Despite the fact that suboptimal kidneys have worse outcomes, differences in waiting times and wait-list mortality have led to variations in the use of these kidneys. It is unknown whether aggressive center-level use of one type of suboptimal graft clusters with aggressive use of other types of suboptimal grafts, and what center characteristics are associated with an overall aggressive phenotype. United Network for Organ Sharing (UNOS) data from 2005 to 2009 for adult kidney transplant recipients was aggregated to the center level. An aggressiveness score was assigned to each center based on usage of suboptimal grafts. Deceased-donor transplant volume correlated with aggressiveness in lower volume, but not higher volume centers. Aggressive centers were mostly found in regions 2 and 9. Aggressiveness was associated with wait-list size (RR 1.69, 95% CI 1.20–2.34, p = 0.002), organ shortage (RR 2.30, 95% CI 1.57–3.37, p < 0.001) and waiting times (RR 1.75, 95% CI 1.20–2.57, p = 0.004). No centers in single-center OPOs were classified as aggressive. In cluster analysis, the most aggressive centers were aggressive in all metrics and vice versa; however, centers with intermediate aggressiveness had phenotypic patterns in their usage of suboptimal kidneys. In conclusion, wait-list size, waiting times, geographic region and OPO competition seem to be driving factors in center-level aggressiveness.
As the prevalence of end-stage renal disease (ESRD) increases, there has been a parallel increase in the size of the deceased-donor kidney transplantation (DDKT) wait-list (1–3). DDKT not only increases patient survival, but also provides a better quality of life and confers a more cost-effective option over long-term dialysis for most patients (3–5). However, the gap between available organs and the number awaiting transplantation widens annually (6). Disappointingly, many patients currently on the wait-list are more likely to die than they are to receive a kidney transplant (7).
Disparity between organ supply and the rising demand for kidneys has led to the use of organs that were previously deemed undesirable for transplantation (8–10). A set of characteristics defining expanded criteria donors (ECD) was implemented by the United Network for Organ Sharing (UNOS) in October 2002 (11–13). Shortly thereafter, a significant increase in the number of ECD kidneys utilized for transplantation was noted. In addition, the use of donation after cardiac death (DCD) kidneys has further increased the donor pool (10,14–16). As centers become more comfortable transplanting suboptimal kidneys, single-center and national database analyses have reported increased utilization of kidneys with elevated terminal creatinines, HCV infection and extended cold ischemia time (CIT) (5,8,13,17–19). Although there are reports discussing the use of these grafts and their outcomes, little is known about whether transplantation of suboptimal grafts is clustered by center and what center-level factors might be associated with these practices (11).
All transplant centers aim to provide patients with suitable organs in a timely manner, and even the use of suboptimal grafts provides better quality of life and improved survival over dialysis for some patients (20). However not all transplant centers use suboptimal deceased-donor organs (1,11,17,21–25). We hypothesized that an “aggressive center phenotype” exists and that some centers have become more aggressive in an effort to help their wait-listed patients achieve transplantation. To better understand this potential effect, we studied patterns of center-level clustering of suboptimal graft utilization in 5 years of recent national data.
Materials and Methods
We analyzed center-level UNOS data for 46 752 adult kidney-only deceased-donor kidney transplants between January 1, 2005 and December 31, 2009. We excluded centers if they had more than 2 years with no deceased-donor transplants or fewer than 10 transplants per year for all 5 years. Ultimately, we analyzed data from 206 US transplant centers.
Defining aggressiveness by organ type
Based on the available literature and clinical judgment, seven categories of suboptimal kidneys were selected for exploration: donor age, hepatitis B virus (HBV) infection, hepatitis C virus (HCV) infection, donor terminal creatinine, DCD, national import and CIT>24 h. Missingness was 0% for age, donor creatinine, DCD and organ share type and <1% for HCV, HBV and CIT >24.
For each category, cumulative distribution functions were plotted for various cut-points, and the area under the cumulative distribution curve (AUC) was used to assess center-level clustering. For example, Figure 1 illustrates that organs from donors over 60 were less clustered than organs from donors over 65 or 70. The most clustered were organs from donors over 70, where 150 centers did not even use these organs, and 50% of these organs were used by only 20 centers. We sought definitions of suboptimal that would not be too clustered (for example, too few centers would be considered aggressive in the category of older donors if we defined suboptimal as donor age ≥ 70, thus making it less likely to find centers that are aggressive in more than one category) but that would not be too widely distributed (for example, too many centers would considered aggressive in the category of older donors if we defined suboptimal as donor age ≥ 60, thus diluting the aggressiveness and making it less likely that aggressive phenotypes would arise). We used the calculated AUC for the cumulative distribution curves described above to represent the clustering factor (CF) for each type of suboptimal organ, and selected CFs in the range of 0.7 < CF < 0.82 (Table 1); for these CFs, approximately 75 centers did not use the given organ type at all, and approximately 45 centers used 50% of the organs of the given type. For example, we selected age ≥ 65 as the ideal suboptimal cut-point in the older donor category.
Table 1. Clustering factors (CF) for various cut-points of categories potentially considered aggressive in deceased donor kidney transplantation
Less evidence of clustering (CF <0.70)
Evidence of Clustering (0.70 ≤ CF ≤ 0.82)
Too much clustering (CF >0.82)
Cumulative distribution functions were plotted for each cut-point (see Figure 1), and the area under these curves is reported as the clustering factor. CF <0.7 was considered not clustered enough for defining a center's aggressiveness, and CF >0.82 was considered too clustered. A CF of 0.7–0.82 was used to define aggressiveness cut-points for each metric, which were then decile-scored, averaged and assigned on a center level.
Nationally imported graft
time > 24 hours
Defining the aggressiveness phenotype
The suboptimal cut-point was established for each graft category as above (Table 1, middle column indicates selected cutoff). Then, category-specific aggressiveness scores for each center (the degree to which the center was aggressive in using suboptimal organs of a given category) were determined according to the relative proportion of center volume comprised grafts of that category. For each of the seven categories, the centers were ordered by the proportion of their deceased-donor transplants composed of kidneys in that category, and assigned a decile score (1 = least aggressive, 10 = most aggressive). The final aggressiveness score was calculated by averaging the individual decile scores for each category at a given center. The distribution of aggressiveness scores was explored, and for the purposes of dichotomous analyses, an aggressiveness score of 7 or greater was classified as “aggressive”. A sensitivity analysis was also performed by scoring aggressiveness based on the number of transplants performed at each center, rather than the proportion of transplants. Correlation between volume-based and proportion-based estimates was high (0.82); therefore, proportions were chosen so that the relationship between volume and aggressiveness could be explored.
Wait-list size was defined as the number of active registrants listed at a center at the beginning of the study period. Organ shortage was defined as the ratio of patients waiting at all centers within an OPO to kidneys recovered by that OPO (annually). Median waiting time for an optimal graft was defined as time to transplantation for any individual who was added to the wait-list between 2000 and 2009 who received an graft that did not meet any of the seven aggressive characteristics (that is, a locally or regionally shared graft from a brain dead donor who was <65 years old with terminal creatinine < 2.5, less than 24 h of CIT and no evidence of HCV or HBV infection). Kaplan–Meier estimates were used to determine median waiting time at aggressive versus nonaggressive centers censoring for live donor transplant, death on the wait-list or transplantation with an aggressive graft. The wait-list size, organ shortage and median waiting times were than categorized into tertiles for regression analysis.
Aggressiveness and graft survival
Overall graft survival was estimated using Kaplan–Meier methodology. One- and 3-year graft survival estimates were compared between aggressive and nonaggressive centers using a log-rank test. Cox proportional hazard models were utilized to determine the unadjusted and adjusted hazards of graft loss and death at aggressive versus nonaggressive centers. All variables from the SRTR program-specific reports, in identical functional forms, were used for risk adjustment.
Univariate comparisons of center characteristics for aggressive and nonaggressive centers were performed using t-tests for pseudonormally distributed continuous variables, rank-sum tests for non-normally distributed continuous variables and chi-squared tests for categorical variables. Generalized linear models (GLM) were used to model the association between aggressiveness and multiple center-level characteristics including median center waiting time, wait-list size and organ availability. To obtain estimates of relative risk, a GLM was fit using a Poisson model with robust variance estimation as previously described (1).
Potential phenotypes were explored by a hierarchical agglomerative cluster analysis with Ward's linkage using the hclust() function in the stats package and graphics functions from the gplots package for R version 2.13.0 (Vienna, Austria). Cluster analysis was chosen to allow for empirical selection of clusters of centers with similar aggressive practices. A hierarchical agglomerative approach allows for the organization of individual entities (in this case transplant centers) into progressively large, more homogenous groups (in this case, centers with similar transplantation practices). Ward's linkage utilizes an analysis of variance approach to minimize the differences between centers within a given cluster. Other than cluster analysis, all other analyses were performed using STATA 11.0/MP for Linux (College Station, TX, USA).
Of 206 eligible transplant centers performing 46 752 adult deceased-donor kidney-only transplants during the study period, the mean annual volume per center was 45.7 (median 36, interquartile range 39.4, range 4.4–173). Decile-based aggressiveness scores were calculated for each empirically derived category (age ≥ 65, HBV core positive, HCV positive, terminal creatinine >2.5, DCD graft age >50, nationally imported kidney and CIT >24 h) (Figure 1 and Table 1), based on the proportion of transplants represented by grafts of each category (Table 2). For each center, an average aggressiveness score was determined (based on the numerical average of the seven category-based scores) (Figure 2). For the binary analyses, a center was classified as “aggressive” if the average aggressiveness score was 7 or higher; based on this definition, 37 centers were defined as aggressive.
Table 2. Overall percentage of utilization of suboptimal kidneys
Median percentage (IQR)*
Columns 2–4 report the median percentage and interquartile range (IQR) of each type of suboptimal organ transplanted from among all deceased-donor transplants at a given center. For example, the median percentage of grafts from donors over the age of 65 transplanted at any given center was 2.9%.
Age > 65
DCD > 50
CIT >24 h
Aggressive center characteristics
Center volume and aggressiveness were associated in centers with volumes up to 80 deceased-donor transplants per year (Figure 3). Aggressive centers had slightly longer median waiting times (4.5 years vs. 4.0 years) (p = 0.03, Table 3). Aggressive centers did appear to be clustered within various geographic regions, particularly in regions 2 (DE, DC, WV, Northern VA, MD, NJ, RA) and 9 (NY, Western VT). No centers in single-center OPOs (defined as having one center perform >80% of an OPOs deceased-donor volume) were aggressive. The median wait-list size also varied between aggressive and nonaggressive centers (393 vs. 278, p = 0.008) as did the organ shortage ratio, with 7.9 candidates waiting for each kidney at aggressive centers versus 6.1 at nonaggressive centers (p < 0.001). Median wait-list size (RR 1.69, 95% CI 1.20–2.34, p = 0.002), organ shortage ratio (RR 2.30, 95% CI 1.57–3.37, p < 0.001) and median waiting time for an optimal graft (RR 1.75, 95% CI 1.20–2.57, p = 0.004) were associated with aggressiveness.
Table 3. Center-level characteristics of aggressive versus nonaggressive centers
1Dominated (>80% of OPO transplant volume) by one center.
2Defined as the number of candidates awaiting transplantation at a given OPO divided by the number of kidneys recovered annually.
3Defined as median waiting time for a nonaggressive graft (a locally or regionally shared graft from a brain dead donor who was <65 years old with terminal creatinine <2.5, less than 24 h of CIT and no evidence of HCV or HBV infection).
4Relative risks were not calculated for region (because there were an insufficient number of centers in some regions making the model unstable) and single-center OPO (nbecause this was perfectly correlated with nonaggressiveness thereby making regression modeling both unstable and unnecessary).
Single-center OPO1 (%)
Median wait list size
1.69 (1.20–2.34, p = 0.002)
Organ shortage ratio2
2.30 (1.57–3.37, p<0.001)
Median wait time for optimal graft3
1.75 (1.20–2.57, p = 0.004)
Median wait time for any deceased-donor graft
1.18 (0.85–1.65, p = 0.3)
Patient and graft loss
Overall graft survival was lower in aggressive centers (75.1% at 3 years vs. 81.7% at nonaggressive centers, p < 0.001) (Figure 4). Unadjusted risk of graft loss at aggressive centers was HR 1.42 (95% CI 1.35–1.49, p < 0.001); after risk adjustment, this was reduced to HR 1.11 (95% CI 1.05–1.18, p < 0.001). Unadjusted risk of death at aggressive centers was HR 1.24 (95% CI 1.16–1.32, p < 0.001); after risk adjustment, this was reduced to HR 1.01 (0.94–1.09, p = 0.8).
The aggressive phenotype
Patterns of aggressiveness were explored using heat map and cluster analysis (Figure 5). The most aggressive centers (shown under the red colorbar on the far right) were, in general, aggressive in all categories. Similarly, the least aggressive centers (shown under the blue colorbar) were, in general, nonaggressive in all categories. There appeared to be phenotypic patterns among centers with intermediate aggressiveness. For example, centers seen under the green colorbar were more comfortable utilizing grafts from older donors (such as donors >65 or DCD grafts from donors >50) and less comfortable utilizing imported grafts or grafts with prolonged CIT. Wait-list size increased as aggressiveness increased, as did the organ shortage ratio and waiting time for an optimal, nonaggressive graft.
We present a unique method to evaluate patterns of suboptimal deceased-donor graft use by transplant centers across the United States. Through the exploration of cumulative distribution plots by various graft characteristics, we identified seven categories of grafts that were only being transplanted by approximately one-quarter of centers (Table 1, center row). Based on the utilization of these types of grafts, we created an aggressiveness score, a composite score that quantifies the overall use of suboptimal grafts at a given center. We identified 37 aggressive centers with the highest utilization of suboptimal grafts. These centers were clustered in the northeast region, with over 30% residing in each of regions 2 and 9. Univariate analysis demonstrated that increased wait-list size, waiting times and organ shortage were associated with aggressiveness. Additionally, not one center in a single-center OPO was found to be aggressive. Finally, cluster analysis demonstrated that while the most aggressive centers were aggressive in all metrics, and vice versa, centers of intermediate aggressiveness demonstrated phenotypic preferences for which categories of suboptimal grafts they transplanted.
As the gap between supply and demand widens, utilization of suboptimal grafts is necessary and occurring at increasing rates (10–12,26). Previous studies have demonstrated acceptable outcomes with various types of suboptimal kidneys (1,8,13,20,26–27). However, in this study we have demonstrated that utilization of suboptimal grafts is not uniform across all transplant centers; in fact, there appears to be clustering of this practice at a relatively small number of centers. For most of our aggressive metrics, fewer than 50 centers contributed to the total transplant volume of any given category of suboptimal kidney. Three major factors appeared to be associated with aggressive practices. Not surprisingly, wait-list size, waiting time for an optimal graft and organ shortage were associated with aggressiveness. Of note, waiting time for any deceased-donor graft did not differ at aggressive versus nonaggressive centers.
Perhaps most interestingly, we found that none of the aggressive centers were from OPOs dominated by a single center. This suggests factors beyond wait-list characteristics, surgeon preference and geographic variation may play a role in the utilization of suboptimal grafts. More specifically, competition with other centers for the same donor pool may also play a role in the utilization of these types of grafts. While the most aggressive centers utilized all types of suboptimal grafts, centers of intermediate aggressiveness demonstrated interesting phenotypic patterns of graft utilization. These patterns are likely driven by center and surgeon comfort or preference with a given organ type (1,11,28–31). For example, there appeared to be centers that were more comfortable using grafts from elderly donors (and therefore utilized grafts from donors >65 and DCD grafts from donors >50 more routinely). Other centers demonstrated comfort with organs with infectious risks (such as from HCV-positive or HBV-core positive donors). While wait-list characteristics may be slightly responsible for this pattern, we hypothesize that surgeon and center preference are likely a driving force, and perhaps certain centers have institutional experience and/or protocols for certain suboptimal organs (17,25,31).
Our inferences are subject to several limitations that warrant discussion. Currently, there are no defined metrics for aggressiveness in kidney transplantation. In order to identify and classify suboptimal kidneys and aggressive transplant centers, decisions were made to classify graft types and center behaviors. These decisions were driven by the data in the UNOS registry, clinical judgment and extensive literature review. Since there is no gold standard for comparison, our cut-points and definitions for aggressiveness are subject to misclassification bias. In other words, choosing alternate cut-points for our aggressive metrics and aggressiveness score may have altered the number of aggressive centers identified. However, a sensitivity analysis using alternate definitions did not yield significant alterations in our univariate regression or cluster analysis, suggesting that our inferences were robust to minor alterations in the definition. Additionally, aggressiveness was defined by metrics that describe center utilization patterns, but do not account for recovery practices within a given OPO. Finally, the center-level and OPO-level characteristics utilized in this analysis might be surrogates for other undefined or unmeasured characteristics of the center or OPO.
Unfortunately, the current supply of deceased-donor kidneys cannot meet the increasing demand for these organs. While the demand is vast, only a small percentage of transplant centers routinely utilize organs from all types of suboptimal donors. Expansion of these utilization practices could help to decrease wait-list sizes, waiting times and wait-list mortality for the appropriate candidates, yet transplant centers differ vastly in their willingness to use suboptimal organs for this purpose. One possible explanation for this phenomenon may be related to the Centers for Medicare and Medicaid Services (CMS) regulations on performance outcomes. With penalties in place for those centers failing to meet certain performance measures, this has led to conflicting interests as centers desire to recover more organs for transplantation, yet must achieve optimal outcomes from suboptimal organs that might not be properly accounted for in performance-based models (32–35). Aggressiveness subsequently threatens the center's existence and may hinder utilization of suboptimal grafts that could have provided a survival benefit for the right patient (14,34).
With the methods developed in this study, we provide a framework for understanding and directly comparing center-level practices regarding suboptimal organ utilization across the United States. If reported to centers, a metric of relative aggressiveness might help a center titrate acceptance behavior to the relative demand for organs at that center. If reported to the public, insights on center practices may help patients with decreased likelihood of receiving an offer, particularly those that are highly sensitized or harder to transplant, make better informed decisions about where to list.
As a study of the United Network for Organ Sharing database, this work was supported in part by Health Resources and Services Administration contract 234-2005-370011C. The content is the responsibility of the authors alone and does not necessarily reflect the views or policies of the Department of Health and Human Services.
The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation.