Federated learning for optimizing machine learning models trained with data from different hospital information systems.

• Grant number: 25/04100-7
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: May 01, 2025
• End date: April 30, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Agreement: European Commission (Horizon 2020)
• Principal Investigator: Paulo Mazzoncini de Azevedo Marques
• Grantee: Hilton Vicente César
• Host Institution: Faculdade de Medicina de Ribeirão Preto (FMRP). Universidade de São Paulo (USP). Ribeirão Preto , SP, Brazil
• Associated research grant: 21/06137-4 - Predicting cardiovascular events using machine learning, AP.R

Abstract

Although several machine learning (ML) models using electronic health record (EHR) data have been developed recently, there is a significant need for external validation of these models. Predictive models may perform well in a center with a population similar to the training data but may perform worse in centers with different patient characteristics. Multicenter studies should be considered to develop generalizable models and discriminate equally well across different cohorts. However, such multicenter studies are often limited if they require sharing patient data in a centralized location. Even if data is anonymized before sharing, there is always some risk of anonymity compromise for specific data types. Federated learning-based approaches circumvent these limitations by sharing models and metrics instead of data, making it possible to improve the generalizability of predictive models while safeguarding patient privacy. The proposed project seeks to apply data harmonization, implementation, validation, adaptation, and integration of machine learning models in the clinical environment to predict the risk of major adverse cardiovascular events (MACE).