INTRODUCTION

Graft outcome is one of the factors that most conditions the subsequent evolution of patients undergoing a liver transplantation. This risk is carefully assessed by transplant teams and depends on donor-related variables, even more so when organ donors are managed using expanded criteria. Furthermore, variables associated with the donation-transplant process, which may differ from one program to the next, are also assessed. Finally, recipient-related variables are also studied, even more so when previously defined transplant indication limits are extended for new recipients. Therefore, indicators have been established to assess both liver graft and recipient survival post-transplantation. These include: MELD and CTP ¹, NDDF ², SOLD ³, DRI ⁴, SOFT ⁵, D-MELD ⁶, BAR ⁷, ET-DRI ⁸, NN-CCR and NN-MS ⁹, NNtop15 ¹⁰, and DLI ¹¹ (Supplementary Table 1). However, significant differences exist between donation-transplantation programs in different countries and organizations, both with regard to donor characteristics and transplant outcome (Supplementary Table 2). These differences are most remarkable for donor age and ethnicity, number of brain-dead donors (DBDs) vs cardiac-dead donors (DCDs), causes of brain death and liver bipartition, among others. All the above mean that donors and recipients are quite different between countries and/or regions. Hence a region/site specific scoring system would be more appropriate for liver graft acceptance and allocation.

Supplementary Table 1 Survival indicators in liver transplantation 

Reference: S: single-center study; MS: multicenter study; UNOS: United Network for Organ Sharing; SRTR: Scientific Registry of Transplants Recipients ET: Eurotransplant; UKTR: United Kingdom Transplant Registry. *Combined transplants and re-transplants excluded. †Re-transplants, acute liver failure, and donation after cardiac death excluded. ‡Split, living donor, and combined transplants excluded. §Recipients < 16 years, heterotopic or auxiliary liver and blood incompatibility excluded.

Study: MELD: model for end-stage liver disease; CTP: Child-Turcotte-Pugh score; NNDF: newly derived discrimination function; SOLD: score of liver donor; DRI: donor risk index; SOFT: survival outcome following liver transplantation; D-MELD: donor-MELD; BAR: balance of risk score; ET-DRI: Eurotransplant donor risk index; NN-CCR: neural network-correct classification rate; NN-MS: neural network-minimum sensitivity; DLI: donor liver index.

Results: ǁExcept when indicated, AUC-ROC (95% CI): area under the ROC curve (95% confidence interval); ¶Linear regression OR (95% CI); **Adjusted survival; ††Kaplan-Meier.

Supplementary Table 2 Activity at OPTN, ET, and ON during 2016* 

DBD: donor brain dead; DCD: donor cardiac dead. Data in absolute value (%). *Based on: OPTN, Organ Procurement and Transplantation Network. Available from: https://optn.transplant.hrsa.gov/data/; ET : Eurotransplant. Available from: http://statistics.eurotransplant.org/; ONT: Organización Nacional de Trasplantes. Available from: http://www.ont.es/infesp/Paginas/Memorias.aspx; Newsletter Transplant 2017. Available from: http://www.ont.es/publicaciones/Documents/. †OPTN2011-2014: mean and 95% confidence interval; ONT2016: median and interquartile range. ‡Data < or ≥ 60 years.

The Registro Español de Trasplante Hepático (RETH) records the experience with liver transplantation in Spain, including systematic data, for all liver transplants.

Therefore, it is a tool that can be used to gain insight into our own experience. The goal of the study was to describe a new liver graft risk indicator based on the RETH results. The indicator combined donor, process and recipient related factors to facilitate decision making on organ acceptance and allocation to specific patients. Attempts were also made to validate and compare this indicator with the American DRI and European ET-DRI in our setting.

MATERIAL AND METHOD

GRI description

The study methodology for the design and validation of the GRI is summarized in Table 1:

• The RETH report includes a Cox proportional hazards regression analysis of the factors associated with overall graft survival. Results are expressed as CoxPH (RR) values with their corresponding 95% confidence intervals (95% CI) (Table 1A).

• Based on the Cox regression equation [ln (λt) = a + b1x1 + b2x2 + ... + bnxn], an indicator may be defined as an exponential (inverse logarithm) of a linear risk score. Three indicators were defined by grouping together variables with statistically significant differences (SSDs) according to the multivariate analysis: donor risk index (dRI), recipient risk index (rRI) and graft risk index (GRI), where GRI = dRI x rRI (Table 1B).

• The new indicator (GRI) was validated and compared to the American DRI and European ET-DRI in our series (Table 1C).

Two events were considered for indicator assessment:

• Event 1. Graft survival (GS): defined as the time elapsed from the transplant to either re-transplantation or recipient death from any cause, whichever comes first.

• Event 2. Failure-free graft survival (ffGS): defined as the time elapsed from the transplant date to the either re-transplantation or recipient death with an associated chronic graft dysfunction, whichever comes first.

Table 1 GRI description: variables included (1A), formula (1B) and examples in comparison with DRI and ET-DRI (1C) 

*RETH Results Report (Memoria de Resultados del RETH), available from: http://www.sethepatico.org. RR: risk ratio; CI: confidence interval; NS: not significant; AS: acute stroke; SALF: severe acute liver failure; TBI: traumatic brain injury; ICU: intensive care unit; UNOS: United Network for Organ Sharing; HCV: hepatitis C virus.

Example of reference in boldface (Risk 1); variables modified from reference in italics. In DRI and ET-DRI calculations variables not included in the GRI were considered to have a value of 0 (risk 1).

RESULTS

Validation series

Data from 600 liver transplants performed in our center were used for the validation of the new indicators, DRI and ET-DRI. The data with regard to the validation series are listed in Table 2.

Table 2 Principal cohort characteristics 

HCV: hepatitis C virus; HIV: human immunodeficiency virus; MELD: model for end-stage liver disease; UNOS: United Network for Organ Sharing; HCC: hepatocellular carcinoma; HBV: hepatitis B virus. *Quantitative variables are expressed as: mean ± standard deviation, median (interquartile range), and range. Qualitative variables are expressed as n (%).

AST: aspartate aminotransferase; ALT: alanine aminotransferase; GGT: gamma glutamyl transpeptidase; HBcAb: hepatitis B virus core antibody; CMV: cytomegalovirus; UW: University of Wisconsin. Location: local, in the city; regional, < 200 km; national, > 200 km. *Quantitative variables are expressed as: mean ± standard deviation, median (interquartile range), and range. Qualitative variables are expressed as n (%).

Clinical outcome variables categorized as in the RETH. DDG: died with dysfunctioning graft; DFG: died with a functioning graft. *Quantitative variables are expressed as: mean ± standard deviation, median (interquartile range), and range. Qualitative variables are expressed as n (%). Kaplan-Meier: mean (95% confidence interval). †Tweny-one patients died while on the re-transplantation waiting list.

GRI logistic function

A Hosmer-Lemeshow analysis showed that the observed probabilities were similar to those expected for the three indicators: DRI (p = 0.883), ET-DRI (p = 0.317) and GRI (p = 0.210). Figure 1 shows the logistic equation curve and the values observed by GRI interval. A GRI of 1 represents a graft loss likelihood of 23.25%; each point increase in the GRI multiplies this graft loss probability by 1.33 (95% CI: 1.24-1.44).

Fig. 1 GRI logistic function: predicted event probability and 95% CI. 

Indicator comparisons: values, graft survival and hazard ratios by stratification

No statistically significant differences in the mean DRI were seen between groups with and without graft survival: 1.54 (95% CI: 1.50-1.58) vs 1.60 (95% CI: 1.55-1.65), respectively. Moderate statistically significant differences were obtained for ET-DRI: 1.51 (95% CI: 1.47-1.55) vs 1.60 (95% CI: 1.56-1.63), p < 0.05, respectively.

Finally, GRI showed highly statistically significant differences between the groups: 4.27 (95% CI: 4.06-4.49) vs 6.46 (95% CI: 5.98-6.93), p < 0.0001.

Differences in graft survival between higher and lower indices are clearly seen for all three indicators. However, GRI had the best discriminating power for survival among intermediate groups of the three indicators. Furthermore, survival diminished progressively down to a GRI ≥ 9, which represents a 1-year graft survival rate of 47.5% (95% CI: 34.6-59.3) and there were highly significant differences versus the standard group (HR: 4.03; 95% CI: 2.63-6.19; p < 0.0001) (Table 3).

Table 3 Adjusted 1-, 3-, and 5-year graft survival and hazard ratio (HR) according to index stratification 

*% events over "n of subgroup". NS: not significant.

ROC curves

The area under the curve (AUC) according to the ROC analysis was 0.54 (95% CI: 0.50-0.59) for DRI and 0.56 (95% CI: 0.51-0.61) for ET-DRI. Thus, their ability to predict graft survival was not above the level of chance. The AUC for GRI was 0.70 (95% CI: 0.65-0.73) with highly significant differences when compared to the other two indices (DeLong test, p < 0.0001) (Fig. 2). With regard to the power of GRI to predict transplant futility, the fact that a GRI score ≥ 10 always forecasted graft loss should be highlighted. The specificity was 100%, the sensitivity was 14.4%, the positive predictive value was 100% and the negative predictive value was 54.2%.

All the results obtained for event 2 (ffGS) are superimposable on those described for event 1 (GS) (unpublished data).

Fig. 2 ROC curves for GRI, DRI and ET-DRI. 

DISCUSSION

Classic indicators to assess cirrhosis prognosis (Child-Pugh) or transplant waiting list mortality (MELD) are not suitable to predict post-transplant patient survival ^1,2. The latest indicators based on neural networks are complex, rely on limited experience, are scarcely generalizable and are also difficult to extrapolate to the daily clinical practice ^9,10. Finally, external validations for some indicators (DRI, MELD, D-MELD and SOFT [7], DRI and ET-DRI [13], DRI and SOFT [10]) do not confirm their validity when applied to scenarios different to their principal intended use, as is the case in our study. The latter point likely results from large differences between countries and programs, in terms of donor population and donation-transplantation process characteristics and not so much with recipient characteristics.

Donor-related variables may be linked to specific graft features and general donor characteristics. Most indicators described thus far share a number of variables including age, cause of brain death, etc. Furthermore, there are some variables where the justification or relation to graft or patient survival does not seem to be based on clear scientific grounds, such as smoking, height, etc. However, even though steatosis is a primary determinant of post-transplant liver function ¹⁴ (particularly within the first year), none of the established indicators include this variable, either directly or indirectly. Kulik et al. ¹⁵ highlighted in a recent study that allografts with a fatty liver are a significant cause of primary graft failure and an excessive mortality after transplantation. Traditionally, high donor sodium levels have been associated with graft failure and primary non-function (PNF) ¹⁶. A study by Sirivatanauksorn et al. ¹⁷ showed that high donor sodium was associated with marginal liver grafts, but only ALT > 65 IU/l was associated with higher PNF rates. More recently, Al-Freah et al. ¹⁸ did not find any donor variables that were associated with PNF, except for donor-recipient ABO mismatch. Finally, these indicators should contemplate specific donor variables. Some of these are considered on an international level under the expanded criteria donor concept. These are also included in Spain in the non-standard risk donor (Donantes de Riesgo no Estándar [DRNE]) category and may include drug abuse, intoxications, tumors, infections, hepatitis B or C, age above 65 years, etc.

With regard to recipient-related variables, allocating a specific organ to its most appropriate recipient is a difficult decision, particularly in certain situations. We agree with Feng et al. ⁴ that grafts are ideally a highly homogeneous group and thus an optimal organ may be allocated to any person, whereas non-ideal grafts make up a highly heterogeneous group with a high risk spectrum. In the latter case, allocating a suboptimal organ to a high-MELD recipient ^19,20 with certain conditions ²¹ or with a higher waiting-list mortality rate ²² gambles with the risks of post-transplant survival versus wait-list mortality.

Finally, with regard to variables related with the donation-transplantation process, the most established indicators include cold ischemic time (CIT) in their design. CIT may in turn be dependent on donor location, available transportation means, surgeon skill and unforeseeable events associated with the surgical technique, etc. A good indicator should already be available at the time that an organ is offered and should not include variables that measure equal or similar factors (e.g., donor location or CIT). Perhaps decision making in medicine in the future, particularly in transplant medicine, will be based on complex artificial neural networks ²³. The analysis of countless variables (big data) and numerous interdependent networks means that we will learn, decide and maybe even get it right or wrong. In the meantime, we still believe that a solution to the donor-recipient matching problem may be facilitated by these types of tools and it will not entail expunging the human component of awareness, conscience and sensitivity.

The present study has a number of limitations: the above indicator is based on a non-predictive analysis of a national registry with over 22,000 transplants; it was validated with a limited series in only one center; the actual weight of certain variables in graft survival will have to be refined and updated (HCV is the most significant instance) and these indices are intended to simplify complex situations dependent on multiple factors, which often reduces quantitative variables to categorical variables.

To conclude, our study highlights that neither DRI nor ET-DRI seem to adequately predict graft risks in our setting. A national GRI might be a very useful tool to categorize graft-recipient matching. Hence, a national study is needed to regularly streamline and update this indicator and to provide a wider validation thereof.