Comparing Unirradiated vs Irradiated White Blood Cell Transfusions

Appendix

In this study we conducted a 2-arm Bayesian adaptively randomized study of irradiated vs unirradiated leukocyte transfusions for patients with life-threatening infections over 50 years of age and patients 50 years of age and younger who had failed the standard treatment four times. The primary objective was to decrease the mortality rate within 30 days by 20% when compared with a radiated Leukocyte transfusion.

The study included patients who have a life-threatening infection and belong to one of the following two groups 1) over 50 years of age and 2) and patients 50 years of age and younger who had failed the standard treatment four times. Since the benefit of unirradiated treatment may only be realized in one age subgroup, we included a treatment-subgroup interaction term in the statistical model specification to allow us to reach different conclusions for each age subgroup. The study was designed so that if there was sufficient evidence that one treatment produced a lower mortality rate then the other treatment within a subgroup the study would be halted for that subgroup.

The first 40 patients were randomized with equal probability. Beginning with the 41st patient the randomization probabilities were based on the probability that a given treatment has the lowest mortality rate within each subgroup. However, the randomization probabilities were constrained to a minimum of 20% and a maximum 0f 80% so that we could continue to gain valuable information about both treatments. Since there was interest in identifying which treatment had the lowest mortality rate as soon as possible, the design included both a futility and superiority monitoring rule and the trial could be stopped for futility or superiority within an age subgroup.

As a secondary outcome, we monitored long-term progression-free survival (PFS). Since we did not want to reduce the long-term progression-free survival rate, beginning with the 50th patient and every 25 patients thereafter we compared progression-free survival of the irradiated and unirradiated treatments within each subgroup. If at any of these interim analyses there was sufficient evidence, greater than 99% probability, that the unirradiated treatment reduces PFS the trial would be stopped for safety reasons.

Due to the complexity of the design, we conducted a simulation study to obtain the usual operating characteristics such as false-positive and power.