Making Rare Diseases Count 2026

Rare diseases remain difficult to identify early because data is fragmented across institutional silos. Traditional AI models rely on integrated data warehouses that are expensive to build, difficult to govern and often unviable in many health systems. What is needed is a fundamentally new architecture that enables analytics across distributed data sources without moving data or compromising privacy. The OSCAR model Global biopharmaceutical company Sanofi’s OSCAR (multimOdal tranSformer foundation model for Clinical Analysis of Rare diseases) algorithm, developed through the cross-industry partnership Project Saturn, demonstrates how multimodal, privacy-preserving AI can address this challenge. In its initial clinical evaluation shared through the American Diabetes Association, OSCAR increased detection efficiency for type 1 diabetes more than 18-fold and corrected diagnostic misclassification for nearly one- third of affected individuals, reducing delays to appropriate treatment.41 The model has been tested computationally and is planned for subsequent validation in clinical practice. The same architecture allows deployment across diverse health systems, from tertiary research networks to emerging national infrastructures in LMICs, contingent upon alignment with local regulations. Continued refinement of OSCAR uses a federated learning architecture, training models across decentralized datasets that remain securely within each participating institution. This enables advanced analysis of clinical, genomic and other data while maintaining data sovereignty and compliance with local privacy standards. Extending to rare diseases The OSCAR model is now being adapted for rare disease detection, starting with Pompe and Fabry diseases, two lysosomal storage disorders with subtle, multisystemic presentations that make them ideal test cases for multimodal AI approaches. By fine-tuning the architecture on rare disease datasets that combine genomic, clinical and laboratory signals, researchers aim to identify early markers that could significantly shorten diagnostic journeys. Deployment pilots are planned for 2026, expanding collaborations with health ministries, patient organizations, research institutions, payers and industry partners. The goal is to create a global federated learning network by 2030, enabling privacy-preserving AI to improve early detection, diagnosis and coordinated care for rare diseases worldwide. Integrating AI with coordinated care Data infrastructure and AI can reveal new insights, but their full potential is realized only when linked to care models that provide practical benefits for patients, caregivers and communities. Coordinated care connects data-driven insights with human support networks, allowing interventions to extend beyond the clinic into daily life. Programmes developed under Project Saturn show how digital tools can empower families to manage health collaboratively, improving adherence, engagement and continuity of care. For rare diseases, where small populations and complex care needs make coordination difficult, such models demonstrate how combining federated AI with relational support can improve outcomes, creating a virtuous cycle of insight and intervention.CASE STUDY 4 Sanofi’s OSCAR AI algorithm and Project Saturn First part of the title: Second part of the title 22