Statistics Colloquium by Jackson Anderson ’24

Abstract:

Interval-censored data occurs frequently in follow-up studies, in which time to the event-of-interest of an individual (T) is only known to belong to an interval (L, R]. To compare the survival distributions of k treatment groups based on interval-censored data, various nonparametric test procedures have been developed. Alternatively, a linear rank-type test has been developed assuming a proportional reversed hazard model. In this work, we propose a new linear rank-type test based on a proportional odds (PO) model, a semi-parametric model which assumes a multiplicative effect of covariates on the odds of a survival event. A PO model might be preferred over the most popular proportional hazards model, when a hazard ratio is not constant over time. A simulation study showed that the proposed test works well. The new test is applied to two sets of real-world data for illustration.