KI-basierte Entscheidungsunterstützungssysteme für die Antibiotikatherapie im Krankenhaus: Determinanten für die Implementierung aus Perspektive beteiligter Versorgungsakteure

Background: The potential of artificial intelligence (AI) and AI-based systems for healthcare has steadily increased in recent years. AI-based clinical decision support systems (CDSS) make it possible to rapidly identify patterns within large datasets and generate robust predictions to support clinical decision-making – including in the context of antibiotic therapy. However, hospitals in Germany currently make little use of AI-based CDSS, partly because their implementation is complex. Objective and methods: The aim is to examine the requirements for successful implementation of AI-based CDSS for antibiotic therapy in hospitals from the perspective of healthcare stakeholders, and to derive future research and action needs. Four studies examined determinants of implementation using both quantitative and qualitative research methods. All peer-reviewed publications draw on the structuring elements of the Human–Organization–Technology(HOT)-fit model and to systematically outline facilitators and barriers that determine the implementation process of AI-based CDSS. Results: The imple-mentation of AI-based CDSS involves organizational, user-related, and technology-related measures, with the characteristics of the systems themselves representing key determining factors. In particular, usability, interoperability, recommendation quality, and system transparency set the course for successful implementation. At the organizational level, real-time-capable infrastructures, readiness for change, resources, as well as political and regulatory frameworks concerning education, responsibility, and financing constitute central requirements. Insufficient readiness for change often results from situational factors, including a perceived lack of need or the complexity of new technologies being considered too high. This may also reflect underlying fears or feelings of being overwhelmed. In addition, the attitudes, knowledge, and perceived clinical benefit among potential users are crucial. At the same time, the boundary between decision support and decision-making requires AI-specific competencies and an ongoing discourse on normative and professional-ethical issues. Conclusion: Implementation processes must take into account organizational structures, cultural and professional norms, as well as social and political factors such as guidelines, laws, and regulations. Interdisciplinary and transdisciplinary approaches are necessary to identify problem areas at an early stage and to anticipate practical as well as ethical challenges.