FC 105: Multidimensional Prognostication Tool for Kidney Transplant Patient Survival: The Mortality Mbox

Abstract

Abstract BACKGROUND AND AIMS Predicting kidney transplant patient mortality has been hampered by registry-based studies and low-level phenotyped cohorts without specific design towards mortality prediction. This represents a limitation for decision making and ultimately patient care. We aimed to build a robust patient mortality prognostication system. METHOD We enrolled 1446 patients transplanted in France between 2004 and 2014 in whom a protocol-based collection including more than 160 parameters from the recipient (past medical history, risk factors) donor and graft, biological and imaging data, was performed on the day of transplantation (TX) and during the first year of transplantation. Multivariable Cox model was used to develop an individual predictive score of mortality, further improved by a Lasso regression in order to retain the strongest predictors of mortality. RESULTS Among the 1446 kidney transplant recipients included, 309 patients died after a median post-TX follow-up time of 7.6 years (IQR 5.40–10.78). Among the 120 parameters, 19 predictors were selected using lasso regression. The strongest predictors of patient survival were (1) baseline recipient factors (age, history of cancer, diabetes mellitus, chronic obstructive pulmonary disease, cardiovascular events: myocardial infarction, stroke or arteritis, supraventricular cardiac rhythm disorder, psychiatric history and VHC status); (2) post-TX parameters (need for dialysis, cardiovascular complications and cancer in the first year of TX); and (3) seven biological variables (HbA1c, C-reactive protein, albumin, gamma-glutamyl transferase, uric acid, neutrophils and urinary protein). The mortality score showed accurate calibration and discrimination at 10 years [C-statistic = 0.81; 95% confidence interval (95% CI) 0.78–0.83]. CONCLUSION We generate the first integrative patient survival score that shows a superior prediction performance when compared to previous prognostic systems, reaching 81% prediction accuracy at 10 years. A reliable survival prediction tool will enable improved patient monitoring and therapeutic decision making to alter the course of an unfavorable patient survival outcome and may provide a surrogate endpoint in clinical trials.