gms | German Medical Science

Introduction: In 2021, the Advisory Council on Health & Care (Sachverständigenrat Gesundheit & Pflege) recommended that Germany should pursue a Learning Health System. In 2023, the Federal Ministry of Health in Germany followed the Advisory Councils' suggestion and proposed key elements of Learning Health Systems [1].

State of the art: Being on the German agenda since 2023, Learning Health Systems were established back in 2006 by the US National Academy of Medicine which still improves and promotes it. It describes how technological advancement can contribute to fast and continuous biomedical and healthcare discoveries using real-world data stored in digital systems like clinical data repositories or electronic health records (EHR). In Germany, different initiatives exist striving to conceptualize, implement and test key elements, e.g., medical information objects of the EHR or the central data hubs of the Medical Informatics Initiatives. They form the backbone of the policymaker's goal of achieving a national Learning Health System.

Despite these national initiatives, Learning Health Systems comprise data interoperability, data, and AI-driven predictive models, and clinical decision support systems (CDSS) aligned for continuous data-driven discovery. When aligned, these components form continuous learning cycles where 1.) healthcare reflects itself in its data, 2.) AI models leverage data, and 3.) AI-driven CDSS presents to healthcare workers what was learned for informed decision-making.

Concept: Learning Health Systems can be realized at various levels, including planting the seed in local initiatives which can grow. The advantages of local implementations are lower technical complexity, flexibility about the medical questions of concern, and less administrative overhead. Initiated by physicians of a wound care hospital department, we established a Learning Health System in 2016 [2] on the clinical problem of diabetic foot ulcers – a chronic wound responsible for 75% of lower leg amputations [3]. The objective of the Learning Health System is the prediction of amputations and, hereby improving their prevention.

Implementation: We established a regular exchange with the team of the wound care unit at Christliches Klinikum Melle, an access to an anonymized routine data set consisting of structured and standardized patient data based on the PEDIS classification [4] and a data-driven learning cycle. It included a risk prediction model, its implementation in a clinical decision support system, and the incorporation of feedback from clinicians. The development of the risk prediction model itself was based on a learning mechanism via diachronic learning enabled through Bayesian Logistic Regression (AUC 0.8) [5]. The clinical decision support system based on the model embraced a visual representation of the distribution of the probability and the risk cut-off point. The tool was implemented at the hospital and supported the physician's prediction in the sense of a second opinion.

Lessons learned: When establishing a Learning Health System as the Federal Ministry of Health aims for, local initiatives can help to establish them in a timely and low-threshold manner. It is key to the initiation to work on physician-driven problems, engage them throughout the implementation and show results early on.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.