Analysis of Bivariate Censored Longitudinal Data: A Case Study

Abstract

Laboratory measurements that are below the limit of detection (LOD) are common in continuous longitudinal data in the biomedical sciences. The presence of these left-censored values, or non-detects (NDs), complicates the statistical analysis of such data since improper treatments of NDs may lead to loss of power or biased results. This thesis aims at investigating the effects of high percentages (>50%) of NDs on quantifying the magnitude of dental Bisphenol A (BPA) exposure, measured by urinary BPA concentrations, before and after treatment in children treated with Bisphenol A glycidyl methacrylate (BisGMA) - based dental materials. Data analysis results using the data handled with different statistical methods, including naïve substitution, Paxton’s random imputation procedure, modified Paxton’s method, and multiple imputation, were compared. In conclusion, our results of all methods suggest that dental treatment using BisGMA-based materials is associated with elevated uBPA concentrations.