Summary/Abstract
Kidney transplant (KT) survival faces challenges from increased patient complexity, use of non-standard donors,
and longer cold ischemic times. Despite refinement in immunosuppression (IS) management over several
decades, nearly all KT patients experience progressive loss of allograft function, leading to eventual graft failure
while still risking life-threatening IS-related complications. Development of evidence-based IS management
practice to optimize long-term KT function has been limited by the need for detailed, longitudinal clinical data for
large representative populations. Our previous NIDDK-supported R01 grant (Choosing IS regimens in kidney
Transplant by Efficacy and Morbidity; CISTEM), leveraged integrated transplant registry, Medicare claims and
national pharmacy clearinghouse data to assess the impact of early IS regimen selection on key post-KT events:
infections, malignancy, new onset diabetes, as well as traditional metrics (acute rejection rate, allograft survival
and patient death). We developed a free web-based interface to assist transplant professionals and patients in
shared-decision making about IS choice at the time of KT. However, IS requires lifelong dynamic adjustments.
We will enhance our prior work with the following Specific Aims in CISTEM2: 1) Recognizing the lack of
longitudinal and granular clinical observations and lab results, we will establish a novel CISTEM2 dataset,
integrating: a) national transplant registry data granular clinical data from 12 transplant centers housed in the
Greater Plains Collaborative (GPC,) a component of the Patient Centered Clinical Research Network
(PCORnet); b) administrative claims; and c) social determinants of health (SDOH) indicators for KT recipients.
GPC utilizes the PCORnet common data model (CMD) for all clinical data, which can be securely linked to
national transplant registry, administrative claims and SDOH indicators. We will expand from the multivariable
propensity and Cox proportional hazard models with time-varying covariates used in CISTEM, to more
responsive and clinical meaningful endpoints such as percentage drop in KT function (via estimated glomerular
filtration rate; eGFR) and the validation of computed phenotypes for key clinical events. 2) We will extend
CISTEM by developing longitudinal machine learning (ML) algorithms to dynamically identify IS strategies that
optimize longer-term KT eGFR, reduce cost, and limit those IS comorbidities that post-transplant patient focus
groups identify as contributors to diminished quality of life. 3) We will validate the predictive models refined in
Aim 1 and the ML models developed in Aim 2, using two additional PCORnet sites (12 more KT centers; Total
N: 40,535 KTs). We will analyze PCORnet CDM from all centers using distributed learning models to refine the
first dynamic, evidence based clinical decision tool for longitudinal IS management after KT. Balancing the risk
of acute rejection, patient and graft survival, and risk of IS-related complications after KT, based on highly
granular, multicenter, longitudinal clinical data from real-world patient experience, will allow patients and
physicians to dynamically optimize IS in a more personalized, patient focused, and cost-effective manner.