A1878
Title: Characterizing the survivor and hazard functions with quantile-parameterized Meta-B distributions
Authors: Bryan McNair - University of Colorado Anschutz Medical Campus (United States) [presenting]
Elizabeth Juarez-Colunga - University of Colorado Anschutz Medical Campus (United States)
Abstract: A quantile-parameterized meta-B distribution (QPM$_m$-B) is a flexible distribution based on any basis distribution and parameterized by quantiles. It achieves arbitrary flexibility by adding additional quantiles to the basis parameter vector. These distributions are placed on a continuum, with the basis distribution on one end and the true distribution on the other. The QPM$_m$-B methodology proposes a nonparametric estimator of the survival function that also allows estimation of the hazard function. A log-logistic distribution is used as the basis function. The QPM$_m$-B methodology can handle complex features of the data in a straightforward manner, including all types of censoring and truncation. Estimation is carried out using maximum likelihood methods, and consistency of estimators is demonstrated. Simulation studies assess the performance of the proposed method for small and large sample sizes. An application of the QPM$_m$-B methodology is illustrated through analysis of a rheumatoid arthritis study involving left-truncation and interval censoring on a reversed time scale by modeling time of first biomarker positivity prior to rheumatoid arthritis diagnosis.
