Caries distribution modelling: an application on the data set from the first National Pathfinder survey of Italian 4-year-old Children

Caries prevalence in Western countries has declined since
the 1970s, so the distribution of caries indices (DMFT/S) became
positively skewed. These changes have had the effect of increasing
the proportion of zeros in the distribution of dental caries
indices. In the literature several models are applied to describe
the nature of the index distribution and applied in multiple regression
techniques to estimate the covariates of caries disease.
The aim of this paper was to determine the best model to estimate
the proportion of caries-free and the dependence of dmfs
index to the influence of childhood socio-demographic factors
(SDF). The data set from the first National Pathfinder survey of
4-year-old Children’s Oral Health in Italy was used. Negative
binomial distribution (NBD) and Poisson distribution (PD), as
the most common models for dental caries, were compared to
the recent related zero-inflated models (ZIP and ZINB). NBD
proved to be more appropriate for analyzing the distribution of
caries data in children than PD (p ! 0.001). Similarly, ZINB
goodness-of-fit was better (p ! 0.001) than that of ZIP, that of
ZIP better than that of PD and, finally, ZINB was better than
NBD. The significant background variables in bivariate analysis
(parent’s nationality and educational level, pre-term birth,
breast-feeding and age of tooth eruption) were considered as
predictors for dmfs in the ZINB regression model. Children
from parents with high educational level (mother or father) or
from a father of Italian nationality had higher probability of being
caries free (0.82 vs. 0.73, 0.83 vs. 0.74, 0.81 vs. 0.61, respectively).
Investigation of caries distribution via new models might
be useful to provide new insight into caries patterns.