CMStatistics 2018: Start Registration
View Submission - CMStatistics
B1273
Title: On bias reduction when estimating the causal effect of pre-emptive kidney transplantation Authors:  Els Goetghebeur - Ghent University (Belgium) [presenting]
Abstract: While causal inference made great progress over past decades, much of the applied literature appears oblivious. We evaluate the effect of Pre-emptive (immediate) Kidney Transplantation (EKT) versus starting on dialysis in patients with end-stage kidney disease (EKD). We show how in disease registers which 1) follow patients from EKD onwards and 2) measure sufficient confounders at the start of EKD, one can not only estimate the total effect of PKD on survival time among the (un)treated but also the survival time lost for each day spent on initial dialysis (relative to starting with transplantation). This is possible through accelerated failure time models even when the switch to delayed transplantation depends on unobserved time-varying covariates. These same methods applied to data from transplant registers however do reap biased estimators due to the truncated nature of data entering conditional on survival up to (delayed) transplantation. We identify various sources of bias in this case, and discuss assumptions and likelihood based methods under which this bias can be resolved. The methods are applied to the Swedish kidney register and supported by simulations.