Harmonization of Electronic Medical Records for Federated Learning: Addressing Challenges in International Healthcare...

For the development and implementation of a federated learning (FL) model, a clear outcome definition and the harmonization of healthcare data present major challenges. Healthcare institutions use different data representation standards. This variability poses significant challenges for harmonizing data across multiple partners.

The aim of this paper was to show how we have standardized datasets from three partners from Austria, Germany and Brazil to be used for a FL model for the prediction of major adverse cardiovascular events (MACE).

Each partner extracted healthcare data from their electronic medical record (EMR) system. The mapping of these data was performed using common international standards.

The final harmonized dataset consisted of demographic, administration, diagnoses, medication and laboratory data represented with uniform international standards across the partners.

This study highlights key challenges of harmonizing healthcare data and provides an example as a solution. In order to confirm reliability of our dataset for predicting MACE using FL, further testing and validation will be performed.