Date of Award

January 2021

Document Type


Degree Name

Master of Public Health (MPH)


School of Public Health

First Advisor

Shuangge S. Ma


Epidemiologic studies are often subject to the existence of immortal time, a period of follow-up during which outcomes cannot occur because of the observed later treatment initiation. Methods such as including or excluding immortal time from the analysis with an ever-treated and never-treated group definition are known to result in biased estimates of treatment effects. Other approaches that also model time zero, such as prescription time distribution matching (PTDM), and grace period and landmark methods, have been proposed to handle immortal time bias. In addition to modeling time zero, approaches like modeling time-varying treatment status, and duplication and nested trial design, have recently been applied to many observational studies and emulated trials to derive the desired causality. Currently, there is no comprehensive comparison available about the performance of all methods, especially when a causal effect is of interest. In this thesis, we conduct extensive Monte Carlo simulations to assess the performance in settings that reflect real-world scenarios. Also, the advantages, applications, and limitations of each method are illustrated when combined with causal inference methods such as inverse probability weighting and emulation approach. We find that the time-varying treatment modeling and sequential trial approaches provide unbiased estimates, whereas other methods result in substantial bias. We suggest that for the specific setting when a grace period is available, it is possible to emulate a target trial using the grace period to mimic the treatment assignment. Compared to other methods, the sequential trial design and the cloning idea with the adoption of artificial censoring during follow-up provide the potential for deriving a per-protocol effect. We summarize the current analytic approaches and identify the possible directions to avoid immortal time bias that widely exists in comparativeeffectiveness research.


This thesis is restricted to Yale network users only. This thesis is permanently embargoed from public release.