Individualized immunosuppression in organ transplant recipients based on real world evidence and artificial intelligence

New York University School Of Medicine, New York NY

Investigators

Abstract

PROJECT SUMMARY Over 250,000 kidney transplant (KT) recipients depend on lifelong immunosuppressive therapy to prevent rejection and graft failure. However, immunosuppression increases the risk of complications such as infections, cancer, and cardiovascular disease (CVD), which together account for >70% of deaths in KT patients. To optimize transplant outcomes while minimizing immunosuppression-associated adverse drug events (ADEs), clinicians aim to choose the most appropriate immunosuppression regimen for the individual recipient's risk profile. Nonetheless, this practice is often empirical and rudimentary because little clinical evidence guides it. Our previous study found substantial disagreements between transplant centers' protocols, suggesting that over- or under-immunosuppression could be prevalent. Efforts to develop such evidence have been hampered by the limited granularity of current KT data. Existing large-scale datasets primarily focus on transplant-specific outcomes like graft failure and death but fail to capture broader clinical data, such as infections or blood glucose levels, which are essential for understanding immunosuppression-associated ADEs. To address this gap, we plan to use Epic Cosmos, a nationwide electronic health record database that launched in 2019 and now covers over 274 million individuals. Our preliminary dataset includes 69,418 KT recipients (39% of the total) between 2014-2022, encompassing detailed longitudinal data such as lab values, diagnoses, procedures, and prescriptions. Additionally, traditional regression-based analyses are ill-equipped for evaluating individual treatment effects (ITE) because they focus on average treatment effects, neglecting individual differences. To overcome this, we propose using meta-learners, a class of machine learning algorithms designed to model ITE based on individual characteristics. These algorithms can also identify the clinical factors that make specific treatments more suitable for the individual recipient. Leveraging the newly introduced database and the cutting-edge inferential method, we propose to create a quantitative framework for immunosuppression individualization with the following aims: 1) to quantify the individualized treatment effects of immunosuppression regimens on immunosuppression-related side effects, 2) to leverage the granular data from Cosmos to improve our proof-of-concept (PoC) meta-learners on transplant outcomes, and 3) to develop a clinical decision support tool for immunosuppression tailoring. This project will create the first framework to address the long-standing challenge of evidence-based immunosuppression tailoring. This framework will be implemented to an open-access web application for clinicals, potentially improving the survival and well-being of over 250,000 KT recipients. Our experience of using meta-learners on large real-world datasets can potentially be leveraged to transform clinical decision- making in diverse medical specialties.