St. James Ballroom
Survey of Methods for Dynamic Prediction of Graft Survival in Kidney Transplant Patients (303897)
*Matthew Buras, Mayo Clinic AZ Department of Biostatistics
Michael Golafshar, Mayo Clinic
Raymond L. Heilman, Mayo Clinic
Andrés Jaramillo, Mayo Clinic
Girish K Mour, Mayo Clinic
Daniel S. Ramon, Mayo Clinic
Morgan A. Whigham, Mayo Clinic
Keywords: survival, prediction, landmark, joint-model, time-dependent, time-varying
Standard survival analyses utilize baseline clinical variables such as donor type (living vs deceased) and recipient age with the goal of accurately predicting graft or overall survival. Laboratory values attained at last follow-up, such as estimated glomerular filtration rate (eGFR), a common measure of kidney function, are also commonly included in a Cox proportional-hazards (Cox PH) model in an attempt to reflect the patients changing health. While this approach does account for some post-transplant information it fails to utilize the vast amounts of data that is accumulated post-transplant over potentially years of follow-up. Several approaches for incorporating data obtained during repeated follow-up are considered - namely landmark analysis, Cox PH with time-dependent covariates, and joint-modeling. Each of these techniques is applied on a database consisting of all kidney transplant patients with donor-specific anti-human leukocyte antigen (HLA) antibody (DSA) data at Mayo Clinic Arizona from October, 2011 to October, 2017. Their results and interpretations are compared and contrasted.