Advanced Analytics in der Gesetzlichen Krankenversicherung: KI-gestützte Analyse von Abrechnungsdaten zur Versorgungsoptimierung

The project demonstrates how AI methods can be used to identify multimorbid patient groups with a primary indication (in this case, diabetes) in routine statutory health insurance data. The basis was anonymized longitudinal data, including quarterly diagnoses (ICD) and drug prescriptions (ATC), costs, and sociodemographic information, which were translated into vectors using neural networks to identify differences and similarities. These embeddings enable new forms of patient grouping and illustrate, for example, correlations between disease patterns and resource consumption. A two-stage HDBSCAN clustering, supplemented by UMAP, maps both snapshots and temporal developments. Pattern mining and vector variance analysis also provided additional evidence of stable and dynamic disease progression. The results prove that anonymized SHI billing data can be used effectively with AI methods to identify outliers and complex care constellations. The methodology thus opens up new approaches for the further development of health services research and more targeted management of care.