The Effect of Antilymphocyte Induction Therapy on Renal Allograft Survival

Numerous mechanisms exist though which anti-lymphocyte antibodies may improve allograft survival [1-5]. However, use of these antibodies in the perioperative period of cadaveric kidney transplantation (induction therapy) has been controversial because these antibodies have uncertain efficacy in prolonging allograft survival and have associated side effects. Most cohort studies [6-9] and all randomized trials [10-16] comparing induction therapy with conventional immunosuppression have failed individually to show that induction therapy has benefit with regard to allograft survival. However, our recent meta-analysis pooling the published data from the randomized trials of induction therapy [10-16] in adults receiving cadaveric renal transplants showed a 31% lower rate of allograft failure at 2 years [17] with induction therapy. That meta-analysis did not allow us to examine the effect of induction therapy beyond 2 years or to examine its effect among subgroups with risk factors for early allograft loss, including elevated panel-reactive antibody levels [18, 19], African-American ethnicity of the recipient [18, 20, 21], previous transplantation [18, 22], delayed allograft function [23-27], diabetes mellitus in the recipient [18], greater donor-recipient HLA mismatch [18, 28-31], and prolonged cold ischemic time [18]. In this meta-analysis, we used individual patient-level data from randomized, controlled trials to confirm the beneficial effect of induction therapy on allograft survival at 2 years, to extend these observations to a longer follow-up period, and to look for differential benefit associated with clinical factors (such as delayed allograft function) previously shown to be associated with a higher rate of allograft failure. Methods Study Identification and Selection The selection of studies has been described else-where [17]. Of 247 potentially eligible studies identified from MEDLINE, 7 met the inclusion criterion of reporting allograft survival data from a randomized trial that compared polyclonal or monoclonal antilymphocyte antibodies used in the period immediately after transplantation with conventional immunosuppression in adult recipients of cadaveric renal transplants. No additional published or unpublished trials were identified. Because the data from two trials [10, 12] were not available, our study is based on data from patients in five previously published randomized, controlled trials [11, 13-16]. No industry funds were used directly in support of our study. Two randomized trials included in this meta-analysis were initially supported with industry funds: The study by Norman and colleagues [14] was paid for by the R.W. Johnson Pharmaceutical Research Institute (Raritan, New Jersey), and the study by Abramowicz and coworkers [13] was supported in part by Cilag Benelux, Belgium. No company was involved, directly or indirectly, in the design or conduction of this meta-analysis or in the decision to submit this manuscript for publication. Data Used for Analysis The baseline individual patient-level information included age, sex, ethnicity, height, weight, presence of diabetes mellitus, history of previous transplantation, panel-reactive antibody levels, number of previous blood transfusions, number of donor-recipient HLA mismatches, cold ischemic time, delayed allograft function, and survival time. Survival time was defined as the interval between placement of the allograft and return to dialysis, retransplantation, death, or end of follow-up, whichever occurred first. For our primary analysis, death was considered an allograft failure. To confirm that our analyses were stable, we also considered death a censoring event rather than an allograft failure. These two approaches produced almost identical results; thus, only our primary analysis is presented here. Because we used individual-level data, follow-up was longer than in our previous meta-analysis. Mean follow-up (SD) ranged from 36.7 16 months to 46.1 11 months (maximum, 60 months). Data from each study were extracted by one of the authors and converted into a common format. Categorical variables were created for age (18 to 34, 35 to 49, 50 to 64, and 65 years of age), previous transplants (zero and nonzero), panel-reactive antibody level (0% to 19%, 20% to 79%, and 80% to 100%), and cold ischemic time (<24 hours or 24 hours) by using the strata used by UNOS (United Network of Organ Sharing) [32]. Two of the categories for panel-reactive antibody level (20% to 79% and 80% to 100%) were combined because of small numbers in those categories. Patients with panel-reactive antibody levels of 20% or more were considered to be presensitized; patients with levels less than 20% were considered to be unsensitized. Because it is known that fewer than 2% of the patients enrolled by Abramowicz and coworkers [13] were diabetic or nonwhite (Vereerstraeten P. Personal communication), all of the patients in that study were considered white and nondiabetic. Delayed allograft function was defined as the need for dialysis during the first week after transplantation in all studies except one [11], in which it was defined as a decrease in creatinine level occurring more than 1 day after transplantation. Each database was examined for missing data, and each observation was checked for internal consistency. The distributions of demographic and prognostic variables at baseline (for example, ethnicity, previous transplantation, or panel-reactive antibody level) were described overall and for each treatment group. Treatment groups were compared by using a chi-square test stratified by study to confirm the success of randomization [33]. Inconsistencies between these results and the results in the published reports were resolved with the authors of the studies. Statistical Analysis All analyses were done by using the intention-to-treat approach; that is, patients were analyzed according to the treatment group to which they had been randomly assigned. Possible selection bias resulting from the inclusion of five of the seven eligible randomized trials was evaluated. Using data from the published reports, we calculated the rate ratios of allograft failure by use of a discrete-time version of the proportional hazards model [34] for both the subset of studies for which individual patient-level data were available [11, 13-16] and the subset for which data were unavailable [10, 12]. The rate ratios compared induction therapy with no induction therapy at 2 years of follow-up. Each rate ratio was compared with the other and with the pooled rate ratio from all seven studies [17]. The unadjusted distribution of allograft survival was estimated for treatment groups within each study individually and for the seven studies overall by using the Kaplan-Meier product-limit estimator [35]. Treatments were compared by using the log-rank test [36], stratified by study, at 2 and 5 years. In addition, Cox proportional-hazards regression was used to estimate the overall effect of induction therapy on allograft survival in both unadjusted and adjusted analyses. Proportional hazards regression was also used to determine other factors associated with survival and to explore for interactions between treatment group and other factors [34]. Initially, a model was fit to the individual patient data for 2 years of follow-up to confirm the results of our previous meta-analysis. The model was then fit by using the entire length of follow-up available in the individual-level data (5 years) to explore the long-term effects of induction therapy on allograft survival. Because we were uncertain whether the effect of induction therapy was constant over time, we looked for a time-by-treatment interaction by using a time-varying indicator variable, distinguishing the periods 0 to 2 years and 2 to 5 years [37]. The cut-point of 2 years was chosen to correspond to the availability of 2 years of follow-up in the published data. Finally, a model was fit for the subgroup of patients whose allografts survived for at least 2 years. Several approaches were used to select variables for inclusion in the multivariable regression models. In general, the models were fit to include the maximal number of candidate variables while minimizing missing data. Variables representing treatment and study were forced into all models. Other predictors of survival were first selected by using both forward and backward stepwise elimination methods. Entry and elimination criteria were set at a value of P = 0.15. In addition, variables were allowed to enter models if their inclusion altered the rate ratio for the treatment variable by 15% or more. Standard methods were used to calculate multivariable rate ratios on the basis of model variable co-efficients [34]. Terms representing interactions between the use of induction therapy and selected variables (recipient ethnicity, delayed allograft function, diabetes mellitus, HLA mismatch, presensitization, previous transplantation, cold ischemic time, and study) were included in separate models. These interaction terms tested whether the benefit of induction therapy differed across categories of these variables. Because data on cold ischemic time, recipient age, recipient sex, delayed allograft function, and number of previous blood transfusions were lacking for a substantial number of patients, these variables were individually examined in models adjusted for use of induction therapy, recipient ethnicity, diabetes mellitus in the recipient, number of HLA-DR mismatches, presensitization, previous transplantation, and study. The same models were also used to evaluate the potential confounding effect of cold ischemic time, recipient age, recipient sex, delayed allograft function, and number of previous blood transfusions on the effect of induction therapy. All P values are two-sided. All analyses, with the exception of those that involved time-varying covariates, were done by using Stata, vers