Background: In conventional IPD-MA of time-to-event data, events of the interest which are yet to be observed are "censored". Censoring is typically assumed to be "non-informative" i.e. the probability of observing the subsequent event of interest is not affected by any characteristics of the study patients. However, an alternative view is that if the patient experienced a different event of interest (e.g. death) he/she cannot be at the risk of developing an outcome that we may be interested in. Therefore, these competing risks (CR) should be taken into account in the analysis of survival data. However, no published method exists on how to conduct IPD MA in the presence of CRs.

Objectives: to develop a method to perform IPD MA in the presence of CRs.

Methods: We performed IPD MA of 9 trials (n=1,111 patients) that compared allogeneic peripheral stem cell transplant (PBSCT) with bone marrow transplant (BMT) in treatment of hematologic malignancies. The event of interest was extensive stage of chronic graft vs. host disease (cGVHD). The analysis was performed with and without taking two CR (death, relapse) into account. Each trial was analyzed separately and data were pooled between trials using an inverse variance method. Standard error in the presence of CR was estimated using Taylor series linear approximation approach.

Results: Probability of developing cGVHD in patients treated with PBSCT vs. BMT without taking CR into account was 57% vs. 37% at 5 years, respectively. When analyzed taking CR into account, cumulative incidence of cGVHD in patients treated with PBSCT was 33% vs. 20% treated with BMT.

Conclusions: IPD MA of time-to-event data may produce dramatic differences in the results depending on whether CRs were taken into account. The users of such evidence should be aware of this. We provide for the first time a method for performing IPD MA in the presence of CRs along with user-friendly program in STATA statistical package.