An example of complex modelling in dentistry using Markov chain Monte Carlo (MCMC) simulation

Area of Science:

• Statistical modeling
• Computational statistics
• Biostatistics

Background:

• Traditional regression analysis struggles with complex datasets, including longitudinal data and measurement errors in covariates.
• Handling multiple observations per subject and covariate errors complicates straightforward statistical evaluation.

Purpose of the Study:

• To introduce and demonstrate Bayesian methods and Markov chain Monte Carlo (MCMC) simulations for analyzing complex data structures.
• To illustrate the application of these methods in a dental research context with non-normally distributed data and measurement errors.

Main Methods:

• Utilized a Bayesian approach with Markov chain Monte Carlo (MCMC) simulations, specifically Gibbs sampling.
• Employed directed acyclic graphs (DAGs) for visual representation of complex model structures.
• Applied freely available BUGS software for model estimation.

Main Results:

• Demonstrated the convenient estimation of complex models using MCMC simulations.
• Explored the impact of measurement error on regression coefficients, highlighting potential underestimation ('regression dilution bias').
• Successfully applied the methodology to a dental dataset involving children followed over several years with complex tooth filling data.

Conclusions:

• Markov chain Monte Carlo (MCMC) methods offer significant value for dentists analyzing complex datasets.
• These methods facilitate the analysis of data exhibiting various forms of complexity, improving statistical rigor in dental research.