We included 15 health care professionals involved in the inpatient surgical care of geriatric patients. Three senior physicians, 3 assistant physicians, and 3 nurses from the departments of urology, visceral surgery, and traumatology (all 3 departments are recruitment centers for the SURGE-Ahead project), as well as 2 senior physicians from the emergency department, 2 physiotherapists, and 2 social service workers, all responsible for surgical patients, agreed to participate. None of the interviewees were involved in the SURGE-Ahead project or in the development of the AI-CDSS.

Participants were recruited at Ulm University Medical Center in Southern Germany. The department of orthopedic trauma surgery has been certified as an AltersTraumaZentrum (center for geriatric traumatology) DGU in cooperation with a local geriatric clinic in 2019. This model of care entails medical visits by a geriatrician on trauma wards. Furthermore, a senior surgeon visits the geriatric clinic. Thus, an exchange exists between the 2 hospitals regarding the care for geriatric trauma patients. Currently, this model exists only for orthopedic trauma surgery and not for other surgical departments.

An interview guideline with 12 questions corresponding to the 3 main topics was prepared by a multidisciplinary research team consisting of geriatricians, surgeons, public health specialists, and ethicists. The questions were formulated on the basis of a prior literature research, the guidelines for conduct of qualitative interviews [20,21], and professional experience of the researchers. The number of questions was limited to 12 due to the expected short time available to our interview partners. Interviews were conducted in a semistructured way. While the preformulated questions set the focal point, ad hoc questions provided a possibility to clarify statements or focus on particularly important issues mentioned by the interviewees [22,23].

Interviews were conducted in person in April 2022 in German. Potential interview partners were contacted personally or via email. If willing to participate, interviewees received a short information sheet about the SURGE-Ahead project.

Ethical approval was waived by the ethics committee of Ulm University, because no information on personal data, health-related data, or data on sexuality of the interviewees was collected (March 21, 2022). All interviewees declared their consent after being informed about the aim of the study, their rights, potential risks, and data protection. Interviewees did not receive any compensation for participation in the study.

Interviews were recorded and transcribed by an external transcription service provider (abtipper.de). After transcription, the interviews were anonymized. Transcripts were checked for accuracy against the recordings by 2 authors (AU, LB) to avoid bias. Occurring flaws in transcription were resolved by mutual agreement.

The statements of the interviewees were coded, extracted, and sorted into main categories, primary codes, and subcodes according to usual procedures of qualitative analysis [24] and thematic analysis [25,26] using the software MAXQDA (VERBI GmbH). Total numbers of the extracted categories and codes are presented. Representative quotes that illustrate various themes were translated from German into English by 2 researchers independently (AU, MLF), compared, and then unified into a final version. Based on these representative quotes, a narrative synthesis of the results was written. Finally responses across categories were also stratified according to the most significant hopes and fears of participants [27]. The final manuscript was written based on COREQ (Consolidated Criteria for Reporting Qualitative Research) [28].

In total, 14 interviews were conducted with 15 professionals. The mean duration of the conducted interviews was 13:51 minutes. The interviewees’ professional experience ranged from 2 to 43 years with a median duration of 15 years. Median time of affiliation with the department (8 years) differed between professions. Nursing management had the longest affiliation of 22 years, and senior physicians in the emergency department had the shortest median affiliation of 1.5 years. For details, see Table 1.

In this qualitative study, we analyzed the status of geriatric care and the expectations for an AI-CDSS supporting patient care by interviewing physicians, nurses, physiotherapists, and social service members. Results were summarized in 3 main categories covering geriatric patient care, the use of an AI-CDSS, and ethical considerations regarding an AI-CDSS. While the biggest hope was that such an application could improve geriatric care including the handling of medical geriatric challenges within a functioning digital service, the loss of individual patient-caregiver relation and a lack of usability were the biggest fears.

Provision of an EHR by an AI-CDSS was the most frequent subcode extracted from the interviews. Currently, health professionals must gather information about patients from different sources (eg, HIS, paper documents, and telephone calls), which is time consuming. One of the biggest advantages of a CDSS could be an overview of relevant individual clinical and social information. However, a higher workload was feared due to the task of entering data. Physicians already spend up to 35% of their time on documentation [29,30]. Because of these shortcomings low acceptance of EHR and CDSS is a common phenomenon in clinical practice [31,32]. Therefore, the involvement of future users in the development of the application [32], ensuring seamless integration with existing clinical information systems, minimizing the need for manual data entry, and designing applications as close as possible to the existing systems have been found to be important for the acceptance of CDSS [17,33].

Interviewees also emphasized how challenging it is to care for geriatric patients due to complex medical conditions and impairments such as cognitive impairment. It has been shown that a multidisciplinary approach is favorable for geriatric patient care and COC decisions [34-37]. Considering the lack of geriatricians [10], geriatric expertise is something the AI-CDSS could provide, and support clinical work, by highlighting relevant issues such as medication interactions or factors increasing the risk for delirium.

The lack of knowledge of optimal COC was considered relevant by all professionals. The discharge destination can have a profound effect on long-term outcomes. A discharge to nursing homes has been associated with lower survival than a discharge home [5]. COC placement suggestions by an AI-CDSS could fill a gap in care for geriatric patients in surgery. Some interviewees even expressed the wish that an AI-CDSS should enable reservation of beds in rehabilitation clinics. For this purpose, new software options for hospitals have already been established and are currently being implemented in several regions in Europe [38].

Time and resource constraints could inhibit overall feasibility and practicality. Some interviewees were concerned whether all health care professionals will have the technical affinity to operate the AI-CDSS, potentially limiting the acceptance of the system [17]. Acceptance of an AI-CDSS is closely connected to explainability, accountability, and trust [39]. Increasing transparency of AI-systems leads to more trust in these systems and thus increases their acceptance [32]. If medical professionals cannot fully understand the recommendation of an AI-CDSS, it might result in conflicting situations [40]. This becomes evident considering the current legislation stating that even when using an AI-CDSS, medical professionals are responsible for the decisions [41]. However, this might change in the future with a CDSS improving and potentially even outperforming health care professionals [42].

In addition, professional caregivers need to be able to explain the use and functionality of the AI-CDSS to patients and their relatives [40,43,44]. In this context, the concept of digital divide seems important, which describes a decreased ability of older adults to use modern technologies compared with their younger counterparts [45]. However, a recent study could not detect interviewed geriatric patents’ reservation toward the use of smart sensors [46].

Regarding the 4 core medico-ethical principles [19], the protection of patient autonomy seemed the most important, which reflects the current consent in medical practice about shared decision-making between patients and clinical professionals [47]. Although an AI-CDSS could support this, the identification of a patient’s values, beliefs, and aims is a prerequisite [48-50]. Overreliance on the AI-CDSS might lead to premature and rigid categorization of the patients according to preprogrammed categories, which are not sublime enough to catch intrinsic differences in a patient’s individual situation. Interviewees emphasized that all recommendations of an AI-CDSS must be reviewed critically by medical professionals to protect the patient’s will and autonomy.

Another central ethical code was the importance of preserving the caregiver-patient relationship—an aspect related to the principle of beneficence [19]. Interviewees demanded that they would not want the AI-CDSS to prejudge a patient. Patient-caregiver relationships were considered important for the recovery and self-efficiency of patients because of associated emotions and their trust-building effect [13]. Research about placebo effects also supports this position [51]. Fear of losing their professional identity was mentioned by some interviewees [52]. The best way to maintain professional identity and a trustful patient-caregiver relationships was seen in an open communication with the patient about the use of a CDSS, followed by informed consent.

Albeit only mentioned by a few participants, data protection and scientific evaluation of a new AI-CDSS are essential regarding the medico-ethical concept of nonmaleficence [53,54]. Especially, data protection seems to be crucial from an ethical perspective to protect sensitive personal patient data from theft, manipulation, or access by third parties [49,55]. In case of geriatric patients, who are often overwhelmed by modern processes of data gathering, storage, and processing, special responsibility falls on the developers of AI-CDSS and clinical professionals collecting the data.

A last aspect of medico-ethical reflection is justice, which was superficially touched upon only during the interviews. It has been argued that developing and maintaining a CDSS is cost-intensive and therefore could be afforded only in high-resource countries with a well-functioning health system and in big hospitals, such as university hospitals [13,56].

One limitation of the study was the rather low number of experts and an unproportionate distribution of health care professionals, with a higher representation of physicians. This does not allow for generalization of the results in a wider perspective. However, 15 interviewees should allow to capture about 85% (17/20) of possible topics, a phenomenon called “thematic saturation” [57], and in a smaller sample of interviewees, it is possible to focus on individual perspectives.

In addition, our information sheet did not explain the term AI to the interviewed health care professionals, so the reflections might have depended a lot on the individual knowledge. We, however, did not want to anticipate too much to broaden perspectives. Finally, the unique clinical information system, the status of hospital readiness for IT, and overall clinical structures could make it difficult to compare the findings with those from other hospitals, both nationally and internationally.

An AI-CDSS was mostly considered to be beneficial, especially by providing health care professionals with an easily accessible data platform focused on geriatric needs and supporting geriatric comanagement and COC decisions. The most common concerns focused on maintaining the patient’s autonomy and preserving the patient-caregiver relationship, as well as smooth integration into existing workflows without extra tasks. Therefore, careful consideration must mainly be given to address technical demands (intuitive usability and easy access for all caregivers involved) and ethical challenges (maintaining the patient’s autonomy, explainability, and supportive character) during the development and implementation of a new AI-CDSS for geriatric patients in surgical departments.