Whilst reviewing an article submitted to a transplant journal recently, my attention was brought to an issue that is often overlooked in survival analysis, called ‘immortal time bias’. The issue arises when the groups to be compared in the analysis are defined by a time-dependent event that can occur any time during follow-up. As standard survival analysis assumes that the risk is present at the start of the observation period, this can lead to biased survival estimates; if the intervention/exposure is beneficial, the survival advantage can be overestimated, and if it is detrimental then the survival disadvantage may be underestimated.
For an excellent description of the problem, and potential solutions, take a look at this excellent article in Transplant International from Gleiss et al. The example given examines the effect of graft survival on post-transplant patient survival, but the problem would also apply to other transplant analyses, such as the effect of acute rejection on graft survival.
A must for anyone undertaking registry analyses!