Department of Biostatistics


Prospective Students



News & Events



Employment Opportunities

Resource Quicklinks

Computing Environment




Causal Inference with Sensitivity Analysis: Methods for Investigating Mediation and Accounting for Death in Observational Studies

 Brian Egleston, PhD Candidate, Johns Hopkins Department of Biostatistics

This dissertation presents causal methods for investigating exposure effects on non-mortality outcomes when death is a competing risk and investigations of mediation. The first two sections discuss identification of the effect of vision loss on emotional distress among the group that would live regardless of vision loss status. The difficulty in identifying this effect is that we do not know the mortality outcomes that participants would have had under a vision loss status different to that observed. A sensitivity analysis methodology is proposed in which effects are identified over a scientifically plausible range of assumptions. A further difficulty in identifying the effect is that important baseline information is often only available for those who survive long enough to be interviewed. This hinders implementation of the causal analysis which requires baseline covariates for the full sample. To account for the potential non-random missingness of the covariates, another sensitivity analysis is proposed in which effects are identified over a range of scientifically plausible assumptions about the effects of the missing covariates on the log odds of death. The final section of this thesis investigates the degree to which a reduction in ocular sunlight UV exposure mediates a relationship between wearing eyeglasses and a decreased risk of cataracts. An estimand is proposed in which causal effects are estimated locally within strata based on potential UV exposure without glasses and the degree to which glasses use reduces UV exposure. We take advantage of the structure of the data in which the counterfactual UV exposures if the participants in the study who wore glasses had not worn glasses are considered observable.

Return to
Upcoming Events List | Return to Home Page
© 2004, The Johns Hopkins University. All rights reserved.
web policies, 615 N. Wolfe Street, Batimore, MD 21205-2179, 410-955-5000