Is modelling dental caries a 'normal' thing to do?

Area of Science:

• Dental Public Health
• Biostatistics
• Epidemiology

Background:

• Caries data often exhibit non-normal distributions, posing challenges for traditional statistical analysis.
• Standard methods may not accurately capture the complexities of caries prevalence and severity.

Purpose of the Study:

• To advocate for and illustrate the application of generalised linear models (GLMs) in analyzing dental caries data.
• To demonstrate that GLMs can accommodate non-normal response distributions, offering a more appropriate analytical framework.

Main Methods:

• Employed three GLMs: normal, Poisson, and negative binomial, for modeling caries severity (dmf/DMF).
• Utilized a binomial model to analyze the dichotomous outcome of caries-free versus caries-present.
• Data from 871 primary school children in Manchester were analyzed.

Main Results:

• A significant covariate influencing caries presence was identified using a binomial model.
• The impact of a key covariate differed between normal and non-normal models when analyzing dmf/DMF data.
• Non-normal models (Poisson, negative binomial) provided distinct insights compared to the normal model.

Conclusions:

• Recommends Poisson or negative binomial models for dmf/DMF response and binomial models for caries-free/caries-present outcomes.
• This approach enables separate estimation of factors influencing caries magnitude versus caries presence.
• GLMs provide a robust framework for nuanced analysis of dental caries data.