A Meta-Analysis of Immunosuppression Withdrawal Trials in Renal Transplantation

Search Strategy

We searched (1) Medline, (2) bibliographies from recent, pertinent publications, and (3) abstracts from scientific meetings conducted in 1998 and 1999. Key words in our Medline search included “prednisone” or “corticosteroids” combined with “withdrawal” or “discontinue” or “conversion” as well as “cyclosporine” or “cyclosporine A” combined with “withdrawal” or “discontinue” or “conversion.” We searched abstracts from international transplantation society meetings published in 1998 or 1999 in either Transplantation Proceedings or Transplantation. Included in these were abstracts from meetings of the International Transplantation Society, the American Society of Transplant Surgeons, and the American Society of Transplantation.

Study Selection

We included only published, randomized, controlled trials in which either prednisone or CsA was withdrawn or randomized trials comparing prednisone withdrawal with CsA withdrawal. We included only trials that reported the incidence of acute rejection and/or renal allograft failure. All end points were analyzed by intent to treat. We did not exclude reports on the basis of language, but all studies were published in English. In several instances, results from some or all patients in a clinical trial were reported more than once. In these instances, we extracted data on end points from the publication with the longest duration of follow-up. We did not include studies of prednisone or CsA avoidance, i.e., studies in which patients in one arm never received prednisone or CsA. Studies that compared prednisone withdrawal or CsA withdrawal with the withdrawal of other immunosuppression, e.g., azathioprine, also were excluded.

Data Extraction

Two investigators (B.L.K. and H.A.C.) independently reviewed all studies. We extracted data on study quality, including whether (1) the study was published in a peer-reviewed journal, (2) it had been stated that institutional review board approval had been obtained, (3) the source of funding was indicated, (4) the study design included a statistical assessment of the sample size needed, (5) inclusion and exclusion criteria were clearly stated, (6) possible differences in patients allocated to treatment and control were examined, (7) a placebo was used to mask investigators and subjects, (8) the allocation process was truly random (compared with a pseudo-random allocation whereby, for example, every other patient was allocated to treatment or control), (9) the randomization technique was clearly described (envelopes, central randomization, etc.), (10) the statistical methods used to analyze the results were described, (11) patient withdrawals were described, (12) results were analyzed by intention to treat, and (13) the study end points were clearly defined. We derived an arbitrary quality index using a formula whereby each of the above quality indices were given one point if the criterion was completely fulfilled, one half of a point if the criterion seemed to be only partially fulfilled, or (in the case of publication in a peer-reviewed journal) two points. The composite study quality index was the total number of points (maximum = 14).

We also extracted data on the proportion of study patients who were first transplants, the proportion who were recipients of cadaveric donor kidneys, whether patient selection was based on the absence of acute rejection and/or stable graft function, and what other immunosuppressive agents were used, e.g., azathioprine and mycophenolate mofetil. In addition, we tabulated the time withdrawal was begun, the time it was completed, and the duration of follow-up (from the time of transplantation to the time of last follow-up or loss of the allograft).

Statistical Techniques

For acute rejection, the duration of follow-up was often relatively short, because most acute rejection episodes occurred soon after withdrawal. Indeed, some studies failed to report the exact duration of follow-up. Therefore, we examined the difference in the proportion of withdrawal and control patients who had acute rejection following randomization: D_i = P_ti - P_ci, where P_ti = r_ti/n_ti, P_ci = r_ci/n_ci, r_ti and r_ci are the number of patients with an acute rejection episode, and n_ti and n_ci are the numbers of patients allocated to the treatment and control groups respectively for the ith study. The variance of D_i was calculated as

Weighted means and confidence intervals (CI) were then calculated for the combined differences between treatment and control groups using a fixed-effects model. The weighted mean treatment effect was calculated as where D_i is the difference between treatment and control for the ith study and w_i = 1/Var(D_i). The 95% CI of this weighted mean treatment effect was calculated as where t_df is the 97.5th percentile of the t distribution for df = 100. We used 100 for estimating the degrees of freedom, i.e., an order of magnitude larger than the number of studies and an order of magnitude smaller than the total number of patients in the combined studies so that t = 1.984.

We looked for homogeneity of treatment effects with the test statistic: where Q is approximately a X² statistic with k-1 degrees of freedom, where k was the number of studies. Q was used to test whether the variance of the treatment effect from the population mean of treatment effects was significantly different from 0 (the null hypothesis).

We also examined the relative risk (RR) of graft failure (defined as either return to dialysis or death with a functioning allograft) between immunosuppression withdrawal and nonwithdrawal control groups. Some studies reported results over a specific time interval, e.g., the proportions of patients with graft survival in withdrawal and control groups after an exact (or mean) period of follow-up. In other words, these studies reported two proportions—the proportion of patients who survived in the withdrawal group and the proportion of patients who survived in the control group—and reported the exact (or mean) duration of follow-up. For these studies, we estimated the RR of graft survival between withdrawal and control as where S_ti and S_ci are the survival probabilities at time x for the withdrawal and control groups, respectively, for the ith study. Variance of R_i was calculated as Var(R_i) = 1/d_1I + 1/d_2i, where

For other studies, data were presented in more detail using tables and/or figures that indicated the proportions of the withdrawal and control groups that survived after two or more discrete time intervals. For these studies, we estimated the RR of graft failure as where d_tji and d_cji were the numbers of grafts lost in the withdrawal and control groups, respectively, in the jth follow-up interval of the ith study. Weighted means, 95% CI, and a χ² test for homogeneity were calculated as described above.

We also examined potential reasons for heterogeneity in study results using linear regression analysis weighted by inverse variance. In this analysis, differences in the proportions of patients with acute rejection and differences in the RR for graft failure were used as dependent variables. Study quality and other study characteristics were used as independent variables. All differences were considered significant for P < 0.05 (two-tail). Analysis was carried out using the Statistical Package for the Social Sciences.

Prednisone Withdrawal

We included 10 studies that examined prednisone withdrawal (Table 1) (2,3,5,6,7,8,9,10,11,12,13,14,15,16,17,18). Seven had been published in a peer-reviewed journal. Nine studies that examined the effects of prednisone withdrawal on acute rejection included 1461 patients (mean ± SD, 162 ± 147) with 29 ± 19 mo of follow-up. Nine studies that examined the effects of prednisone withdrawal on graft survival included 1899 patients (mean ± SD, 211 ± 186) with 28 ± 19 mo of follow-up.

For acute rejection, the pooled mean difference between treatment and control was 0.14 (95% CI = 0.10 to 0.17, P < 0.001), indicating that the proportion of patients with acute rejection after prednisone withdrawal was significantly greater compared with controls (Figure 1). The test for homogeneity (χ² = 13.7, P > 0.05) indicated that the individual studies were relatively homogeneous. The RR for graft failure was 1.40 (1.09 to 1.71, P = 0.012), indicating that significantly more patients lost their grafts in the prednisone withdrawal group compared with controls (Figure 2). The test for homogeneity (χ² = 12.0, P > 0.05) indicated that the individual study results were relatively homogeneous. Although the results were relatively homogeneous, we nevertheless sought possible reasons for differences in results using regression analysis. There were no statistically significant correlations between the differences in acute rejection and any of the study or patient characteristics. None of the patient or study characteristics correlated with the effect of prednisone withdrawal on graft survival.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Effects of prednisone withdrawal on acute rejection. Shown are weighted differences in the proportion of patients with acute rejection between withdrawal and control for individual studies (♦) and the pooled results of all studies (⋄). The area to the right of the vertical dashed line indicates a greater proportion of patients with acute rejection after withdrawal compared with control. Horizontal lines indicate 95% confidence intervals (CI), and failure of 95% CI to cross the vertical dashed line indicates P < 0.05. The number of patients uniquely identifies each study (see Table 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Effects of prednisone withdrawal on graft failure. Shown are weighted relative risks (RR) of graft failure after withdrawal versus control for individual studies (♦) and the pooled results of all studies (⋄). The area to the right of the vertical dashed line indicates a greater RR of graft failure after withdrawal versus control. Horizontal lines indicate 95% CI, and failure of 95% CI to cross the vertical dashed line indicates P < 0.05. The number of patients uniquely identifies each study (see Table 1).

CsA Withdrawal

Thirteen studies examined CsA withdrawal (Table 2) (5,6,7,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36). Seven had been published in peer-reviewed journals, and six had not. The 10 studies that examined the effects of CsA withdrawal on acute rejection included 1049 (mean, 105 ± 83) patients with 50.7 ± 33.9 mo of follow-up. The 12 studies that provided adequate data on graft survival after CsA withdrawal included 1151 (mean, 96 ± 80) patients with 45.1 ± 32.9 mo of follow-up.

For acute rejection, the pooled mean difference between CsA withdrawal and control was 0.11 (0.07 to 0.15, P < 0.001), indicating that the proportion of patients with acute rejection after CsA withdrawal was 11% higher than that of controls (Figure 3). The test for homogeneity (χ² = 64.9, P < 0.001) indicated that the individual study results were heterogeneous. To explore possible reasons for this heterogeneity in results, weighted, univariate regression analysis was carried out to determine which, if any, patient or study characteristics correlated with the effect of CsA withdrawal on acute rejection. The study quality index (r = -0.698, P = 0.025), whether the study was published in a peer-reviewed journal (r = -0.696, P = 0.025), and the number of patients studied (r = -0.823, P = 0.003) each correlated inversely with the difference in the rate of acute rejection. However, these characteristics were significantly correlated with each other. It is difficult to tell whether study quality and sample size were independent correlates to the study results. However, in a multivariate analysis, sample size was a better predictor and study quality did not correlate with differences in acute rejection once sample size was taken into account (data not shown).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Effects of cyclosporine withdrawal on acute rejection. Shown are weighted differences in acute rejection between withdrawal and control for individual studies (♦) and the pooled results of all studies (⋄). The area to the right of the vertical dashed line indicates a greater proportion of patients with acute rejection after withdrawal compared with controls. Horizontal lines indicate 95% CI, and failure of 95% CI to cross the vertical dashed line indicates P < 0.05. The number of patients uniquely identifies each study (see Table 2), except for the two studies with n = 64, which are in the same top-to-bottom order as in Table 2.

The RR for graft failure was 1.06 (0.82 to 1.29, P = 0.646), indicating no difference in the risk for graft failure in the CsA withdrawal group compared with controls (Figure 4). However, the test for homogeneity (χ² = 43.0, P < 0.001) again indicated that the individual study results were heterogeneous. In an unweighted regression analysis, the study quality index was inversely proportional to the RR for graft failure (r = -0.645, P = 0.024), suggesting that in studies of lower quality, CsA withdrawal had a more adverse effect on graft survival. Whether studies were published in peer-reviewed journals also correlated inversely with the RR of graft failure from CsA withdrawal (r = -0.721, P = 0.008) in unweighted regression analysis. However, when the regression analysis was weighted by inverse variance, neither study quality (r = -0.367, P = 0.241) nor peer review (r = -0.416, P = 0.178) significantly influenced the results of studies that examined the effects of CsA withdrawal on graft survival. Thus, it seems that the influence of study quality and peer review on the results was largely abolished when the variability in results was accounted for by weighting with inverse variance.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Effects of cyclosporine withdrawal on graft failure. Shown are weighted RR of graft failure after withdrawal versus control for individual studies (♦) and the pooled results of all studies (⋄). The area to the right of the vertical dashed line indicates a greater RR of graft failure after withdrawal versus control. Horizontal lines indicate 95% CI, and failure of 95% CI to cross the vertical dashed line indicates P < 0.05. The number of patients uniquely identifies each study (see Table 2), except for the two studies with n = 64, which are in the same top-to-bottom order as in Table 2.

Studies that were or were not peer-reviewed were combined separately (Figure 5). The pooled RR for graft failure among five studies (n = 189 patients) that were not published in peer-reviewed journals was 2.07 (1.30 to 2.83, P = 0.006; χ² test for homogeneity = 39.4, P < 0.001), indicating that these study results were heterogeneous. In contrast, the pooled RR for graft failure among the seven studies (n = 962 patients) published in peer-reviewed journals was only 0.95 (0.70 to 1.20, P = 0.682). The χ² test for homogeneity = 3.5 (P > 0.05), indicating that the results in these seven studies were more homogeneous. Thus, the heterogeneity in results seemed to be found in studies that were not published in peer-reviewed journals, which also correlated with a low study quality index.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Study quality and the effects of cyclosporine withdrawal on graft failure. Shown are weighted RR of graft failure after withdrawal versus control for individual studies that were (♦) or were not (⋄) published in a peer- reviewed journal. The area to the right of the vertical dashed line indicates a greater RR of graft failure after withdrawal versus control. Horizontal lines indicate 95% CI, and failure of 95% CI to cross the vertical dashed line indicates P < 0.05. The number of patients uniquely identifies each study (see Table 2), except for the two studies with n = 64, which are in the same top-to-bottom order as in Table 2.

We also compared the pooled results of studies with less than 48 or more than 48 mo of follow-up. The pooled RR of graft failure for the six studies (n = 519 patients) with less than 48 mo (17 ± 7 mo) of follow-up was 1.43 (0.97 to 1.89, P = 0.066). The χ² was 34.9 (P < 0.001), indicating significant heterogeneity. The pooled RR of graft failure for the six studies (n = 632 patients) with at least 48 mo (70 ± 20 mo) of follow-up was 0.92 (0.64 to 1.20, P = 0.569). The χ² was 5.1 (P > 0.05), indicating that these studies with longer follow-up were relatively homogeneous. Thus, the studies of short duration seemed to contribute to the heterogeneity of results and were more likely to report that CsA withdrawal negatively influenced graft survival compared with studies of longer duration.

CsA Versus Prednisone Withdrawal

Three studies compared CsA withdrawal with prednisone withdrawal, and two of these were published in peer-reviewed journals (Table 3) (5,6,7,37,38). The three studies included 259 (86 ± 35) patients who were followed for 43.9 ± 6.0 mo. Each examined both acute rejection and graft survival. There was no difference in the proportion of patients with acute rejection between CsA and prednisone withdrawal; pooled difference = 0.04 (-0.07 to 0.144, P = 0.516). The χ² was 6.2 (P > 0.05), indicating that the results were homogeneous. There was a nonsignificant trend for graft failure to be less for CsA withdrawal compared with prednisone withdrawal (0.63 [range, 0.08 to 1.18], P = 0.190; χ² = 3.5, P > 0.05).