A model to predict survival following liver retransplantation


  • Hugo R. Rosen,

    Corresponding author
    1. Division of Gastroenterology/Hepatology, Portland Veterans Affairs/Oregon Health Sciences University
    • Address reprint requests to: Hugo R. Rosen, M.D., Division of Gastroenterology/Hepatology, Oregon Health Sciences University, Portland Veterans Affairs Medical Center, 3710 SW U.S. Veterans Hospital Rd., P.O. Box 1034, P3-GI, Portland, OR 97207. fax:(503) 273-5348
    Search for more papers by this author
  • Joseph P. Madden,

    1. Health Data Research Inc., Portland, OR
    Search for more papers by this author
  • Paul Martin

    1. Hepatology Section, University of California at Los Angeles, Los Angeles, CA
    Search for more papers by this author


In the current era of critical-organ shortage, one of the most controversial questions facing transplantation teams is whether hepatic retransplantation, which has historically been associated with increased resource utilization and diminished survival, should be offered to a patient whose first allograft is failing. Retransplantation effectively denies access to orthotopic liver transplantation (OLT) to another candidate and further depletes an already-limited organ supply. The study group was comprised of 1,356 adults undergoing hepatic retransplantation in the United States between 1990 and 1996 as reported to the United Network for Organ Sharing (UNOS). We analyzed numerous donor and recipient variables and created Cox proportional-hazards models on 900 randomly chosen patients, validating the results on the remaining cohort. Five variables consistently provided significant predictive power and made up the final model: age, bilirubin, creatinine, UNOS status, and cause of graft failure. Although both hepatitis C seropositivity and donor age were significant by univariate and multivariate analyses, neither contributed independently to the estimation of prognosis when added to the final model. The final model was highly predictive of survival (whole model χ2 = 139.63). The risk scores for individual patients were calculated, and patients were assigned into low-, medium-, and high-risk groups (P < .00001). The low degree of uncertainty in the probability estimates as reflected by confidence intervals, even in our high-risk patients, underscores the applicability of our model as an adjunct to clinical judgment. We have developed and validated a model that uses five readily accessible “bedside” variables to accurately predict survival in patients undergoing liver retransplantation.