Open issues and unmet needs in healthcare include the enhancement of the statistical power of the clinical outcomes along with the development of prediction models for effective disease management, the detection of prominent factors for disease progression and the identification of targeted therapies. In this work, we deploy a computational pipeline that uses data curation and ontology matching to curate and align heterogeneous data structures to enhance the statistical power of the outcomes. Then, we use federated learning to develop disease prediction models across harmonized cohort data that are stored in private cloud databases. A preliminary case study was conducted for the first time on three European cohorts on primary Sjögren’s Syndrome (pSS) yielding harmonized data with 90% average overlap along with a federated lymphomagenesis progression model with accuracy 0.848, sensitivity 0.833 and specificity 0.849.
