SEARCH

SEARCH BY CITATION

Keywords:

  • Cold ischemia time;
  • deceased donor transplantation;
  • DonorNet 2007;
  • equity

Abstract

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

In 2007, UNOS released DonorNet 2007® (DN07) in hope of improving allocation equity and efficiency. We hypothesized that hard-to-place organs might be less efficiently handled through this regimented process. We analyzed associations between DN07 and center-level equity, number of refusals per organ and cold ischemia time (CIT). A total of 8244 kidney transplants between 1/2006 and 12/2006 (pre-DN07) were compared with 6029 transplants between 5/2007 and 2/2008 (post-DN07). Distribution equity was assessed by the Gini coefficient, changes in the number of refusals and CIT by negative binomial regression and discard rates by logistic regression. We estimated quantile-specific differences in CIT by bootstrapping. We found no significant change in center-level distribution equity after DN07. Number of refusals per organ increased by 20% (adjusted rate ratio 1.121.201.28, p < 0.001) at the patient level and 11% (ARR 1.071.111.16, p < 0.001) at the center level. Regression models of CIT showed no global change in CIT associated with DN07, but those kidneys with the longest CIT pre-DN07 had statistically significantly longer CIT post-DN07. The discard rate also increased significantly (ARR 1.061.111.17, p < 0.001). DN07 has not improved equity or efficiency in allocation of deceased donor kidneys, and may be harming the allocation of hard-to-place kidneys.


Background

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

On January 1, 2007, the United Network for Organ Sharing (UNOS) released DonorNet 2007®* (DN07; a registered trademark of UNOS) for voluntary use. Use of DN07 became mandatory on April 30, 2007 (1). The stated goal of DN07 was to ‘facilitate and expedite organ placement’ (2). To accomplish this goal, a system of electronic organ offers was put into place for organs allocated outside of an organ procurement organization (OPO), in other words for organs refused by all centers in a given OPO. Through this system, a ‘national list’ is generated for the organ, and offers must be made to centers in the order of this list, with only three open offers allowed at a time and up to 1-h allowance to decline an offer. Thus, as an extreme example, if the first center that would take a marginal kidney is 100th on the list, it is conceivable that 33 rounds of 1-h offers might have to pass before the offer is made to that center.

We suspect that, prior to DN07, OPOs had developed good heuristics for quickly allocating marginal organs, including directed offers to centers with a high likelihood of accepting the offers. Although, in general, heuristic-based systems are inexact and possibly less equitable, we hypothesized that the heuristics in place before DN07 were more efficient than the new regimented system yet similarly equitable because of the tendency for only certain centers to use hard-to-place kidneys.

To test our hypothesis, we analyzed data on kidney transplants before and after the implementation of DN07, comparing center-level distribution equity, number of refusals, cold ischemia time (CIT) and discard rate in the 12-month period before the voluntary release of DN07 to those metrics in the 9 months after use of DN07 became mandatory.

Methods

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Study population

Kidney offers and transplants in the year prior to DN07 (January 1, 2006 to December 31, 2006, n = 8244 transplants) were compared with those in the 9-month period following mandatory implementation of DN07 (May 1, 2007 to February 1, 2008, n = 6029 transplants). Transplants occurring during the 4-month period of voluntary DN07 use between January 1, 2007 and April 31, 2007 were excluded. Also excluded were kidneys typically allocated in an expedited manner: pediatric recipients; multiorgan transplants and zero-mismatch transplants.

Center-level distribution equity

We evaluated the effect of DN07 on equity of distribution to transplant centers using the Gini coefficient, a measure of inequality (3–5), and the Lorenz curve, a graphical representation of inequality (6). The Lorenz curve displays the cumulative proportional distribution of a value (in this case, a center's proportion of all import kidneys received), sorted from lowest to highest (see Figure 1). In the case of perfect equality, the Lorenz curve would fall along the diagonal reference line. The Gini coefficient is equal to twice the area between the reference line and the Lorenz curve, and ranges from 0 (same number of kidneys distributed to each center) to 1 (all kidneys distributed to the same center). Thus, an increase in the Gini coefficient over time would correspond to decreased equality, whereas a decrease would correspond to increased equality. The Gini coefficient was computed for pre- and post-DN07 transplants, and a 95% confidence interval for the change from pre- to post-DN07 was calculated via jackknifing. The change in the Gini coefficient was estimated for the entire dataset and in various subgroups (local/regional/national exports, no ECD/ECD, donor age quintiles, quintiles of number of refusals of an organ and distance quintiles). Lists of the top 20 centers by number of imported kidneys received before and after the implementation of DN07 were compiled to evaluate whether the same centers received the most imported kidneys before and after DN07. To account for the allocation of more kidneys to centers with longer waiting lists, and thereby comparing center-level equity rather than center-level equality, the same analysis was repeated with the number of organs received normalized to the size of the waiting list on July 2, 2006 (middle of pre-DN07 study period) and August 20, 2007 (middle of post-DN07 study period).

image

Figure 1. Lorenz curves demonstrating inequality of kidney imports by the center, pre- and post-DN07. Perfect equality is represented by the diagonal reference line; the area between the reference line and the curve is half the Gini coefficient of inequality. The upper graph shows Lorenz curves for the raw number of kidney imports by center; the lower graph shows curves for the number of imports normalized by wait list size.

Download figure to PowerPoint

Refusals

To evaluate whether DN07 changed the average number of refusals per kidney, the number of refusals was modeled as count data and found to be overdispersed relative to the Poission distribution. Therefore, negative binomial regression was used to compare the number of refusals before and after DN07. Refusals were counted and compared at the patient level (number of patients turned down before the final patient who received the kidney) and the center level (number of centers which refused the kidney before the final center which received the kidney). This analysis was performed over the whole dataset and in various subgroups as described above (local/regional/national exports, no ECD/ECD, donor age quintiles and quintiles of the distance the organ traveled).

Cold ischemia time

Negative-binomial regression models were also used to compare CIT before and after DN07 as described above. A quantile–quantile (Q–Q) plot of CIT pre- and post-DN07 was used to compare differences in the shape of the entire CIT distribution, rather than focusing only on the mean change. Extreme values of CIT (namely those with CIT > 70 h) were investigated as possible data entry errors. Of 15 transplants with CIT reported as >70 h, 2 occurred pre-DN07 (both 72-h CIT) and 13 occurred post-DN07 (range 71–90 h).

To further investigate these findings, several additional analyses were performed. First, several models were constructed to evaluate the effect of various covariates in addition to DN07 on CIT. These included a univariate model, a multivariate model that adjusted for donor age and ECD characteristics, and a multivariate model that adjusted for all of the above as well as distance the organ traveled, quintile of individual refusals and regional/national export. Second, models were also constructed using subgroups of only nationally exported kidneys, and only kidneys in the fifth quintile of individual refusals (those with the most refusals). Third, Q–Q plots of CIT pre- and post-DN07 were repeated with all extreme values (CIT > 70 h) dropped (data not shown), confirming that our inferences were not sensitive to these outliers. To assess the comparability of donor quality pre- and post-DN07, we compared demographic characteristics of donors from the pre-DN07 era to those from the post-DN07 era, for both all donors and the 5% of donors with the longest CIT.

Finally, the apparent DN07-associated increase in CIT was also evaluated by a nonparametric bootstrap method. One thousand bootstrapped samples were drawn with replacement from the dataset, each of size equal to the original dataset. For each repetition, for each of 200 quantiles, we calculated a quantity we called Δp representing the change in CIT pre- to post-DN07 for that quantile. Thus, a positive Δp represents an increase in CIT associated with the implementation of DN07 for that quantile, and a negative Δp represents a decrease. We used the bootstrapped estimates of Δp at each quantile to create a 95% confidence interval for Δp associated with DN07 across the range of pre-DN07 CIT.

Discard rate

The rate of discarded kidneys pre- and post-DN07 was assessed using a dataset of 24 450 kidneys recovered for transplantation during the same time period. Discard was considered when a kidney was recovered but not transplanted. An odds ratio (OR) of discard associated with DN07 was calculated in univariate and multivariate models using logistic regression; however, since an OR overestimates the risk ratio, we also modeled the RR of discard associated with DN07 using Poisson regression with a robust variance estimator as per the method of Zou (7). RR was calculated in a univariate and a multivariate model controlling for donor blood type, gender, cause of death, age quartile, race and serum creatinine level.

Statistical analysis

All statistical tests were two-sided with an α of 0.05. Model diagnostics for the regression models of the number of refusals and CIT included Q-normal plots to assess normality of residuals and box plots of residuals pre- and post-DN07 to assess homoscedasticity. Statistical programming was performed using Stata 10.1 (StataCorp LP, College Station, TX), except for the Q–Q plot and Δp bootstrapping which were performed using R 2.7.2 (The R Foundation for Statistical Computing, Vienna, Austria). Gini coefficients were calculated using the INEQUAL7 module for Stata by Philippe Van Kerm (8), and Lorenz curves were created based on code from the sg30 module for Stata by Edward Whitehouse (9). Confidence intervals are reported as per the method of Louis and Zeger (10).

Results

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Study population characteristics

The study periods before and after DN07 were similar in proportion for ECD kidneys, import status and donor age (Table 1). The rate of transplants per day decreased slightly from 22.65 to 21.84, a decrease of 3.6%; the total number of kidneys transplanted from May 1 to December 31 (the time period for which we have data for both eras) decreased from 5619 to 5432, a decrease of 3.3%.

Table 1.  Characteristics of the study population: deceased donor kidney transplants before and after DonorNet 2007 (DN07)
 Pre-DN07Post-DN07p (χ2)
N82446029 
Transplant date range1/01/06–12/31/065/01/07–2/01/08 
Transplants per day (mean)22.65  21.84   
Expanded criteria donors
 Non-ECD79.80%79.53% 
 ECD19.71%19.11%0.52
 Missing ECD 0.49% 1.36% 
Disposition
 Local80.05%80.06% 
 Regional 7.36% 7.91%0.32
 National12.59%12.03% 

Center-level distribution equity

Equity in the distribution of kidneys was largely unchanged by DN07. The Lorenz curves of the cumulative distribution of imports by center were similar before and after the change (Figure 1), and 15 of the top 20 centers for kidney imports before DN07 stayed in the top 20 after DN07 was implemented (Figure 2). The Gini coefficient of kidneys received per center, representing a measure of inequality of distribution, showed no statistically significant change (−0.0160.0100.036) (Table 2, second column). Almost all of the many subgroups for which Gini coefficients were computed showed no statistically significant change. There was a statistically significant decrease in the Gini coefficient for the first quintile of the number of refusals (−0.072−0.039−0.007), an increase for the third quintile of refusals (0.0160.0500.083) and an increase for the fourth quartile of the distance the organ traveled (0.0170.0490.080).

image

Figure 2. Change in ranks of the top 20 transplant centers by the number of imported kidneys, in order (top = highest volume of imported kidneys), pre- and post-DN07. The number represents the pre-DN07 rank of each center; the order of the right column shows how these centers ranked post-DN07. For example, the second ranked center by number of imported kidneys pre-DN07 became the fourth ranked center post-DN07; similarly, the third ranked center pre-DN07 became the first ranked center post-DN07. Fifteen of the top 20 in one list also fall in the other; centers falling in only one list are shown in bold. For centers listed in the right column which were not ranked in the top 20, the pre-DN07 rank is shown; for example, the 24th ranked center pre-DN07 became the 6th ranked center post-DN07.

Download figure to PowerPoint

Table 2.  Change in the Gini coefficient of inequality as a result of DonorNet 2007 (pre-DN07 to post-DN07), by subgroup, with 95% confidence intervals
 Change in the Gini coefficient, pre- to post-DonorNet
RawNormalized by wait list size
  1. Figures in bold are statistically significant at the 95% level (p < 0.05). Subgroups of donor age, number of refusals and the distance the organ traveled are shown by quintiles (Q1 = quintile with lowest values, Q5 = quintile with highest values).

All kidneys−0.0160.0100.036−0.0190.0170.053
Local−0.0090.0130.036−0.0170.0220.061
Regional−0.0500.0050.060−0.0230.0320.086
National−0.068−0.0070.054−0.0350.0260.087
No ECD−0.0140.0120.037−0.0130.0250.063
ECD−0.0270.0120.0510.0090.0560.102
Donor age Q1−0.0340.0090.051−0.0250.0430.112
Donor age Q2−0.0210.0200.061−0.0330.0340.101
Donor age Q3−0.0150.0220.0590.0000.0460.092
Donor age Q4−0.0220.0130.048−0.0130.0340.080
Donor age Q5−0.0280.0130.054−0.0200.0240.069
No. of refusals Q10.0720.0390.007−0.056−0.0000.055
No. of refusals Q2−0.0030.0370.076−0.0350.0150.064
No. of refusals Q30.0160.0500.083−0.0370.0360.109
No. of refusals Q4−0.0050.0340.0730.0080.0620.117
No. of refusals Q5−0.063−0.0160.031−0.055−0.0000.054
Distance Q1−0.0280.0040.036−0.0300.0140.058
Distance Q2−0.028−0.0010.027−0.0090.0300.070
Distance Q3−0.0290.0010.032−0.0320.0520.136
Distance Q40.0170.0490.080−0.0140.0220.058
Distance Q5−0.0500.0100.070−0.0270.0230.072

The Gini coefficient normalized by waiting list size also showed no statistically significant change. The coefficient for all donations increased by −0.0190.0170.053 (Table 2, third column). Again, for three of the subgroups, the change in the Gini coefficient was statistically significant at the 95% confidence level. However, since the categories with a statistically significant change in the normalized Gini were all different from the categories in the raw Gini, and since no clear pattern emerges from the stratification by quintiles and no correction was done for multiple comparisons, the practical significance of these changes is likely minimal.

Refusals

The average number of refusals per kidney appears to have increased in the post-DN07 period (Table 3). For the entire dataset, a 20% increase in the number of refusals per kidney was observed after the policy change (rate ratio 1.121.201.28, p < 0.001). The increase was statistically significant for both ECD (rate ratio 1.161.321.51, p < 0.001) and non-ECD kidneys (rate ratio 1.081.161.25, p < 0.001). Although the point estimate for the increase in the number of refusals was positive for all substrata, the change was not statistically significant for the first and third quartiles of donor age, or for the lower three quartiles of the distance the organ traveled (Table 3, second column). A similar pattern was observed in center refusals; although the proportion of the increase was in general lower, since the general magnitude of the number of refusals was higher when counting patients than when counting centers, it was still statistically significant (Table 3, third column).

Table 3.  Ratio of average number of individual refusals per kidney post-DN07 to pre-DN07 for various subgroups, by negative binomial regression
  Risk ratio of refusals post- to pre-DN07 (95% CI)
Individual refusalsCenter refusals
  1. Subgroups of donor age and distance the organ traveled are shown by quintiles (Q1 = quintile with lowest values, Q5 = quintile with highest values).

All kidneys 1.121.201.281.071.111.16
Non-ECD 1.081.161.251.031.081.12
ECD 1.161.321.511.101.201.31
Q10.901.041.200.910.991.08
Q21.271.471.711.061.161.27
Donor ageQ30.881.021.180.941.021.12
Q41.071.231.421.041.131.24
Q51.171.341.541.171.281.39
Q10.941.071.221.001.081.17
Q20.891.011.161.011.091.17
DistanceQ30.911.041.180.880.941.01
Q41.181.371.580.961.051.14
Q51.131.281.461.151.251.36

Cold ischemia time

No statistically significant change in the mean CIT associated with DN07 was observed. The point estimate for the estimated coefficient was very near 1, with tight confidence intervals, in the univariate (RR 0.981.001.02, p = 0.90) as well as multivariate models with donor age and ECD status included (RR 0.981.001.02, p = 0.83); with donor age, ECD status and export status (RR 0.991.001.02, p = 0.42) and the full model which added refusal quintile and distance (RR 0.991.011.03, p = 0.20). Results were similar in a model restricted to only nationally exported kidneys (RR 0.971.001.05, p = 0.71) and only the fifth quintile of refusals (RR 0.971.001.04, p = 0.81). These results suggest that, on average among all kidneys transplanted, DN07 had negligible influence on CIT.

However, the Q–Q plot of CIT pre- versus post-DN07 suggested an increase in CIT in the post-DN07 era for those kidneys where CIT was prolonged in either era (CIT > 36 h) (Figure 3). In other words, kidneys with long CIT would be expected to have even longer CIT after DN07. As established by bootstrap analysis, having a pre-DN07 CIT > 40 h was associated with a statistically significant positive Δp parameter (Figure 4). This graph suggests that CIT was unchanged for most kidneys, but that the longest CIT times were an additional 20–30 h longer post-DN07 than pre-DN07.

image

Figure 3. Quantile–quantile plot of cold ischemia time pre- versus post-DN07, showing the difference in distributions of CIT pre-DN07 (vertical axis) versus post-DN07 (horizontal axis) among kidneys with the longest CIT.

Download figure to PowerPoint

image

Figure 4. Bootstrapped 95% confidence intervals of change in CIT (Y-axis) post-DN07, by pre-DN07 CIT (X-axis), showing that kidneys with long CIT were associated with increases in CIT of 5–25 h after the implementation of DN07. The points represent the actual data: the difference between each pre-DN07 CIT and the corresponding post-DN07 CIT at the same percentile. The solid black line represents the median of 1000 bootstrapped repetitions of the points, and the gray lines represent a 95% confidence interval for the points.

Download figure to PowerPoint

There were some slight changes in donor demographics between the pre- and post-DN07 eras (Table 4). Mean donor serum creatinine increased from 1.14 to 1.17 mg/dL (p = 0.01) among all donors; an increase from 1.30 to 1.38 mg/dL in the 5% of kidneys with the longest CIT was not statistically significant (p = 0.33). The proportion of Hispanic donors increased from pre- to post-DN07 while the proportions of Caucasian and African-American donors decreased (p = 0.02 for all donors, 0.04 for the 5% with the longest CIT). A slight increase in the proportion of donations after cardiac death was statistically significant in all donors (11.2–13.0%, p = 0.001) but not in the 5% of kidneys with the longest CIT (10.7–12.2%, p = 0.58). No statistically significant change in the mean donor age or the proportion of donors with hypertension or diabetes was observed.

Table 4.  Donor demographics pre- and post-DN07, for both all kidneys and the 5% of kidneys with the longest CIT
 All donors5% of kidneys with the longest CIT
Pre-DN07Post-DN07pPre-DN07Post-DN07p
  1. Age and serum creatinine are reported as mean (SD).

Donor age39.4(17.0)39.3(16.9)0.640.6(19.0)42.3(16.0)0.2
Donor creatinine1.14(0.72)1.17(0.98)0.011.30(0.84)1.38(1.10)0.3
COD
 Anoxia17.60%19.06% 49.60%43.32% 
 CVA43.34%41.53%0.126.81%34.82%0.02
 Head trauma36.56%36.32% 0.00%0.81% 
 Other0.33%0.28% 0.00%0.00% 
Race
 White69.29%67.16% 68.36%63.97% 
 Black14.03%13.97% 16.89%13.77%0.04
 Hispanic13.48%15.33%0.0212.06%20.65% 
 Asian2.29%2.50% 1.34%1.21% 
 Other0.91%1.04% 1.34%0.40% 
Hypertension28.85%28.26%0.535.12%38.06%0.5
Diabetes7.47%7.40%0.911.80%9.72%0.4
Donation after cardiac death11.17%12.97%0.00110.72%12.15%0.6

Discard rate

Pre-DN07, 2303 of 13 890 kidneys (16.6%) recovered for transplantation were eventually discarded; post-DN07, 1979 of 10 560 kidneys (18.7%) were discarded. The crude RR of discard associated with DN07 was 1.061.121.19 (p < 0.001); in a multivariate model adjusting for donor blood type, gender, cause of death, age, rate and serum creatinine, the ARR was 1.061.111.17 (p < 0.001). The OR of discard associated with DN07 was 1.081.151.23 in a univariate model and 1.091.181.26 (p < 0.001) in a multivariate model.

Discussion

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Since the initiation of DN07, we have not been able to detect any change in equality or equity in center-specific distribution of deceased donor kidneys. DN07 was associated with a statistically significant increase of about 20% in patient-level refusals before a kidney was accepted for transplantation, and of about 11% in center-level refusals. While we found no overall change in the mean CIT associated with DN07, the highest percentile post-DN07 CITs (those above 36 h) were statistically significantly longer than the highest percentile pre-DN07 CITs, suggesting that the efficiency of easy-to-match kidneys is relatively unchanged but the efficiency of hard-to-match kidneys has worsened. The increase in the kidney discard rate associated with DN07 lends further support to this conclusion.

As far as we know, this is the first statistical analysis of these organ allocation metrics (equality, equity, refusals and CIT) since the introduction of DN07. Feedback to UNOS from the transplant community suggests that DN07 seems to have resulted in an increase in unwanted offers (11). This is consistent with our findings and also with our hypothesis that pre-DN07, OPOs had identified centers that were willing to accept marginal kidneys and had made these offers directly to those centers rather than in a systematic top-to-bottom manner; since post-DN07, OPOs are required to run the entire list in order and, as such, centers uninterested in marginal kidneys who in the past were skipped by OPOs aware of their practices are now receiving these offers by default.

This study is limited to the first 9 months of post-DN07 transplants. Implementation of a complex new system may be associated with decreases in the efficiency at startup which fade over time; it is possible that kidney allocation will improve under DN07 as OPOs get used to the system. We did purposefully exclude the 4-month rollout period during which OPOs were first using DN07 for this reason. While inferences from only 9 months of transplants might be somewhat limited, the sample size is certainly adequate for many comparisons (over 6000 transplants). Furthermore, we feel that an analysis of early postimplementation outcomes might help guide future modifications.

We feel that DN07 is a powerful concept which needs a few practical modifications. For an organ likely to be accepted quickly by transplant centers, sequential electronic offers will likely work well and ensure an equitable distribution. For an organ likely to be accepted by very few transplant centers, an efficient system for skipping uninterested centers is required to balance the relative inefficiency of sequential offers (when compared to heuristic-based targeting of aggressive centers as practiced pre-DN07). Indeed there is a system in place for specifying general organ acceptance criteria, but our findings suggest that this system is not being efficiently utilized. One potential problem is the relatively rudimentary specifications that a center can provide (i.e. ranges for certain donor criteria). Perhaps an expansion of these criteria (for example, specifying acceptable ranges for certain characteristics, such as terminal creatinine, stratified by other characteristics, such as age) will be less limiting to centers with complex acceptance criteria. Another potential problem is the lack of enforcement of these criteria. It seems reasonable that a center that specifies a donor age range of 5–80 but declines 100 offers in a row from donors over 70 should be asked to tighten their age range criteria. If organ criteria are honestly and accurately specified in advance, marginal kidneys will likely reach the aggressive centers as quickly as they did before DN07, or possibly even more quickly.

We have seen no evidence of increased efficiency or equity in kidney allocation over the first 9 months of DN07. Analyses similar to ours using other organs would present a more complete picture of the effectiveness of DN07; improvements may be seen in those areas. However, insofar as DN07 was created to improve equity and efficiency in allocation, this study suggests that it has not achieved these goals.

Footnotes
  • *

    DonorNet is a registered trademark of UNOS.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

The advice of Dr. Brian Caffo (Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health) is gratefully acknowledged. Also, we very much appreciate the advice and data support of Katarina Linden (Research Department, United Network for Organ Sharing). The OPTN is supported by Health Resources and Services Administration contract 234-2005-370011C. The analyses described here are the responsibility of the authors alone and do not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.

References

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References