Since the primary endpoint was known 1 week after randomization, and there was an ethical requirement to minimize the number of patients in the study, a sequential design was chosen. Patients were recruited until there was sufficient evidence, based on the primary end point, to reject the null hypothesis that there was no difference in preservation treatments (12). Two previous studies of MP in DCD donor kidney transplants suggested a reduction in DGF from 78% and 85% down to 57% and 62%, respectively, although other publications have suggested a rate of 20% is possible with machine perfusion (13,14). We therefore used a triangular test with boundaries set to permit a 90% chance of detecting a change in the incidence of DGF from 80% on CS to 60% on MP, using a 5% significance level.
The design properties indicated an expected sample size of 142, with a 10% chance that more than 220 patients would be needed, if there were no treatment difference. This compares to a sample size of 209 patients for the equivalent fixed sample size study. The initial analysis took place after the recruitment of 60 patients, which we considered the minimum number of participants for the trial to have credence; further analyses took place after every subsequent 20 patients had reached the primary endpoint, since this interval was considered a satisfactory compromise between realizing the benefits of a sequential design and the logistical difficulties in monitoring the trial. The ‘Christmas tree correction’ (12) was used to adjust for the discrete monitoring, and an intention to treat analysis was used throughout. The data were analyzed using the PEST4 software program (Planning and Evaluation of Sequential Trials, University of Reading, Reading, UK).
Because of the paired nature of the design, most of the comparisons between treatments at 7 days and 3 months after transplant were undertaken using McNemar's test for categorical variables and the paired t-test for continuous variables. Comparisons based on subsets of recipients, namely time to last dialysis and creatinine reduction ratios, respectively used multiple linear regression and logistic regression adjusted for the height, weight, gender, age, ethnicity and center of the donor. At 12 months after transplant data on a small number of recipients were incomplete reducing the number of complete paired data, and so unpaired analyses were used. Survival times were compared using the log rank test. Univariate analyses were supplemented by risk-adjusted analyses based on multiple linear regression, logistic regression and Cox regression, as appropriate. Recipient age, HLA mismatch level, incidence of DGF, donor weight, height, gender and age were all adjusted for when investigating differences in the incidence of acute rejection. In addition to the above factors, recipient weight, height, gender, ethnicity, duration of dialysis pretransplant, type of dialysis, graft number, sensitization status, serum creatinine pretransplant, serum urea, serum albumin, serum calcium, use of a dopamine infusion, angiotensin-converting enzyme inhibitor, angiotensin II receptor blocker, calcium channel blocker as well as transplant center were all adjusted for when modeling differences in time to last dialysis, serum creatinine and eGFR. Adjusted p-values were broadly similar to those from the unadjusted analyses.