Employment status at transplant influences ethnic disparities in outcomes after deceased donor kidney transplantation

Deceased donor (DD) kidney transplantation (KT) from African American (AA) donors is associated with shorter allograft survival compared to DDKT from donors of other races/ethnicities. Donor African ancestry is included as a risk factor in the calculation of the Kidney Donor Risk Index (KDRI), a measure of DD organ quality used to generate the Kidney Donor Profile Index in the US kidney allocation system [1, 2]. Similarly, AA recipients of DDKT have poorer outcomes, regardless of the race/ethnicity of the donor [3, 4].

Causes of ethnic differences in DDKT outcomes remain unclear; they are likely multifactorial, with inherited, environmental, and socioeconomic factors contributing to donor- and recipient-level effects. Several reports highlighted the adverse impact of genetics, poverty, geography, and lack of education on access to kidney transplantation and outcomes after engraftment [3, 5‐10]. We demonstrated more rapid allograft failure after kidney transplantation from DDs with apolipoprotein L1 gene (APOL1) high-risk genotypes. We suggested that using APOL1 genotyping instead of DD race might refine the KDRI by increasing the number of good quality kidneys for waitlisted recipients [11‐15]. We and others reported genetic variants that affect AA DDKT outcomes either independently or through their interaction with APOL1 kidney-risk variants [16‐19]. Beyond APOL1, several biological factors independently contribute to, or interact with non-biological factors leading to poorer outcomes among AA DDKT recipients. For example, given fewer AA donors and greater allelic variation at the HLA locus, potential AA recipients are disadvantaged in an allocation system that includes HLA matching. Despite recognizing these limitations and related changes, AA wait longer for kidney transplantation, an important modifiable risk factor for adverse outcomes [20‐22]. The situation is compounded by complex interactions between donor and recipient characteristics impacting long-term outcomes.

Herein, we attempt to measure the effects of recipient- and donor-specific factors and their interaction on observed racial/ethnic disparities by studying partner kidneys from DDs that are, by definition, genetically identical and were transplanted into recipients of different races. Analyses were restricted to AA and European American (EA) donors and recipients for ease of comparison. This approach provides better control for donor-level confounding factors, including donor-level genetic risk and race/ethnicity, on recipient outcomes after transplantation [1, 23].

These analyses used donor and recipient data in the Scientific Registry of Transplant Recipients (SRTR) for kidneys procured and transplanted between October 1, 1987, and June 30, 2016. Analyses were restricted to AA or EA DDs who had both partnered kidneys transplanted, one to an AA recipient and the other to an EA recipient, yielding four groups of donor/recipient pairs (DRP): AA/AA, AA/EA, EA/AA, and EA/EA. This matched design better controlled for confounding by donor-related genetic, organ-specific, or socioeconomic factors and facilitated comparison of recipient-level factors contributing to observed racial disparities in outcomes. Donors or recipients < 18 years of age were excluded.

The primary outcome was death-censored time to kidney allograft failure, determined by the interval between transplantation dates and allograft loss. In patients with a functioning allograft, the final observation date was censored for death with function or at last follow-up before March 5th, 2016. A secondary outcome treating death as a competing risk (CR) was also considered. In this case, the final observation date was censored at death for individuals who died with a functioning allograft or at the most recent follow-up before March 5th, 2016, for living individuals with functioning allografts.

A split-half hypothesis-free analysis approach was applied where a random survival forest (RSF) model was fit in a randomly selected subset of the data representing 50% of the data to rank variables and their interaction with DRP based on their variable importance (VIMP) measure [24, 25]. RSF models implementing the conditional VIMP measures are robust to multicollinearity between predictors and are well-suited to detect interaction effects, which are of particular importance here [26, 27]. Analyses were repeated on the second half of the data and then on the complete data after observing strong reliability between the results obtained in the two subsets. Therefore, effect sizes and interaction effects with the DRP were estimated in the combined dataset using the top-ranked variables based on VIMP. This approach minimized the loss of statistical power caused by splitting the data into subsets [28]. Cox Proportional Hazard (CPH) models were fitted for death-censored allograft survival (DCAS) and the Fine and Gray model when death was considered a CR to allograft survival to obtain effect size estimates. The sandwich estimator was used to obtain a robust estimation of the covariance matrix associated with the parameter estimates to account for the correlation between allograft failure rate and time to failure of kidneys donated by a single individual to two recipients. Lin and Wei reported that this sandwich estimator was consistent and robust to several misspecifications of the Cox model [29]. Proportional hazard assumptions were checked by visual inspection of the log-log curve and assessing the Schoenfeld and martingale residuals [30]. Models were fitted separately following missing data imputation, which was performed within the RSF framework because RSF based-imputations have demonstrated high degree of robustness even in the presence of non-random missingness patterns [31, 32]. Ten imputed datasets were created, and the result obtained with these datasets were combined using established approaches [33‐35]. Analyses were performed in SAS 9.4 and R 4.1. The RandomForestSCR package was used to fit Random Forest models for DCAS and the competing risk model [36].

The cohort consisted of 47,182 kidney transplants from 3872 AA and 19,719 EA DDs. Tables 1 and 2 display distributions of demographic variables and clinical characteristics for donors and recipients, respectively. Data are presented as median (Q₁, Q₃) for continuous and N (%) for categorical variables. All comparisons in these Tables were statistically significant (p < 0.0001).

AA and EA DDs had comparable body mass index (BMI) and KDRI. Relative to EA DDs, AA DDs were more likely to be male (64.6% vs. 59.1%), younger (median age 35.0 vs. 40.9 years), cytomegalovirus (CMV) IgG antibody-positive (75.3% vs. 56.2%), and diabetic (5.2% vs. 4.3%). However, AA DDs were less likely to be smokers (60.3% vs. 68.9%) or expanded-criteria donors (12% vs. 14.4%) (Table 1).

Independent of the race/ethnicity of the DD, AA recipients received their transplant at a younger age (median 48.0 vs. 51.0 years), were more likely to have been on dialysis (61.3% vs. 50.9%), and had longer dialysis vintage (4.2 vs. 3.1 years). In addition, AA recipients were less likely to have received a prior transplant (11.2% vs. 15.2%) ordie with a functioning allograft (17.1% vs. 23.8%), but more likely to experience DGF (30.5% vs.21.7%) and had higher rates of acute rejection (1.8% vs. 1.2%) (Table 2). However, rates of immunosuppression medication use and the proportion of KT recipients needing induction therapy were comparable. Supplementary Table 1 show the demographics and clinical characteristics distribution by donor and recipient race.

Fig. 1 displays unadjusted death-censored allograft survival for KT recipients by DRP. Figure 1A shows the unadjusted allograft survival; differences in allograft survival outcomes are apparent between recipients based on race; the top two curves represent DCAS in EA recipients, and the bottom two curves display DCAS in AA recipients. Hazard ratios (HRs) (95% CI) for EA/EA, AA/EA, and EA/AA DRPs, relative to AA/AA pairs, were 0.56 (0.53, 0.60), 0.65 (0.59, 0.70), and 0.96 (0.91, 1.02), respectively. Figure 1B shows unadjusted recipient survival, with mortality treated as a competing risk to allograft failure. At first glance, this graph suggests slightly higher recipient survival rates among AA/AA and EA/AA, compared to AA/EA and EA/EA DRP. However, it is important to keep in mind that AA recipients are approximately 3 years younger than EA recipients. Causes of graft failure did not vary between AA and EA recipients, except for the rate of non-compliance to immunosuppression medication, which was 11.9% among AA recipients, compared to 9.2% for EA recipients.

The five-year DCAS rate improved among all four DRPs during the observation period (Supplementary Table 2). Five-year allograft survival rates in transplants performed after 2010 vs. before 2001 were (0.74 (0.52, 0.90) vs. 0.64 (0.60, 0.67) for AA/AA DRPs, 0.85 (0.76, 0.94) vs. 0.74 (0.71, 0.77) for AA/EA, 0.83 (0.81, 0.86) vs. 0.64 (0.63, 0.65) for EA/AA, and 0.89 (0.87, 0.92) vs. 0.78 (0.77, 0.79) for EA/EA transplantations. Results of the random forest models, which inform the interaction tests that were subsequently performed can be found in Supplementary Table 3.

CPH models showed statistically significant interaction effects between the DRP with the transplant era (0.02), KDRI (p = 0.0009), and EPTS (p < 0.0001) for DCAS.

The CR analysis helped clarify these results; it showed statistically significant interactions between the DRP and KDRI (p < 0.001) for allograft survival, and between the DRP with the KDRI (p < 0.0001), EPTS (p = 0.009), employment status (p < 0.0001) and transplant era (p < 0.0001) with kidney recipient mortality. Table 3 shows HRs for overall DCAS according to employment status and assuming no change in KDRI and EPTS. With employment EA/EA DRPs saw consistent improvement over time; for transplantations performed after 2010, HRs ranged from 0.42 (0.37, 0.47) to 0.46 (0.41, 0.51) for employed recipients and from 0.52 (0.48, 0.58) to 0.57 (0.52, 063) for unemployed recipients. Similar improvements were also observed with AA/EA pairs. However, for EA/AA DRPs, significant improvement in the overall DCAS was observed only post-2010 DDKTs, and the overall improvement was significantly smaller; HRs were 0.78 (0.66, 0.92) for EA/AA DRPs, compared to 0.42 (0.38, 0.47) for EA/EA’s.

Table 4 shows HRs for the effect of DRP, KDRI, EPTS, and transplant era and employment status on recipient mortality with allograft failure as a CR. For transplantations performed before 2001 and assuming no change in KDRI and EPTS over time, reductions in mortality were observed among all four DRPs for employed DDKT. HRs for the post 2010 transplant era were 0.24 (0.13, 0.43), 0.27 (0.17, 0.45), 0.20 (0.14, 0.28), 0.24 (0.19, 0.32) for AA/AA, AA/EA, EA/AA and AA/AA DRPs, respectively. In contrast, HRs for mortality were higher among unemployed recipients; 0.50 (0.29, 0.87), 0.55 (0.35, 0.87), 0.32 (0.24, 0.42), and 0.49 (0.43, 0.57) among these 4 DRPs, assuming no change in KDRI and EPTS. Figure 2 shows the disparity in recipient mortality according to employment status and DRP.

Donor characteristics contribute to racial disparities in outcomes following DDKT [2, 23, 37]. The present study evaluated recipient factors potentially affecting ethnic disparities in DDKT outcomes using a unique donor-matched design that controlled for genetic differences in transplanted kidneys, which allowed us to limit the impact of donor characteristics on DDKT outcomes, including many donor factors not available in the OPTN registry.

The analysis included 47,182 total kidney transplantations, 3872 involving AA DDs. As such, it is the most extensive analysis of its kind. Transplants resulting from the four possible DRPs had different DCAS, with EA recipients having better overall allograft survival than AA, independent from DD race/ethnicity. Analyses suggest that multiple factors contribute to kidney allograft outcomes. Some of the reported associations were described previously, including the well-known effects of DGF, serum creatinine at hospital discharge, recipient age, KDRI, EPTS, immunosuppressant medication, transplant era, donor age, etc. [7, 38, 39] However, these effects are not modified by the DRP.

Employment status, KDRI, and EPTS interacted with DRP to affect DDKT outcomes. Unemployed recipients had worse DDKT allograft survival and mortality. Employment status was obtained before kidney transplantation. Recipients who reported working a full-time or a part-time job was considered employed; all others were considered unemployed, independently of the reason for not working. The HR estimates among unemployed recipients were almost twice those observed among employed recipients for mortality, although there was a minor overlap between confidence intervals in some cases.

Employment status at transplantation was the only socioeconomic variable that showed significant interaction effects with the DRP. The absence of independent effects of educational attainment and insurance status probably reflects the careful screening process of potential recipients by transplant programs. In contrast, employment status is rarely invoked as a reason to preclude active status of wait-listed transplant candidates in the US, despite its potential adverse effect on the ability to afford medications or access health insurance, especially after expiration of the 36-month post-transplant coverage provided by the Center for Medicare and Medicaid Services. The newly passed Immuno Bill indefinitely extends Medicare coverage of immunosuppressive drugs for KT recipients and may help reduce disparities in long-term allograft survival. However, employment status may be a broader measure of social determinants of health with a clear association between unemployment, job loss, and retirement with poor outcomes.

In contrast, employment contributes to better physical health [40‐42]. Unemployed individuals, independent of race/ethnicity, more often report feelings of depression and anxiety and high blood pressure, and tend to have higher rates of stroke, heart attack, and heart disease [43‐45]. Unlike the composite scores considered in these analyses, employment status is a modifiable factor. Specific steps can be taken to understand how it affects outcomes among DDKT recipients and mitigate its effects.

Some measures reported in these analyses (e.g., KDRI and EPTS) are relatively new and were not previously part of the kidney allocation process. However, their utilization in these analyses ensures that comparisons across transplant eras are appropriate. KDRI includes donor race and other donor demographic and clinical characteristics. EPTS depends on recipient age, diabetes status, prior organ transplantations, and previous time on dialysis. Including these scores, the DRP, and the other variables in these models may have induced some collinearity. However, the random forests models are robust to multicollinearity. The KDRI score for AA donors is multiplied by a factor of 1.2, regardless of donor age, sex, and presence of other comorbidities. However, AA deceased donors were more likely to be younger and males such that the distributions of KDRI scores were comparable between AA and EA donors. The inclusion of these variables in the models was meant to help determine how socioeconomic and social determinants of health factors, which may interact with these scores, affect kidney transplant outcomes among AA and EA recipients.

Limitations of this report include potential underreporting in the SRTR database of various outcomes (e.g., DGF), mischaracterization of race and ethnicity, and viral infections, whose effects on KT outcomes were not initially recognized [46]. Analyses used registry data that were not collected for research purposes; therefore, some variables (e.g., employment status, medication use) may be incomplete and might not have been rigorously collected. However, it is unclear when the ongoing prospective APOL1 Long-term Kidney Transplantation Outcomes (APOLLO) study will accumulate enough events to address these questions [47]. These analyses provide some preliminary results that can be explored in other datasets.

Also, the study compared DDKT outcomes over more than 30 years, such that the standard of care and ways that measurements were collected and reported to the SRTR may have changed over time. However, focusing on four transplant eras should reduce these effects and their likelihood for confounding. These analyses were performed in a non-random subset of the SRTR data that may not have provided a representative sample of the distribution of outcomes observed among all DDKT recipients. For multiple reasons, including a greater need for kidney transplants in AA, lower rate of living kidney donation among AA, higher rates of HLA matching among individuals with recent African ancestry, waitlisted AA are more likely to receive AA DDKTs. Therefore, AA/AA DRP represents a significant proportion of all DDKTs [7, 48, 49].

AA recipients of kidney transplants from AA DDs had significantly shorter kidney allograft survival than EA recipients of AA DD kidneys and AA recipients of EA DD kidneys. Mortality among DDKT recipients remains high, especially among unemployed recipients, and does not appear to have changed since the early 2000s among unemployed AA recipients. Unemployment is associated with poorer outcomes among DDKT recipients, independent of race/ethnicity; however, its effects appeared to be consistently worse for AA DDKT recipients. Thus, improving outcomes for transplant recipients will require an improved understanding of the mechanisms by which socioeconomic factors, such as unemployment, adversely affect outcomes in the United States.