Data on long-term outcomes of users of inhibitors of the mammalian target of rapamycin (mTORI) are lacking in kidney transplantation. In an analysis of 139 370 US kidney transplant recipients between 1999 through 2010, we compared clinical outcomes among users of mTORIs versus calcineurin inhibitors (CNI) in their primary immunosuppresive regimen. During the first 2 years posttransplantation, primary use of mTORIs without CNIs (N = 3237) was associated with greater risks of allograft failure and death compared with a CNI-based regimen (N = 125 623); the hazard ratio (HR) of the composite outcome ranged from 3.67 (95% confidence interval [CI], 3.12–4.32) after discharge to 1.40 (95% CI 1.26–1.57) by year 2. During years 2–8, primary use of mTORIs without CNIs was independently associated with greater risks of death (HR 1.25; 95% CI, 1.11–1.41) and the composite (HR 1.17; 95%CI, 1.08–1.27) in fully adjusted analyses. The results were qualitatively unchanged in subgroups defined by medical history, immunological risk and clinical course during the index transplant hospitalization. In a propensity-score matched cohort, use of mTORIs was associated with significantly worse outcomes during the first 2 years and greater risks of death (HR 1.21; 95% CI, 1.05–1.39) and the composite (HR 1.18; 95% CI, 1.08–1.30) in years 2–8. Compared with CNI-based regimens, use of an mTORI-based regimen for primary immunosuppression in kidney transplantation was associated with inferior recipient survival.
Kidney transplantation improves survival and quality of life in patients with kidney failure (2004). The introduction of calcineurin inhibitors (CNI) in the 1980s as the backbone of immunosuppressive regimens dramatically reduced rates of acute rejection and improved allograft and recipient survival following kidney transplantation (2010). Despite their efficacy in preventing acute rejection, nephrotoxicity of CNIs may jeopardize long-term allograft survival, and adverse effects on blood pressure, lipids and glycemia may increase cardiovascular risk (2004). Inhibitors of the mammalian target of rapamycin (mTORI) were developed as alternative immunosuppressive agents and approved for use in kidney transplantation in 1999 (1999, 2001, 2000).
Some short-term clinical trials suggested beneficial effects of mTORIs on kidney function relative to CNIs (2002, 2001). Although other trials reported greater risk of acute rejection among mTORI users (2007, 2011, 2011), adequately powered, long-term trial data comparing the impact of mTORIs versus CNIs on hard clinical outcomes are lacking. In a single-center, prospective, observational study of prevalent kidney transplant recipients, we reported that use versus nonuse of mTORIs was associated with significantly increased risk of mortality (2012). However, we could not determine whether excess mortality was directly attributable to mTORIs use or was driven by the clinical decision to convert from CNIs to mTORIs when kidney function was already deteriorating. The purpose of the current study was to mitigate this source of bias by testing the hypothesis that de novo use of mTORIs compared with CNIs in the primary immunosuppressive regimen prescribed at the time of discharge from the index transplant hospitalization is associated with greater long-term risks of allograft failure and mortality in kidney transplant recipients in the United States.
Materials and Methods
Sources of data
The United Network for Organ Sharing (UNOS) dataset was the primary source for exposure, covariate and outcome data. The United States Renal Data System (USRDS) served as an additional source to capture more complete comorbidity data. We used the unencrypted transplant recipient registration identification (TRR_ID) to merge the two national registries, as has been done previously (2011). The study was approved by UNOS, USRDS, the National Institute of Diabetes and Digestive and Kidney Diseases, the Health Resources and Services Administration and the institutional review board of the University of Miami Miller School of Medicine, which waived the requirement for informed consent.
Based on UNOS data as of March 4th, 2011, we identified all adult and pediatric first-time kidney-only transplants that occurred between September 16th 1999 and December 31st 2010 (N = 153 669), which coincided with the Food and Drug Administration's approval of sirolimus for use in kidney transplantation. UNOS ascertains the primary maintenance immunosuppressive regimen at the time of discharge from the index transplant hospitalization. Therefore, we excluded patients who experienced death or allograft failure prior to discharge (N = 6563) and patients who had incomplete information on maintenance immunosuppressive drugs at discharge (N = 7434). We also excluded 302 patients with missing information on time-to-end-points. The final analytic sample consisted of 139 370 patients.
Exposure to mTOR inhibitors
We divided patients into three groups according to exposure to mTORI in their primary maintenance immunosuppressive regimen: mTORI without CNI (N = 3237), defined as use of sirolimus (99%) or everolimus (1%) but not cyclosporine or tacrolimus; CNI (N = 125 623), defined as use of cyclosporine or tacrolimus without sirolimus or everolimus; and CNI with mTORI (N = 10 510), defined as receiving a combination of CNI and mTORI. In order to conservatively manage potential bias due to changes in therapy in response to worsening clinical course, we used an intention-to-treat analysis that maintained individuals in their originally assigned de novo treatment group throughout follow-up.
The primary outcomes were times to death-censored allograft failure, death and their composite. For each outcome, we censored at loss to follow-up or at year 8 posttransplantation due to small numbers of subsequent events. In analyses of death, we censored at allograft failure because data on subsequent mortality are not uniformly ascertained by UNOS. The 603 patients reported as having experienced allograft failure and death on the same day were considered as having had events in analyses of each individual outcome and as a single event in analyses of the composite outcome.
We included the following covariates in the multivariable models: recipient and donor demographics (age, gender, race, ethnicity); recipient education; etiology of kidney failure (diabetes, hypertension, polycystic kidney disease, glomerulonephritis, other, unknown); dialysis duration prior to transplantation (years); donor type (living, deceased, expanded criteria); human leukocyte antigen (HLA) mismatch categories (0, 1, 2, ≥3); panel reactive antibody (PRA) category (0–10, 10–100%); cold ischemic time (hours); use or nonuse and type of induction therapy (Il-2 receptor blockers, thymoglobulin, other); other agents in the primary immunosuppressive regimen (antimetabolites, steroids); delayed graft function (need for dialysis during the first week posttransplantation); acute rejection during the index transplant hospitalization; transplant center volume and year of transplantation. In addition, we examined recipients’ history of the following comorbid conditions ascertained prior to transplantation from the USRDS and UNOS databases: diabetes, hypertension, cardiovascular disease, cancer, current tobacco use, body mass index and functional status.
We computed unadjusted incidence rates of outcomes within the exposure groups using Poisson distribution. We examined univariate relationships between exposure groups and outcomes using Kaplan–Meier estimates with 95% confidence intervals (CI). Because examination of scaled Schoenfeld residuals (1994) from univariate Cox models of the primary exposure revealed that the proportionality assumption was violated for each outcome, multivariable adjusted analyses required that we use a time-segmented model. During years 0–2 of follow-up, we compared outcomes among exposure groups using a semivarying coefficient Cox model, which accommodated the nonproportional hazards in the early posttransplant period and censored patients who had not experienced death or allograft loss at the end of year 2. Since the proportionality assumption was met thereafter, we used a constant effect Cox model to analyze risks of outcomes during years 2–8 among living patients with functioning allografts at year 2. We used Cox proportional hazards models to hierarchically adjust for potential confounders in years 2–8. We coded missing covariate data by generating additional “missing” categories.
To test the robustness of the results in years 2–8, we conducted multivariable-adjusted stratified and restricted analyses across a set of past medical, immunological and clinical transplant characteristics. Because delayed graft function and acute rejection during the initial hospitalization predict inferior long-term transplant outcomes (1992) and may influence the choice of primary immunosuppression regimen at discharge, we performed analyses stratified by these covariates. Furthermore, since episodes of acute rejection during the first year may impact subsequent allograft and patient survival, we adjusted for this potential mediator variable in supplementary analyses (1993). To account for differences in utilization of immunosuppressive regimens over time and across centers, we adjusted for transplant center volume and year of transplantation and performed analyses stratified by these covariates. Additionally, we adjusted for and performed analyses clustered by transplant centers’ proclivity for mTORI use, defined as above or below the median frequency of mTORI prescription relative to the individual centers’ total transplant volumes. In sensitivity analyses, we restricted the study population to transplants performed prior to 2008 because USRDS comorbidity data were not available thereafter, and since the most recent transplant outcomes may not have been fully ascertained by UNOS at the time the dataset was locked for this analysis.
To further address potential confounding by indication, we derived a propensity score that summarized the predicted probability of being prescribed mTORIs without CNIs using a logistic regression model that included the following covariates: recipient demographics; etiology of kidney failure; dialysis duration; recipient comorbidities; donor risk factors; human leukocyte antigen mismatch level; plasma-reactive antibodies category; cold ischemic time; induction therapy; delayed graft function; acute rejection during the index transplantation hospitalization; transplant center volume and year of transplantation. We matched 1 mTORI-treated patient to 10 non-mTORI-treated patients by their propensity score ± 0.0024 (0.1 × standard deviation of the propensity score) to generate a subcohort of 2939 mTORI-treated and 29 390 mTORI-untreated patients. Any mTORI-treated patient for whom fewer than 10 matches could be found was dropped from the subcohort, thereby excluding patients at the extremes of the propensity score distribution (2010). To account for the matched nature of the data and the clustering as a result of comparing 1 mTORI-treated patient to 10 CNI-treated patients in each matched stratum, we used the extended McNemar test (2003) to compare categorical variables, and the Rosner's Wilcoxon signed rank test (2006) to compare continuous variables between matched groups. Similar to the main analysis, we used the Kaplan–Meier method to plot time-to-events, performed time-segmented Cox regression to analyze data during years 0–2, and Cox proportional hazards to summarize hazards during years 2–8.
Analyses were performed with SAS 9.2 (SAS Institute, Cary, NC, USA) and R 2.14.2. P values < 0.05 were considered statistically significant.
Baseline characteristics according to the exposure groups are shown in Table 1. A CNI-based immunosuppressive regimen without mTORI was most common (90.1%), followed by combination therapy with CNI + mTOR (7.6%), and then mTORI without CNI (2.3%). Patients in the mTORI without CNI group were more likely than those in the other groups to carry a diagnosis of diabetes or cardiovascular disease, to have received an expanded criteria donor organ, an organ with longer cold ischemic time and to have higher PRA levels. The prevalence of delayed graft function and acute rejection during the index transplant hospitalization was significantly higher in users of mTORI without CNI compared to the other groups (p < 0.001). Within the first year posttransplant, 14.2% of the patients in the mTORI without CNI group and 10.6% in the dual therapy group were treated for acute rejection compared to 9.6% in the CNI group (p < 0.001).
Table 1. Patient characteristics according to treatment status in the overall study population
CNI without mTORI (N = 125 623)
mTORI without CNI (N = 3237)
CNI + mTORI (N = 10 510)
CNI = calcineurin inhibitors; mTORI = inhibitors of the mammalian target of rapamycin.
Values are % and medians (interquartile range).
College education, %
Etiology of kidney failure
Polycystic kidney disease, %
Unknown or missing, (%)
Current smoking, %
Body mass index >30 kg/m2, %
Cardiovascular disease, %
Dialysis duration, years
Expanded criteria donor, %
Human leukocyte antigen mismatches, %
Most recent panel-reactive antibodies >10%, %
Cold ischemia time, hours
Delayed graft function, %
Acute rejection during initial hospitalization, %
Transplant center volume, no. of transplants since 1999, %
Other maintenance immunosuppressive drugs
Mycophenolate mofetil (%)
Il-2 Receptor blockers (%)
There were 16 273 episodes of death-censored allograft failure (31.6 per 1000 person-years; 95% CI, 31.1–32.1), 13 017 deaths (25.3/1000 person-years; 95% CI 24.9–25.7) and 28 687 composite events (55.7/1000 person-years; 95% CI 55.1, 56.4) in the overall cohort. Based on their nonoverlapping 95% CIs, unadjusted rates of death-censored allograft failure, death and their composite were significantly higher in the mTORI without CNI group compared with the CNI and the CNI + mTORI groups (Figure 1A–C).
Since the proportional hazards assumption was violated during years 0–2, we examined rates of outcomes during the first 2 years separately from years 2–8. Unadjusted hazards ratios (HR) of death-censored allograft failure, death and their composite were significantly higher in the mTORI without CNI group versus the CNI group throughout years 0–2, but were highest in the early posttransplant period and gradually diminished to a nadir by year 2 (Figure 1D–F). For death-censored allograft failure, the HR ranged from 4.30 (95% CI 3.49–5.29) after discharge from the index transplant hospitalization to 1.35 (95% CI 1.17–1.56) by year 2; for death, the HR ranged from 2.71 (95% CI 2.17–3.39) to 1.52 (95% CI 1.29–1.78); and for the composite outcome, the HR ranged from 3.67 (95% CI, 3.12–4.32) to 1.40 (95% CI 1.26–1.57).
During years 2–8, when the proportionality assumption was met, crude and minimally adjusted risks of death-censored allograft failure, death and their composite remained significantly higher among the mTORI without CNI group compared with the CNI group (Table 2). However, adjustment for other immunosuppressive medications and induction agents attenuated the results for death-censored allograft failure. In contrast, the risks for allograft failure-censored mortality and the composite outcome were unchanged. In the fully adjusted model, the mTORI without CNI group demonstrated a 1.25-fold greater risk of death and a 1.17-fold greater risk of the composite outcome compared with the CNI group. Sensitivity analyses that restricted the study population to transplants performed prior to 2008 yielded qualitatively similar findings (Table S1). Adjustment for acute rejection within the first year of transplantation, a potential mediating factor for allograft loss, further attenuated the effect of mTORI on death-censored allograft failure during years 2–8 but did not alter the risk of mortality (Table S2). Fully adjusted analyses also confirmed inferior patient survival in the mTORI without CNI group compared with the CNI group during years 0–2 (data not shown).
Table 2. Risks of death-censored allograft failure, death and their composite according to treatment status in the overall study population
CI = confidence intervals; CNI = calcineurin inhibitors; mTORI = inhibitors of the mammalian target of rapamycin.
Model 1 adjusts for recipient demographics, including age, sex, race, ethnicity and education.
Model 2 adjusts for covariates in Model 1 plus etiology of kidney failure, dialysis duration, recipient comorbidities, including hypertension, diabetes, cardiovascular disease, cancer, body mass index >30, current smoking and poor functional status.
Model 3 adjusts for covariates in Model 2 plus the following donor characteristics: donor type (living, deceased, expanded criteria donor), age, race, ethnicity and gender, donor body mass index >30.
Model 4 adjusts for covariates in Model 3 plus the following characteristics: human leukocyte antigen mismatch level, plasma-reactive antibodies category, cold ischemic time, induction therapy and type, antimetabolites, steroids, delayed graft function, acute rejection during the index transplantation hospitalization, transplant center volume and year of transplantation.
Allograft failure, death-censored
CNI without mTORI
33.7 (32.9, 34.4)
mTORI without CNI
39.4 (35.2, 44.0)
CNI + mTORI
39.7 (37.5, 42.1)
Death, allograft failure-censored
CNI without mTORI
26.6 (25.9, 27.2)
mTORI without CNI
36.2 (32.3, 40.6)
CNI + mTORI
26.7 (24.9, 28.6)
Composite of allograft failure or death
CNI without mTORI
59.1 (58.2, 60.1)
mTORI without CNI
74.4 (68.7, 80.7)
CNI + mTORI
65.3 (62.4, 68.3)
Stratified and restricted analyses of death during years 2–8
Clinicians’ selection of the primary maintenance immunosuppressive regimen may be driven by donor characteristics, recipient characteristics and events that occur during the index transplant hospitalization. Multivariable-adjusted analyses of death stratified by recipients’ history of diabetes, prior cardiovascular disease and etiology of kidney failure, donor type, PRA and cold ischemic time categories, year of transplantation and transplant center volume yielded results that were qualitatively similar to the primary analyses in the overall population (Table 3). Similarly, results of multivariable-adjusted analyses of death restricted to individuals without a history of cancer were also unchanged (Table 3). Finally, adjustment for and clustering by the transplant center's proclivity to use mTORI's did not change the results (Table S1).
Table 3. Stratified analyses of death during years 2–8
Estimates are for the effect of mTORI group compared to the CNI group.
With the exception of the variable used for stratification, models were adjusted for recipient demographics (age, sex, race, ethnicity, education); etiology of kidney failure; dialysis duration; recipient comorbidities (hypertension, diabetes, cardiovascular disease, cancer, body mass index >30, current smoking, poor functional status); donor risk factors (living, deceased, expanded criteria donor, age, race, ethnicity, gender, donor body mass index > 30); human leukocyte antigen mismatch level; plasma-reactive antibodies category; cold ischemic time; induction therapy and type; antimetabolites; steroids; delayed graft function; acute rejection during the index transplantation hospitalization; transplant center volume; and year of transplantation.
Etiology of kidney failure
Polycystic kidney disease
Panel-reactive antibodies, %
Cold ischemic time, hours
Transplant center volume
Delayed graft function
Acute rejection prior to discharge
Delayed graft function and acute rejection during the index transplant hospitalization lead to impaired allograft function that may deter clinicians from using CNIs in favor of mTORIs. Among individuals who did not develop these complications, risk of death in the mTORI without CNI group remained significantly elevated compared with the CNI group (Table 3). Finally, acute rejection during the first year following transplantation is a risk factor for allograft failure and mortality that occurred significantly more frequently in the mTORI without CNI group. When we repeated the multivariable-adjusted analysis of death during years 2–8 restricted to those who never developed acute rejection during the first year, the mTORI without CNI group experienced similarly greater risk of mortality compared with the CNI group (HR 1.24; 95% CI, 1.07–1.43) as in the overall cohort.
Propensity score-matched analyses
Propensity score matching successfully balanced the distributions of patient characteristics across users versus nonusers of mTORIs (Table S3). The propensity score-matched analyses recapitulated the findings from the overall cohort throughout the study period (Figure 2A–C). During years 0–2, the HR for death-censored allograft failure ranged from 3.10 (95% CI 2.44–3.94) after discharge from the index transplant hospitalization to 1.21 (95% CI 1.02–1.42) by year 2; for death, the HR ranged from 2.33 (95% CI 1.75–3.10) to 1.29 (95% CI 1.08–1.55); and for the composite outcome, the HR ranged from 2.79 (95% confidence interval [CI], 2.31–3.36) to 1.24 (95% CI 1.09–1.40). During years 2–8, the mTORI without CNI group demonstrated a 1.12-fold greater risk of death-censored allograft failure (95% CI, 0.98–1.28), a 1.21-fold greater risk of death (95% CI, 1.05–1.39) and a 1.18-fold greater risk of the composite outcome (95% CI, 1.08–1.30) compared with the CNI group.
In this large historical cohort study of United States kidney transplant recipients with up to 8 years of longitudinal follow-up, risk of death was increased among patients who received mTORIs without CNIs in their primary immunosuppressive regimen compared with those who received CNI-based regimens. The risk was highest during the first 2 years posttransplantation, but remained elevated through year 8. Intermediate risk was observed in the group that received a combination of mTORIs and CNIs. Whereas the mTORI group was more likely to manifest greater prevalence of certain risk factors for allograft failure and mortality than the CNI group, their higher risk of death extended to strata without these risk factors and was supported by similar results derived from propensity score-matched analyses. While it might be argued that familiarity with the optimal dosing and side effect profile of mTORIs grew over time since their introduction into practice, an important finding is that the mTORI-associated risk of mortality was similar in recent and earlier years of its availability. Although differences in factors that we were unable to measure may have affected the results, these data suggest that use of mTORIs in kidney transplantation is associated with inferior recipient survival.
Early clinical trials reported beneficial effects of mTORIs on serum creatinine relative to CNIs (2002, 2001), but these short-term studies did not evaluate hard clinical end-points. The few that did were insufficiently powered and had limited duration of follow-up (2006). Although a metaanalysis revealed no differences in patient or allograft survival according to use versus nonuse of mTORIs (2006), strict entry criteria of the contributing trials likely resulted in substantial differences between study participants and the general population of unselected kidney transplant recipients that we studied. More recent randomized trials with longer duration of follow-up reported increased risks of adverse outcomes associated with an mTORI-based versus a CNI-based maintenance immunosuppressive regimen in both kidney and liver transplantation (2007, 2011, 2011, 2011, 2011, 2009, 2012). One conversion study of liver transplant recipients was discontinued prematurely and prompted an FDA alert because of increased risk of death in association with conversion to sirolimus (2012, 2012, 2012). Our findings support increased risk of adverse outcomes in association with use of mTORIs in the primary immunosuppressive regimen and contemporize the results of a previous observational study conducted during an earlier era of mTORI use in kidney transplantation (2007).
The results of this study are also consistent with our previous finding of increased mortality in association with mTORI use in a single-center, observational study of 993 Hungarian kidney transplant recipients who were studied beginning 6 years posttransplant (2012). That study of prevalent recipients was limited by survivor bias, in which participants had to live long enough to receive therapy, and by confounding by indication, because patients converted from CNIs to mTORIs due to a failing allograft. Although we acknowledge that mTORI use as de novo therapy in kidney transplantation in the United States is already low, the reason we specifically focused the current study on the primary immunosuppressive regimen was to address these methodological limitations (2003). We used an intention-to-treat design, which yields conservative effect estimates since crossing over in either direction biases results to the null. For example, allograft loss caused by CNI nephrotoxicity that occurred after conversion to an mTORI as a salvage maneuver was properly attributed to the CNI group. Conversely, survivorship among mTORI users who later switched to CNIs following resolution of delayed graft function would have been credited to the mTORI group by our analytical approach. Although we acknowledge that we cannot exclude residual confounding by factors we were unable to capture in the propensity score-matched analyses, our results are generally consistent with recent studies that reported greater tendency of adverse outcomes among mTORI users (2011, 2011, 2009).
Other limitations of this registry study of de novo primary immunosuppression are that we could not study the doses or achieved plasma levels of mTORIs and CNIs, and we had to study mTORI as a class because of the low rate of everolimus use. Thus, our findings may not be generalizable to mTORI use later in the course of transplantation and under conditions of careful monitoring of drug levels, and whether the results apply to everolimus use is not known. Another unique group for whom our findings may not apply is kidney-transplant recipients with previous squamous cell cancers, in whom sirolimus has been shown to have beneficial antitumor effects (2012).
Additionally, we were unable to identify whether the adverse effects on outcomes resulted from toxicity due to mTORI or loss of a protective effect that was incurred by withholding CNI. For example, mTORI can induce proteinuria and raise cholesterol levels (2005, 2008), but data on these factors were not available. It is also unclear whether the observation of the highest risk of adverse outcomes during the first 2 years posttransplant that waned with time is the result of known early toxicities of mTORI, such as delayed wound healing and greater risk of lymphocele (2011), an effect of patients gradually crossing over from mTORI to CNI (2009), or residual confounding by indication. Interestingly, we confirmed the results from prior randomized trials of a significantly higher rate of acute rejection during the first year posttransplant in the mTORI without CNI group (2007, 2011, 2011), and adjustment for acute rejection, which may be on the causal pathway of the association between mTORI exposure and allograft loss, further attenuated the findings for death-censored allograft failure. However, neither adjustment for acute rejection nor inclusion of other immunosuppressive medications altered the findings for risks of death and the composite outcome. Moreover, the risk of mortality was significantly greater among mTORI users who did not experience rejection. These data suggest additional mechanisms beyond higher susceptibility to acute rejection contribute to greater long-term risk of mortality among mTORI users.
Randomized studies are the gold standard for comparing different therapeutic strategies. However, careful selection of a limited number of trial participants can obscure clinically meaningful effects that may emerge later when use of a new agent is broadened to patients who may have been excluded from previous trials (2002, 2003). Furthermore, insufficient duration of follow-up is an especially important limitation of clinical trials in kidney transplantation because complications that occur during the early transplant period may have remote long-term consequences that evade detection in short-term trials. For example, early episodes of acute rejection lead to greater net immunosuppression exposure, which is an independent risk factor for infection, malignancy and cardiovascular disease that may first manifest many years later (2007, 1988, 2011). This emphasizes the need for longer term trials or, in their absence, additional pharmacoepidemiological studies to further evaluate the long-term outcomes associated with mTORI use.
TI was supported by NIH grant K23DK087858. MW was supported by NIH grants R01DK076116 and R01DK081374.
This work was supported in part by Health Resources and Services Administration contract 234-2005-37011C. 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 US Government.
The data were supplied by the USRDS; the conclusions and opinions are those of the authors and do not represent the USRDS or NIH.
The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation.