Journal Logo

Original Clinical Science—General

Validation of the Prognostic Kidney Donor Risk Index Scoring System of Deceased Donors for Renal Transplantation in the Netherlands

Peters-Sengers, Hessel MSc1; Heemskerk, Martin B.A. PhD2; Geskus, Ronald B. PhD3,4; Kers, Jesper MD, PhD5; Homan van der Heide, Jaap J. MD, PhD1; Berger, Stefan P. MD, PhD6; Bemelman, Frederike J. MD, PhD1

Author Information
doi: 10.1097/TP.0000000000001889


In the article, Validation of the Prognostic Kidney Donor Risk Index (KDRI) Scoring System of Deceased Donors for Renal Transplantation in the Netherlands (Transplantation. 2018;102:162–170), one of the author affiliations was missing.

The correct list of affiliations is as follows:

Hessel Peters-Sengers, MSc,1 Martin B.A. Heemskerk, PhD,2 Ronald B. Geskus, PhD,3,4,5 Jesper Kers, MD, PhD,6 Jaap J. Homan van der Heide, MD, PhD,1 Stefan P. Berger, MD, PhD,7 and Frederike J. Bemelman, MD, PhD1

1 Department of Nephrology, Academic Medical Center, University of Amsterdam, Amsterdam, the Netherlands.

2 Dutch Transplant Foundation, Leiden, the Netherlands.

3 Nuffield Department of Medicine, University of Oxford, Oxford, United Kingdom.

4 Oxford University Clinical Research Unit, Wellcome Trust Major Overseas Programme, Ho Chi Minh City, Vietnam.

5 Department of Clinical Epidemiology, Biostatistics and Bioinformatics, Academic Medical Center, University of Amsterdam, Amsterdam, the Netherlands.

6 Department of Pathology, Academic Medical Center University of Amsterdam, Amsterdam, the Netherlands.

7 Department of Nephrology, University of Groningen, University Medical Center Groningen, Groningen, the Netherlands.

Transplantation. 102(7):e359, July 2018.

In 2015, the median waiting period for a kidney from a deceased donor was more than 3 years, and 74 patients died while waiting for a kidney transplantation in the Netherlands.1 To address the organ shortage, accepting older deceased donors with more comorbidities has become common practice. However, this policy increases the risk of delayed graft function, prolonged hospitalization, rejection, and graft failure.2,3 The development of a reliable tool to aid transplant professionals to estimate the quality of a specific donor organ would be highly useful in selection of the donors and help to increase the donor pool safely.

To assess the quality of a deceased donor kidney, a variety of prediction tools have been proposed.4-12 Up to now, the most widely used clinical tool to assess the quality of a deceased donor is the Kidney Donor Risk Index (KDRI).5 The KDRI is a continuous risk scoring system based on 10 donor factors and 4 transplant factors; however, the donor-only KDRI was the version that was implemented because these factors are generally known at the time of donor organ offer. The reference donor (KDRI, 1.00) corresponds to a 40-year old non-African American man; height, 1.70 m; weight, 80 kg, with a serum creatinine of 1.0 mg/dL, without diabetes, hypertension or a cerebrovascular cause of death, who is hepatitis C virus (HCV) negative and brain dead. In the United States, the KDRI ranges from 0.5 to 4.0, and higher values are associated with a lower graft survival. The KDRI was developed and internally validated based on US data, including 69 440 adults, ABO-compatible, solitary, first-time deceased donor kidney recipients from 1995 to 2005 in a multivariable Cox proportional hazards regression model, stratified by recipient age and diabetes, and transplant center.

The applicability of the KDRI, however, has not been assessed in many other populations. For such a model to be used in practice in other countries, like the Netherlands, it must be externally validated. Two key aspects of evaluating model performance of the KDRI are (1) the ability to distinguish between kidneys with a low and high risk for graft failure (discrimination) and (2) the accuracy of prediction between predicted and actual observed graft survival (calibration). Discrimination of the KDRI in the United States (internal validated) resulted in a concordance (c)-statistic of 0.62. To put this in perspective, the KDRI with a c-statistic of 0.62 ranks only 62% of the grafts correctly to outcome. Recently, the KDRI was validated in the UK registry with comparable c-statistic of 0.63; however, calibration was not reported, and the focus was on developing a new simplified 5-factor UKKDRI (also with a c-statistic of 0.62).7 The strategy chosen in this article is to compare the reported model of the KDRI by Rao et al (2009) based on US data with the (external) data from the Dutch cohort, referred to as the derivation and validation data set, respectively, a method that is known as a fully independent external validation.13,14 Because the Dutch cohort is of a more recent period (2002-2012), this is both external and temporal validation.

In this registry-based cohort study, we sought to address 3 research questions to determine external validation: (1) What is the discriminative ability of the KDRI in the more recent Dutch cohort? (2) How accurately does the KDRI predict graft survival? (3) Are the factors of the KDRI correctly specified for the Dutch situation, and do other donor factors deliver added value?


Study Population

The Dutch Organ Transplantation Registry (NOTR) records kidney transplantation follow-up data from all 8 transplantation centers in the Netherlands. Criteria for selecting patients were like those of Rao et al (2009). We included recipients aged 18 years or older, of a first kidney from a donations after brain death or a donation after circulatory death (DCD) category III (donation after controlled circulatory death)15 donor aged 18 years and older between January 5, 2002, and January 5, 2012. Multiorgan transplants and DCD donor kidneys from categories I and II (donation after uncontrolled circulatory death) were excluded. The final follow-up date was September 10, 2015.


We evaluated graft survival, similar as Rao et al, which is defined as patient death or graft loss leading to dialysis treatment (whichever came first). Follow-up was assessed at 0.5, 1, 3, 5, and 7.5 years after transplantation.


The Dutch cohort was used to validate the KDRIfull 14-variable model. The KDRI prediction equation is:

where I is equal to 1 if the condition is true and I is equal to 0 if the condition is false.

The KDRI calculator from the United States16 is normalized by a scaling factor using the following formula: [KDRI(normalized)= KDRI(Rao)/scaling factor]. The scaling factor is the median value of the KDRI for all donors from the previous calendar year in the United States. The same applies to the Kidney Donor Profile Index, a cumulative percentage scale of the relative risk scale of the KDRI compared with the chosen reference population, which is the prior calendar year in the United States (eg, a donor with Kidney Donor Profile Index of 0% has a KDRI less than all donors in the reference population).17

From 2002 to 2012, only 32 (1.0%) transplantations were either double or en-bloc in the Netherlands, and we categorized these as double transplants because no difference between these 2 categories is made in the NOTR database. As the NOTR does not record ethnicity, we classified all recipients in our population as white. For assessment of the KDRIdonor-only, we excluded the 4 transplant factors (HLA-B and HLA-DR mismatches, cold ischemic time, en bloc, and double transplantation), leaving 9 donor factors for external validation (with exclusion of ethnicity).

Calibration of Predictors of the KDRI

We assessed calibration with the calibration slope with the regression coefficient of the KDRI (from equation 1 without exponential transformation) in Cox regression analysis. A calibration slope more than 1 indicates that predicted probabilities do not vary enough, and if less than 1, then predicted risks are on average too low for low-outcome risks and too high for high-outcome risks. Second, a calibration plot of estimated 1-, 3-, 5-, and 7.5-year survival probabilities of KDRI quintiles on Ln(−Ln(.) scale from the Dutch cohort were compared with the United States, extracted from Figure 2 by Rao et al (2009).13,14 This plot visualizes differences in survival probabilities between the US and the Dutch cohort, also known as calibration-in-the-large. If the plot is scattered around the 45° line through the origin, the model is valid. If not, it indicates whether the “relative risks” or “baseline risks” of the KDRI are different in the Dutch population. Third, to assess the accuracy of prediction of the KDRI over time, we compared the predicted graft survival curves with the observed Kaplan-Meier curves of the KDRI quintiles. This is also known as “partial validation approach,” in detail described by Royston (2015).18 Because the KDRI and KDRI quintiles are calculated from the published information, the resulting graph of observed and predicted survival probabilities shows how well the “external’” KDRI combined with an “internal” reestimation of the baseline cumulative survival is calibrated on the validation data set.

Calibration plot on Ln(−Ln(.)) scale with the survival estimates from Table 2. Survival on the left-hand side of the diagonal indicates higher graft survival in the US, on the right-hand side of the diagonal indicates higher graft survival in the Netherlands. The original KDRI survival estimates were extracted from Figure 2 in the study by Rao et al (2009). Colors correspond to follow-up times in graft survival, dots correspond to KDRI quintiles. Graft survival was defined as patient death or graft loss leading to dialysis treatment (whichever came first).

Discrimination of the KDRI

The discriminative ability of the model was tested with Harrell’s C stat for Cox models. A c index of 0.5 represents no predictive discrimination on graft survival, and an index of 1 represents perfect ability to distinguish patients.19 Because Harrell's C is no longer an overall quantity with stratum variables, we chose different follow-up times (0.5, 1, 3, 5, 7.5 years). Furthermore, we checked time-dependent sensitivity and specificity at different cutoffs of the KDRIfull, described by Blanche et al.20 Third, we plotted Kaplan-Meier curves for KDRI risk groups. These were based on the quintiles of the KDRI index by Rao et al, (2009): 0.45 to less than 0.79, 0.79 to less than 0.96, 0.96 to less than 1.15, 1.15 to less than 1.45, 1.45 or greater.

Fit and (mis)Specification of KDRI Factors

Finally, we checked the model misspecification with a joint test (difference of −2*Log L) for all the donor and transplant factors indexed in the KDRI by running a Cox regression, whereas the coefficient of the log(KDRI) is constrained to equal 1 via an offset term.13 If the joint test is not significant, there is no overall evidence of lack of fit. If significant, lack of fit of each KDRI factor was tested. Furthermore, it was tested if the following factors are valuable additions to the KDRI: (1) use of inotropic drugs before donation, (2) donor atherosclerosis (massive vs none to moderate), (3) donor smoking (yes vs no), and (4) HLA-A mismatch levels (0, 1 (reference), 2). The degree of atherosclerosis is used as an easy and quick criterion to assess donor quality (none, mild, moderate, massive).

Data Analyses

This study conforms to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis Or Diagnosis statement, and corresponds to a type 4 external validation study.21 The KDRI equation (1) was exactly computed in the Dutch cohort as reported by Rao et al (2009). To preserve inferences about the transplant population, we chose to impute missing values (see Table S1, SDC, for detailed description on handling missing information). All analyses were stratified by recipient age (<45, 45 to < 55, 55 to < 65, 65+) and recipient diabetes. Significance level was set at 5%. Data are presented as mean ± SD (for parameters with skewed distribution, median (interquartile ranges), or as numbers (percentages). Analyses were conducted using R (version 3.2.4)22 with the rms package (version 4.5-0), the timeROC package (version 0.3), and mitools (version 2.3). This study was approved by the Dutch Transplant Foundation.


Characteristics of the Study Cohort

Table 1 shows the characteristics of a total of 3201 recipients meeting inclusion criteria from Rao et al (2009). A higher percentage of DCD donors were discarded in the selection process for transplantation than donations after brain death donors (27.6% vs 4.9%, respectively, see Figure S2, SDC, Figure 1 shows the distribution of the KDRI from the derivation study from the United States as well as the validation in the Dutch cohort. The median Dutch KDRI was increased to 1.21 (Rao et al, reported median KDRI 1.05) and comparable with the year 2012 in United States (median of 1.24).23 The minimum KDRI did not differ (0.5 in both cohorts), but the maximum observed KDRI was lower in the Dutch cohort (3.0 vs 4.2).

Descriptive statistics of KDRI risk factors in Dutch population (n = 3201 transplantations)
Distribution of KDRI in US and NL donors.

Calibration of Predictors of the KDRI

Due to the higher KDRI in Dutch transplanted donors, also, a different distribution among the KDRI quintiles was found. Table 2 compares the graft survival rate, attributable to graft loss and return to dialysis, or to patient death, per the original KDRI quintiles for a nondiabetic recipient aged 45 years to younger than 55 years. Compared with the KDRI quintiles reported by Rao et al (2009), the survival probabilities at the first year were slightly lower in the Dutch cohort in all KDRI quintiles, but were comparable at the third year. At 5 and 7.5 years, the observed survival in the Dutch cohort is considerably higher compared with the US cohort (see Figure 2).

Survival percentages of the original KDRI for the US cohort and the Dutch cohort of deceased donor kidney transplantations

The KDRIfull calibration slope was 0.98 (standard error, 0.13) and KDRIdonor-only calibration slope was 0.96 (standard error, 0.14), both not significantly different from 1 (P = 0.877 and P = 0.779, respectively), indicating that predictions (on average) of both KDRIs were almost identical in the Dutch cohort. Figure 3 shows that observed and expected survival of the KDRIfull quintiles are reasonably calibrated; graft failure predictions in the highest KDRI quintile (0.45- < 0.79) were somewhat overestimated, and to some extent, underestimated in the third KDRI quintile (0.96- < 1.15). Within the KDRIfull top quintiles (≥1.45), calibration of observed and expected survival show less agreement; predicted graft failure lines of higher KDRIfull top quintiles were slightly overestimated compared with observed survival.

Calibration of observed (smooth line) vs predicted (dashed line) graft survival, per (A) the KDRI quintiles, and (B) the quintiles of the KDRI-top quintile only (1.45+). Graft survival was defined as composite endpoint of patient death or graft loss leading to dialysis treatment (whichever came first).

Discrimination of the KDRI

Figure 4 illustrates graft survival in the Dutch validation set per the originally reported KDRI quintiles for a nondiabetic recipient aged 45 years to younger than 55 years. For kidneys within the lowest quintile of the KDRI (0.45- < 0.79), 5-year graft survival was 87.3%, compared with 71.2% in the highest KDRI quintile (>1.45). Four examples of the KDRI (extracted from Rao et al, 2009) with predicted graft survival are shown in Table 3. Table 4 shows Harrell's C of the KDRIfull in the Dutch cohort at different follow-up times. At 5 years, Harrell's C of the KDRIfull was 0.63 (95% confidence interval [CI], 0.62-0.64). Harrell C of the KDRIdonor-only at 5 years was 0.62 (95% CI, 0.61-0.63). Harrell's C of the KDRIfull and KDRIdonor-only for death-censored graft survival at 5 years was also 0.63 (95% CI, 0.61-0.65) and 0.62 (95% CI, 0.60-0.63), respectively. Harrell's C of the KDRIfull and KDRIdonor-only was slightly higher for 5-year mortality—including death after graft loss—0.68 (95% CI, 0.67-0.68) and 0.68 (95% CI, 0.67-0.68), respectively.

Predicted graft survival of KDRI quintiles in the Dutch cohort corresponding to a transplanted patient who is between 45 and 55 years of age with no diabetes. Graft survival was defined as patient death or graft loss leading to dialysis treatment (whichever came first).
Calculating KDRI examples in the Dutch cohort of deceased donor kidney transplantations
Discrimination of the KDRIfull and the KDRIdonor-only in the Dutch cohort

Table 5 shows estimated sensitivities and specificities for KDRIfull score cutpoints. Arguably, the rate of false-positives given a cutoff value for the KDRI at time of transplantation should be as low as possible; considering the decline of a donor offer that would have proper function after transplantation will result in increased waiting time. However, this could lead to an undesirable rate of false-negatives cases: acceptance of the donor kidney resulted in graft failure which was not detected by the KDRI. A hypothetical (relatively high) cutoff point of 2.0 for the KDRI would lead to 3.4% false positives (eg, donor kidneys that are wrongly declined), but then only a marginal 9% of the graft failures would have been correctly identified at 5 years.

Sensitivities and specificities of KDRIfull score cutpoints

Fit and (mis)Specification of KDRI Factors

A joint test of all KDRI factors in Cox regression—including the KDRI as offset term—indicated overall evidence of lack of fit (χ2[13] = 46.5, P < 0.001). We found misspecification of the following donor factors: age (P = 0.002; if below 18 years, P = 0.002; if higher than 50 years, P = 0.009), weight if below 80 kg (P = 0.017), and cold ischemia time (P < 0.001) (summarized in Figure 5). Coefficient of donor weight (with 5 kg increase if <80 kg) was reversely associated with graft failure in the Dutch cohort (see Figure S1, SDC, Cold ischemia time was significantly underestimated from the KDRI for brain-death donors as well as for circulatory-death donors, with an increased underestimated risk of graft failure for recipients of circulatory-death donors as cold ischemia time increased. If this interaction was added to the KDRI, cold ischemia time above 24 hours (versus < 12 hours) did not reach significance on a multiplicative scale (P = 0.059) (see Figure 5). When adding other variables to the KDRI, the only valuable addition to the KDRI was ‘use of inotropic drugs before donation’ (P = 0.040), present in 30.4% of the donors. The other added donor factors were no valuable additions to the KDRI: donor atherosclerosis (massive vs none to moderate), donor smoking (yes vs no), and HLA-A mismatch levels (0, 1, 2).

(Mis)specification of KDRI factors and additional factors in the Dutch cohort. From the point of view of successful validation, the “best” result is that all the coefficients of factors are 0, meaning that there are no differences between the βs estimated in the KDRI by Rao et al, (2009) and the Dutch cohort. * P values indicate whether a factor is significantly different from 0.

Validation of KDRI With Recipient Age and Diabetes

Increasing KDRI—modeled with a 3-knot restricted cubic spline—was rather linearly associated with graft failure (P = <.001) than nonlinear (P = 0.880). Interaction analyses revealed that the effect of KDRI on graft survival significantly decreased with every 5-year increase of recipient age (hazard ratio, 0.86; 95% CI, 0.79-0.95, P = 0.002). Figure 6 shows that the slope of the log(KDRI) was more gradually associated with graft failure at 5 years in elderly (65+) recipients than in younger recipients, which is attributable to a higher (death-censored) graft loss. Opposite to Rao et al, (2009) recipient diabetes was no significant risk factor for graft failure in our cohort (hazard ratio, 1.10; 95% CI, 0.93-1.29, P = 0.272), and hazards were proportional over time by inspecting the Schoenfeld residuals.

Interaction of log(KDRI) with recipient age groups, depicted in 5-year graft survival, 5-year death-censored graft survival, and 5-year patient survival. In contrast to previous methods depicted in Figures 2 till 5, KDRI and recipient age groups were treated as covariates.


The KDRI shows reasonable calibration, where the KDRIfull is slightly more accurate than the KDRIdonor-only. The calibration plot shows that the observed and expected survival probabilities of the KDRIfull are reasonably consistent. Within the KDRI top quintiles (KDRI ≥ 1.45), calibration of observed and expected survival show less agreement. The KDRI shows comparable, but only modest discrimination in the Dutch setting, with a c-stat of 0.63 for KDRIfull and 0.62 for KDRIdonor-only. A lack of fit was found for 3 donor factors: age, weight, and cold ischemia time. The coefficients of donor age were lower, except if donor age was younger than 18 years. The coefficient of cold ischemia was higher for Dutch recipients, especially if donors were from circulatory death. Coefficient of donor weight was reversely associated with graft failure. It was not possible to find a KDRI cutoff point with both good sensitivity and specificity; a high cutoff point would give a false-positive rate that comes with too high false-negative rate.

We observed a shift of the median of Dutch deceased donors to higher KDRIs as compared with the US cohort. This could be explained by the increasing tendency to accept kidneys from older donors with comorbidities because the US cohort included transplantations between 1995 and 2005, whereas the Dutch cohort included transplantations performed between 2002 and 2012. Between 2002 and 2012, we also observed a significant gradual increase of the KDRI of 0.02 per year (95% CI, 0.013-0.022, P < 0.001). From 2002 to 2012, median donor age increased from 48 to 54 years, the proportion of circulatory death versus brain death donor kidneys increased (30.2%-51.0%), and a higher incidence of hypertension (18%-28%) and diabetes (1.5%-5.0%) was observed. On the other hand, we observed a decrease of the median cold ischemic time from 2002 to 2012 (19.3-15.3 hours).

Strengths of our study are the national cohort in a high-quality database with good follow up, the availability of data on mortality after graft loss, and a comprehensive analysis of the KDRI. Instead of deriving a new risk tool for graft failure based on Dutch data, we preferred to externally validate the current most widely adapted KDRI from the United States which is based on a far larger data set than ours. Some limitations should be considered. In Rao et al (2009), results were stratified on recipient age for every single year, whereas this was not possible in the current study, and therefore recipient age was categorized into groups. This study was limited by the absence of data about ethnicity, which is not registered. HCV positivity, en bloc, and double transplantation are yet rare in the Netherlands, and thus should be validated when these are a more common practice. Because these factors are not the chosen reference category in the KDRI, this does not constitute a problem for validation.

Other limitations are related to the KDRI itself. The composite outcome of the KDRI is graft survival, defined as patient death or graft loss leading to dialysis treatment (whichever came first). KDRI factors may be more accurate if modeled with death-censored graft survival, but leading to model selection problems. Also, the results are subject to selection of donor kidneys by transplant professionals and to the selection of recipients by allocation. There are reasons to decline a kidney offer even before calculating the KDRI. For example, specific damage (eg, complex cysts, trauma [eg, contusion], or anatomical abnormalities of the kidney) are not accounted for, as well as the likelihood of transmission of diseases, and specific factors from the DCD procedure, such as the first warm ischemia time. Stratification among others on recipient diabetes and age, as Rao et al, (2009) stated in the methods, do not contribute to the discrimination index, but may change the predictive value of KDRI factors. And so, the predictive value of the KDRI may change considerably if the allocation strategies change or if recipient characteristics change. Also, the use of stratification provides a snapshot view of discriminative ability, because survival curves may cross. The follow-up time at which the discrimination index was calculated in Rao et al (2009) was not reported. Despite this, discrimination of the KDRI is equal at different timepoints compared with the United States, in a different population in the Eurotransplant region.

Graft survival was lower in the Dutch population at 1 year and considerably higher at 7.5 years, and recipient diabetes was not a risk factor in the Dutch cohort. Discrepancies between the Dutch and the US population have been reported previously, also depending on life expectancy, access to dialysis, length of waiting time on the transplantation list, and health insurance regulations.24,25 The discrepancies in graft survival could explain the misspecification of some donor factors in the KDRI. It may also be that the original KDRI parameter is too small or too large due to a different case-mix of donors in the US cohort; however, the donor case-mix as well as the reference level of other recipient factors other than age and diabetes were not reported. Therefore, the cause of misspecification of factors remains unknown.

The KDRI with a discrimination of 0.63 will make 37% incorrect outcomes, making it clearly limited for adequate individualized decisions. There are many (dynamic) posttransplant factors of the recipient that contribute to graft failure that are hard to capture at baseline, whereas the KDRI consist of only donor and transplant factors. However, including these recipient factors that are known at the time of transplantation will also not improve the current KDRI at the desired level of greater than 0.8 discrimination for the individual recipient.26 Most of the prediction scores in transplant nephrology have a low C-statistic which may reflect the attainment of a certain threshold of prediction in the complex transplant setting. Therefore, the KDRI seems most suitable to assist allocation for longevity matching between cohorts of donors and recipients, for example, the policy of the United States by allocating the top 20% of low KDRI kidneys to the top 20% of candidates with best prospect of graft survival.

Instead of creating an alternative prediction model, it would be interesting to assess an updated KDRI that include the following: (1) correct misspecified donor factors in the Dutch population and include additional valuable donor factors, (2) include interactions of KDRI and recipient characteristics (such as age), and (3) include surrogate outcomes and their (possible) interactions with the KDRI. We observed an interaction of the KDRI and recipient age: the association between the KDRI and graft failure was higher for younger aged recipients. These results are consistent with those of Hernandez et al27 and Heaphy et al28 who showed that higher KDRI donor kidneys (thus lower quality) have a more perceptible effect in lower-risk recipients who are younger with no history of diabetes. Inclusion of surrogate (intermediate) transplant outcomes, for example, delayed graft function and acute rejection which are associated with graft failure,29 as well as other recipient characteristics, such as dialysis vintage and previous transplantation, could contribute to a more accurate KDRI. However, for validation purposes, we applied the same strategy as described by Rao et al (2009), to stratify the analyses on recipient age and diabetes.

The current study shows directions for an updated more accurate KDRI, such as the need to include interaction of donor type and cold ischemia time which support previous research.2 Although the KDRI performs equally well, our results show the limitations of the KDRI for selecting donor kidneys for transplant patients. Future studies should focus on how the KDRI should be adapted to be used for allocation for longevity matching between donors and recipients in the Eurotransplant region.


1. Nederlandse Transplantatie Stichting. Annex of stats and figures 2015 (In Dutch: Cijferbijlage 2015). 2016.
2. Summers DM, Johnson RJ, Hudson A, et al. Effect of donor age and cold storage time on outcome in recipients of kidneys donated after circulatory death in the UK: a cohort study. Lancet. 2013;381:727–734.
3. Snoeijs MG, Schaubel DE, Hene R, et al. Kidneys from donors after cardiac death provide survival benefit. J Am Soc Nephrol. 2010;21:1015–1021.
4. Nyberg SL, Matas AJ, Kremers WK, et al. Improved scoring system to assess adult donors for cadaver renal transplantation. Am J Transplant. 2003;3:715–721.
5. Rao PS, Schaubel DE, Guidinger MK, et al. A comprehensive risk quantification score for deceased donor kidneys: the Kidney Donor Risk Index. Transplantation. 2009;88:231–236.
6. Schold JD, Kaplan B, Baliga RS, et al. The broad spectrum of quality in deceased donor kidneys. Am J Transplant. 2005;5:757–765.
7. Watson CJ, Johnson RJ, Birch R, et al. A simplified donor risk index for predicting outcome after deceased donor kidney transplantation. Transplantation. 2012;93:314–318.
8. Karpinski J, Lajoie G, Cattran D, et al. Outcome of kidney transplantation from high-risk donors is determined by both structure and function. Transplantation. 1999;67:1162–1167.
9. Remuzzi G, Cravedi P, Perna A, et al. Long-term outcome of renal transplantation from older donors. N Engl J Med. 2006;354:343–352.
10. 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;74:1281–1286.
11. De Vusser K, Lerut E, Kuypers D, et al. The predictive value of kidney allograft baseline biopsies for long-term graft survival. J Am Soc Nephrol. 2013;24:1913–1923.
12. Munivenkatappa RB, Schweitzer EJ, Papadimitriou JC, et al. The maryland aggregate pathology index: a deceased donor kidney biopsy scoring system for predicting graft failure. Am J Transplant. 2008;8:2316–2324.
13. Royston P, Altman DG. External validation of a Cox prognostic model: principles and methods. BMC Med Res Methodol. 2013;13:33.
14. van Houwelingen HC. Validation, calibration, revision and combination of prognostic survival models. Stat Med. 2000;19:3401–3415.
15. Kootstra G, Kievit JK, Heineman E. The non heart-beating donor. Br Med Bull. 1997;53:844–853.
16. Organ Procurement and Transplantation Network. KDPI calculator. Updated 2017. Accessed November 14, 2016.
17. Israni AK, Salkowski N, Gustafson S, et al. New national allocation policy for deceased donor kidneys in the United States and possible effect on patient outcomes. J Am Soc Nephrol. 2014;25:1842–1848.
18. Royston P. Tools for checking calibration of a Cox model in external validation: prediction of population-averaged survival curves based on risk groups. Stata J. 2015;15:275–291.
19. Harrell FE Jr, Lee KL, Califf RM, et al. Regression modelling strategies for improved prognostic prediction. Stat Med. 1984;3:143–152.
20. Blanche P, Dartigues JF, Jacqmin-Gadda H. Estimating and comparing time-dependent areas under receiver operating characteristic curves for censored event times with competing risks. Stat Med. 2013;32:5381–5397.
21. Collins GS, Reitsma JB, Altman DG, et al. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD). Ann Intern Med. 2015;162:735–736.
22. R: a language and environment for statistical computing [computer program]. R Core Team; 2014.
23. Organ Procurement and Transplantation Network (OPTN). A guide to calculating and interpreting the Kidney Donor Profile Index (KDPI). Published March 7, 2016. Accessed December 15, 2016.
24. Rose C, Schaeffner E, Frei U, et al. A lifetime of allograft function with kidneys from older donors. J Am Soc Nephrol. 2015;26:2483–2493.
25. Gondos A, Dohler B, Brenner H, et al. Kidney graft survival in Europe and the United States: strikingly different long-term outcomes. Transplantation. 2013;95:267–274.
26. Molnar MZ, Nguyen DV, Chen Y, et al. Predictive score for posttransplantation outcomes. Transplantation. 2017;101:1353–1364.
27. Hernandez RA, Malek SK, Milford EL, et al. The combined risk of donor quality and recipient age: higher-quality kidneys may not always improve patient and graft survival. Transplantation. 2014;98:1069–1076.
28. Heaphy EL, Goldfarb DA, Poggio ED, et al. The impact of deceased donor kidney risk significantly varies by recipient characteristics. Am J Transplant. 2013;13:1001–1011.
29. Irish WD, Ilsley JN, Schnitzler MA, et al. A risk prediction model for delayed graft function in the current era of deceased donor renal transplantation. Am J Transplant. 2010;10:2279–2286.

Supplemental Digital Content

Copyright © 2018 Wolters Kluwer Health, Inc. All rights reserved.