gms | German Medical Science

In this talk, we will describe a planned collaboration between French and German institutions towards the establishment of a data-driven analysis platform for the rare disease Osteogenesis Imperfecta. The collaboration is led by the Institut Imagine in Paris with participation of the Chair of Medical Informatics at Erlangen University and the Department of Pediatrics at Cologne University Hospital. The partners have submitted a second-stage grant proposal within a binational funding call by ANR/DFG which is currently under review. The project outline is described in the following:

The objective of AI4OI (Artificial Intelligence for Osteogenesis Imperfecta) is to build a supra-national learning system for rare diseases and test it on bone diseases. For rare diseases, more than in other areas, each case counts, making data sharing a necessity. Distributed data collection (a data collection distributed over many centers) and federated infrastructure are means to maximize access to data across centers while maintaining data privacy and control by data providers.

Our objective is to develop such federated infrastructure to re-use electronic health record (EHR) data for research purposes. In Germany, a national data integration infrastructure is being developed by the BMBF-funded Medical Informatics Initiative (MII) to enable to share EHR data at the national level, using tools that transform data into a common format. In France, the Imagine Institute has developed a system called “Dr Warehouse” that automatically extracts all patient related phenotypic information from their EHRs.

The partners will design a common data model based on the MII HL7 FHIR Core Dataset and develop machine learning algorithms on harmonized data in a federated manner, including the application of federated random forests and gradient boosting decision trees to model phenotypes as well as response to treatment. This approach will be applied to Osteogenesis Imperfecta (OI), a genetic bone disorder characterized by increased bone fragility and susceptibility to bone fractures, with highly heterogeneous severity. Bisphosphonate therapy has been proposed to patients but the response to treatment is uncertain. We will train a model to predict response to bisphosphonates on retrospective data available at children’s hospital University of Cologne and the Assistance Publique-Hopitaux de Paris/ Imagine Institute. The consortium will collaborate with the European Research Network on bone diseases (ERN-BOND) to validate and extend the approach to a larger set of centers. Due to the involvement of the colleagues from Cologne in the European patient organization for OI (Pr Semler is chair of Osteogenesis Imperfecta Federation Europe), the perspective of patients will be included in the project. The planned project will collaborate closely with the MII and its use cases, including the application of results from the previous CORD (Collaboration on Rare Diseases) and ongoing PrivateAIM (Privacy-preserving Analytics in Medicine) projects.

The authors declare that they have no competing interests.

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