Title: Bayesian estimation of a time-varying bivariate distribution from censored data
Authors: Bradley Barney - University of Utah (United States) [presenting]
Garritt Page - Brigham Young University (United States)
Louise Lawson - Kennesaw State University (United States)
Reese Clark - Pediatrix Medical Group (United States)
Miguel de Carvalho - School of Mathematics, University of Edinburgh (Portugal)
Abstract: Motivated by the desire to construct reference growth curves for body mass index (BMI) in preterm infants, emphasis is on estimation of a time-varying bivariate density. The two inputs to BMI calculation, weight and length, are first jointly modeled because of the lesser frequency of observing length in the available data. From this joint distribution, the BMI distribution is induced. Bayesian implementations of several methods for the bivariate estimation are considered: structured additive distributional regression with a copula, nonparametric density estimation, and quantile regression. Existing approaches are extended to adjust the estimation for censoring, reflecting a prominent feature of the collected data. The performance of these methods are assessed in a simulation study, and the results from the motivating application are presented and discussed.