The Contribution of Donor Quality to Differential Graft Survival in African American and Caucasian Renal Transplant Recipients
Although a number of factors contributing to the disparity in graft survival between African American (AA) and Caucasian kidney transplant recipients have been described, the role of donor quality is less well understood. This study was undertaken to determine the impact of donor quality differences on this disparity, based on review of UNOS (United Network for Organ Sharing) data on deceased donor renal transplantation from 2000 to 2010. Donor quality was determined by the kidney donor risk index (DRI), and was compared between AA and Caucasian recipients. There were 33,405 Caucasians and 22,577 African Americans in the study, with mean DRI of 1.17 versus 1.27 (p < 0.001), respectively. In analysis 2,446 recipients of each race matched by propensity scoring (based on medical, socioeconomic and immunologic covariates), mean DRI was 1.25 for Caucasians and 1.28 (p = 0.02) for AA. The hazard ratio (HR) for graft failure associated with AA race was 1.8 (p < 0.001) on unadjusted analysis, and decreased to 1.6 (p < 0.001) after matching for DRI. These results indicate a significant disparity in quality of kidneys received by African Americans, which propensity analysis indicates is partially explained by differences in medical, immunologic and socioeconomic factors. Furthermore, this difference in donor quality partially accounts for poorer graft survival in African Americans.
body mass index
donation after cardiac death
donor risk index
extended donor criteria
hepatitis C virus
human leukocyte antigen
panel reactive antibody
receiver operating characteristic
United Network for Organ Sharing
Kidney transplantation remains the gold standard for treatment of patients with end stage renal disease when compared to long-term maintenance dialysis, which has been demonstrated across ethnic groups (1). As the field of transplantation has matured a number of disparities in access (2–4) to transplantation, as well as in posttransplant outcomes (4–8) between African American and Caucasian recipients have come to light. Several authors have proposed factors contributing to these disparities, including socioeconomic (5,8,9), genetic (10), immunologic (11,12) and even pharmacokinetic (13) issues.
Issues of donor quality also play a major role in posttransplant outcomes. Kidneys from African American donors, for example, have been shown to be associated with an increased risk of graft loss (6,14,15), especially when transplanted into other African Americans (16). The importance of donor quality in posttransplant outcomes is especially clear when looking at kidneys from expanded criteria donors, which pose a 70% greater risk of graft failure than lower risk organs (17). Despite the known importance of donor quality, there is a relative paucity of large-scale studies examining racial disparities in terms of donor quality in a systematic manner (18).
We undertook this study to determine the presence of and any possible contributing factors to differences in donor quality between African American and Caucasian first time recipients of deceased donor kidney transplants, as well as to ascertain the effect of any donor quality disparity on posttransplant outcomes.
Patients and Methods
Selection of patients for analysis
After receiving an institutional review board exemption, an analysis of the United Network for Organ Sharing/Organ Procurement and Transplantation Network database as of March 31, 2011 was performed. The target population for this study was Caucasian or African American first-time recipients of deceased donor kidney alone transplants from January 1, 2000 through December 31, 2009.
Assessment of donor quality
The primary goal of this study was to determine disparity in donor quality between African American and Caucasian recipients. The kidney donor risk index (DRI) derived by Rao et al. (19) was chosen as the measure for donor quality for this study as it avoids the creation of arbitrary cutpoints in continuous data, and that it provides more granular information than more traditional measures such as the expanded donor criteria. The DRI includes donor age, race, history of hypertension and diabetes, serum creatinine, cause of death, height and weight, donation after cardiac death status, hepatitis C status, B and DR locus mismatch level, cold ischemic time and whether an enbloc or double kidney transplant is to be performed. The score represents the estimated risk of graft failure relative to a hypothetical standard donor with a DRI of 1.0. For example, a DRI of 1.3 confers a 30% greater risk of graft failure than the reference donor.
Comparisons were made between the African American and Caucasian recipient cohorts. Continuous variables were summarized as mean/standard deviation and analyzed using Student's t-test while categorical variables were summarized as count/percentage and analyzed using chi-squared or Fisher's exact test, where appropriate. Graft survival was calculated as the time from transplant to graft failure, with censoring at the time of death with a functioning graft or last follow-up (right-censoring) in the case of Cox proportional hazards analysis (20). The cumulative incidence of graft failure in African Americans versus Caucasians was calculated by competing risks analysis (21) and compared using a modified chi-squared statistic as described by Gray (22). In this analysis, death and graft failure were treated as competing risks, while patients still alive at the end of follow-up with a functioning graft were censored.
To control for confounders related to recipient race that may influence observed differences in DRI, a 1:1 propensity-matched (23) cohort of African American and Caucasian recipients was created using a nearest neighbor algorithm (24). Matching was without replacement. Propensity scores were developed through logistic regression, with immunologic (HLA-A, -B and -DR antigens, peak PRA and ABO antigen), socioeconomic (payment source, educational achievement, employment status and UNOS Region) and medical (age, recipient bmi, etiology of renal failure, recipient diabetes status, weight, gender, recipient HCV status, recipient CMV status and days on the waiting list) factors as independent variables and recipient race as the dependent variable. The result of this model is to assign each patient a propensity score that describes the probability of the patient either being African American or Caucasian based upon the independent variables entered in the model. Differences in the overall DRI as well as individual components were then assessed in the propensity-matched cohort. To determine the relative contribution of social versus medical/immunologic confounders, separate propensity-matched cohorts were also created based on the social and medical/immunologic covariates listed above.
To explore the contribution of DRI to disparities in graft survival between African Americans and Caucasians, several Cox regression models were examined in sequence. First univariable models were fitted with recipient race and DRI respectively as the sole predictor variables to determine the baseline unadjusted hazard for graft failure posed by these factors. Next, to control for other potentially relevant covariates, a multivariable Cox regression model was fitted using purposeful variable selection (25,26). A full model was initially fitted with all the covariates used to estimate the propensity score listed above. Those factors which were either significant at the p < 0.1 level or resulted in a greater than 5% change in the hazard ratio associated with African American race were retained in the model after iteratively removing those that did not meet these criteria. The donor risk index was then included as a covariate in the third and final multivariable model. The difference between the hazard ratios corresponding to African Americans was then examined in the progressively adjusted models. Reductions in the hazard ratio would indicate that the additional covariates partially explain the observed disparity in graft survival between African American and Caucasian recipients.
As a final test of the contribution of DRI disparity to the difference in graft survival between African Americans and Caucasians, graft survival was analyzed in a cohort of African Americans and Caucasians matched on DRI. The matching process was similar to that used in the propensity matching scheme, namely 1:1 matching based on a nearest neighbor algorithm without replacement. The hazard ratio for African American race in this matched cohort was then compared to the unadjusted hazard ratio for African American race in the overall cohort. The percent of excess hazard explained by matching on DRI was calculated as (HRunmatched−HRmatched)/(HRunmatched−1)*100%. (27) The interaction between DRI and recipient race was also examined by performing Cox proportional hazards regression of graft survival stratified by DRI in the overall cohort. The three strata analyzed were DRI <1, 1≤DRI<1.5 and DRI ≥1.5.
Sensitivity analysis was also performed to determine whether significant bias may have been introduced into the analysis by exclusion of patients with missing data that did not allow for calculation of the DRI. First, chi-squared analysis was performed to determine whether patients with missing DRI data differed significantly by race. DRI for the patients with missing component variables was then calculated using multiple imputation, whereby five imputations were performed in which missing DRI component variables were predicted based on remaining DRI variables that were available for each patients. The five imputations were then combined to give estimates of the mean and standard deviation for the DRI in subset of patients with missing data. The imputed DRIs were then compared to the DRI for African Americans and Caucasians with complete data to determine whether the patients with missing data were significantly different from the subset with complete data. Cox proportional hazards analysis was then performed to determine whether graft failure in the patients with missing DRI data was significantly different from the patients who had complete data. All statistical analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC, USA), with the exception of competing risks analysis, which was performed in R (The R Project for Statistical Computing).
Overall recipient and donor characteristics
There were 55 982 recipients included in the study with complete DRI data collection, of whom 33 405 (59.67%) were Caucasian and 22 577 (40.33%) were African American. There were 3916 Caucasians and 2534 African Americans in the original dataset that had missing data for DRI and thus were not included in the analysis. The majority of patients analyzed were male (61.02%) at a mean age of 50.74 years at the time of transplant. Mean time on the waiting list was 734.02 days. Peak PRA was a mean of 12.79%. Remaining descriptive statistics of the recipients in this study are detailed in Table 1.
Table 1. Overall recipient characteristics
|Male gender||20 582 (61.6%)||135 77 (60.1%)|| <0.001 |
|Hepatitis C positive||1 022 (3.1%)||2 178 (9.8%)|| <0.001 |
|Cytomegalovirus positive||17 373 (54.2%)||16 079 (74.4%)|| <0.001 |
| EnBloc transplant ||478 (1.5%)||321 (1.4%)||0.726|
| Double kidney transplant ||484 (1.5%)||374 (1.7%)||0.050|
|Age (mean, std)||48.3 (14.1)||52.4 (15.0)|| <0.001 |
|Peak panel reactive Antibody (median, IQR)||0 (7)||0 (14)|| <0.001 |
|A locus mismatch (median, IQR)||1 (1)||2 (1)|| <0.001 |
| B locus mismatch (median, IQR) ||1 (1)||2 (1)|| <0.001 |
| DR locus mismatch (median, IQR) ||1 (2)||1 (1)|| <0.001 |
|Total HLA mismatch (median, IQR)||4 (3)||5 (1)|| <0.001 |
|Height (cm, mean, std)||170.2 (14.5)||171.1 (13.6)|| <0.001 |
|Days on the waitlist (median, IQR)||478 (702)||756 (973)|| <0.001 |
The average age of donors was 38.14 years, with 59.70% being males (n = 33 423) and 7496 (13.39%) African Americans. Terminal creatinine was a mean of 1.11 mg/dL. Cause of death was cerebrovascular accident in 21 717 (38.79%). Donation after cardiac death was uncommon (8.19% of all donors) as was utilization of HCV positive donors (2.91%). Mean cold ischemic time was 18.59 hours. The mean DRI was 1.21. Remaining donor characteristics are summarized in Table 2.
Table 2. Donor characteristics
|Donor risk index||1.17 (0.43)||1.27 (0.45)|| <0.001 |
| Age (years, mean, std) ||38.1 (17.2)||38.3 (16.8)||0.186|
| Terminal Cr (mg/dL, mean, std) ||1.08 (0.85)||1.14 (0.97)|| <0.001 |
| Weight (kg, mean, std) ||77.4 (23.6)||77.7 (23.7)||0.175|
| Height (cm, mean, std) ||169.3 (17.7)||169.2 (17.7)||0.450|
| Cold ischemic time (hours, mean, std) ||18.4 (8.9)||18.9 (9.5)|| <0.001 |
|Male gender||19 848 (59.4%)||13 575 (60.1%)||0.092|
| African American race ||2 896 (8.67%)||4 600 (20.37%)|| <0.001 |
| History of hypertension ||8 077 (24.18%)||5 963 (26.41%)|| <0.001 |
| History of diabetes ||1 884 (5.64%)||1 405 (6.22%)|| 0.004 |
| Death from stroke ||12 865 (38.51%)||8 952 (39.21%)||0.097|
| Donation after cardiac death ||2 631 (7.88%)||1 954 (8.65%)|| 0.001 |
| Hepatitis C positive ||402 (1.20%)||1 226 (5.43%)|| <0.001 |
Differences in donor risk index by recipient race
African American recipients received kidneys from higher risk donors as measured by most components of the donor risk index (Table 1, 2). Specifically, African Americans were more likely to receive a kidney from an African American donor (20.37% vs. 8.67%; p < 0.001). Utilization of hepatitis C positive donors was also significantly higher among African American compared to Caucasian recipients (5.43% vs. 1.20%; p < 0.001). There was also a statistically significant, though likely not clinically significant, increase in cold ischemic time for African American recipients (18.86 vs. 18.41 hours; p < 0.001). Overall, donor risk index for African American recipients was significantly higher than for Caucasian recipients (1.27 vs. 1.17; p < 0.001). The distribution of the DRI in African American and Caucasian recipients is presented as supporting information (Figure S1). There was no significant difference in the percentage of African Americans and Caucasians who received organs from extended criteria donors (17.66%, n = 3986 vs. 17.98%, n = 6005; p = 0.330).
The propensity model performed well in discriminating African American from Caucasian recipients, with an area under the receiver operating characteristic (ROC) curve of 0.96 (Figure S2). Matching on propensity score yielded a cohort of 2446 Caucasian and 2446 African American recipients (distribution of propensity scores pre- and postmatching are presented in Figure S3 and Figure S4, respectively). The African American and Caucasian groups were well matched on etiology of renal failure (p = 1.0), ABO type (p = 0.99), UNOS region (p = 0.99), A1 (p = 0.92), A2 (p = 0.96), B1 (p = 0.87), B2 (p = 1.0), DR1 (p = 0.80) and DR2 (p = 0.99) haplotypes, payment source (p = 0.98), HCV positivity (6.17% vs. 6.13%; p = 1.0), educational achievement (p = 0.99), diabetes status (p = 0.80), age (mean 52.4 vs. 53.0; p = 0.1), peak PRA (mean 14.8 vs. 15.0; p = 0.29), weight (mean 83.9 vs. 84.4kg; p = 0.29) and days on the waiting list (mean 823 vs. 835; p = 0.531).
In the propensity-matched cohort, there remained a minor difference in DRI between the African American (1.28) and Caucasian (1.25) cohorts (mean difference 0.03, 95% CI 0.0005–0.06; p = 0.02) (Figure S5). Comparison of donor risk index components between the two groups are outlined in Table 3. Notably, African Americans are still significantly more likely than Caucasians to receive a kidney from an African American donor (15.74% vs. 13.41%, OR 1.21, 95% CI 1.03 – 1.41; p = 0.02). Furthermore, utilization of HCV positive donors remained significantly higher in African American recipients (3.97% vs. 1.96%, OR 2.06, 95% CI 1.45–2.93; p < 0.001). In the cohort matched on social factors only, there were 12 726 recipients of each race. The mean DRI for African Americans in this group was 1.30 compared to 1.23 in the Caucasians (mean difference 0.07, 95% CI 0.06–0.09; p < 0.001). In the cohort matched on medical/immunologic factors only, there were recipients of each race. The mean DRI for African Americans in this group was 1.23 compared to 1.19 in the Caucasians (mean difference 0.04, 95% CI 0.02–0.06; p < 0.001).
Table 3. Differences in donor risk index components in a propensity-matched cohort of African American and Caucasian recipients
|Donor age (mean, std)||39.45 (16.71)||39.67 (16.76)||0.64|
|Donor weight (kg, mean, std)||79.00 (24.42)||79.17 (24.41)||0.80|
|Donor height (cm, mean, std)||169.4 (17.7)||169.1 (18.0)||0.61|
|Cold ischemic time (hours, mean, std)||19.12 (10.99)||18.75 (9.56)||0.21|
|B locus mismatch (median, iqr)||2 (1)||2 (1)||0.05|
|DR locus mismatch (mean, std)||1 (1)||1 (1)|| 0.03|
|Donor terminal creatinine (mg/dL, mean, std)||1.16 (0.83)||1.14 (0.70)||0.45|
|African American donor||328 (13.41%)||385 (15.74%)|| 0.02|
|Diabetic donor (n, %)||164 (6.70%)||196 (8.01%)||0.08|
|Hypertensive donor (n, %)||704 (28.78%)||706 (28.86%)||0.95|
|CVA as cause of death (n, %)||984 (40.23%)||974 (39.82%)||0.77|
|Hepatitis C positive donor (n, %)||48 (1.96%)||97 (3.97%)||<0.001|
|DCD donor (n, %)||245 (10.02%)||273 (11.16%)||0.19|
|Double kidney transplant (n, %)||33 (1.35%)||50 (2.04%)||0.06|
|En bloc transplant (n, %)||33 (1.35%)||35 (1.43%)||0.81|
Effect of the recipient race and donor risk index on graft survival
Both DRI and recipient ethnicity significantly correlated with graft survival in the univariable models. The hazard ratio for each one point increase in DRI in the over the baseline of 1.0 in the unadjusted model was 2.2 (95% CI 2.071–2.239; p < 0.001). One-, 5- and 10-year cumulative incidence of graft failure for African Americans was 7.5% (95% CI 7.2%–7.9%), 25.2% (95% CI 24.5%–25.9%) and 43.9% (95% CI 42.4%–45.3%) compared to 5.6% (95% CI 5.3%–5.8%), 14.8% (95% CI 14.4%–15.3%) and 25.7% (95% CI 24.8%–26.6%) (p < 0.001) for the Caucasian cohort. The hazard ratio for African Americans relative to Caucasians in the unadjusted model was 1.8 (95% CI 1.7–1.9; p < 0.001). When graft failure was modeled with terms for both race and DRI, the hazard ratio for African American race reduced to 1.7 (95% CI 1.5–1.7; p < 0.001). The median follow-up for the overall, African American and Caucasian cohorts was 37.2 months, 35.9 months and 41.5 months, respectively. The numbers of graft failures (not including death with a functioning graft) in the overall, African American and Caucasian cohorts were 10 132, 5245 and 4887, respectively.
The multivariable Cox regression model for graft survival based on the purposeful selection algorithm included recipient race and age, days on the waiting list, peak PRA, recipient weight, cause of renal failure, employment status, UNOS region of transplant, HLA DR antigen, payment source, CMV status, educational achievement and recipient hepatitis C status. The hazard ratio for African Americans relative to Caucasians in this model was reduced to 1.4 (95% CI 1.3–1.6; p < 0.001). After entering DRI as a covariate in the above multivariable model, the hazard ratio for African Americans was further reduced to 1.3 (95% CI 1.2–1.4; p < 0.001). The percent of excess hazard explained by adding DRI into the multivariable model is 25%. The serial attenuation of the hazard ratio associated with African American race after adjusting for covariates followed by DRI is summarized in Table S1.
As a further test of the importance of DRI disparities on the differential graft survival seen with African American and Caucasian recipients, matching for donor risk index yielded a cohort of 22 466 African Americans and 22 466 Caucasians with a mean DRI of 1.26 for each group (p = 1.0). The hazard ratio for African Americans relative to Caucasians in this matched cohort was reduced to 1.6 (95% CI 1.5–1.7; p < 0.001), giving a percent of excess hazard explained of 25%.
Finally, examination of the interaction between DRI and race in terms of graft survival was undertaken using stratified analysis as outlined above. In the lowest DRI stratum (DRI < 1.0), the HR associated with African American Race was 2.1 (95% CI 1.9–2.2; p < 0.001). In the middle stratum (1.0 ≤ DRI < 1.5), the HR associated with African American race decreased to 1.6 (95% CI 1.5–1.7; p < 0.001). In the highest DRI stratum (DRI ≥ 1.5), the HR associated with African American race was further reduced to 1.4 (95% CI 1.3–1.5; p < 0.001).
Sensitivity analysis for missing DRI data
There were 2534 (10.09%) African Americans with missing DRI data compared to 3916 (10.49%) Caucasians (p = 0.106), indicating that patients with missing DRI were equally distributed by race. Using multiple imputation as described above, the mean DRI for African Americans with missing data was predicted to be 1.26 (0.465) compared to 1.27 (0.445) for those with complete data (p = 0.486). The mean imputed DRI for Caucasians with missing data were 1.15 (0.482) compared to 1.17 (0.542) for those with complete data (0.058). Thus, the patients with missing data did not differ significantly in terms of DRI from those with complete data. In terms of graft failure, Caucasians with missing DRI had statistically similar risk of graft failure compared to those with complete DRI (HR for missing DRI = 1.015, 95% CI 0.936–1.101; p = 0.716). African Americans with missing DRI data had significantly worse graft failure than their counterparts with complete DRI data (HR for missing DRI = 1.114, 95% CI 1.033–1.200; p = 0.005). Thus, by not including the patients with missing data, our analysis of disparity in graft survival between African Americans and Caucasians may actually be conservative.
This study demonstrates that African American recipients receive significantly “riskier” organs as defined by the DRI than their Caucasian counterparts, and that this disparity cannot completely be explained by socioeconomic, immunologic, or other medical factors. The propensity-matched analysis indicates that medical/immunologic factors account for a greater degree of the disparity in DRI than do social factors. Furthermore, this disparity on organ donor quality has a significant impact on graft survival. The negative effects of the donor quality disparity are illustrated by our analysis of graft survival, in which the percent of excess hazard explained by matching on DRI was 25%. The analysis of graft survival between African Americans and Caucasians indicates that the contribution of DRI to the observed higher risk of graft failure in African Americans is most prominent in the higher DRI strata. Gaining a clear understanding of the reasons underlying the lack of access to quality organs faced by African Americans thus becomes crucial if outcomes are to be improved.
The propensity-matched model of DRI demonstrates that multiple immunologic, medical and socioeconomic factors account for most, but not all, of the disparity in DRI. Two of the most important contributors to the disparity in organ quality appear to be increased use of organs from African American and hepatitis C positive donors in African American recipients. Although matching for HLA type and region did reduce the disparity in usage of African American donor kidneys, the odds of an African American versus a Caucasian receiving a kidney from an African American donor remained 1.2 after matching. That such a difference remains after extensive matching suggests the influence of unmeasured factors which merit further study, such as the role of minor blood group antigens. Other potential explanations include that African Americans recipients and donors may live in closer proximity to each other, or perhaps that transplant surgeons preferentially allocate organs to recipients of the same race.
The fact that African Americans remain more likely to receive a kidney from a hepatitis C positive donor than their Caucasian counterparts after matching for recipient hepatitis C status presents a similar quandary. Given that the patients were also matched on educational achievement, it may be assumed that there existed a similar level of sophistication on the part of recipients in both groups. Despite this, African Americans apparently were more likely to receive a kidney from a hepatitis C positive donor than their Caucasian counterparts. This continued differential suggests an opportunity for a more complete informed consent process.
The underlying causes for the disparity in kidney donor quality faced by African Americans are likely multifactorial. Adjustment for socioeconomic, immunologic and clinical factors closes the gap somewhat, giving evidence to their significant role in the disparity. Even when taking these factors into consideration, African Americans still receive significantly more “risky” organs than their Caucasian counterparts. Given the known barriers to transplantation faced by African Americans (4,28), transplantation with a lower quality organ may be preferable to remaining on dialysis; however, such a decision requires thoughtful and thorough informed consent and patient education. Furthermore, the disparities in deceased donor organ quality faced by African American recipients underscores the importance of efforts to increase living donation among the African American community. Successful examples of such educational efforts can be seen in the work of Callender and Foster (8,29).
If the disparity in posttransplant outcomes between African Americans and Caucasians is to be remedied, then we clearly need to improve our efforts to understand and address the donor quality gap. It is upon the transplant community to ensure that the distribution of higher risk organs to African Americans as illustrated above is not the result of a failure of transparency on our part. Patients place a tremendous amount of trust in their physicians because of our “expert” status. We must ensure that such trust is not misplaced by taking extraordinary efforts to ensure that patients are properly educated to make the best decisions for themselves. Organ donor quality and its effect on posttransplant outcomes should be a mandatory part the discussion held with the patient when determining whether to accept a particular organ offer. Failure to clearly outline the risks posed by donor factors would be to perpetuate the disparity in the quality of organs accepted by and for African American patients. As a prescriptive measure, we echo Rao's call to include the DRI, or a similar measure of organ quality, in both the allocation and consent processes (19). In the end, the decision of whether or not to accept a particular organ must lie with the patient, and be based on full disclosure of all relevant information and informed consent. Only when we have fully disclosed and discussed issues of organ quality with the patient can we say that informed consent has truly taken place. That such an informed consent process can be effective is evidenced by the fact that African Americans and Caucasians accepted organs from extended criteria donors at similar rates in this study. Were potential recipients made as aware of the risks of high DRI kidneys as they are of the risks of ECD kidneys, then perhaps the gap in DRI between African Americans and Caucasians would be mitigated.
In this study we have outlined several factors that contribute to this gap in what is, to our knowledge, the largest study to systematically address donor quality issues in African American recipients. The disparities underscored by our study, such as use of hepatitis positive and African American donors, certainly have more subtle effects on graft survival than more traditional measures of donor quality such as the expanded donor criteria, which were not significantly different in the matched cohort. When taken altogether, however, these small differences combine to produce a significant effect. It is through acknowledging and paying appropriate attention to all the small details of patient care that optimal outcomes are achieved. Furthermore, it may be preferable in the future to preferentially allocate higher risk organs to patients with shorter life expectancy in order to mitigate the effects of shorter expected graft longevity.
Despite these data, there remain other unmeasured factors yet to be determined that merit future study. It will require further vigilance on the part of the transplant community going forward to ensure that one of these factors is not the failure to disclose important issues of donor quality. Limitations of this study are largely a product of its observational nature. We have included a number of potential confounders that can potentially explain observed differences in DRI in our model; however, there may be other factors which were not included and may contribute to residual confounding. The most important of these is the inability to control for specific immunosuppression protocols, which is the factor which likely has the greatest impact on graft survival, and may have differing effectiveness across racial groups. Other factors may include minor blood group antigens and willingness on the part of the patient (of either race) to accept suboptimal organs as an alternative to remaining on dialysis (30). We have also used multivariable modeling of graft survival in order to control for the effects of confounders. Though this approach has been widely used in the literature, there is the possibility that such models do not completely remove the effects of confounding variables. Another weakness of our Cox regression model is that the hazard associated with DRI and African American race appear to change with time, thus making our estimates less precise than a model where race and DRI were treated as time dependent. Though such analysis would add precision to a model intended to predict survival, that was not the goal of the current manuscript. The Cox regression analysis was of death censored graft failure, which tends to overestimate risk in the setting of competing risks. Were death instead treated as a competing risk, the observed difference in graft survival between African American and Caucasians may differ from the findings in the current manuscript.
Another potential weakness is our use of the DRI as a surrogate for organ quality, as it differs from traditional markers such as the expanded donor criteria. One particular potential weakness of the DRI is that it does not completely isolate donor factors, in that HLA mismatch is a product of both the donor and the recipient. Nonetheless, we feel that DRI is the most appropriate proxy for donor quality for the reasons outlined previously in this manuscript. Finally, there were a number of patients in the original dataset for whom complete information was not available to calculate the DRI, and inclusion of these patients may potentially lead to somewhat different outcomes. Although our sensitivity analysis indicates that the patients with missing values were similar to those with complete data, the possibility of bias introduced by exclusion of patients with missing data still remains.
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, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government.
The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation. This manuscript was neither prepared nor funded by any commercial organization.