A Bayesian methodology to improve prediction of early graft loss after liver transplantation derived from the Liver Match study

作者：Angelico Mario; Nardi Alessandra; Romagnoli Renato^*; Marianelli Tania; Corradini Stefano Ginanni; Tandoi Francesco; Gavrila Caius; Salizzoni Mauro; Pinna Antonio D; Cillo Umberto; Gridelli Bruno; De Carlis Luciano G; Colledan Michele; Gerunda Giorgio E; Costa Alessandro Nanni; Strazzabosco Mario

来源：Digestive and Liver Disease, 2014, 46(4): 340-347.

DOI：10.1016/j.dld.2013.11.004

• Donor-recipient match
• Donor Risk Index
• Graft failure
• Hepatitis C
• Risk factors
• Transplantation outcome

Background: To generate a robust predictive model of Early (3 months) Graft Loss after liver transplantation, we used a Bayesian approach to combine evidence from a prospective European cohort (Liver-Match) and the United Network for Organ Sharing registry. %26lt;br%26gt;Methods: Liver-Match included 1480 consecutive primary liver transplants performed from 2007 to 2009 and the United Network for Organ Sharing a time-matched series of 9740 transplants. There were 173 and 706 Early Graft Loss, respectively. Multivariate analysis identified as significant predictors of Early Graft Loss: donor age, donation after cardiac death, cold ischaemia time, donor body mass index and height, recipient creatinine, bilirubin, disease aetiology, prior upper abdominal surgery and portal thrombosis. %26lt;br%26gt;Results: A Bayesian Cox model was fitted to Liver-Match data using the United Network for Organ Sharing findings as prior information, allowing to generate an Early Graft Loss-Donor Risk Index and an Early Graft Loss-Recipient Risk Index. A Donor-Recipient Allocation Model, obtained by adding Early Graft Loss-Donor Risk Index to Early Graft Loss-Recipient Risk Index, was then validated in a distinct United Network for Organ Sharing (year 2010) cohort including 2964 transplants. Donor-Recipient Allocation Model updating using the independent Turin Transplant Centre dataset, allowed to predict Early Graft Loss with good accuracy (c-statistic: 0.76). %26lt;br%26gt;Conclusion: Donor-Recipient Allocation Model allows a reliable donor and recipient-based Early Graft Loss prediction. The Bayesian approach permits to adapt the original Donor-Recipient Allocation Model by incorporating evidence from other cohorts, resulting in significantly improved predictive capability.

• 出版日期2014-4