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

Kidney transplant recipients (KTRs) experience fluctuating insulin needs during the first 90 days post-transplant due to metabolic and medication changes. Inadequate monitoring and fragmented data can hinder effective glucose control, increasing the risk of complications such as organ rejection. To address these challenges, we present an end-to-end data pipeline architecture developed through a clinical collaboration between IBM and Cleveland Clinic. Our initial analysis of retrospective electronic health records (EHR) and continuous glucose monitoring (CGM) data highlighted the need to integrate additional factors—meal intake, activity levels, and sleep quality—for comprehensive insulin dosing models. In response, we customized a Health Guardian (HG) web application to collect patientreported data via a food diary and questionnaires in both clinical and home settings. The integrated system combines EHR data (demographics, laboratory measurements, medication history), continuous glucose readings, and patient-reported metrics in a PostgreSQL database. To improve performance, we streamlined back-end analytics by transitioning from a complex message queue system, Orbit Service, to a direct API handler connection with IBM API Hub. This architectural change improves system stability, enables real-time monitoring, and facilitates proactive error detection, reducing disruptions and enhancing the participant experience. The framework supports both machine learning model development for researchers and clinical dashboard for physicians to optimize diabetes management during the critical post-transplant period.