Multilevel Models for Survival Analysis in Dental Research

This dissertation, "Multilevel Models for Survival Analysis in Dental Research" by Chun-mei, May, Wong, 王春美, was obtained from The University of Hong Kong (Pokfulam, Hong Kong) and is being sold pursuant to Creative Commons: Attribution 3.0 Hong Kong License. The content of this dissertation has not been altered in any way. We have altered the formatting in order to facilitate the ease of printing and reading of the dissertation. All rights not granted by the above license are retained by the author. Abstract: Abstract of thesis entitled "Multilevel models for survival analysis in dental research" Submitted by Wong Chun Mei, May for the degree of Doctor of Philosophy at The University of Hong Kong in July 2005 Introduction: Correlated interval-censored or grouped survival data with at least a 3-level dependence structure arise naturally in dental research. Simple models and estimation methods for the analysis of such data are not available in the literature and are warranted. Aim: To develop flexible multilevel random effects models for correlated grouped survival data. Methods: Three models were developed in this research: (i) a fully parametric multilevel random effects regression model in analyzing correlated interval-censored survival data; (ii) semi-parametric multilevel random effects regression models for correlated grouped survival data and; (iii) the extension of the semi-parametric models to allow the regression coefficients to vary over time in order to capture the possibly time-varying treatment effect. Estimation of the parameters was carried out using Monte Carlo Markov Chain (MCMC) approach with non-informative prior in a Bayesian framework to mimic the maximum likelihood (ML) estimation in a classical frequentist approach. Estimation of the intra-cluster correlations among the survival times was also considered. Data from a clinical trial investigating the effectiveness of topical fluoride agents in arresting active dentin caries in Chinese preschool children (SDF study) and another clinical trial on the survival of atraumatic restorative treatment (ART) restorations placed on permanent teeth (ART study) were used throughout the research. All analyses were performed using the software WinBUGS 13.0, 10000 simulations (after 5000 burn-in) were generated from the posterior distributions of the parameters. Results: Results from the multilevel random effects models showed that the intra-cluster correlation among the arrest times of dentin caries lesions in tooth surfaces from the same child and the correlation among the failure times of the different ART restorations from the same child were fairly strong (corr 0.60). child Hence, analyses without considering this strong association would be inappropriate and the statistical inference might not be valid. The semi-parametric multilevel random effects models were found to be more flexible than the fully parametric models because they accommodate time-dependent covariates naturally and can be extended to incorporate time-varying regression coefficients to investigate the possibly time-varying covariate effects. For the SDF study data, it was found that the treatment effects of the topical fluoride agents with caries removal prior to the application faded out rapidly after the first interval. Thus multilevel models assuming constant treatment effects over time may not be appropriate here. Conclusion: Three flexible multilevel random effects models were proposed. ML estimation of the parameters is extremely difficult as the likelihood function does not have an explicit form in general or even intractable particularly in the analysis of correlated interval-censored or grouped survival data. The use of the MCMC approach with non-informative prior in a Bayesian framework to mimic the ML estimation in a frequentist approach in multilevel modeling of correlated interval-censored or grouped survival data can be easily applied with the use of the s