systems are to be used, the existing structure of the data, their continuity, quantity, and quality must be taken into account. Classification systems such as the European Nursing Care Pathways (ENP), the International Classification of Nursing Practice (ICNP) or the Systematized Nomenclature of Medicine (SNOMED), with structured data on nursing and medical diagnoses and measures, offer significantly simpler evaluation options than freetext documentation, but have so far found only limited use in R&D projects. The implementation of standards, especially in the area of interoperability, was considered to be very important for successful product development in the long term. The use of open protocols and data exchange formats enables integration into existing or future infrastructures. This is particularly relevant in regard to the frequently formulated requirement of cross-sector data exchange; a lack of integration of stand-alone solutions into existing systems within the care facility was perceived as disadvantageous and can thus represent an obstacle to market introduction. In this context, the technical aspects of data protection and data security should also be addressed as early as the planning phase of R&D projects.
Considering social and ethical aspects, the experts pointed out, that, next to incorporating lessons-learned from previous projects to identify ethical and social challenges that might affect a new project, research activities supporting the identification of ethical or social questions that arise among the actors within the research field involved and addresses them with regard to technology development and use, should be incorporated in the project. Recurring ethical issues concerned undesirable effects, such as the creation of new inequalities through new technologies in care and the representativeness of the persons with whom a new technology is tested and of the data from which an AI system learns. While the dimensions of gender and skin color in particular are prominently discussed in general AI research here, nursing-specific dimensions are largely unknown to date. For nurses, the comprehensibility and explainability of AI was highlighted as particularly important in decision support. The concept of human-in-the-loop as the involvement of humans in the decision-making and implementation process of algorithmic decisions and ultimately their option not to implement or act on the suggestion of an AI system, was considered essential for nursing.
In addition to project-specific considerations, structures for the transfer of knowledge and the networking of actors and projects in the thematic field should be created. By implementing communities and eco-systems that go beyond the boundaries of individual projects, translation and dissemination of practical findings that may contribute to nursing education and the scientific community could be supported. If open-source program code is made available in online communities, existing programs or elements of them can not only be used by other developers but can also be extended or improved by them as part of their work. By using internationally recognized software libraries and programming languages, developers may prevent the development of niche solutions. As civil society initiatives also call for the fundamental disclosure of program code from publicly funded projects, a care specific platform would not only promote interoperability and provide targeted solutions for recurring problems but could also support public involvement and transparency.