It is critical to balance waitlist mortality against posttransplant mortality.
Our objective was to devise a scoring system that predicts recipient survival at 3 months following liver transplantation to complement MELD-predicted waitlist mortality.
Univariate and multivariate analysis on 21 673 liver transplant recipients identified independent recipient and donor risk factors for posttransplant mortality. A retrospective analysis conducted on 30 321 waitlisted candidates reevaluated the predictive ability of the Model for End-Stage Liver Disease (MELD) score.
We identified 13 recipient factors, 4 donor factors and 2 operative factors (warm and cold ischemia) as significant predictors of recipient mortality following liver transplantation at 3 months. The Survival Outcomes Following Liver Transplant (SOFT) Score utilized 18 risk factors (excluding warm ischemia) to successfully predict 3-month recipient survival following liver transplantation.
This analysis represents a study of waitlisted candidates and transplant recipients of liver allografts after the MELD score was implemented. Unlike MELD, the SOFT score can accurately predict 3-month survival following liver transplantation. The most significant risk factors were previous transplantation and life support pretransplant. The SOFT score can help clinicians determine in real time which candidates should be transplanted with which allografts. Combined with MELD, SOFT can better quantify survival benefit for individual transplant procedures.
The MELD (Model for End-Stage Liver Disease) scoring system (1,2) has transformed liver allograft allocation in the United States since it was implemented for prioritization of transplant candidates in 2002 (3). The MELD score is an accurate predictor of waitlist mortality, as demonstrated in the pioneering study by Wiesner et al. (4), with a c-statistic (5) of 0.83 when used to predict 3-month mortality of candidates on the waitlist. The score substituted effectively for candidate stratification based on subjective assessment. However, the MELD score is a poor predictor of mortality following transplantation (4,6,7). This observation was confirmed by Desai et al. in their analysis, which reports a c-statistic of only 0.54 with the use of the MELD to predict 3-month recipient mortality following liver transplantation (7). When mortality of recipients on the waitlist is compared with the highest and the lowest MELD scores, there is a 300-fold difference, in contrast to the 2-fold difference in survival of patients after liver transplantation (8). Methods other than the MELD score, such as the Child–Pugh score, also had a poor ability to predict posttransplant survival (6). The inability of existing methods to predict posttransplant survival prevents clinicians from effectively selecting potential recipients for transplantation from MELD scores alone.
Because candidate factors alone are not predictive of survival following transplantation, a new model is required to accurately predict posttransplant survival. The lack of consideration of donor risk factors is one limitation of the existing standard (transplanting patients with a MELD greater than 15) (8). Recently, the donor risk index (DRI) has been proposed as a method to stratify outcomes associated with graft selection (9). However, the lack of contribution from recipient factors gives the DRI alone a poor predictive value (c-statistic 0.53 based on its application to the United Network for Organ Sharing [UNOS] database). In the present analysis, we combined both donor and recipient risk factors in constructing the Survival Outcomes Following Liver Transplantation (SOFT) score to accurately predict recipient posttransplant survival at 3 months. This score would then allow clinicians to balance waitlist mortality at 3 months as predicted by the MELD score against 3-month mortality following liver transplantation as predicted by the SOFT score to determine which candidates should undergo liver transplantation. Since MELD has been proven to be an accurate predictor of 3-month waitlist mortality (4), we constructed the SOFT score to complement the MELD score by predicting 3-month posttransplant mortality. The SOFT score along with the MELD score allows clinicians to make a real-time go or no-go decision on a particular allograft.
The SOFT score can also be used to avoid wasteful transplants where predicted survival is below acceptable standards. Furthermore, as the critical liver allograft shortage fuels more aggressive practices to utilize increasingly marginal donor allografts, the SOFT score can establish risk limits for particular liver transplant candidates.
We performed a retrospective analysis of UNOS deidentified patient-level data of all recipients of liver transplantation between March 1, 2002, the date of implementation of the MELD prioritization system, and August 1, 2006. Our analysis employed the liver registry with data collected by the Organ Procurement and Transplantation Network. We included all transplant recipients aged 18 years or older. Donor and recipient characteristics were reported at the time of transplant. Follow-up information was collected at 6 months and then yearly after transplantation. Patients undergoing combined or multivisceral transplants (n = 1402) and recipients of a live-donor transplant (n = 1163) were excluded from the study. All patients were followed from the date of transplant until either death (n = 6004) or the date of last known follow-up (n = 15 669); we analyzed 21 673 recipients.
We performed a retrospective analysis of waitlisted candidates to determine their mortality rate at 3 months and to reassess the predictive value of the MELD score. We excluded patients under the age of 18 years (n = 2383). The analysis included patients with an initial date of registration between March 1, 2002 and August 1, 2006. Patients who were transplanted within 3 months upon registering on the waitlist were excluded (n = 14 721). This exclusion was only used for our waitlist mortality analysis, since we needed 90 days at-risk to determine 3-month waitlist mortality. This was not an exclusion used in our analysis of posttransplant survival. The final analysis included 30 321 waitlisted candidates.
Data were analyzed using a standard statistical software package, Stata® 9 (Stata Corp, College Station, TX). Continuous variables were reported as a mean ± standard deviation and compared using the Student's t-test. Contingency table analysis was used to compare categorical variables. Results were considered significant at a p-value of <0.05. All reported p-values were two-sided. The primary outcome measure was patient death. Time to death was assessed as time from the date of transplantation to the date of death. Kaplan–Meier analysis with log-rank test and logistic regression were used for time-to-event analysis. Three-month survival was the dependent variable and the risk factors were the independent variables in the logistic regression analysis. Patients lost to follow-up (n = 573) or alive (n = 15 669) on October 7, 2007 were censored at the date of last known follow-up.
The recipient and donor risk factors considered in this analysis are listed in Table 1. The characteristics that were present in the plurality of transplants were used as the reference groups. Serum creatinine was utilized instead of calculated creatinine clearance because serum creatinine is readily accessible for rapid assessment of donor allograft quality. This analysis only included recipients who were transplanted after the MELD scoring system was instituted for liver allocation in 2002, resulting in high entry completion (99.9%). Since MELD scores were analyzed as a recipient risk factor, recipient creatinine, bilirubin and international normalized ratio (INR) were not included as individual predictors. Separately, the components of the MELD score were each significant predictors of 3-month posttransplant mortality: total recipient bilirubin ≥ 8 mg/dL (OR 1.2), INR ≥ 2.5 (OR 1.2), creatinine between 1.5 and 2.0 mg/dL (OR 1.3) and creatinine ≥ 2.0 mg/dL (OR 1.5). Patients with malignancy were known to have cancer prior to transplantation and did not reflect incidentally discovered cancer at transplantation.
Table 1. Risk factors considered in univariate and multivariate analysis
Donor risk factors
Recipient risk factors
CVA = cerebrovascular accident; AST = aspartate aminotransferase; SGOT = serum glutamic-oxaloacetic transaminase; ALT = alanine aminotransferase; SGPT = serum glutamate-pyruvate transaminase; UNOS = United Network of Organ Sharing; TIPS = transjugular intrahepatic portosytemic shunts; COPD = chronic obstructive pulmonary disease.
*Covariates from the SRTR 1-year patient survival model.
Deceased donor after cardiac death*
ABO incompatible transplant*
Age 0–10 years*
Diagnosis—acute hepatic necrosis*
Age 10–20 years*
Diagnosis—cholestatic liver disease*
Age 20–30 years*
Diagnosis—metabolic liver disease*
Age 45–55 years*
Age 55–60 years*
Age 60–70 years*
Portal vein thrombosis at transplant*
Age > 70 years*
Age > 70*
Height > 75th%tile*
Height < 25th%tile*
Cause of death CNS tumor*
Incidental tumor found at transplant*
Cause of death anoxia*
Intensive care unit pretransplant*
Cause of death cerebral vascular accident*
Admitted to hospital pretransplant*
Cause of death other*
Life support pretransplant*
Cold ischemia time 0–6 h*
Previous abdominal surgery*
Cold ischemia time 12–16 h*
Cold ischemia time 16–20 h*
Cold ischemia time >20 h*
Partial or split liver*
Body mass index 30–35
Creatinine > 1.5 mg/dL
Body mass index > 35
Creatinine > 2.0 mg/dL
Hepatitis B (core Ab positive)
Hepatitis C (positive serology)
One previous transplant
Resuscitation following cardiac arrest
Two previous transplants
AST or SGOT < 90 IU/L
History of angina or coronary artery disease
AST or SGOT > 140 IU/L
ALT or SGPT < 60 IU/L
ALT or SGPT > 100 IU/L
ALT or SGPT > 100 IU/L
Albumin 2.0–2.5 g/dL
Diabetes mellitus (type unspecified)
Albumin < 2.0 g/dL
Insulin-dependent diabetes mellitus
Dialysis prior to transplantation
Hypertension—less than 10 yr duration
UNOS status 1
Hypertension—greater than 10 yr duration
MELD score < 9
History of alcohol dependency
MELD score 10–19
History of cigarette use > 20 pack years
MELD score 20–29
History of cocaine use in the past
MELD score 30–39
History of IV drug use
MELD score > 40
Encephalopathy at transplant
Hepatitis B (core Ab positive)
1–2 yr on waitlist
Hepatitis C (positive serology)
>2 yr on waitlist
Total bilirubin 1–1.8 mg/dL
Transplant performed between 4-1-1994 to 1-1998
Total bilirubin > 1.8 mg/dL
Transplant performed between 1-1-1998 to 1-1-2002
Deceased donor—three or more inotropic agents
Hx of peripheral vascular disease
Warm ischemia time ≤30 min
Hx of COPD
Warm ischemia time 60–70 min
Portal bleed 48 h pretransplant
Warm ischemia time 70–80 min
Any previous malignancy
Warm ischemia time 80–90 min
Variceal bleeding within 2 weeks of registration
Warm ischemia time >90 min
Spontaneous bacterial peritonitis pretransplant
Pulmonary embolus within 6 months of registration
TIPS at transplant
Logistic regression analysis determined the predictors of patient death at 3 months posttransplantation. Donor and recipient variables were first analyzed with univariate analysis and are listed in Table 1. Variables found to be significant in univariate analysis were then subjected to multivariate analysis. Points were assigned to each risk factor based on its odds ratio for patient death at 3 months. One point was awarded to each risk factor for every 10% increase in risk for death at 3 months. Negative points were also awarded for every 10% decrease in risk for death at 3 months.
We assigned four risk groups based on MELD-predicted 3-month waitlist mortality. We formulated two distinct scores: the preallocation score to predict survival outcomes following liver transplantation (P-SOFT) designed to evaluate patients on the waitlist and the score to predict survival outcomes following liver transplantation (SOFT) that includes both donor and recipient factors to evaluate transplants at the time of transplantation. Model discrimination was assessed using the area under the receiver operating characteristic (ROC) curve.
The study population included 21 673 patients. Analysis included 41 504 years-at-risk for the liver transplant recipients. Mean graft survival was 4.2 years. Mean follow-up was 2.0 years. Demographic and clinical characteristics are summarized in Table 2.
Table 2. Demographic characteristics of donors and recipients
NA = not applicable to this group of patients; CVA = cerebrovascular accident.
52.0 ± 10.2
41.1 ± 17.6
% African American
172.4 ± 10.6
171.7 ± 11.0
83.2 ± 19.5
77.9 ± 19.1
1.9 ± 1.8
1.4 ± 1.1
1.4 ± 1.5
20.6 ± 9.6
Cause of liver failure
• Acute hepatic necrosis
• Cholestatic liver disease
• Metabolic liver disease
• Hepatitis C
• Hepatitis B
• Alcoholic cirrhosis
Cold ischemia time (hours)
7.7 ± 3.6
Cause of death
Data entry rate
The data entry completion for variables that were significant in univariate analysis is listed in Table 3. A majority of variables are well populated. Exceptions include hepatitis C Virus (HCV, 87.1%), cold ischemia time (86.8%), warm ischemia time (75.7%) and portal bleed 48 h pretransplant (50.7%). Recipients with missing entries were not dropped, but rather, added to the reference group under the assumption that the missing reports were randomly distributed. Given the large number of risk factors analyzed, this was necessary to preserve the total number of patients studied.
Table 3. Summary of donor and recipient risk factors
Percent entry filled
Percent of patients
COD = cause of death; Cr = serum creatinine; CIT = cold ischemia time; WIT = warm ischemia time; CVA = cerebrovascular accident; %tile = percentile; BMI = body mass index; HCV = hepatitis C Virus; ALT = alanine aminotransferase; ICU = intensive care unit; UNOS = United Network for Organ Sharing; MELD = model for end-stage liver disease.
Height > 25th%tile
Weight > 25th%tile
COD (anoxia, trauma)
Cr < 1.5
Cr > 1.5
No cocaine use
CIT 6–12 h
CIT < 6 h
WIT 30–60 min
WIT 60–70 min
WIT 70–80 min
WIT 80–90 min
WIT > 90 min
Age > 70
Height > 25th %tile
Height < 25th %tile
BMI < 30
BMI > 35
All other diagnoses
Acute hepatic necrosis
No prior transplants
1 Previous transplant
2 Previous transplants
No previous abdominal surgery
Previous abdominal surgery
ALT < 100
ALT > 100
Albumin > 2.5
2.0 < Albumin < 2.5
Albumin < 2.0
Non-UNOS status 1
UNOS Status 1
MELD score: (10–19)
MELD ≤ 9
MELD ≥ 40
Not on life support Pre transplant
Life support pretransplant
Encephalopathy at transplant
No portal vein thrombosis
Portal vein thrombosis at transplant
No incidental tumors
Incidental tumor found at transplant
< 1 year on waitlist
1–2 years on waitlist
Not ventilator dependent
No portal bleed within 48 h pretransplant
Portal bleed within 48 h pretransplant
No variceal bleeding within 2 weeks of registration
Variceal bleeding within 2 weeks of registration
No ascites pretransplant
No SBP pretransplant
No PE within 6 months of registration
PE within 6 months of registration
Univariate and multivariate analysis
Table 1 lists all of the risk factors that were considered. Risk factors that were significant in univariate analysis were then subjected to multivariate analysis. The significant risk factors in multivariate analysis are presented in Table 3. The most significant risk factors were two previous transplants (OR 2.4, confidence interval (CI) 1.52–3.67), warm ischemia time > 90 min (OR 2.3, CI 1.58–3.41), one previous transplant (OR 1.9, CI 1.56–2.24), life support (OR 1.9, CI 1.54–2.35) and warm ischemia time 80–90 min (OR 1.8, CI 1.10–3.00).
Table 3 summarizes donor and recipient risk factors and their assigned points. Table 4 presents two different risk scores: the preallocation score to predict survival outcomes following liver transplantation (P-SOFT) and the score to predict survival outcomes following liver transplantation (SOFT). The SOFT score is formulated from combining the recipient waitlist score in addition to donor factors and cold ischemia times. Warm ischemia was excluded since it cannot be reliably predicted prior to transplantation. Table 5 illustrates the population distribution and odds ratios based on the group with less than five points. The odds ratios for 3-month survival for groups with 6–15 points, 16–35 points, 36–40 points and > 40 points were 2.2, 5.8, 15.4 and 30.5 respectively. Figures 1 and 2 illustrate the Kaplan–Meier curves and life-table analysis of immediate patient survival post liver transplantation based on risk point totals from the P-SOFT and SOFT scores. Using the SOFT score, the 3-month patient survival of recipients with <5 points was 97%, 6–15 points was 94%, 16–35 points was 84%, 36–40 points was 62% and >40 points was 38%. These groups were then labeled according to 3-month mortality risk, with <5 points designated low risk, 6–15 points low-moderate risk, 16–35 points high-moderate risk and 36–40 points as high risk. Calculations of area under the ROC curves for 3-month survival showed P-SOFT and SOFT score values of 0.69 (CI 0.67–0.70) and 0.70 (CI 0.69–0.71), respectively.
Table 4. P-SOFT and SOFT scores
BMI = body mass index; MELD = model for end-stage liver disease.
Preallocation score to predict survival outcomes following liver transplantation (P-SOFT)
• Age > 60
• BMI > 35
• One previous transplant
• Two previous transplants
• Previous abdominal surgery
• Albumin < 2.0 g/dL
• Dialysis prior to transplantation
• Intensive care unit pretransplant
• Admitted to hospital pretransplant
• MELD score >30
• Life support pretransplant
• Portal vein thrombosis
• Ascites pretransplant
Score to predict survival outcomes following liver transplantation (SOFT)
• P-SOFT score
Total from above
• Portal bleed 48 h pretransplant
• Donor age 10–20 years
• Donor age > 60 years
• Donor cause of death from cerebral vascular accident
• Donor creatinine > 1.5 mg/dL
• National allocation
• Cold ischemia time 0–6 h
Table 5. Risk groups generated from P-SOFT and SOFT scores
Percentage of patients
Odds ratio (CI)(low risk is Ref)
26.56 (16.75– 42.11)
Warm ischemia times
Prolonged warm ischemia times may indicate surgical inexperience or difficult operative circumstances and are likely found only in exceptional cases. This factor was removed from the SOFT score since it cannot reliably be predicted prior to transplantation.
Eleven percent of the recipients had hepatocellular carcinoma. Since this analysis only used calculated MELD scores, exceptional MELD points awarded for a cancer diagnosis did not affect the analysis.
MELD as a predictor of posttransplant survival
The MELD score is a poor predictor of posttransplant mortality as demonstrated in Figure 3. When the MELD score is used as a model to predict 3-month posttransplant survival, the c-statistic was 0.63 (0.62–0.65).
One-year, 3-year and 5-year posttransplant survival
The SOFT score is also an accurate predictor of 1-year, 3-year and 5-year posttransplant survival as demonstrated in Figure 4.
The 3-month posttransplant mortality for the SOFT score and the 3-month waitlist mortality for the MELD score are represented in Figure 5. The MELD score entry completion for waitlisted patients was 99.9%. The c-statistic for 3-month waitlist survival is 0.83 (CI 0.82–0.83), consistent with previous published studies (4).
It is critical to balance pre- and post-transplant mortality rates by considering graft, recipient and operative factors to determine whether to accept a liver allograft for a particular candidate. We report here the first model to fully consider the effect of graft selection, recipient factors and operative impact with the highest reported c-statistic in the literature. The MELD score has only proven to be an accurate predictor of pretransplant mortality. Wiesner et al. analyzed 2 271 candidates on the waitlist before concluding that the MELD score had a c-statistic of 0.83 when used for predicting 3-month mortality while on the waitlist (4). Our analysis of 30 322 waitlisted candidates confirmed this finding, with a c-statistic of 0.83 when only the MELD score was used to predict 3-month mortality of candidates on the waitlist. Attempts to employ the MELD score alone to predict posttransplant mortality have yielded inaccurate results (6,7). In the present analysis, the MELD score alone yielded a c-statistic of 0.63 when used to predict 3-month survival following transplantation. The Kaplan–Meier survival curves, stratified by the MELD score in Figure 3, illustrate the poor predictive value of the MELD score. In the model proposed by Desai et al., retransplantation, recipient age, mechanical ventilation and dialysis were used to predict posttransplant recipient survival at 3 months and yielded a c-statistic of 0.65 (7). Our newly formulated survival outcomes following liver transplantation (SOFT) score that combined donor and recipient risk factors resulted in a c-statistic of 0.70 as a predictor of 3-month posttransplant survival. The SOFT score is therefore the most accurate predictor to date of 3-month recipient survival following liver transplantation.
The preallocation score to predict survival following liver transplantation (P-SOFT) differs from the SOFT score since it utilizes only 14 out of 19 risk factors that are available when the candidate is on the waitlist. It is designed to evaluate a candidate prior to liver allograft allocation and results in a c-statistic of 0.69 as a predictor of 3-month recipient survival following liver transplantation. The SOFT score includes donor and recipient factors but is dominated by recipient factors, which comprise 14 out of 18 included risk factors. Seven donor factors, cold ischemic times and portal bleeds are also considered in the SOFT score (Table 4). The SOFT score does not correlate with the MELD score as shown in Figure 6. Although cold ischemic times are determined in the course of transplantation, we propose that they may be estimated when a donor allograft is offered. The overall result of the SOFT score may ultimately guide the clinician to either accept or decline the offer based on the projected risk group into which the recipient is placed. The DRI is inadequate in this respect since there is no consideration for recipient risk factors.
In order to determine which recipients, defined by MELD risk score, should be transplanted, we compared waitlist mortality predicted by the MELD score to posttransplant mortality defined by the SOFT score (Figure 5). On the basis of these results, we suggest that candidates with a MELD score ranging from 17 to 19 points should only receive low-risk SOFT transplants; candidates with a MELD score of 20–29 points should receive low or low-moderate risk SOFT transplants; candidates with a MELD score of 30–39 points should receive low, low-moderate or high-moderate risk SOFT transplants; and candidates with the highest waitlist mortality risk with a MELD score of greater than 40 should receive low, low-moderate, high-moderate or high-risk SOFT transplants. These recommendations likely do not apply to patients with hepatic cancers since the benefit of early removal of tumor must also be considered in addition to the MELD and SOFT scores.
This study emphasizes that transplants in patients with a SOFT score of >40 are likely futile since the predicted posttransplant mortality is greater than any waitlist mortality as predicted by the MELD score. Since patients with acute hepatic failure have an exceptionally high waitlist mortality, which is not accurately predicted by the MELD score alone, the futility of a SOFT score >40 does not apply to this group of patients.
The Scientific Registry for Transplant Recipients (SRTR) offers a risk-adjusted model to project patient survival after liver transplantation. The model combines an extensive list of donor and recipient covariates and is an effective auditing tool since a transplant program's observed outcomes can be compared to expected outcomes (10,11). Unlike the SRTR risk-adjusted model, the proposed SOFT risk score can be used to predict outcomes for a particular recipient and donor allograft prior to transplantation in order to compare against waitlist mortality predicted by the MELD score. The multivariate analysis in this study included all of the variables considered in the SRTR risk-adjusted model. In comparison, the SRTR model for 1-month posttransplant survival had an index of concordance of 0.66.
Since the passage of the National Transplantation Act of 1984, data entry has been mandatory for all US transplant centers. Nevertheless, all patient registries often suffer from variability in data entry. Findings from this study use large cohorts of patients and are unlikely to be significantly impacted by small amount of missing data. Fields contained within this data base were generally well populated with a 95–99% data entry rate for the majority of variables. The fact that center-specific factors could not be appropriately accounted for is a significant limitation.
This analysis represents the largest study of waitlisted candidates and transplant recipients of liver allografts after the MELD Score was implemented for allocation in 2002. Survival after liver transplantation must be greater than survival on the waitlist to justify liver transplantation. The SOFT score, newly formulated from the same cohort of patients, can accurately predict 3-month survival following liver transplantation. It can then be compared with MELD-predicted waitlist mortality to determine which patients should be transplanted. The SOFT score can also be used to improve donor–recipient matching.
This study was funded by National Institutes of Health training grant T32HL07854–11 (AR) and by Health Resources and Services Administration contract 231–00-0115.