ZDIN-Zahlen im Detail

To develop a decision support system for pediatric cardiology case conferences, the anonymization of 4,000 freetext medical case reports is required. This paper presents an anonymization strategy using LLM-AIx, a tool for structured information extraction based on large language models (LLM). The three-step process involves automatic extraction of personally identifiable information (PII) from the reports, evaluation of the results against a manually annotated ground truth, and replacement of identified PII with surrogate values, including controlled date shifting. Initial tests with six example reports revealed challenges regarding handling multiple attribute occurrences and consistent replacements. Future work will focus on full pipeline implementation and mapping clinical information to standardized terminologies such as SNOMED CT.

Die Nutzung elektronischer Patientenakten (ePA) zur Datenspende bietet großes Potenzial für die medizinische Forschung, stellt jedoch hohe Anforderungen an Information, Kommunikation und Akzeptanz in der Bevölkerung. Im Rahmen eines interdisziplinären Workshops anlässlich der 69. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e.V. diskutierten in Dresden Expert:innen aus Medizinischer Informatik, Wissenschaftskommunikation und Patientenvertretung gemeinsam mit dem Publikum zentrale Fragestellungen zur Kommunikation der Datenspende an Bürger:innen. 

Ziel war es, Bedingungen zu identifizieren, unter denen eine informierte und selbstbestimmte Entscheidung über die freiwillige Nutzung von Gesundheitsdaten für Forschungszwecke möglich ist. Im Fokus standen dabei die Fragen, wer über die Datenspende informiert, wie Vertrauen in Institutionen gestärkt, Nutzen und Risiken verständlich vermittelt und Zielgruppen angemessen erreicht werden können. Impulsvorträge und Diskussionen zeigten, dass es einer Vielzahl niedrigschwelliger, transparenter und zielgruppengerechter Informationsangebote durch glaubwürdige Akteure bedarf. Dabei gilt es, digitale Gesundheitskompetenz zu fördern, Zugangshürden abzubauen und Bürger:innen nicht nur technisch, sondern auch kommunikativ zu befähigen. 

Der Beitrag verdeutlicht, dass eine wirksame Kommunikation zur Datenspende eine zentrale Voraussetzung für gesellschaftliche Akzeptanz und eine gerechte Teilhabe an digitaler Gesundheitsforschung darstellt.

The 24th Special Topic Conference (STC 2025) of the European Federation for Medical Informatics (EFMI) was held in Osnabrück, Germany from 20–22 October 2025. The Conference was co-hosted by Osnabrück University of Applied Sciences, the German Society of Medical Informatics, Biometry and Epidemiology (GMDS) and EFMI.
The theme of STC 2025 was “Good Evaluation – Better Digital Health”. In an era of rapid digital transformation, the conference highlighted that rigorous evaluation is essential to ensure the safety, effectiveness and sustainability of health IT innovations. While evaluation is often carried out at the very end of the development process, good evaluation starts at the outset of a project with an investigation of the context within which the implementation will take place and its determinants. In the best case, the evaluation will follow a longitudinal approach to capture the full impact of the implementation regardless of whether it is positive or negative. With the advent of powerful digital systems, including those making use of AI technology, evaluation is needed more than ever to prevent harm and ensure effectiveness and meaningfulness on a micro, meso and macro level. As digital interventions have multiple determinants and implications, evaluation methods have to consider the full spectrum of dimensions, ranging from technical and organisational aspects to human, social, ethical and
educational perspectives.
These proceedings present current scientific trends in the evaluation of digital health and AI-enabled systems in patient care. Some contributions address methodological topics such as evaluation paradigms and techniques, research models and frameworks that constitute the foundation of reliable and valid results. Others point to critical areas of investigation such as technical quality, including interoperability, as well as data protection and security. Non-technical areas of paramount importance, first and foremost ethics, are equally addressed. Many papers take the perspective of human and organisational factors combined with social and political aspects in their dual role as potential determinants and outcomes. Finally, educational aspects are covered by a number of papers addressing competencies and tools for enhancing education in the era of AI.
While these topics have been scientifically investigated by biomedical and health informatics for many decades, the latest AI developments impose a new imperative on the evaluation of digital health. These proceedings reflect the current debates and developments from a scientific point of view with the goal of having an impact on practice.
The proceedings are published by IOS Press in the series Studies in Health Technology and Informatics. Volumes in this series are submitted (for evaluation) for indexing by MEDLINE/PubMed; Web of Science: Conference Proceedings Citation Index – Science (CPCI-S) and Book Citation Index – Science (BKCI-S); Google Scholar; Scopus; and EMCare.

Study objectives: Chronic wounds represent a significant economic and personal burden. For their successful treatment, the causes must be known and treated. Wounds caused by pyoderma gangrenosum (PG), a rare inflammatory skin disease, are often misdiagnosed. This study, therefore, aims to develop a machine learning model capable of differentiating PG from other wound types, focusing on chronic leg wounds to address this diagnostic challenge.

Methods: We used 3674 wound photographs from three specialised wound centres with the four most common types of foot and leg ulcers and the rare inflammatory differential diagnosis PG. The convolutional neural network classifier ConvNeXt 'B' was pretrained on LAION2B, ImageNet12k and ImageNet 1k before being trained and fine-tuned on an 85:15 train, validation split.

Results: The model achieved an overall high accuracy in multiclass classification of the chronic wounds (unbalanced accuracy 90%, balanced accuracy 87%). The sensitivity for identifying PG was 94%, while the sensitivity forother chronic wound types was 97% for diabetic foot ulcers (DFU), 92% for venous leg ulcers (VLU), 78% for mixed leg ulcers and 74% for arterial leg ulcers.

Discussion: The machine learning model effectively differentiates PG from the most common leg and foot ulcers and was very accurate for classifying DFU and VLU. A higher rate of misclassifications occurred for the other vascular ulcers, that is, mixed and arterial leg ulcers. This aligns with the challenges in clinical practice.

Conclusion: Despite the limited number of wound images, this novel multiclass wound classification model accurately identified PG and differentiated leg and foot ulcer subtypes, providing a foundation for a diagnostic support system.

This special issue of MIBE comprises contributions submitted to the Special Topic Conference “Good Evaluation – Better Digital Health” of the European Federation of Medical Informatics (EFMI). EFMI Special Topic Conferences (STC) are scientific events focussing on a dedicated topic in biomedical and health informatics and are addressing particularly young and early career scientists. This year, the STC is hosted by the German Association of Medical Informatics, Biometry and Epidemiology (GMDS) and organised by Osnabrück University of Applied Sciences. Publishing contributions to a European conference illustrates MIBE’s general offer to colleagues from Europe and beyond to publish their work in this journal. As a first step, posters submitted as a three-page poster contribution to the STC were considered for publication and underwent the same rigorous peer review process as all contributions to the conference before they were proposed for publication in MIBE. Here they were peer reviewed again.

Nine out of 20 poster contributions submitted were accepted and published in this special issue covering topics of the theme “Good Evaluation – Better Digital Health”. Evaluation is a centre piece of the software and systems life cycle and starts – in contrast to usual anticipations – not at the end when the system has been rolled out but at the very beginning of the development process. It thus allows the integration of empirical patterns against which the testing can be performed such as user expectations and requirements. The articles “Co-creating a cancer screening dashboard with screening invitees and experts“ by Oldhoff-Nuijsink et al. [1] and “Building a digital platform for collaborative second opinions in rare disease: Integrating AI and healthcare networks for improved care“ by Lima et al. [2] illustrate the meaningfulness of early evaluations. Conversely and even a step earlier, evaluations of patients’ behaviour can stimulate the needs to foster long-term medication adherence through digital methods as was shown by Kim et al. in their study “Adherence to hormonal therapy in breast cancer patients: EHR-based retrospective data analysis” [3].

Evaluations and tests that are derived from frameworks, theories and concepts have a different background, however, need not contradict more empirically grounded evaluation targets. It is therefore useful to identify new frameworks as in the case of Hülsmann et al. in “Identifying assessment frameworks for digital public health interventions: First results of a scoping review“ [4] or to augment existing ones as was proposed by Grashof et al. in “A new perspective on eHealth acceptance: Combining health-related factors with the Technology Acceptance Model“ [5].

The emergence of more and more AI models for use in healthcare increases the necessity to conduct proper evaluation studies before such models can be integrated into systems for clinical use. The study by Galland-Decker et al. “Performance evaluation of Meditron3-70B in medical coding: Current limitations and integration perspectives for clinical practice” [6] investigated the performance of a medical large language model to generate SNOMED and ICD codes from vignettes and hinted at current flaws.

Evaluating a system often means evaluating the context in which the system is embedded in. This also applies to AI systems and their rootedness in models that have been developed on a specific set of data to be applied to fresh clinical data. The study by Slob et al. “FAIVOR – a push-button system for AI validation within the hospital“ [7] describes a tool that serves as an AI model repository and allows the evaluation of a given model on local data.

Evaluating applications and systems is often accompanied by educational measures to really obtain better digital health which means nothing less than better use from better digital systems. Two articles address education and training: “Developing competencies in health informatics: Blended teaching method “by Mannevaara et al. [8] and “AI and AR for inclusive health education“ by Focsa [9].

The studies presented in this special issue give rise to longitudinal measures to evaluate the output, outcome and impact of digital systems to finally appraise whether an application really contributed to better digital health.