Prediction of Three-Year Mortality After Deceased Donor Kidney Transplantation in Adults with Pre-Transplant Donor and Recipient Variables

Background

Kidney transplantation is the best replacement therapy for patients with kidney failure [1]. In comparison to dialysis, kidney transplantation offers advantages like higher quality of life and longer survival [1]. Meanwhile the disparity between the number of patients who need a kidney graft and appropriate donors is still growing [2], as is the demand for expanding the number of donors. This situation can force identification of potentially hazardous donors and recipients with inherent covariables that pose high risks for unfavorable outcomes after transplantation. The United Network of Organ Sharing has implemented a definition of Expanded Criteria Donors that are defined by being older than 60 years of age or by age 50–59 years, plus at least 2 criteria out of the following 3: cerebrovascular accident as cause of death, serum creatinine greater than 1.5 mg/dL, or history of hypertension [3]. Eurotransplant organization reacted to the increasing waiting list by developing the Eurotransplant Senior Program (ESP) which includes donors aged ≥65 years [4].

The Kidney Transplant Morbidity Index was used by Pieloch et al. to determine the 3-year graft and patient survival rate by recipient’s pre-transplant comorbidities [5]. Laging et al. recently proposed the Rotterdam Comorbidity in Kidney Transplantation Score to predict post-transplant mortality risk [6]. Interestingly, in the population investigated by Laging et al., 50% of those patients with the highest comorbidity scores survived more than 10 years [6]. Patients with comorbidities are likely those patients with a greater long-term survival benefit afforded by transplantation when compared to dialysis. This notion has been further underlined by the recent findings published by Sørensen et al. which demonstrated a survival benefit for kidney transplantation despite high comorbidity [7].

The current study aims to identify risk factors for patient mortality and prognostic factors for 3-year post-transplantation mortality based on pre-transplant donor and recipient variables excluding comorbidities that cannot be altered at the time of organ allocation to ensure optimal transplant benefit by improved donor organ allocation.

Material and Methods

SETTING AND DATA COLLECTION:

A university hospital in Germany within the Eurotransplant community provides the setting. This single center retrospective analysis has been based on a comprehensive clinical data base which has been complemented by additional retrospective data from clinical charts for the purpose of this study.

ETHICS STATEMENT:

This study has been approved by the Ethics Committee at Hannover Medical School (reference number 2375-2014). Patients gave general informed consent for the analysis of their data in medical research. All data were anonymized prior to research.

INCLUSION AND EXCLUSION CRITERIA:

The inclusion and exclusion criteria are summarized in Figure 1. It includes all consecutive deceased-donor kidney transplants performed at Hannover Medical School between the January 1, 2000 and December 31, 2012. Pediatric (age ≤17 years) with combined transplants as well as simultaneously performed double kidney transplants were excluded. Study Cohort 1 was used to identify independent risk factors for survival (multivariable Cox regression modelling). Study Cohort 2 was defined after additional exclusion of survivors who had a period of less than 3 years for follow-up with the goal to assess those independent risk factors for survival identified in Study Cohort 1 as prognostic factors for observed 3-year morality using multivariable logistic regression modelling (Figure 1).

DEFINITIONS OF VARIABLES:

The investigated variables on the urgency of kidney transplantation are defined by the waiting list status immediately prior to transplantation according to the organ allocation rules established by the German Medical Council (Bundesärztekammer) [8]. These allocation rules are executed by Eurotransplant for Germany [8]. Patients listed as high urgency were defined with an imminent lack of access for either hemodialysis or peritoneal dialysis; severe (uremic) polyneuropathy, inability to cope with dialysis with a high risk for suicide; severe bladder problems (hematuria, cystitis, etc.) due to kidney graft failure. Patients declared as highly-immunized were those who suffered from an end-stage renal disease and who were transplantable with a panel reactive allo-antibodies (PRA) range of ≥85%. Immunized listed recipients who had an end-stage renal disease, were transplantable and had a measured PRA range of ≥6% to <85%. Marked as transplantable were those recipients with an end-stage renal disease who were transplantable and had a PRA range of <6% [9].

STUDY END-POINTS:

Long-term patient survival (Study Cohort 1) and 3-year observed survival status versus death (Study Cohort 2) regardless of graft function were defined as primary study end-points. For those patients who were lost to clinical follow-up, the German legal registration offices provided us with information on their current survivor status. As we have reported before, all changes of address and all deaths have to be reported to the legal registration offices in Germany. This information is accessible for our institution within the current legal framework in Germany [10].

STATISTICAL METHODS:

The complete data set from Hannover was used for a brain storming session that included experienced kidney transplant specialists in order to define pre-transplant prognostic factors that are commonly known prior to transplantation. These factors were used as candidate variables for the prognostic model design. The next step was statistical evaluation of all the potential prognostic factors using univariable Cox regression analysis with the goal to determine the relevance of variables for long-term survival.

Variables with more than 5% missing values were submitted to an assessment of a potentially significantly different distribution of missing data between patients with 3-year mortality and those without 3-year mortality using the chi-squared test. For the assessed variables with more than 5% missing values a significant difference (P<0.05) in distribution of missing data could not be detected. Patients with missing values for variables that were critical for prognostic modelling were eliminated.

The influence of significant categorical variables on survival over time was further assessed in exploratory analyses using Kaplan-Meier curves and log rank tests (data not shown).

Principal component analyses were applied for better understanding of the underlying data structure and avoiding multi-collinearity in regression. Principal component analyses and multivariable Cox regression analyses were performed for donor and recipient variables [11].

The Shapiro-Wilk W test was used to assess normal distribution of variables. Not normally distributed variables were included into multivariable regression using only their quartiles which were used as nominal variables (Supplementary Table 1).

In Study Cohort 1, all uncorrelated variables with P values ≤0.250 in univariable Cox regression analysis were included in multivariable regression modelling as previously described [10]. An initial stepwise backwards likelihood elimination process of the least significant variables was performed. A threshold of >20% change between each of the steps in one or multiple betas of the investigated variables was chosen for the anticipation of potentially significant factor interactions [10].

The finally reached multivariable Cox regression model in Study Cohort 1 with pre-operative donor and recipient variables was used for the construction of a 3-year mortality prognostic model. Identified variables with significant independent influence on long-term survival were used for multivariable logistic regression analysis for the purpose of building a prognostic model for 3-year mortality using stepwise backwards likelihood elimination.

Lack of fit of the derived prognostic model was assessed with the Hosmer-Lemeshow test. Additional evaluation of the derived prognostic model included determination of the area under the receiver operating characteristic curve (AUROC) to assess the sensitivity and specificity of the model’s predictions of 3-year mortality after transplantation (bootstrap 95% CI: 1000 iterations; random number seed: 978). AUROCs >0.700 are widely regarded as a prerequisite for clinically useful prognostic models [12,13]. The best Youden index (Youden index=sensitivity+specificity−1) [14] was used to determine the cutoff value with the best sensitivity and specificity for the prediction of 3-year mortality with the logit of the developed prognostic model. The relevance of this cutoff value for long-term survival was investigated with Kaplan-Meier analysis using the log rank test.

JMP Pro 11.0 Software (SAS Institute, Cary, NC, USA) was used to perform statistical analyses with P values <0.050 defined as significant.

Results

CLINICAL AND DEMOGRAPHIC CHARACTERISTICS AND DESCRIPTIVE STATISTICS:

The hospital mortality rate in Study Cohort 1 was 1.5% and in Study Cohort 2 it was 1.6%. A total of 359 patients (23.2%) in Study Cohort 1 died during follow-up and 332 patients (27.8%) in Study Cohort 2 died. Further details of the observed pre-transplant donor and recipient variables in Study Cohort 1 and Study Cohort 2 are summarized in Supplementary Tables 1–5.

RISK FACTOR ANALYSIS WITH UNIVARIABLE COX REGRESSION ANALYSIS:

Tables 1 and 2 summarize the influence of observed pre-transplant variables on long term survival as evaluated by univariable Cox regression.

INDEPENDENT RISK FACTORS FOR LONG-TERM SURVIVAL:

While recipient weight, recipient body mass index, the number of HLA A mismatches, as well as the indications of congenital anomalies of the kidney and urinary tract and IgA nephropathy had a significant impact on earlier death in the univariable Cox regression; however, the significance of these factors for earlier death could not be confirmed in multivariable Cox regression modelling (Table 1). Several donor variables, including donor age, last potassium, last urea, hypertension reported, smoking, as well as respirational donor cause of death had a significant influence on earlier recipient death in univariable Cox regression. The significance of these influences could not be confirmed in multivariable analyses (Table 2).

The urgency of the waiting list status highly immunized-KI did not display a significant impact on earlier death in the univariable Cox regression (Table 1) but gained an independently significant influence on earlier death in multivariable regression modelling (HR: 2.579; 95% CI: 1.272–4.631; P=0.011) (Table 3).

The final result of multivariable Cox regression modelling demonstrated that the following variables had a statistically significant and independent impact on the risk of earlier death after kidney transplantation: Urgency of waiting list status highly immunized, urgency of waiting list status high-urgency, recipient diabetic nephropathy, recipient age in years ≤42.1 years (quartile 1), recipient age in years 42.2–2.8 years (quartile 2), recipient age in years 52.9–62.8 years (quartile 3), cold ischemic time in hours ≤11.8 hours (quartile 1), cold ischemic time in hours 11.9–15.3 hours (quartile 2) and 0, 1, or 2 HLA DR mismatches (Table 3).

Factor interactions could not be detected in multivariable Cox regression modelling for donor and recipient variables during stepwise backwards likelihood elimination. The final multivariable model is summarized in Table 3.

PROGNOSTIC FACTORS FOR 3-YEAR POST-TRANSPLANT MORTALITY:

Prognostic factors for 3-year mortality demonstrated an AUROC larger than 0.700 (AUROC=0.748, bootstrap 95% CI=0.689–0.788) (Figure 2). This model demonstrated no significant lack of model fit (P=0.132) and was defined as follows:

Lin[1]=–1.957

The sensitivity of prediction of 3-year mortality with this model was 50.8% and the specificity 86.1% with an overall correctness of prediction 68.5%. Sample size calculation for external validation of the proposed prognostic model for 3-year mortality with a power >80% was determined to require a total of 8464 cases with 847 cases with 3-year mortality estimated to be at 10.0%.

Significantly worse long-term survival for those patients with a predicted risk of 3-year mortality greater than 15.7% as had been determined with the proposed prognostic model for 3-year mortality was detected when compared to those patients with a lesser predicted risk of 3-year mortality in Study Cohort 2 (P<0.001, log rank test). Due to the inability to calculate the predicted risk, patients without data for variables that were contained in the proposed prognostic model were excluded (n=63) (Figure 3). Statistically significant effects of the number of HLA-DR mismatches regarding patient survival (P<0.001, log rank test) were calculated using Kaplan-Meier analysis as shown in Figure 4.

Discussion

This study identified independent pre-transplant donor and recipient risk factors for patient mortality. The developed prognostic model for 3-year mortality based on these results is potentially clinically useful for recipient counselling and donor organ acceptance decisions (AUROC >0.700, Figure 2) and has a cutoff with a highly significant influence on earlier death after transplantation (Figure 3). The derived prognostic model demonstrated that older recipients with longer cold ischemic time, who suffer from diabetic nephropathy and who are either highly immunized or urgent recipients can be transplanted with a lower risk of early mortality, if they are transplanted with a donor kidney without any HLA-DR mismatches. The highly significant impact of the number of HLA-DR mismatches on patient survival is shown in Figure 4. Transplantation of patients with more favorable recipient risk profiles could be justifiably transplanted with donor kidneys that result in 1 or 2 HLA-DR mismatches. Such a decision should be based on the individual weighing of recipient risk factors (Table 3). The proposed prognostic model and the underlying Cox regression model both provide tools for such a weighing of individual recipient risk profiles.

It is striking that this study externally confirms the prognostic relevance of an older recipient age and pre-existing recipient diabetes for an increased risk of mortality after kidney transplantation in a European cohort as has been published before in a report from a large registry trial from the United States which has proposed a predictive score for post-transplant mortality [15].

It has been reported before that both, the age of the donor and the age of the recipient have an influence on patient survival after transplantation, although these results have not been explicit. Dempster et al. for example found no significantly increased recipient mortality at one year after transplantation even when donors were 65 years or older [16]. In this study donor age demonstrated a significant impact on early recipient mortality in univariable analysis (Table 2) which could not be confirmed in multivariable Cox regression analysis. This is likely due to the fact that older recipients tended to receive older donor kidneys. Dempster et al. found in a similar setting where older patients tended to receive older donor kidneys more complications after transplantation defined as delayed graft function, kidney failure in the first year after transplantation and higher serum creatinine at one year after transplantation [16]. In this context, McCaughan et al. have shown that patients with graft failure who needed to return to dialysis had worse survival rates when compared to patients who underwent dialysis and never had transplantation or when compared to patients who had functional kidney grafts [17]. Additionally, Frei et al. detected delayed graft function as a relevant risk factor for graft and patient survival [18]. This study has shown that cold ischemic time is an independent and significant risk factor for patient survival. Long CITs are known to increase the risk of delayed graft function and early graft failure [18].

Recipient age has been identified to have a significant influence on post-transplant survival. These findings agree with clinical experience and could be explained with increased comorbidity of older patients [5] as well as with decreasing life expectancy that naturally decreases with increasing age. Because of an increasing frequency and percentage of older recipients and donors, Eurotransplant established the ESP which allocates kidneys from deceased donors older than 65 years to recipients in the same age range by keeping the CIT as short as possible by ignoring HLA matching [8]. The findings of this study describe an increased mortality risk for patients who were transplanted with higher numbers of HLA-DR mismatches (Table 3, Figure 4). Therefore, the practice of ignoring HLA matching in the ESP should be regarded with great caution.

Jacobi et al. have proposed to define patients transplanted in the ESP as a high-risk population who need careful evaluation and selection for transplantation and close clinical surveillance after transplantation [19]. Frei et al. could not find a negative influence on graft and patient survival for patients transplanted in the ESP in comparison to standard allocation [18]. In contrast to the aforementioned results, the current study identified higher recipient age as an independent risk factor for 3-year mortality. The first, second and third quartile of recipient age were independent and significant protective factors in the proposed prognostic model for 3-year mortality demonstrating that lower age quartiles were more protective when compared to the fourth quartile of recipient age (≥62.9 years) (Table 3). This study clearly showed that recipient age had a non-linear influence on early mortality risk after kidney transplantation with increasing risk of earlier death per unit of older age (Supplementary Figure 1).

The Dempster et al. study showed a higher mortality rate in the first year after transplantation for older patients [16]. While older recipients have been shown to be at higher risk for complications after transplantation [16,19,20], the only alternative to transplantation would be dialysis, which has been shown to have even worse results concerning survival, quality of life, and economic factors [16,19,21,22].

The study by Orlandi et al. did not find recipient age as an independent risk factor for negative outcomes but found recipient diabetic state was a relevant and independent risk factor for earlier death [21]. This is particularly important, as the increasing diabetes prevalence in the population leads to an increase in the frequency of diagnosed end-stage renal disease cases [22]. Foucher et al. found that age-related mortality after kidney transplantation was not significantly increased, whereas the diabetic state of the recipient was shown to be a risk factor for excess mortality when compared to a general population [23]. These findings are in line with our findings that a recipient’s diabetic nephropathy was a highly significant risk factor for early mortality after kidney transplantation with a hazard ratio of 3.471 (95% CI: 2.476–4.751) independent of the recipient age at transplantation (Table 3).

This study showed that cold ischemic time was a relevant risk factor for early death. Especially for older patients who receive an expanded criteria organ may be negatively influenced by long cold ischemic time [24]. This is why the ESP aims to keep cold ischemic time as short as possible [19]. Van der Vliet and Warlé found cold ischemic time to be an independent risk factor for delayed graft function and acute rejection, but not for long-term outcomes [24]. Frei et al. showed that every hour of cold ischemic time increased the risk of graft loss by 3% [18]. However, Jacobi et al. and Giessing et al. could not find any negative impact of longer cold ischemic time, not even using expanded criteria for donor kidneys by comparing the outcome of successfully and subsequently transplanted kidneys from one donor with just low differences in cold ischemic time [19,25]. The results of our study clearly point to the clinical relevance of cold ischemic time for post-transplant patient survival.

Concerning HLA mismatches, 1 or 2 HLA DR mismatches had a statistically significant impact on survival (Figure 4) which was confirmed in multivariable Cox regression modelling (Table 3). This result is in line with previously published findings [26]. Laging et al. found that all HLA mismatches were relevant factors for graft survival [28]. Furthermore, Frei et al. revealed higher rates of acute and late rejection for ESP patients with shorter cold ischemic time and explained these findings by more HLA mismatches leading to antibody-mediated rejection as a consequence [18]. We propose, based on our findings as well as previously published reports to consider HLA DR mismatches for donor kidney allocation while keeping the cold ischemic time as short as possible. This concept has been realized in the Eurotransplant Senior DR-compatible Program (ESDP), which includes full HLA DR compatibility and reduced cold ischemic time [8] in comparison to the ESP [4].

The proposed prognostic model for 3-year mortality requires external validation with data from other centers before allocation rules can be adapted. Sample size calculation based on the results of this study revealed that the data of a total of 8464 transplanted patients with an estimated 3-year mortality rate of 10.0% would be needed for external validation of the proposed prognostic model with a power >80%.

This study investigated the independent influences of pre-transplant recipient and donor risk factors on post-transplantation survival beyond recipient comorbidity. The recently defined Kidney Transplant Morbidity Index with its demonstrated significant influence on 3-year patient survival [5] and the Rotterdam Comorbidity in Kidney Transplantation Score used to predict post-transplant mortality risk [6] were intentionally not used as analyzed risk factors in this study. Laging et al. showed that patient death was significantly influenced by cardiovascular disease, other organ transplantation, and total comorbidity scores [6]. However, in the population investigated by Laging et al., 50% of the patients with the highest comorbidity scores survived more than 10 years. Laging et al. suggested that a high comorbidity score should not be seen as a contraindication for kidney transplantation [6]. In addition, patients on the waiting list for kidney transplantation with comorbidities that increased post-transplant mortality risk were those patients with greater long-term survival benefit afforded by transplantation when compared to continued dialysis [6]. This notion has been further underlined recently by Sørensen et al., who demonstrated a survival benefit in kidney transplantation despite high comorbidity [7]. Thus, patients with high comorbidity should not be excluded from kidney transplantation. The current study showed how donor kidneys could be matched to recipients to reduce the 3-year mortality risk while the recipients’ comorbidity burden could not be possibly reduced at the time when donor organ offers are made and a decision on the acceptance of such an offer for an individual patient is made responsibly.

Predicting an unfavorable outcome using the proposed prognostic model, allows the offered donor organ to be used for more favorable donor-recipient combinations, while keeping urgency aspects in mind. This weighing of options has profound ethical implications in the dimension of distributive justice. The current study clearly showed that HLA-DR mismatches should be taken into account, even though they are not available before listing. Unfavorable combinations of pre-transplant donor and recipient variables and increased recipient risk profiles should at least trigger heightened clinical vigilance after transplantation.

The presented study had several limitations including a possible center-bias which may have influenced the findings in this single-center study. A further limitation of the current study was that cold ischemic time can only be estimated prospectively by transplantation surgeons prior to actual transplantation for each patient. However, in our clinical experience, the estimation of the quartiles of cold ischemic time, which were identified as significant factors in the proposed prognostic model, would usually be possible with sufficient accuracy.

Conclusions

The main conclusion of this study was that especially older, highly immunized, or high urgency transplantation candidates with anticipated longer cold ischemic times, who are transplanted with the indication of diabetic nephropathy, should not receive donor organs with 1 or 2 HLA DR mismatches. The proposed prognostic model was able to weigh the risk of 3-year post-transplant mortality that was associated with different individual expressions of these identified risk factors. In case of predicting an unfavorable outcome with the proposed prognostic model, the offered donor organ could be used for more favorable donor-recipient combinations, while keeping urgency aspects in mind. This weighing has profound ethical implications in the dimension of distributive justice.