RESUMO
A recent study using US national registry data reported, using Cox proportional hazards (PH) models, that split-liver transplantation (SLT) has improved over time and is no more hazardous than whole-liver transplantation (WLT). However, the study methods violated the PH assumption, which is the fundamental assumption of Cox modeling. As a result, the reported hazard ratios (HRs) are biased and unreliable. This study aimed to investigate whether the risk of graft survival (GS) in SLT has really improved over time, ensuring attention to the PH assumption. This study included 80,998 adult deceased donor liver transplantation (LT) (1998-2015) from the Scientific Registry Transplant Recipient. The study period was divided into 3 time periods: era 1 (January 1998 to February 2002), era 2 (March 2002 to December 2008), and era 3 (January 2009 to December 2015). The PH assumption was tested using Schoenfeld's test, and where the HR of SLT violated the assumption, changes in risk for SLT over time from transplant were assessed. SLT was performed in 1098 (1.4%) patients, whereas WLT was used in 79,900 patients. In the Cox PH analysis, the P values of Schoenfeld's global tests were <0.05 in all eras, which is consistent with deviation from proportionality. Assessing HRs of SLT with a time-varying effect, multiple Cox models were conducted for post-LT intervals. The HR curves plotted according to time from transplant were higher in the early period and then decreased at approximately 1 year and continued to decrease in all eras. For 1-year GS, the HRs of SLT were 1.92 in era 1, 1.52 in era 2, and 1.47 in era 3 (all P < 0.05). In conclusion, the risk of SLT has a time-varying effect and is highest in the early post-LT period. The risk of SLT is underestimated if it is evaluated by overall GS. SLT was still hazardous if the PH assumption was considered, although it became safer over time.