A Deep Model of the Incidence of Dental Caries on Proximal Surfaces

As a component of an analysis of the benefits of alternative frequencies of bitewing radi ographs to detect dental caries, the authors developed and validated a model to generate an individual's probability distribution for new carious lesions in a year. The model postulates two sources of variability in caries incidence-differences in individuals underlying caries susceptibilities and a random component. The model is used to examine the nature of caries risk over time. The large random fluctuations in an individual's caries susceptibility from year to year, combined with the random nature of caries attack, makes it difficult to predict future caries experience from the individual's caries experience in the recent past. By modeling the process giving rise to observed incidence data rather than focusing directly on the observed data, i.e., by developing a deep rather than a surface model, the authors have elucidated underlying disease dynamics and provided a basis for generalizing from the particular data used to develop the model.