Recent Developments in Joint Modelling of Mixed Types of Outcomes
Chair: Xin (Cindy) Feng
Organizer: Xin (Cindy) Feng
Sponsor: Biostatistics Section
[Tuesday, June 13, 2017 15:30-17:00]
15:30-16:00
Charmaine B. Dean (University of Western Ontario)
Joint Analysis of Longitudinal Zero-heavy Panel Count Outcomes with Application to Understanding Desistance from Criminal Activity
Regression models for zero-inflated count data often need to accommodate within-subject correlation and between-individual heterogeneity; frequently random effects models are utilized for incorporating complex correlation structures. In cases where several longitudinal zero-heavy count outcomes are jointly considered, excess zeros may arise from several distinct sources. This is the case for a study of the patterns and mechanisms of the process of desistance from criminal activity. Methodological challenges in the analysis of longitudinal criminal behaviour data include the need to develop methods for multivariate longitudinal discrete data, incorporating modulating exposure variables and several possible sources of zero-inflation. There are additional complications such as some outcomes being prohibited during time in a secure facility; as well as intervention carry-over effects for some outcomes, and that the underlying process generating events may resolve for some individuals.
16:00-16:30
Dongsheng Tu, Hui Song (Dalian University of Technology), Yingwei Peng (Queen’s University)
Joint Modeling of Longitudinal Proportional Measurements and Survival Times with a Cure Fraction
In cancer clinical trials and other medical studies, both longitudinal measurements and data on a time to an event (survival time) are often collected from the same patients. Joint analyses of these data would improve the efficiency of the statistical inferences. We propose a new joint model for the longitudinal proportional measurements which are restricted in a finite interval and survival times with a potential cure fraction. A penalized joint likelihood is derived based on the Laplace approximation and a semiparametric procedure based on this likelihood is developed to estimate the parameters in the joint model. A simulation study is performed to evaluate the statistical properties of the proposed procedures. The proposed model is applied to data from a clinical trial on early breast cancer.
16:30-17:00
Elizabeth Juarez-Colunga (University of Colorado Denver), Brandie Wagner (University of Colorado Denver), Edith Zemanick (University of Colorado Denver)
Joint Analysis of a Longitudinal Binary Outcome and Recurrent Events with Application to Cystic Fibrosis Outcomes
In cystic fibrosis, chronic Pseudomonas aeruginosa (Pa) infection is associated with worse clinical outcomes including more frequent pulmonary exacerbations (PE). PE are themselves a leading cause of morbidity in CF and important endpoints in CF clinical trials. This talk discusses joint models that address the challenge of understanding the co-dependence of progression of Pa infection and recurrent PE over time. Using data from the ongoing Early Pseudomonas Infection Control study, we propose a joint model built within the frameworks of hidden Markov chain models to model progression of Pa, and dynamic recurrent event models to model the recurrence of PE. We address additional challenges in the motivating study including missing and misalignment of covariates, the dynamic aspect of the recurrence of PE, and the latent aspect of the different Pa states of infection.