A re-analysis of caries rates in a preventive trial using Poisson regression models.

The analysis of caries incidence in clinical trials has several challenging features: (1) The distribution of the number of caries onsets per patient is skewed, with the majority of patients having few or no cavities; (2) the number of surfaces at risk varies (i) over time and (ii) between patients, due to eruption and exfoliation patterns, dental diseases, and treatments; (3) surfaces within a patient differ in their caries susceptibility, and (4) caries onsets within a patient are correlated due to shared host factors. Recent statistical developments in the area of correlated data analyses permit incorporation of some of these characteristics into the analyses. With Poisson regression models, the expected number of caries onsets can be related to the number of surfaces at risk, the time they have been at risk, and surface- and subject-specific explanatory variables. The parameter estimated in these models is an epidemiological measure of disease occurrence: the disease incidence rate (caries rate) or the rate of change from healthy (sound) to diseased (carious). Differences and ratios of these rates provide standard epidemiological measures of excess risk. To illustrate, Poisson regression models were used for exploratory analyses of the Ylivieska xylitol study.