CFE 2019: Start Registration
View Submission - CMStatistics
Title: Right-censoring bias correction for growth curve linear mixed models Authors:  Dominique-Laurent Couturier - University of Cambridge (United Kingdom) [presenting]
Stephane Guerrier - University of Geneva (Switzerland)
Maria-Pia Victoria-Feser - University of Geneva (Switzerland)
Abstract: Tumour growth inhibition studies typically involve analysing tumour sizes measured regularly over a period of time. The aim is usually to detect differences in growth rate between experimental conditions. Many methods have been considered. Some summarise each growth curve into a single measure and compare the location parameter of these statistics between different experimental conditions by means of Welsh tests. Others consider mixed/longitudinal models, taking into account the time and within tumour dependence of the observations to provide a parametric fit on all collected data. As animals are culled when their tumour size exceeds a legal upper limit or when the discomfort level is considered too high, such data are often right censored, leading to biased growth estimates. The objective is to develop a method allowing one to correct the bias of growth curve linear mixed models in the presence of right censoring due to a fixed upper tumour size limit. Simulations show that the iterative bootstrap bias corrected estimator we developed for random intercept and slope mixed models allows us to obtain unbiased growth rate estimates as well as confidence intervals showing coverages close to the nominal value.