Rate of Renal Graft Function Decline After One Year Is a Strong Predictor of All-Cause Mortality

Authors

  • O. Moranne,

    Corresponding author
    • Service de Néphrologie, Département de Santé Publique, Hôpital Pasteur et Larchet, CHU de NICE, France
    Search for more papers by this author
  • N. Maillard,

    1. Service de Néphrologie, Dialyse, Transplantation Rénale, Laboratoires d'Explorations Fonctionnelles Rénales, Hôpital NORD, CHU de Saint-Etienne, Université Jean MONNET, PRES Université de LYON, France
    Search for more papers by this author
  • C. Fafin,

    1. Service de Néphrologie, Département de Santé Publique, Hôpital Pasteur et Larchet, CHU de NICE, France
    Search for more papers by this author
  • L. Thibaudin,

    1. Service de Néphrologie, Dialyse, Transplantation Rénale, Laboratoires d'Explorations Fonctionnelles Rénales, Hôpital NORD, CHU de Saint-Etienne, Université Jean MONNET, PRES Université de LYON, France
    Search for more papers by this author
  • E. Alamartine,

    1. Service de Néphrologie, Dialyse, Transplantation Rénale, Laboratoires d'Explorations Fonctionnelles Rénales, Hôpital NORD, CHU de Saint-Etienne, Université Jean MONNET, PRES Université de LYON, France
    Search for more papers by this author
  • C. Mariat

    1. Service de Néphrologie, Dialyse, Transplantation Rénale, Laboratoires d'Explorations Fonctionnelles Rénales, Hôpital NORD, CHU de Saint-Etienne, Université Jean MONNET, PRES Université de LYON, France
    Search for more papers by this author

Corresponding author: Moranne Olivier

moranne.o@chu-nice.fr

Abstract

The slope of GFR associates with an increased risk for death in patients with native CKD but whether a similar association exists in kidney transplantation is not known. We studied an inception cohort of 488 kidney transplant recipients (mean follow-up of 12 ± 4 years) for whom GFR was longitudinally measured by inulin clearance (mGFR) at 1 year and then every 5 years. Association of mGFR at 1 year posttransplant and GFR slope after the first year with all-cause mortality was studied with a Cox regression model and a Fine and Gray competing risk model. While in Crude analysis, the mGFR value at 1 year posttransplant and the rate of mGFR decline were both associated with a higher risk of all-cause mortality, only the slope of mGFR remained a significant and strong predictor of death in multivariate analysis. Factors independently associated with a more rapid mGFR decline were feminine gender, higher HLA mismatch, retransplantation, longer duration of transplantation, CMV infection during the first year and higher rate of proteinuria. Our data suggest that the rate of renal graft function decline after 1 year is a strong predictor of all-cause mortality in kidney transplantation.

Abbreviations
ACE inhibitor

angiotensin converting enzyme inhibitor

CKD

chronic kidney disease

CMV

cytomegalovirus

CNI

calcineurin inhibitor

eGFR

estimated glomerular filtration rate

GFR

glomerular filtration rate

mGFR

measured glomerular filtration rate

Introduction

Transplant physicians have long been interested in improving renal graft function. Over the past decades, many therapeutic interventions have been developed aiming to augment—at least, preserve—GFR in renal transplant recipients [1]. This has been translated into an increasing number of transplant clinical trials using renal function as a primary or secondary endpoint [2]. The interest toward renal function in this context stems from the association which is consistently observed between the level of graft function, assessed either by serum creatinine concentration or by creatinine-based GFR estimates, and graft survival [3]. Albeit still controversial when applied to an individual patient [4, 5], the paradigm “the higher GFR, the longer graft survival” is largely accepted by the transplant community and continues to stimulate the quest toward less nephrotoxic therapeutic strategies.

Beyond a sole effect on graft survival, a better graft function might even be beneficial in terms of transplant recipients’ survival. Indeed, renal transplant patients are by definition considered to be affected by chronic kidney disease (CKD) and its related complications [6]. In the general population, data have accumulated suggesting that CKD—defined by a decreased GFR—is an independent risk factor for cardiovascular and all-cause mortality [7]. Likewise, an impaired renal function at a fixed time point after transplantation, usually within the first year, associates with patients’ death [3, 11]. By analyzing a large sample of transplant recipients from the United States Renal Data System [14], Schnitzler and colleagues have recently reported that the adjusted relative risk of death increased by at least 27% in patients with a MDRD-estimated GFR below 45 mL/min/1.73 m2 at 1 year posttransplant [15].

Lately the limits inherent to the static definition of CKD in the general population have been stressed. Two studies have shown in patients with native CKD that, compared to a single GFR value observed at a given time point, the rate of GFR decline is a stronger and independent risk of mortality [10, 16]. By capturing the progressive nature of CKD, the change of GFR over time might thus permit to further refine risk stratification for patients with CKD. Whether this notion holds true for renal transplant patients is uncertain. The trajectory of posttransplant CKD is particularly heterogeneous from one patient to another and somewhat unique with roughly 50% of patients experiencing no change or even improvement of GFR over time [17]. Although CKD progression has been associated to graft survival [18], it is not known whether the rate of decline in kidney function after the first year posttransplant is associated with an increased risk for death in renal transplant patients.

Herein, we have tested the hypothesis of an association between the variation of GFR after 1 year postrenal transplant and the risk of mortality.

Material and Methods

Study population

We analyzed all single renal transplantation performed in the transplant center of Saint-Etienne between January 1989 and December 2000 (n = 611). We selected patients having a functioning graft at 1 year posttransplant and for whom inulin-mesured GFR (mGFR) was available at 1 year. Renal inulin clearance was done according to the continuous infusion method as previously described [19]. Among the 529 (86%) patients with a functioning graft at 1 year posttransplant, 41 (7.7%) were excluded, 14 because of an inulin clearance below <15 mL/min/1.73 m2 and 27 (5%) because 1 year posttransplant inulin clearance was not available. Beyond the first anniversary of transplantation, mGFR was routinely done every 5 years for all kidney recipient patients. Patients were followed up to death or to 31 December 2010. Flowchart of the study cohort is presented in Figure 1. The study protocol was approved by the Institutional Review Board of our institution and was conducted in accordance with good international clinical practice guidelines.

Figure 1.

Flowchart detailing the exclusion and inclusion criteria for the creation of the population study cohort.

Change in mGFR (dependent variable)

The natural association between the rate of GFR decline and the initiation of dialysis can act as an informative censor during follow-up of the cohort. For this reason, annual variations in mGFR (i.e. slope) for the whole study population and for each patient were calculated using a joint random effects model (JREM). The JREM combines linear mixed random effects models with a Cox regression model in order to model the change of mGFR over time conditionally to the occurrence of dialysis. This model permits to reduce the bias of underestimation of the slope as previously reported [20]. Mixed random effects models, which are recommended for the analysis of the unbalanced repeated quantitative value, permit to take into account the different number of mGFR performed between patients [21].

Importantly patients with only one single inulin measurement were considered for the analysis using the JREM. This approach allows us to predict the slope of patients with a single measurement by modeling the values of inulin clearance from all participants and tends 1- to improve the precision of the slope's estimate and, 2- to limit the selection bias that would result from the exclusion of patients with only one single measurement [22].

The linear hypothesis for the slope of mGFR was tested with the addition of a fixed quadratic and cubic effect term, and normality and homoscedasticity of the residuals were examined using graphic methods. Linear slope hypothesis was accepted and the model with better goodness of fit was obtained by analyzing mGFR on the log scale as in other studies using this biomarker [23]. Statistical analysis was first done with mGFR transformed to the logarithmic scale and the results were then reexpressed as percent annual change (% variation/year) or absolute annual change (mL/min/1.73 m2/year). We imputed an mGFR value of 8 mL/min/1.73 m2 to the patients who returned to dialysis.

Covariates (independent variables)

Data for covariates were extracted from electronic database and from paper records. We collected data susceptible to influence patient's survival and mGFR slope. For recipients, age, gender, primary cause of renal disease (glomerular disease, polycystic kidney disease, hypertension, other, unknown), diabetes, duration of pretransplant dialysis, period of transplantation (before or after 1995), hepatitis C status and first versus subsequent kidney transplantation were collected; for donor, age, gender and type of donor (deceased/living) and, regarding the transplantation procedure, cold ischemia times, HLA matching, panel reactive antibodies (PRA), delayed graft function (defined as need for dialysis therapy in the first week posttransplant), episode of acute rejection during the first year posttransplant (Corticosteroid sensitive or not) and occurrence of CMV disease. Additionally, other clinical and biological 1 year posttransplant variables were considered: body mass index, systolic and diastolic blood pressure; ratio of proteinuria to urine creatinine, antihypertensive treatment (type), immunosuppressive treatment and treatment with statin. Proteinuria was also longitudinally collected at the time of each inulin clearance and considered as a time-varying covariate.

Mortality (outcome)

In our center like in most centers in France, transplant patients have a full medical check-up at least annually in their original transplantation center until the loss of the graft. These particularly close follow-up and relationship between the transplant patient and the medical team of the transplant center permits to easily trace patient's outcome. For patients who have returned to dialysis, the patients, their relatives or their dialysis centers were contacted by phone during the first weeks of January 2011.

Overall, we were able to clarify the life status and the date of death at the 31st of December 2010 of all our patients except 21 of them who were non-French citizens: 9 of them have returned to dialysis in their own country; the remaining 12 patients were lost to follow-up while they still had a functioning graft.

Only all-cause mortality was considered.

Relation between covariates and GFR slope

We estimated the mean effect of baseline characteristics on mGFR slope in univariate and multivariate analysis in order to identify independent factors associated with mGFR slope.

Relation between “dialysis” and “mortality

Start of dialysis and death are competing events susceptible to blur the association between the decline of renal function and all-cause mortality. In addition, there is a possible over-risk of mortality for dialysis patients compared to transplant patients with a functioning graft [26]

In order to better account for this relation, we carried out two distinct but complementary regression analyses:

  1. A classical competing risk regression in which “dialysis” was considered as an informative censor (dialysis-censored cohort). For this approach, we used a Fine and Gray analysis [27] that takes into account patients who reached the event “dialysis” before the event “death” for the calculation of subhazard ratio of mortality. This analysis permits to circumvent the bias associated to the Cox analysis, which excludes patients once they have reached the event “dialysis” for the calculation of HR.
  2. An alternative approach taking into account the occurrence of mortality until the end of follow-up (12-31-2010) for the entire population, including patients who have returned to dialysis (dialysis-uncensored cohort). In this approach, the time-dependent event “dialysis” was considered as an intermediate event in the causal chain leading to mortality. As such, “dialysis” is no more integrated into the multivariate analysis because of a risk of overadjustment. For this approach, we used a classical Cox regression analysis.

Statistical Analysis

Continuous and categorical variables were compared using ANOVA and chi-square tests, respectively.

Using a two-stages analysis procedure, we firstly estimated with JREM the fixed effect of mGFR (with the restricted maximum likelihood—REML) and the subject-specific predictor of mGFR slope with a Bayesian method (best linear unbiased predictor [BLUP]). In a second stage, we estimated the hazard ratios (HR) and 95% confidence intervals (95% CI) for all-cause mortality of 1 year posttransplant mGFR and individual-specific predictors of mGFR slope with a Cox regression model. One-year posttransplant mGFR and mGFR slope were analyzed as quartiles class and also in a nonlinear P-splines of degree 3 because of the absence of predefined hypothesis regarding this relationship.

Scaled Schoenfeld residuals were computed and plotted against time to evaluate potential violations of the proportional hazards assumption for Cox model. SubHRs and HRs for all-cause mortality were calculated for the dialysis-censored and the dialysis-uncensored cohort. In the Cox regression analysis, the nested models were compared using the log-likelihood ratio test.

HRs and subHRs (SHRs) of all-cause mortality associated to 1 year posttransplant mGFR and to mGFR slope after 1 year posttransplant were assessed in crude analysis, then in nested models adjusted for confounding factors.

The multivariate analysis for JREM, Fine and Gray and Cox model incorporated variables found to significant at p ≤ 0.20 in the unadjusted analysis.

Several sensitivity analyses were done to confirm the association with all-cause mortality. First, estimation of the individual mGFR slope was tested for different levels of mGFR imputed to patients starting dialysis (5, 10 mL/min/1.73 m2 and no imputation). Second, we repeated this analysis after stratifying patients across the different CKD stages at 1 year posttransplantation and according to donor's age and sex, recipient's age and sex, diabetes and proteinuria.

A 95% CI of an HR that did not include unity was considered statistically significant. All p-values were two-tailed, and values <0.05 were considered significant. Analyses were performed with SAS software V 9.1 (SAS Institute, Cary, NC, USA) and R (JMpackage [28]; www.r-project.org).

Results

Demography

We built an inception cohort of 488 renal transplant recipients whom GFR was longitudinally followed by measuring inulin clearance (mGFR) (Figure 1). Mean follow-up was 12 ± 5 years during which the median [IQ] number of mGFR by patient was 3 [2–4]. Among the 488 patients of the cohort, 69 (14%) and 122 (25%) patients have had only one and two mGFR, respectively. Few inulin clearances were missed by patients free of dialysis and not dead at the time of the scheduled inulin clearance (n = 113 out of 1433 inulin measurements (8%)).

The mean [Interquartile Range] of mGFR decline was −1.9 [−2.2; −1.5] mL/min/1.73 m2/year or −4.3 [−5.1; −3.5] when expressed as percent annual change.

Demographic and clinical characteristics of donors and recipients with a functioning graft at 1 year posttransplantation for the whole population and by quartiles of mGFR slope are presented in Table 1.

Table 1. Demographic and clinical characteristics of the study population
  Quartile of mGFR slope 
ParameterWhole cohortQ1Q2Q3Q4Global test p
  1. PRA = panel reactive antibodies; CNI = calcineurin inhibitor; ACE inhibitor = angiotensin converting enzyme inhibitor.

N488122122113131 
mL/min/1.73 m2/year [range][+1.8 ; −8.0][+1.8; −0.46][−0.47; −1.8][−1.9; −3.6][−3.6; −8.0] 
% Annual change [range][+3.7; −17][3.7; −1.0][−1.0; −4.0][−4.0; −7.0][−7.0; −17] 
Characteristics      
Recipients (n)488     
At the time of transplant      
Age (year; mean ± SD)45.4 ± 12.942.8 ± 12.245.1 ± 12.551.2 ± 11.943.3 ± 13.0<0.0001
Male gender, n (%)339 (69%)90 (74%)81 (66%)82 (73%)86 (66%)0.39
Cohort period: 1989–1994, n (%)268 (55 %)54 (44%)66 (54%)67 (59%)81 (62%)0.03
Cause of renal disease-–n (%)      
Glomerulopathy160 (33%)33 (27%)38 (31%)35 (31%)54 (41%) 
Diabetes5 (1%)01 (1%)0 (0%)4 (3%) 
Polycystic kidney disease68 (14%)18 (15%)18 (15%)22 (19%)10 (8%) 
Hypertension38 (8%)8 (7%)11 (9%)8 (7%)11 (8%) 
Interstitial nephritis37 (8%)11 (9%)8 (7%)10 (9%)8 (6%) 
Others121 (25%)36 (30%)29 (24%)29 (26%)27 (21%) 
Unknown59 (12%)16 (13%)17 (14%)9 (8%)17 (13%) 
Diabetes, n (%)30 (6%)3 (2%)8 (7%)9 (8%)10 (8%)0.26
Hepatitis C infection, n (%)37 (8%)5 (4%)7 (6%)12 (11%)13 (10%)0.16
Preemptive transplantation, n (%)43 (9%)12 (10%)9 (7%)9 (8%)13 (10%)0.85
Pretransplant dialysis duration (months, mean ± SD)39.5 ± 44.035.9 ± 36.647.7 ± 54.742.1 ± 50.432.7 ± 29.60.05
Retransplantation, n (%)76 (16%)16 (13%)21 (17%)18 (16%)21 (16%)0.84
Donors      
Age (year; mean ± SD)37.8 ± 13.832.5 ± 12.336.7 ± 13.642.2 ± 14.040.1 ± 13.7<0.0001
Male, n (%)353 (72%)92 (75%)84 (69%)81 (72%)96 (73%)0.71
Living donor, n(%)18 (4%)8 (7%)4 (3%)06 (5%)0.06
Transplantation—      
Baseline and 1 year posttransplantation      
Cold ischemia time > 24 h, n (%)276 (57%)60 (50%)65 (54%)72 (64%)79 (61%)0.10
HLA mismatch > 3, n (%)323 (66%)85 (70%)76 (62%)70 (62%)92 (70%)0.34
DR mismatch ≥1, n (%)326 (67%)85 (70%)83 (68%)72 (64%)86 (66%)0.78
PRA > 80%39 (9%)6 (5%)12 (11%)12(11%)9 (7%)0.34
CMV Infection, n (%)254 (52%)54 (44%)61 (50%)58 (51%)81 (62%)0.04
Delayed graft function, n (%)127 (26%)25 (20%)30 (25%)29 (26%)43 (33%)0.16
1 year-rejection, n (%)      
Steroid-sensitive117 (24%)24 (20%)32 (26%)28 (25%)33 (25%) 
Steroid-resistant85 (17%)16 (13%)17 (14%)20 (18%)32 (24%)0.11
Induction treatment, n (%)150 (31%)40 (33%)44 (36%)30 (27%)36 (27%)0.33
Recipients—      
1 year posttransplantation      
Body mass index24 [21; 26]23 [22–25]23 [21–26]24 [22–27]24 [21–26]0.09
(kg/m2; median [IQ])      
Systolic blood pressure (mmHg; mean ± SD)136.6 ± 17.0132.7± 15.0135.3 ± 18.7140.4± 16.9138.3 ± 16.30.004
Diastolic blood pressure78.0 ± 11.376.8 ± 11.377.1± 12.079.3 ± 10.078.8 ± 11.60.28
(mmHg; mean ± SD)      
≥ 1 antihypertensive drugs387 (80%)95 (78%)89 (73%)93 (82%)110 (84%)0.014
Inulin clearance47.2 ± 19.261.0 ± 19.149.1 ± 17.039.7 ± 14.639.2 ± 17.1<0.0001
(mL/min/1.73 m2; Mean ± SD)      
Urinary protein creatinine ratio233 ± 66593.8 ± 375.3168.0 ± 480.3146.1 ± 275.6505.4 ± 1054.4<0.0001
(mg/g; mean ± SD)      
Proteinuria <0. 300 (mg/day)353 (72%)103 (84%)95 (78%)87 (77%)68 (52%) 
300 ≤ Proteinuria <0. 1000, n (%)61 (13%)8 (7%)7 (6%)14 (12%)32 (24%) 
1000 ≤ Proteinuria, n (%)25 (5%)1 (1%)6 (5%)2 (2%)16 (12%) 
Not available49 (10%)10 (8%)14 (11%)10 (9%)15 (11%)<0.0001
Immunosuppressive regimen      
without CNI15 (3%)3 (2%)4 (3%)2 (3%)5 (4%) 
with CNI473 (97%)119 (98%)118 (97%)110 (97%)126 (96%)0.92
ACE inhibitor, n (%)87 (18%)23 (19%)14 (11%)21 (19%)29 (22%)0.16
Calcium channel inhibitor293 (60%)80 (66%)65 (53%)67 (59%)81 (62%)0.25
Diuretics, n (%)172 (35%)21 (17%)39 (32%)57 (50%)55 (42%)<0.0001
Betablocker, n (%)108 (22%)24 (20%)25 (20%)25 (22%)34 (26%)0.63
Statin, n (%)21 (4%)6 (5%)6 (5%)6 (5%)3 (2%)0.62

The proportion of patients across the different CKD stages at 1 year posttransplant was 4%, 19%, 60% (28% and 31%) and 19% for stage 1, stage 2, stage 3 (3a and 3b) and stage 4, respectively.

Characteristics associated with 1 year posttransplant mGFR value and 1 year posttransplant GFR slope (Tables 1 and 2)

In multivariate analysis, the following factors were significantly associated with lower 1-year posttransplant mGFR values: longer transplantation duration, older donors, history of diabetes, higher HLA mismatch, rejection within the first year, lower BMI, higher urinary protein rate and diuretics use at 1 year posttransplantation.

Table 2. Multivariate analysis, using a joint model, of factors associated with 1-year posttransplant mGFR and 1-year posttransplant GFR slope
  mGFR at 1 yearSlope %/years
N = 488NEstimate[CI95%]p-ValueEstimate[CI95%]p-Value
  1. PRA = panel-reactive antibodies; CNI = Calcineurin inhibitor; ACE inhibitor = angiotensin converting enzyme inhibitor; ATG = antithymoglobulin.

Recipients at transplantation       
Age48839.13[26.10; 58.66]0.98−6.33[−15.57; 2.97]0.37
Gender       
Male33938.43[25.64; 57.60]Ref−5.44[−14.67; 3.88]Ref
Female14939.87[26.62; 59.70]0.37−7.89[−17.09; 1.36]0.006
Cohort period       
1989–199426836.68[24.52; 54.86]Ref−7.95[−17.13; 1.28]Ref
1995–200122041.77[27.81; 62.75]0.005−5.38[−14.65; 3.97]0.01
Diabetes       
No45843.02[29.01; 63.78]Ref−6.16[−15.03; 2.77]Ref
Yes3035.61[23.29; 54.46]0.02−7.17[−17.03; 2.77]0.61
Hepatitis C infection       
No45139.59[26.64; 58.84]Ref−5.35[−14.39; 3.75]Ref
Yes3738.69[25.42; 58.89]0.75−7.98[−17.55; 1.68]0.11
Dialysis seniority p for trend0.51p for trend0.37
Preemptive transplantation4440.35[26.70; 60.98]Ref−7.93[−17.33; 1.54]Ref
<12 months11337.70[25.03; 56.79]0.34−6.18[−15.50; 3.23]0.25
>12 months33139.42[26.37; 58.93]0.32−5.89[−15.06; 3.34]0.78
Living donor       
No47038.00[25.51; 56.61]Ref−5.14[−14.08; 3.87]Ref
Yes1840.31[26.15; 62.15]0.58−8.19[−18.22; 1.93]0.23
Retransplantation        
No41244.27[42.5; 46.24]Ref.−5.44[−14.72; 3.91]Ref
Yes7643.32[39.1; 47.92]0.70−7.89[−17.11; 1.40]0.04
Donor at the transplantation       
Age48836.93[24.80; 55.00]<.0001−7.09[−16.17; 2.06]0.16
Graft at the transplantation       
Cold ischemia       
≤ 24 h20839.19[26.29; 58.42]Ref−6.57[−15.66; 2.60]Ref
> 24 h27639.09[25.97; 58.84]0.95−6.76[−16.08; 2.63]0.83
HLA mismatch       
≤ 316537.24[24.92; 55.65]Ref−5.54[−14.73; 3.71]Ref
> 332341.13[27.36; 61.84]0.03−7.79[−17.06; 1.55]0.03
DR mismatch       
016239.02[26.01; 58.54]Ref−6.44[−15.66; 2.86]Ref
>032639.26[26.22; 58.77]0.89−6.89[−16.11; 2.39]0.64
PRA p for trend0.46p for trend0.50
0 ≤ PRA < 20%32038.11[25.49; 56.99]Ref−6.63[−15.82; 2.63]Ref
20 ≤ PRA < 80%9838.15[25.41; 57.28]0.98−6.10[−15.43; 3.30]0.63
80% ≤ PRA3942.77[28.13; 65.01]0.12−9.13[−18.59; 0.40]0.11
Not available3137.74[24.66; 57.77]0.90−4.80[−14.46; 4.94]0.29
CMV infection       
No23438.90[25.96; 58.28]Ref−5.83[−15.03; 3.45]Ref
Yes25439.39[26.31; 58.97]0.74−7.51[−16.72; 1.78]0.04
Delayed graft function       
No36139.14[26.19; 58.51]Ref−6.31[−15.45; 2.91]Ref
Yes12739.14[26.05; 58.81]1.00−7.02[−16.33; 2.35]0.46
Rejection within the first year p for trend<.001p for trend0.39
No28645.35[30.23; 68.01]Ref−7.36[−16.62; 1.98]Ref
Not requiring ATG treatment11737.86[25.26; 56.74]<.001−6.01[−15.20; 3.25]0.18
Requiring ATG treatment8534.92[23.19; 52.58]<.001−6.63[−15.95; 2.77]0.51
Induction treatment       
No33838.71[25.93; 57.79]Ref−6.16[−15.26; 3.01]Ref
ATG or anti CD2515039.58[26.29; 59.57]0.63−7.17[−16.53; 2.26]0.33
Recipients at 1 year       
Body mass index (kg/m2)48839.68[26.64; 59.09]0.006−6.87[−15.95; 2.27]0.08
Systolic blood pressure (mmHg)48839.05[26.16; 58.30]0.06−6.68[−15.82; 2.54]0.73
Biology at 1 year       
Urinary protein rate (mg/day; Mean ± SD) p for trend0.0002p for trend0.003
Proteinuria < 30035345.04[30.08; 67.44]Ref−4.11[−13.21; 5.07]Ref
300 ≤ Proteinuria < 10006136.55[24.23; 55.13]0.003−7.54[−16.84; 1.85]0.008
1000 ≤ Proteinuria2536.57[23.84; 56.08]0.02−10.54[−20.52; 0.46]0.003
Not available4938.99[25.79; 58.96]0.02−4.48[−13.96; 5.08]0.80
Treatments at 1 year       
Immunosuppressive regimen p for trend0.04p for trend0.15
With CNI44143.39[29.38; 64.09]Ref−5.04[−13.91; 3.89]Ref
Without CNI3041.14[27.07; 62.52]0.51−8.69[−18.33; 1.02]0.06
Not available1733.59[21.47; 52.55]0.01−6.26[−16.45; 4.01]0.60
ACE inhibitor       
No40138.41[25.63; 57.55]Ref−6.37[−15.55; 2.88]Ref
Yes8739.89[26.58; 59.87]0.45−6.96[−16.26; 2.41]0.60
Calcium channel inhibitor       
No19537.62[25.11; 56.37]Ref−6.26[−15.50; 3.06]Ref
Yes29340.72[27.19; 61.00]0.05−7.07[−16.26; 2.18]0.36
Diuretics       
No31642.18[28.45; 62.54]Ref−6.59[−15.58; 2.47]Ref
Yes17236.32[23.99; 54.99]0.0004−6.74[−16.17; 2.77]0.87
Betablocker       
No38039.42[26.44; 58.78]Ref−6.47[−15.57; 2.70]Ref
Yes10838.86[25.79; 58.57]0.76−6.86[−16.21; 2.56]0.70

Factors associated with a more rapid mGFR decline were longer transplantation duration, feminine gender, higher HLA mismatch, retransplantation, CMV infection and higher urinary protein rate at 1 year.

One year posttransplant mGFR, mGFR decline beyond the first year and the risk for death

During the mean (±SD) follow-up of 12 ± 5 years, 136 patients (28%) of the cohort returned to dialysis and 139 died (including 32 patients after starting dialysis). Twelve patients with a functioning graft were lost to follow-up after 9 ± 5 years. Outcome after dialysis was available for all patients except for nine non-French citizens. Rate of all-cause mortality over the follow-up period was 23 per 1000 person-years. Causes of death are presented in Table 3. The three leading causes were cardio-vascular, cancer and infection.

Table 3. Causes of mortality before and after starting dialysis
Causes of mortalityDeath before dialysis (N = 107)Death after dialysis (N = 32)
Cardio-vascular35 (32%)11 (37%)
Cancer33 (30%)3 (10%)
Infection17 (15.5%)3 (10%)
Hepatopathy6 (5.5%)1 (3%)
Malnutrition3 (2.5%)1 (3%)
Diabetes3 (2.5%)-
Respiratory1 (0.9%)-
Undetermined9 (8.5%)13 (40%)

In crude analysis using a competing-risks regression (Fine and Gray analysis) applied to the dialysis-censored cohort, we found that compared with patients who experienced lower mGFR decline (Quartile 1: [3.7; −1%] mL/min/1.73 m2/year), those who experienced mGFR slope between ([−1; −4%]: Q2) and ([−4; −7%]: Q3) exhibited a significant increased risk of mortality (Q2: HR [IC95%]: 3.42 [1.74–6.72]; Q3: 6.36 [3.33;12.14). Of note the risk for death was diminished for patients with higher mGFR decline (Q4: HR 2.35 [1.02–5.38]) but still significant probably due to the higher frequency of dialysis in this group (88%) (Table 4). A statistically significant trend test was however found for an increased mGFR decline and a higher risk of mortality in the dialysis-censored cohort (p<0.0001).

We then applied a Cox regression analysis taking into account the occurrence of death until the end of follow-up for the entire cohort, including patients who have returned to dialysis (dialysis-uncensored cohort).

With this approach presented in Table 4, we found a similar trend with a higher risk of mortality for the highest quartile of mGFR decline.

Table 4. Crude and adjusted subhazard ratio (sHRs [95% CI]) of all-cause mortality (censored for dialysis) according to quartiles of mGFR slope (Fine and Gray model)
Death censored for dialysis
VariableN = 488Dialysis N = 136Death before dialysis N=107Sub hazard ratio sHR [95%CI]p
  1. *Period of time for transplantation, pretransplant dialysis duration, HLA mismatch, PRA, delayed graft function, living donor, CMV infection, acute rejection, recipient's age and sex, donor's age and sex, causes of renal disease, diabetes; at 1 year posttransplantation: BMI, systolic blood pressure, use of ACE inhibitors, use of diuretics, use of betablockers.

Crude analysis
Quartiles of mGFR slope   p for trend<0.0001
Q1: (3.7; −1 %)/mL/min/1.73 m2/year122011Ref 
Q2: (−1; −4.0 %)/mL/min/1.73 m2/year1220353.42 [1.74 ;6.72]0.0003
Q3: (−4.0; −7.0 %)/mL/min/1.73 m2/year11320506.36 [3.33 ;12.14]<0.0001
Q4: (−7.0; −17 %)/mL/min/1.73 m2/year131116112.35 [1.02 ;5.38]0.04
−2 LogV = 1168     
Quartiles of 1 year posttransplant mGFR   p for trend0.02
Q1: (15–33) mL/min/1.73 m212658321.97 [1.13;3.43]0.02
Q2: (34–45) mL/min/1.73 m212330301.52 [0.87;2.63]0.14
Q3: (46–58) mL/min/1.73 m21192423Ref 
Q4: (59-122) mL/min/1.73 m212024221.09 [0.60;1.99]0.77
−2 LogV = 1202     
Model 1: Groups of mGFR Slope adjusted for baseline mGFR
Quartiles of mGFR slope   p for trend0.0002
Q1: (3.7; −1 %)/mL/min/1.73 m2/year122011Ref 
Q2: (−1; −4.0 %)/mL/min/1.73 m2/year1220353.56 [1.76;7.19]0.0004
Q3: (−4.0; −7.0 %)/mL/min/1.73 m2/year11320506.57 [3.25;13.30]<0.0001
Q4: (−7.0; −17 %)/mL/min/1.73 m2/year131116112.35 [1.01;5.47]0.05
Quartiles of 1 year posttransplant mGFR   p for trend0.93
Q1: (15–33) mL/min/1.73 m212658321.29 [0.73 ; 2.29]0.37
Q2: (34–45) mL/min/1.73 m212330301.21 [0.70;2.10]0.49
Q3: (46–58) mL/min/1.73 m21192423Ref 
Q4: (59–122) mL/min/1.73 m212024221.48 [0.80;2.71]0.21
−2 LogV = 1166     
Model 2: Model 1 adjusted for covariates*
Quartiles of mGFR slope   p for trend0.001
Q1: (3.7; −1 %)/mL/min/1.73 m2/year122011Ref 
Q2: (−1; −4.0 %)/mL/min/1.73m2/year1220354.19 [1.78 ;9.76]0.0009
Q3: (−4.0; −7.0 %)/mL/min/1.73 m2/year11320506.68 [2.91 ;15.35]<0.0001
Q4: (−7.0; −17 %)/mL/min/1.73 m2/year131116112.34 [0.89;6.19]0.09
Quartiles of 1 year posttransplant mGFR   p for trend0.16
Q1: (15–33) mL/min/1.73 m212658321.051 [0.535;2.068]0.87
Q2: (34–45) mL/min/1.73 m212330301.278 [0.668;2.446]0.37
Q3: (46–58) mL/min/1.73 m21192423Ref 
Q4: (59–122) mL/min/1.73 m212024222.091 [0.994;4.399]0.03
−2 LogV = 1053     

With regard to 1 year posttransplant mGFR, in crude analysis, we observed a statistically significant higher risk of mortality for quartile 1 mGFR (15–33 mL/min/1.73 m2), compared to mGFR quartile 3 (46–58 mL/min/1.73 m2) in both the dialysis-censored and uncensored cohorts (Tables 4 and 5).

Table 5. Crude and adjusted hazard ratio (HR [95% CI]) of all-cause mortality not censored for dialysis according to quartiles of mGFR slope (Cox regression)
Death for dialysis-uncensored follow-up
VariableN = 488Dialysis N = 136Death N = 139Hazard ratio HR [95% CI]p
  1. *Period of time for transplantation, pretransplant dialysis duration, HLA mismatch, PRA, delayed graft function, living donor, CMV infection, acute rejection, recipient's age and sex, donor's age and sex, causes of renal disease, diabetes; at 1 year posttransplantation: BMI, systolic blood pressure, proteinuria, use of ACE inhibitors, use of diuretics, use of betablockers

Crude analysis
Quartiles of mGFR slope   p for trend<0.0001
Q1: (3.7; −1 %)/mL/min/1.73 m2/year122011Ref. 
Q2: (−1; −4.0 %)/mL/min/1.73 m2/year1220333.84 [1.89;7.79]0.0002
Q3: (−4.0; −7.0 %)/mL/min/1.73 m2/year11320526.87 [3.49;13.52]<0.0001
Q4: (−7.0; −17 %)/mL/min/1.73 m2/year131116434.75 [2.39;9.46]<0.0001
−2 LogV = 1564/AIC=1570     
Quartiles of 1 year posttransplant mGFR   p for trend0.0002
Q1: (15–33) mL/min/1.73 m212658542.39 [1.50;3.82]0.0003
Q2: (34–45) mL/min/1.73 m212330341.23 [0.73;2.06]0.42
Q3: (46–58) mL/min/1.73 m21192426Ref. 
Q4: (59–122) mL/min/1.73 m212024251.08 [0.63;1.88]0.77
−2 LogV = 1598/AIC=1592     
Model 1 : Quartiles of mGFR Slope adjusted for baseline quartiles of mGFR
Quartiles of mGFR slope   p for trend0.0004
Q1: (3.7; −1 %)/mL/min/1.73 m2/year122011Ref. 
Q2: (−1; −4.0 %)/mL/min/1.73 m2/year1220333.73 [1.82;7.66]0.0003
Q3: (−4.0; −7.0 %)/mL/min/1.73 m2/year11320526.34 [3.12;12.92]<0.0001
Q4: (−7.0; −17 %)/mL/min/1.73 m2/year131116434.11 [2.00;8.41]0.0001
Quartiles of 1 year posttransplant mGFR   p for trend0.23
Q1: (15–33) mL/min/1.73 m212658541.64 [1.01 ; 2.67]0.04
Q2: (34–45) mL/min/1.73 m212330341.008 [0.60;1.69]0.97
Q3: (46–58) mL/min/1.73 m21192426Ref. 
Q4: (59–122) mL/min/1.73 m212024251.34 [0.77 ; 2.34]0.29
−2 LogV = 1558/AIC=1570     
Model 2 : Model 1 adjusted for covariates*
Quartiles of mGFR slope   p for trend0.001
Q1: (3.7; −1 %)/mL/min/1.73 m2/year122011Ref. 
Q2: (−1; −4.0 %)/mL/min/1.73 m2/year1220334.78 [2.19 ;10.43]<0.0001
Q3: (−4.0; −7.0 %)/mL/min/1.73 m2/year11320526.30 [2.89;13.72]<0.0001
Q4: (−7.0; −17 %)/mL/min/1.73 m2/year131116435.36 [2.44;11.75]<0.0001
Quartiles of 1 year posttransplant mGFR   p for trend0.93
Q1: (15–33) mL/min/1.73 m212658541.65 [0.97;2.82]0.07
Q2: (34–45) mL/min/1.73 m212330341.06 [0.61;1.85]0.83
Q3: (46–58) mL/min/1.73 m21192426Ref 
Q4: (59–122) mL/min/1.73 m212024251.93 [1.02;3.65]0.04
−2 LogV = 1308/AIC=1380     

Because the association of mGFR slope and risk of all-cause mortality is presumably confounded by the baseline mGFR value and other risk factors, we computed nested cox models adjusted for several covariates.

In a first model, we investigated the relationship of mGFR slope with mortality after adjustment on 1-year posttransplant mGFR (Tables 4 and 5). The association between groups of mGFR decline and all-cause mortality tended to weaken but the tendency remained statistically significant. With cubic spline regression analysis applied to the dialysis-uncensored cohort, we still found a significant association in univariate analysis for mGFR at 1 year and mGFR slope with mortality (Figure 2). However after adjustment on GFR slope, the 1-year mGFR was no longer significantly associated to mortality (Figure 2).

Figure 2.

Association of mGFR at 1 year and mGFR decline beyond the first year posttransplant with all-cause mortality (Cox regression with cubic spline analysis in the dialysis-uncensored cohort).

In a model adjusting on all covariates, analysis of mGFR as quartile or as continuous nonlinear variable showed a persistent association between mGFR slope and mortality. In this model, 1-year posttransplant mGFR was again not significantly associated with mortality.

When regression analyses were stratified by baseline CKD stages at 1 year posttransplant groups of mGFR slope were associated with similar risk for all-cause mortality (Supporting data).

There was no evidence of interaction between groups of change in mGFR and age, sex and use of ACE inhibitors at 1 year posttransplant (Supporting data).

Discussion

In this inception cohort of renal transplant recipients, we found an independent association between declining graft function after the first year of transplantation and elevated risk for all-cause mortality. This association was observed regardless of the level of GFR measured at 1 year posttransplant and was even better than the association observed between 1 year posttransplant mGFR and all-cause mortality.

We identified several characteristics independently associated with mGFR slope such as proteinuria, HLA mismatch and CMV infection that can represent therapeutic targets in order to slow the mGFR decline and try to limit the mortality excess risk.

While an association between change in GFR over time and an increased risk for death has been reported in the general population and more recently in native CKD patients [7, 8], this is the first time, to our knowledge, that this observation is made in renal transplant patients. The existence of such an association in the transplant patient may first seem evident but is actually not so intuitive. The transplant CKD patient is not a typical CKD patient especially regarding the trajectory of renal function changes over time (stability or even improvement for almost 50% of renal transplant recipients [17]. Moreover incidental events susceptible to strongly impact GFR slope (rejection, drug toxicty, infections) may occur more frequently throughout the course of the transplantation compared to native CKD.

In spite of those transplant-related specificities, we confirm that, similar to the nontransplant population, renal transplant patients who exhibit faster progression of CKD are at higher risk of mortality. Our data support the notion that change in kidney function may, therefore, also be an important prognostic marker in the setting of transplantation. Despite the existence of some degree of colinearity between GFR slope and the 1 year posttransplant mGFR value, our model including those two factors suggest that the association between CKD progression and mortality is independent and stronger than the one between the fixed 1-year posttransplant GFR value and mortality.

In our population, the mean mGFR decline was −4.3%/year. This rate may appear much higher than what is traditionally reported in renal transplant population [17, 29, 30]. However, we used a direct measure of GFR while the overwhelming majority of data on the GFR slope in renal transplant patients comes from creatinine-estimated GFR, an evaluation known to underestimate the slope [31]. Using a reference method of GFR measurement, Gera et al. have reported a comparable slope after 1 year posttransplant (−3.9%/year) [31].

In regard to the evaluation of GFR, we believe that the exclusive reliance on inulin-measured GFR is one of the strengths of our study. Along with the fact that we analyzed a large, rather unselected renal transplant population with long-term follow-up, the use of a reference method to measure GFR permit to better capture the relation between renal graft function and patient outcome [32]. This has however to be balanced with the rather long intervals (every 5 years) between each GFR measurement that can lead to a loss of information regarding the evaluation of the GFR slope during the interval.

Given the poor performance of creatinine-based GFR equations in transplant patients [19], it will be interesting to test whether the association between the GFR decline and mortality remains true when GFRs are estimated with commonly used equations based on creatinine or alternative GFR biomarkers.

Importantly, by tracking patients’ outcome after they had returned to dialysis, we have been able to confirm the strong association between mGFR decline and all-cause mortality even after (re-)starting dialysis probably because of a higher risk of mortality for these patients compared to those who kept a functioning graft. This observation underscores the necessity to consider postdialysis outcome when evaluating potential risk factors for mortality in renal transplant patients.

We did not observe a perfect dose–response relationship between the GFR slope and mortality due to a higher risk of death associated to GFR slope quartile 3 compared to quartile 4. This may have to do with the high proportion of patients who restart dialysis in this quartile within this particular case a protective effect but transient, of dialysis. While dialysis is likely to be overall and in the long run, associated with an overrisk of mortality, it may initially benefit the transplant patient with end-stage renal disease.

There are several limitations to our study that need to be considered when interpreting our data. First of all, this is a single-center study cohort that will need to be confirmed in other renal transplant populations. In this respect, since our patients are almost exclusively Caucasians, our results may not be generalizable to non-Caucasians populations. Of note, a very low proportion (less than 10%) of our patients were diabetic at the time of transplantation. Diabetes prevalence is expected to be much higher in other country like in the United States.

Although we have been able to adjust our analysis on several known and putative factors associated with death in transplant patients, we missed two important cardiovascular comorbidities, i.e. smoking status and hypercholesterolemia (even though for the latter, we try to minimize the bias by adjusting on statin treatment). Additionally, we cannot rule out the possibility of other unknown residuals confounders at 1 year and during follow-up. While our multivariate analysis was adjusted on time-varying proteinuria, others potentially relevant covariates were only collected at one fixed time point and not considered as time-dependent factors (e.g. anti-HLA antibodies).

The absence of information on postdialysis vital status for nine patients constitutes a potential—albeit probably minimal—bias of underestimation of the risk of mortality once patients return to dialysis. Another potential bias of our study is regarding the use or modification of treatments made by physicians for patients with worsening renal graft function. The observational nature of our study does not permit to account for all possible treatments adjustments to declining GFR.

We restricted our analysis to all-cause mortality because of insufficient information and statistical power to study-specific causes of death. Further research is needed to study whether our findings extend to specific causes of death, particularly to cardio-vascular death.

Finally, due to the observational design of our study, we cannot establish a causal relationship between mGFR evolution and mortality risk in renal transplant patients.

In conclusion, in renal transplantation, the rate of graft function decline after the first year posttransplant is an independent risk factor for all-cause mortality that might exhibit a better prognostic value than the isolated value of GFR observed at 1 year posttransplant. While we do acknowledge that this observation does not automatically qualify “GFR slope” as a pertinent and valid biomarker in renal transplantation, we believe that “GFR slope” is potentially far less flawed than a single fixed value of GFR obtained at 1 year posttransplant.

Our data call for interventional trials evaluating strategies of graft protection according to their ability to slow the decline of GFR along with their impact on long-term mortality.

Disclosure

The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation.

Ancillary