A Cloud-Based, Multi-Modal Data Architecture for Post-Transplant Diabetes Management

Effective glucose management in kidney transplant recipients (KTRs) is complicated by dynamic insulin requirements and fragmented clinical data. Existing insulin dosing models based solely on electronic health records (EHR) and continuous glucose monitoring (CGM) data fail to account for the effects of high-dose glucocorticoid tapering, which significantly impacts glucose control. Additionally, few platforms can seamlessly integrate structured clinical data with patient-reported inputs, limiting their utility for real-time decision-making and AI-driven insulin predictions. To address these gaps, we conducted a two-phase clinical study in collaboration with IBM and Cleveland Clinic. In phase 1, we retrospectively analyzed EHR and CGM data to examine how glucocorticoid tapering influenced insulin requirements. This analysis highlighted the need to incorporate additional factors—such as meal intake, activity levels, and sleep quality—to improve predictive modeling. In phase 2, we developed a customized Health Guardian (HG) web application to collect patient-reported measures. We also refined t he HG data pipeline to integrate EHR, CGM, and patient-reported data into a centralized PostgreSQL database, enabling real-time visualization through both clinician and patient interfaces. System performance was improved by replacing the complex Orbit Service message queue with a direct API handler via IBM API Hub, enhancing reliability, monitoring, and error detection. This paper focuses on the platform’s architecture and implementation. We have developed a flexible infrastructure designed to support future integration with AI/ML models for insulin prediction and glucose pattern detection. Initial efforts are centered on collecting, and validating high-quality multi-modal data streams. This end-to-end framework lays the foundation for more personalized insulin management, improved clinician insight, and increased patient engagement in the post-transplant setting.