Transplant survival rates were also estimated according to various values of the UKKDRI for 3 months, and 1, 3, and 5 years and are illustrated in Figure 2. The estimates were based on a standardized recipient who was white, aged 50 years, on dialysis before transplantation, with glomerulonephritis, and receiving a 000 HLA-A, B, DR mismatched transplant with a cold ischemia time of 17 hr. The difference in transplant survival at 5 years between donor kidneys with the highest and lowest UKKDRI is estimated to be over 40%.
Comparison of the UKKDRI With the USKDRI
We next compared the UKKDRI with the USKDRI (2) for each transplant in the validation dataset. Adding either the UKKDRI or the USKDRI to a Cox model fitted to the validation data set that contains all the significant recipient and transplant factors, the change in −2 log L reveals that both indices are highly significant prognostic indicators (P<0.0001) for the UK data. For models that contained UKKDRI or USKDRI, the c-statistic was 0.62 and 0.63, respectively, so that the predictive ability of the two models is practically identical. However, the UKKDRI has the advantage of simplicity as it is derived from just five donor factors, whereas the USKDRI incorporates 15 factors.
To provide a clinical tool for predicting the likely outcome of DD kidney transplantation, we have derived a UKKDRI based on donor factors readily available to the transplant surgeon at the time of donor organ offer. Of the 19 donor variables considered for inclusion in the UKKDRI, the two most important were donor age group and hypertension, of which donor age had the largest influence on transplant outcome.
Neither terminal donor serum creatinine nor death from CVA predicted transplant outcome which contrasts with findings from the SRTR analysis where these were two of the four factors used to define expanded criteria (1). Our finding that terminal serum creatinine is not independently associated with transplant outcome is not perhaps surprising; any potential effect is likely to be dominated by donor age, because renal function deteriorates progressively with age. Moreover, where kidneys were used from donors with a high serum creatinine, it is probable that the decision to proceed was based on additional information suggesting acute recoverable renal injury and a recent near-normal premorbid creatinine.
Donation after circulatory death was not found to be an adverse factor for transplant outcome in the United Kingdom in this analysis, nor in another recent UK analysis (5). Similarly, donor diabetes did not predict poor transplant outcome although this may be because of the small number of donors with diabetes. In the first analysis of the SRTR database, donor diabetes was not an adverse factor, but subsequent analysis of a larger data set showed it to be a significant risk factor (1, 2). It was notable that in this study, neither a history of smoking nor donor ethnicity was associated with transplant outcome, although the number of ethnic minority DD was small.
Dividing the transplant follow-up period into three distinct posttransplant epochs allowed us to assess the role of individual donor factors on transplant outcome in more detail. Donor age was the most important predictor of poor transplant outcome in all three epochs studied. Donor history of hypertension remained a significant risk factor affecting medium and long-term outcome (3 months onward) but was not associated with early posttransplant outcome.
Three other donor factors had significant effects on outcome in different epochs. Donor weight had a deleterious effect in the early posttransplant period but favored long-term graft survival. Obese donors have more perirenal fat, and one potential explanation for the deleterious effect of increased donor body weight in the short term may be a rewarming effect at the time of organ retrieval. Perinephric fat does not cool down quickly during in situ cold perfusion and may rewarm the kidneys after perfusion stops contributing to ischemic injury. In addition, dissection in an obese donor is more challenging and there is a greater risk of kidney damage that may also affect early outcome.
The duration of inpatient stay before organ donation, particularly stay in intensive care, is generally believed to be associated with poor transplant outcome, but surprisingly, this variable has not been included in previous registry studies of kidney, liver, and pancreas transplantation (2, 6, 7). In this study, length of hospital stay before organ donation was related to transplant outcome only in the early posttransplant period. The use of adrenaline (epinephrine) in the donor at the time of organ donor referral also influenced transplant outcome, but only between 3 months and 3 years. It is not clear why the use of adrenaline or the length of hospital stay had effects on transplant outcome that were limited to their respective epochs.
When establishing the UKKDRI, we adjusted for recipient and transplant-related factors, but chose to incorporate in the index itself only factors that characterized the donor kidney, and not recipient or transplant-related factors, such as cold ischemic time that are unknown at the time of organ offer for transplantation. There was no evidence of any interaction effect between UKKDRI and recipient factors, so the risk index can be used independently to aid understanding about donor organ quality at the time of offering and in gaining informed consent. It can also be used in conjunction with other factors in risk-adjusted outcome analyses.
In summary, the UKKDRI provides a simple, clinically useful tool that allows prediction of transplant outcome. It will aid transplant surgeons and others in organ allocation and gaining fully informed consent from potential transplant recipients.
MATERIALS AND METHODS
We analyzed data submitted to the UK Transplant Registry held by National Health Service Blood and Transplant. The registry records mandatory data on all kidney transplants performed in the United Kingdom supplied by all 23 adult UK kidney transplant centers.
Data from 7620 adult recipients of adult DD kidney transplants performed between January 1, 2000, and December 31, 2007, were analyzed. A number of donor variables were studied, all of which were routinely available to the recipient center at the time of the kidney offer for transplantation: donor age group (18–39, 40–59, and 60+ years), gender, ethnicity, donor type (donation after brain death or circulatory death), terminal creatinine (the last value measured before the offer), cause of death (trauma, intracranial hemorrhage, and other), weight, height, body mass index, abdominal girth, length of time in hospital, urine output in the hr before the offer, whether on adrenaline (epinephrine), noradrenaline (norepinephrine) or vasopressin, and history of cardiothoracic disease (unspecified), smoking, diabetes, or hypertension.
To develop a valid donor risk model, it was necessary to adjust for recipient and transplant factors that may influence transplant survival. Factors considered were recipient age, renal disease (glomerulonephritis, polycystic kidney disease, diabetes, “not reported,” or “other”), gender, ethnicity (white, Asian, black, or other), preemptive transplantation, HLA mismatch (according to the four levels used in the UK 2006 Kidney Allocation Scheme  but ignoring the defaulting of rare antigens to more common counterparts), year of transplant, cold ischemia time, donor-recipient cytomegalovirus match, and donor-recipient gender match.
To develop a donor risk index, the transplant dataset was randomly divided into a modeling dataset comprising 4570 (60%) transplants and a validation dataset comprising 3050 (40%) transplants. The randomization and all other statistical analyses were carried out using SAS (version 9.1, SAS Institute Inc., NC).
The distribution of donor, recipient, and transplant factors in the modeling and validation data sets were compared using a chi-square test. In cases where there were large numbers of missing values, a P value is also given for comparing the two data sets excluding the unknowns. Where continuous variables were grouped, the median values were compared using the Mann-Whitney U test.
The outcome variable was transplant survival, defined as time from transplant to the earlier of graft failure or patient death. Donor factors influencing transplant survival were investigated using Cox proportional hazards regression, adjusting for significant recipient and transplant factors identified. Survival times were stratified by transplant center to allow for possible differences between centers. The coefficients of the significant donor factors in the Cox model (that also included the relevant recipient and transplant factors) were used to construct the donor risk index.
Transplant survival was considered over the full follow-up period of up to 9 years (overall survival) and, to identify donor factors that are important in different time periods, in three distinct posttransplant epochs: 0 to 3 months (censoring survival times over 90 days); 3 months to 3 years (omitting observations with survival times under 90 days and censoring survival times over 1095 days); and over 3 years (omitting observations with survival times less than 1095 days). Data on two transplants with missing survival times were excluded, as were transplants where the donor weight and number of days in hospital were missing (n=160).
Alternative models were compared using the log-likelihood ratio statistic (−2 log L) and Akaike's Information Criterion. The extent of any nonlinearity in continuous variables was examined using martingale residuals (9) and by grouping the variable and plotting log HRs for each group against the mid point. The linearity of the UKKDRI was also examined by testing for nonlinearity across the four quartiles of the distribution of UKKDRI values.
Predictive ability of Cox models was assessed using a c-statistic due to Gönen and Heller (10). This statistic is based on the observed survival times of all pairs of patients. The two times in a pair are said to be concordant if the patient with the lower risk score (the linear part of the Cox model) has the longer survival time, and the c-statistic is the estimated proportion of pairs for which this is the case. The value of the statistic, c, is equal to 0.5 when a model has no utility.
Comparison of the UK and US Donor Risk Indices for UK Data
The KDRI of Rao et al. (2) was also applied to the UK validation dataset to compare its performance with that of the newly derived UKKDRI. To render the Rao formula applicable to our data, black ethnic origin was substituted for African American ethnicity, and donors who died of intracranial hemorrhage, intracranial thrombosis, or “intracranial event, type unclassified” were classified as deaths from CVA. Values of USKDRI were calculated after reestimating coefficients of variables in the USKDRI using the UK data.
The authors thank all the transplant centers in the United Kingdom who contributed data on which this study is based, and to the reviewers whose helpful comments have improved the presentation of the results.
1. Port FK, Bragg-Gresham JL, Metzger RA, et al.. Donor characteristics associated with reduced graft survival
: An approach to expanding the pool of kidney donors. Transplantation 2002; 74: 1281.
2. Rao PS, Schaubel DE, Guidinger MK, et al.. A comprehensive risk quantification score for deceased donor kidneys: The kidney donor risk index. Transplantation 2009; 88: 231.
3. Nyberg SL, Matas AJ, Kremers WK, et al.. Improved scoring system to assess adult donors for cadaver renal transplantation. Am J Transplant 2003; 3: 715.
4. Schold JD, Kaplan B, Baliga RS, et al.. The broad spectrum of quality in deceased donor kidneys. Am J Transplant 2005; 5: 757.
5. Summers DM, Johnson RJ, Allen J, et al.. Analysis of factors that affect outcome after transplantation of kidneys donated after cardiac death in the UK: A cohort study. Lancet 2010; 376: 1303.
6. Feng S, Goodrich NP, Bragg-Gresham JL, et al.. Characteristics associated with liver graft failure: the concept of a donor risk index. Am J Transplant 2006; 6: 783.
7. Axelrod DA, Sung RS, Meyer KH, et al.. Systematic evaluation of pancreas allograft quality, outcomes and geographic variation in utilization. Am J Transplant 2010; 10: 837.
8. Johnson RJ, Fuggle SV, Mumford L, et al.. A New UK 2006 National Kidney Allocation Scheme for deceased heart-beating donor kidneys. Transplantation 2010; 89: 387.
9. Therneau T, Grambsch P, Fleming T. Martingale-based residuals for survival models. Biometrika 1990; 77: 147.
10. Gönen G, Heller B. Concordance probability and discriminatory power in proportional hazards regression. Biometrika 2005; 92: 965.
Kidney transplantation; Deceased donation; Graft survival
Supplemental Digital Content
© 2012 Lippincott Williams & Wilkins, Inc.