Title: How serious is bias in effect estimation in randomised trials with survival data given risk heterogeneity and informative censoring?
Abstract: It is often assumed that randomisation will prevent bias in estimation of treatment effects from Randomised Controlled Trial but this is not true when using models such as the semiparametric Proportional Hazards model for survival data, when there is underlying risk heterogeneity. Also, informative censoring of survival data is recognised as another source of bias. Here, in a novel approach, informative censoring and risk heterogeneity are viewed in a common framework as phenomena giving rise to selection effects and thus bias. New formulae for estimation of the bias attributable to heterogeneity alone and for the bias when there is also informative censoring are derived; it is shown that the combined bias can sometimes be less than under simple censoring since the selection effects can act in opposite directions.
When tested against simulated data, the new formulae are shown to be very accurate for prediction of bias in Proportional Hazards and Accelerated Failure Time analyses. These data are also used to investigate the performance of Accelerated Failure Time models which explicitly model risk heterogeneity (‘frailty models’) and G estimation, which are shown to remove the bias when there is simple censoring but not with informative censoring.
The talk will also consider whether, in reality, there is heterogeneity and whether the biases are likely to be serious or trivial.