Objectives: The utilization of liver allografts could be optimized if nonacceptance is predicted. This study aimed to evaluate the prognostic ability of an updated Discard Risk Index in Eurotransplant.
Materials and Methods: Potential deceased donors from January 2010 to December 2015 who had been reported to Eurotransplant were included in our analyses. Liver utilization was defined by transplant status as the primary outcome to evaluate the performance of the Eurotransplant-developed Discard Risk Index.
Results: Of 11?670 potential livers, 9565 (81%) were actually transplanted. Donor sex, age, history of diabetes, drug abuse, use of vasopressors, body mass index category, serum sodium, cause of death, donor type, and levels of C-reactive protein, bilirubin, aspartate and alanine aminotransferases, international normalized ratio, and gamma-glutamyl transpeptidase were associated with discard and combined in the Eurotransplant-developed Discard Risk Index. Correlation between the two Discard Risk Indexes was high (r = 0.86), and both achieved high C statistics of 0.72 and 0.75 (P < .001), respectively. Despite strong calibration, discard rates of 0.8% for overall donors and 6% of donors after circulatory death could be predicted with 80% accuracy.
Conclusions: The Eurotransplant-developed Discard Risk Index showed a high prognostic ability to predict liver utilization in a European setting. The model could therefore be valuable for identifying livers at high risk of not being transplanted in an early stage. These organs might profit the most from modified allocation strategies or advanced preservation techniques.
Key words : Liver transplantation, Organ allocation, Organ procurement
Because of the shortage of available liver allografts, wait list mortality is an important issue in liver transplant. In 2015, 2589 patients were listed for liver transplant and almost 600 patients (20%) were delisted or died while waiting for transplant in the Eurotransplant region. In that same year, approximately 20% of all livers that were reported for allocation were not used for transplant.1
To improve the efficiency of liver utilization, it would be useful to predict which livers will be discarded. Some of the reasons for discarding organs may be modified or better assessed during the allocation phase. For example, a modifiable risk factor is cold ischemic time, which could be minimized by changing allocation algorithms.2 It can also be an useful tool for the (selective) use of advanced preservation techniques like normothermic regional perfusion3-5 or machine perfusion.6 These techniques might especially be useful for high-risk organs to assess their function before transplant and potentially improve it.
To identify high-risk livers in an early stage, only factors known at the time of offering can be used to indicate which livers are at risk of being discarded. Such an effort has been made by Rana and colleagues7 by developing the Discard Risk Index (DSRI). The DSRI includes 15 factors that are associated with liver utilization: donor type (donation after circulatory death [DCD] or donation after brain death [DBD]), age, body mass index (BMI; calculated as weight in kilograms divided by height in meters squared), Centers for Disease Control (CDC) high risk, cause of death, race, sex, hepatitis B core antibodies (HBcAb) status, hepatitis C virus antibody (HCVAb) status, history of diabetes, history of hypertension, and latest laboratory values (sodium, aspartate aminotransferase [AST], alanine aminotransferase [ALT], and total bilirubin). The DSRI had a reported area under the receiver operating characteristic (AUROC) curve of 0.80 in the United Network for Organ Sharing (UNOS) database. This was internally validated in a cohort within the same region.
In this study, our aim was to validate the prognostic ability of a Eurotransplant-updated DSRI (ET-DSRI) and to analyze factors associated with the acceptance of livers for transplant in a European setting to further improve the predictive performance.
Materials and Methods
No ethical or patient consent statement was required according to national ethical guidelines. Potential deceased liver donors from January 1, 2010 to December 31, 2015 who had been reported to Eurotransplant were included in our analyses. Excluded from our study were those potential donors from countries not participating in Eurotransplant, donors <10 years of age, donors who had withdrawn or were without any consent for liver donation, donors with malignancies found at procurement or during transplant, donors who had no organs transplanted, and donors who were DCDs with an agonal phase >1 hour (at an agonal phase >1 hour, the liver is considered not viable for transplant in Eurotransplant).8 We excluded these donors to ensure a group of potential livers donors without absolute contraindications for transplant. Liver donations that were not allocated for reasons other than those described above were also included in the study population. This was done to evaluate the true potential number of livers and to minimize a potential pre-reporting selection bias in our analyses.
Data included in the model
For continuous variables, missing variables were imputed by the median value for gamma-glutamyl transpeptidase (GGT; n = 258, 2% missing, median 42 U/L), serum sodium (n = 68, 1% missing, median 147 mmol/L), AST (n = 168, 1% missing, median 47 U/L), ALT (n = 80, 1% missing, median 33 U/L), bilirubin (n = 286, 2% missing, median 0.5850 mg/dL), international normalized ratio (INR; n = 1337, 11% missing, median 1.15), and C-reactive protein (CRP; n = 718, 1% missing, median 110 mg/L). All laboratory values were last values known before transplant. Categorical variables were considered absent when missing, not tested, or unknown. This applied to a medical history of smoking (n = 1493, 13%), drug abuse (n = 3750, 32%), and (treated) malignancies (n = 6072, 52%). For factors that were already incorporated in the DSRI, similar cutoff values for continuous variables were used in developing the ET-DSRI.
The primary outcome of this study was liver utilization, defined as the organ being either trans-planted or not transplanted. The DSRI was calculated for all included donors as previously described by Rana and colleagues.7 The factors race, CDC high risk, and history of hypertension were not available and therefore set at references values (no CDC high risk, not African American, and no history of hypertension). In Eurotransplant, race is not registered for ethical and legal reasons, whereas CDC high risk and a history of hypertension are not standardly collected.9
Reasons for discarding procured livers
For all livers that are procured but not transplanted, a form is filled out at the Eurotransplant Allocation Department and is registered in the electronical donor log. The form and the donor log include the reason for discarding, location where the organ was sent, and the name of the doctor or transplant center involved. Both sources were analyzed for all organs that were discarded (anonymized for doctor and transplant center).
The allocation process of donors was visualized in a flow diagram, and utilization was evaluated per year and by donor country. The data were randomly assigned in a training (75%) and validation test set (25%). Risk factors for liver utilization were identified in a multivariable logistic regression analysis with backward selection by Akaike Information Criterion (AIC) in the training set. From these results, the ET-DSRI model was developed to predict liver utilization. The correlation between the DSRI and ET-DSRI was evaluated by Pearson test. Subsequently, the performance of both models was compared by discrimination and calibration. Discrimination was defined by the AUROC curve. Calibration was analyzed with the Hosmer-Lemeshow test to test for goodness of fit for logistic regression models. The test assesses whether the observed event rates match expected event rates.
For both models, separate analyses were done for all donors and for DBDs and DCDs. Risk groups were defined using increments of 10% in the quantiles of the risk scores. Reasons for discarding procured livers were also analyzed.
Median values of continuous variables were compared with a Kruskal-Wallis test, and categorical variables were compared with chi-square tests. Kaplan-Meier curves were analyzed by log-rank testing. P < .05 was considered statistically significant, and all analyses were done with SPSS version 24.0 and R software version 3.3.1.
During the study period, 14?253 donors were reported to Eurotransplant; of these, 11?760 donors (83%) were included in the analyses. Inclusion and exclusion criteria and the subsequent allocation process are schematically shown in Figure 1. Eligible donors had a median donor age of 54 years and about 10% were DCDs. However, this 10% overall rate for DCDs varied significantly between countries because DCD procedures are only legally allowed in The Netherlands, Belgium, and Austria. Overall, the highest (absolute) number of donors was reported by Germany followed by Belgium, The Netherlands, and Austria (Table 1).
Of all included livers, 81% (9565/11?760) were used for transplant. As shown in Table 1, transplanted livers versus nontransplanted livers were younger (54 vs 56 years old; P < .001), less often from DCDs (6% vs 26%; P < .001), less often with a history of diabetes (9% vs 12%; P < .001), and had significantly lower laboratory values (AST, ALT, and GGT; P < .001). Overall utilization rate decreased from 84% in 2010 to 80% in 2015 over the study period (P < .001; Figure 2, top left). Significant differences in utilization were also observed between countries (P < .001; Figure 2, top right). Overall, utilization varied from around 90% in Germany to 55% in Hungary. However, the use of DCDs had a significant influence. When only DBDs were considered, overall utilization in The Netherlands and Belgium increased from 63% to 89% and from 84% to 87%, respectively (Figure 2, bottom).
Risk factor analysis and development of the Eurotransplant Discard Risk Index
From the statistical analysis (multivariable logistic regression analysis with backward selection by AIC), the following donor factors were included in the model to predict nonutilization: male sex, higher donor age category, history of diabetes, malignancy, drug abuse, use of vasopressors, BMI category, serum sodium (>160 mmol/L), cause of death category, DCD, a lower CRP and a higher bilirubin, and levels of AST, ALT, INR, and GGT. These factors associated with liver utilization were combined in the ET-DSRI model (Table 2 and Figure 3).
Discriminative values of the Discard Risk Index and the Eurotransplant Discard Risk Index
The DSRI and ET-DSRI scores were distributed normally both in the training and in the validation test set. The correlation between both scores was relatively high (r = 0.86). In the training set, the DSRI achieved an AUROC of 0.73. This was significantly lower than for the ET-DSRI, which achieved an AUROC of 0.77 (P < .001; Figure 4, left). In the validation set, the AUROCs for DSRI and ET-DSRI were 0.72 and 0.75 (P < .007), respectively (Figure 4, right). In subset analysis of DBDs in the validation set, the DSRI and the ET-DSRI achieved AUROCs of 0.68 and 0.70 (P = .014), respectively. In the DCD analysis, AUROCs of 0.69 and 0.67 (P = .695) were observed in the validation set for DSRI and ET-DSRI, respectively.
Calibration of the Discard Risk Index and the Eurotransplant Discard Risk Index
The logistic curve indicates the relation between the estimated outcome (discard) based on the models’ scores and predicted outcomes. For the DSRI and the ET-DSRI, this is shown in Figure 5. We observed a better calibration for the ET-DSRI, especially with the higher risk scores. However, both models tended to overestimate the chance of nonutilization, as indicated by a statistically significant Hosmer-Lemeshow test for the DSRI (P < .001) and for the ET-DSRI (P = .01). Overestimation seems especially to be apparent in the upper 10%. When this subgroup was excluded, the ET-DSRI was well calibrated (P = .56), whereas the DSRI still had a statistically significant calibration error (P < .001). Separate analyses for DBDs (Figure 6) and DCDs (Figure 7) were also performed. In the DBD population, the DSRI performed slightly better than in the overall population but still was not calibrated well (P = .03). However, the ET-DSRI showed good calibration (P = .11) in the DBD population. In the DCD population, both the DSRI (P = .37) and the ET-DSRI (P = .26) estimated utilization adequately. Despite the relatively high calibration, identifying a group of donors that will be discarded with high accuracy is only possible for a small percentage of all donors because only 20% of donors are discarded. In the donors with the highest 10th percentile ET-DSRI scores, the observed probability of discarding did not exceed 60%. For 0.8%, 2%, and 4% of all donors in the validation set, discarding of livers could be predicted with the ET-DSRI with only 80%, 70%, and 60% accuracy, respectively. This could be improved in the DCD subset, where overall discard rate was higher. In this selection, liver discarding could be predicted with 80%, 70%, and 60% accuracy in 6%, 20%, and 36% of all donors, respectively.
Reasons for discarding of organs
In the study period, 485 of 11?760 livers (4%) were procured but not transplanted. For 442 (91%) of these livers, there was at least 1 reason registered for discarding of the organ (Table 3). Organs were most frequently discarded for organ-specific reasons like steatosis and/or fibrosis (60%) or (expected) long cold ischemic time (11%). Procurement-related injuries were also mentioned for discarding of livers (3%).
The decision to decline a liver for transplant may be simple for organs with absolute contraindications; however, decisions can be more complicated for extended criteria livers. Such organs may be considered less suitable for transplant in one transplant center but acceptable for another. Such decisions are not always objective and may be influenced by recent (personal) experiences, general beliefs, or local protocols. This study has objectified the process of accepting a liver for transplant. This enables us to assist in the allocation process of a specific group of high-risk livers and make extra efforts to further optimize their use.
Our results identified 15 factors that are associated with liver utilization in Eurotransplant. These factors were combined in the ET-DSRI. The prognostic performance of this model can be considered good for a clinical model10 with an AUROC of 0.75 and is significantly higher compared with 0.72 for the (original) DSRI by Rana and colleagues in the validation set. Factors that were in the DSRI but not in the ET-DSRI included HCVAb and HBcAb. The higher prevalence of hepatitis C in the United States and lower numbers in this study compared with those shown in Rana and colleagues may explain why hepatitis in the European setting was not confirmed as a factor associated with utilization.7,11,12 This might also explain why hepatitis B was not included in the ET-DSRI despite a higher prevalence of hepatitis B in Europe.13,14 Factors that were included in the ET-DSRI but not in the DSRI were GGT, INR, lower CRP, history of drug abuse, and use of vasopressors.
Our results indicated that significant differences exist between factors associated with the acceptance of livers and factors associated with posttransplant outcomes. This is interesting because the decision to accept or decline livers ought to be based on their expected function after transplant. Well-known models that aim to predict outcomes after liver transplant, such as the donor risk index from Feng and associates,15 the Eurotransplant donor risk index,2 the Survival Outcomes Following Liver Transplantation score,16 the Balance of Risk score,17 and a previous donor-recipient model,18 have not included factors like high transaminases, high bilirubin, and a medical history of drug abuse. In addition, studies on the effects of some of these factors have not found an impact on posttransplant outcomes. This applies for example for dopamine (vasopressor) in the donor,19 a history of drug abuse,20,21 and recipient sex.22 The differences are most likely a result of the selection process that takes place before transplant. Because organs with certain risk factors are not accepted for transplant, these risk factors are not present anymore in outcome analyses. Models based on datasets of transplanted livers are therefore less suitable to predict liver utilization.
Interestingly, the utilization rate of available donors has decreased during our study period (from 84% to 80%). Stricter acceptance criteria may explain this development, although an overall increase of donors with more risk factors seems to be more likely driving this development.23-25 This has previously been shown for donor age26 and steatosis,25 but also the number of DCDs has increased significantly. Donation after circulatory death is one of the explanations for significant differences in utilization between Eurotransplant countries. Although DCD is also practiced in Austria, it is mostly done in The Netherlands and Belgium. In these countries, liver transplants from DCDs increased from 16 to 71 (12% to 42%) and from 23 to 79 (11% to 30%) in 2010 and 2019, respectively.27 Because of higher discard rates for DCDs,28,29 The Netherlands and Belgium were in the highest utilization range in a DBD subset analysis. Even then, significantly low utilization rates were observed in Hungary and in a lesser degree in Austria. It is difficult to specifically address one issue to explain this due to the assumed multifactorial nature. It seems unjust to suggest these countries consider stricter acceptance criteria as no distinction was made where the organ was transplanted (own country or abroad). Logistical reasons seem more likely to explain the low utilization rate. Because of the geographical location and limited flight options in the evening and at night, potential acceptances in bordering countries are more complicated for Hungary and also for Austria due to expected cold ischemic times. The use of the ET-DSRI could be useful in this matter, as (private) transport options can be on standby if high ET-DSRI organs are offered.
With regard to all reasons for discarding a liver that has already been procured, steatosis and/or fibrosis of the liver was the most frequently mentioned. This factor is important for outcomes after transplant30,31 but is usually not well documented in the information that is available at time of the offer. To have this information, a biopsy still seems to be the gold standard over other noninvasive modalities.32-34 In high-risk livers, such biopsies may provide valuable information for transplant centers interested in marginal organs and may avoid procurement of livers of unacceptable quality.35 The ET-DSRI can be helpful to identify these high-risk livers.
In this study, the DSRI showed a lower predictive ability than in the original study, which had included data from the UNOS region. This is likely influenced by the significant differences between both regions with regard to characteristics of livers reported for allocation7 and in transplanted livers.36 With regard to characteristics of livers reported for allocation, Rana and colleagues reported a median donor age of 42 years in the UNOS database (US region) compared with 53 years old in the Eurotransplant region. Other factors such as diabetes (12% vs 9%), HCVAb (5% vs 1%), higher BMI (28 vs 26), and higher DCD rate (11% vs 10%) were more frequently present in donors from the United States. With regard to characteristics of transplanted livers, differences between UNOS and Eurotransplant were observed in donor age (41 vs 54 years old), diabetes (11% vs 9%), BMI (27 vs 25), DCD (5% vs 6%), and female sex (40% vs 47%). The distinct differences between the United States and Europe may be caused by regulations on center-specific outcomes in the United States and/or by epidemiological differences. Center-specific policies may discourage the acceptance of marginal organs for transplant, and epidemiological differences may be influenced by the opioid crises,37,38 the rate of obesity,39 and the higher rate of homicides40 that seem to be more apparent in the United States. Regardless of the exact mechanism, differences in acceptance criteria contributed to the DSRI achieving a lower predictive performance in a European setting. In addition, the prognostic performance of the DSRI may be impaired by the unavailability of 3 factors that were incorporated in the DSRI in our data set: donor race, CDC risk, and history of hypertension.
An important limitation in the study of Rana and colleagues (as well in this study) is the unavailability of biopsy results in the dataset.7 Level of GGT, identified as a risk factor for liver utilization in our study, could be of interest in this matter. This factor has been shown to be associated with outcome2 and liver acceptance and has an association with (liver) steatosis.41
Within Eurotransplant, decisions on which donor organs are suitable for allocation are made in close collaboration with all parties involved in transplantation. Such a decision is likely subjected to the local or national experience with transplanting extended criteria organs, the donors per million inhabitants, and number of patients on wait lists (relative availability). To avoid the loss of potentially transplantable livers in the process of donor reporting, we suggest that all livers, even those with a low chance of acceptance, should be reported for allocation. Especially for these livers, the ET-DSRI might be useful to prevent organ loss. Additional measures could be undertaken, including (1) biopsy results being known at the time offering, (2) modification of allocation algorithms, and (3) the (selective) use of advanced preservation techniques. Biopsy results known at the time of offering could provide crucial additional information and may prevent transplant centers from declining an organ in a (too) late phase of the allocation process.25,35,42 Second, a more aggressive mode of offering a high-risk organ would allow more centers to consider the offer and could prevent additional cold ischemic time. Finally, these organs represent a group that might benefit the most from the use of (expensive) advanced preservation techniques.43 The risk of transplant may be mitigated by assessing their function pretransplant and could decrease the harmful effects of ischemic injury. With such measures, the use of available livers might be maximized to further decrease wait list mortality.
The ET-DSRI showed higher prognostic ability to predict liver utilization in a European (Eurotransplant) setting compared with the DSRI. This model was shown to be a valuable tool for identifying livers at high risk of not being transplanted in an early stage. This model can be used to identify organs that require efforts for additional information and could be used to select organs that may profit the most from modified allocation strategies or advanced preservation techniques.
Volume : 19
Issue : 11
Pages : 1163 - 1172
DOI : 10.6002/ect.2021.0228
From the 1Medical Staff Office, 4Allocation Duty Office, and 5Medical Director Office, Eurotransplant International Foundation, Leiden, The Netherlands; the 2Department of Surgery, Division of Transplantation, and the 3Department of Statistics, Leiden University Medical Center, Leiden, The Netherlands; the 6Department of Surgery, Division of Transplantation, Medical University of Vienna, Vienna, Austria; and the 7Department of General, Visceral, Transplantation, Vascular and Thoracic Surgery, Hospital of the University of Munich, Munich, Germany
Acknowledgements: The authors have not received any funding or grants in support of the presented research or for the preparation of this work and have no declarations of potential conflicts of interest.
Corresponding author: Jacob D. de Boer, Medical Staff Eurotransplant/LUMC, Leiden, The Netherlands
Phone: +31 (0)6 12 45 64 60
Table 1. Demographics of Eligible Donors by Transplant Status
Figure 1. Flow Diagram of the Allocation Process
Figure 2. Utilization of Reported Livers
Table 2. Multivariable Logistic Regression Analysis Results With Backward Selection by Akaike Information Criterion Included in the ET-DSRI(Training Set)
Figure 3. Formula for the Eurotransplant Discard Risk Index Model
Figure 4. Analysis of Area Under the Receiver Operating Characteristic Curve for the Discard Risk Index and the Eurotransplant Discard Risk Index
Figure 5. Calibration of the Validation Dataset
Figure 6. Calibration of the Discard Risk Index and the Eurotransplant Discard Risk Index in the Validation Dataset for Donations After Brain Death
Figure 7. Calibration of the Discard Risk Index and Eurotransplant Discard Risk Index in the Validation Dataset for Donations After Circulatory Death
Table 3. Reasons for Discarding Accepted and Procured Livers (N = 485)