Dissertation Proposal
October 18 (Fri) 11 a.m.
Alavi Commons Room, 6625 Everett Tower
Comparison of Hazard, Risk, and Odds Ratios as Effect Size Estimates in the Survival Data Two-Sample Problem
Benedict Dormitorio
Department of Statistics
Western Michigan University
A common goal in survival analysis is to estimate the effect size of a particular treatment or therapy.
For example, a clinical trial may want to estimate the effect of a new surgical method in the recovery
time of patients. The default method frequently used in analysis is Cox proportional hazards regression (PH).
Cox PH regression uses time-to-event of each subject and estimates effect size using hazard ratios.
However, data on time-to-event may not always be available, for example in some epidemiological or
retrospective studies. When time-to-event is unknown but classified into intervals,
Poisson regression (PR) may be used to estimate and compare the relative risk (or rate) between
treatment and control groups. When information on time-to-event is not available at all and the data
only indicates if an event has occurred or not, logistic regression (LR) may be used to estimate the odds ratio.
This research investigates the relative performance of the three methods in estimating effect size of treatment.
A simulation study is conducted under various survival distributions and varying lengths of follow-up times.
We plan to characterize distributional features under which each method performs well, and subsequently propose
diagnostics for relative applicability of each method.
All statistics students are expected to attend.
.
Past colloquiums