3 questions for...
Prof. Dr. Dagmar Krefting
Mentor
Dagmar Krefting is a professor of medical informatics and also director of the Institute of Medical Informatics at Göttingen University Hospital. Her research focuses, on the one hand, on collaborative infrastructures for cross-institutional biomedical research networks, such as the German Centres for Cardiovascular Diseases and for Child and Adolescent Health, the University Medicine Network, and various consortia within the Medical Informatics Initiative. She also develops methods for multimodal biosignal processing, including for research into sleep-related disorders. These methods encompass statistical techniques, ‘classical’ non-linear time series analysis and machine learning. When using AI methods, the focus is on trustworthiness, e.g. explainability, generalisability and reproducibility
1.
Laboratory data, clinical trials and genetic analyses generate enormous amounts of data every day. What conditions need to be put in place to ensure that this data can be put to good use in research and healthcare?
Firstly, the data must be semantically interoperable. This implies that the meaning of the data and the contextual information must be capable of being processed reliably and automatically. This can be achieved through technical means and subject-matter expertise. However, the biggest obstacle remains the completely heterogeneous and local assessment under data protection law of the permissibility of processing for research and healthcare provision. Here, I can only hope that, with the European Health Data Space, we will have harmonised legal frameworks that are then accepted by all parties involved.
2.
Digital research infrastructures are also intended to indirectly improve patient care. How can the transfer of data-driven research into clinical practice be successfully achieved?
In general, the costs involved in developing digital tools for healthcare are low compared to, for example, drug development, but they are still significantly higher than in other sectors. Here, there is a considerable need for support for start-ups and spin-offs in complying with regulations, for example under the Medical Devices Act and the AI Regulation. This also includes innovation-friendly access to training and validation data within digital research infrastructures
3.
When we think about the future: what developments in data standards, interoperability and governance are crucial for the scalable, cross-organisational use of AI in medicine?
It is no coincidence that the majority of approved AI applications in medicine are found in image analysis – thanks to DICOM, we have had an established standard in this field for many years. I see the main problem as being that proprietary data formats and siloed solutions still seem to be the preferred market strategies for healthcare information system manufacturers, despite numerous legislative initiatives. If only the existing standards were implemented and utilised, we would already have far more trustworthy AI procedures in healthcare – and the learning healthcare system could then develop dynamically from there.
