Estimating causal treatment effects for post-randomization marker data with failure event censoring