IADR Abstract Archives
Challenges and Opportunities for Bayesian Inference on Oral Health Data
Objectives: To present the ontological challenges in dealing with longitudinal cohort data. To illustrate the use of Bayesian Factor (BF) for contingency table analysis and Bayesian logistic regression as an alternative to the classical p-value.
Methods: Data derived from three waves of the Growing Up in Ireland survey were merged and exploratory data analysis completed using R (version 4.0.2). A dashboard was created using the Shiny web application framework to provide aggregate summaries of dental and nutrition related variables. Contingency tables were analysed using BayesFactor package (BFCT) to test the association of previous dental problem visits at 3 or 5 years of age with dental fillings at 9 years of age. Both the classical generalised linear model and Bayesian logistic regression models for dental problem visits, based on known risk indicators as independent variables, were also generated. A simple logistic regression model predicting dental fillings based on a crude cariogenic index was estimated.
Results: Descriptive summaries indicated both an increase in dental problem visits (from 5% to 16%) and toothbrushing prevalence twice a day or more (from 52% to 75%) between 3 and 9 years of age. A test of association produced a BF >1,000 indicating extreme evidence for the non-independence of dental problem visits at 5 years of age and the probability of having a dental filling at 9 years of age. Not all socio-demographic and dietary variables could be included in the regression analysis as this would have over-specified the model. A crude cariogenic index allowed simple modelling of dietary factors.
Conclusions: Challenges related to ontology created difficulties for longitudinal analysis of dental health behaviours. Using Bayes factors for inferential statistics provided more meaningful evaluation of evidence for the probability of having negative dental health outcomes given dental problem visits at a younger age.
Methods: Data derived from three waves of the Growing Up in Ireland survey were merged and exploratory data analysis completed using R (version 4.0.2). A dashboard was created using the Shiny web application framework to provide aggregate summaries of dental and nutrition related variables. Contingency tables were analysed using BayesFactor package (BFCT) to test the association of previous dental problem visits at 3 or 5 years of age with dental fillings at 9 years of age. Both the classical generalised linear model and Bayesian logistic regression models for dental problem visits, based on known risk indicators as independent variables, were also generated. A simple logistic regression model predicting dental fillings based on a crude cariogenic index was estimated.
Results: Descriptive summaries indicated both an increase in dental problem visits (from 5% to 16%) and toothbrushing prevalence twice a day or more (from 52% to 75%) between 3 and 9 years of age. A test of association produced a BF >1,000 indicating extreme evidence for the non-independence of dental problem visits at 5 years of age and the probability of having a dental filling at 9 years of age. Not all socio-demographic and dietary variables could be included in the regression analysis as this would have over-specified the model. A crude cariogenic index allowed simple modelling of dietary factors.
Conclusions: Challenges related to ontology created difficulties for longitudinal analysis of dental health behaviours. Using Bayes factors for inferential statistics provided more meaningful evaluation of evidence for the probability of having negative dental health outcomes given dental problem visits at a younger age.