Non-adult dental age assessment: correspondence analysis and linear regression versus Bayesian predictions

This study focuses on the evaluation of factors influencing the quality (accuracy and reliability) of non-adult dental age assessment from radiographic stages of permanent teeth (excluding the third molar). We used four distinct cross-sectional samples of 1,528 healthy children: 3 of known geographic origin (Ivory Coast, Iran and France) and 1 additional sample of children whose grandparents originated from a different continent. Two different methods of calculations are compared: the correspondence analysis combined with linear regression (CAR) and Bayesian predictions (with no independence assumption). Our results indicate that the quality of age assessment does not seem to depend predominantly on the use of geographic-specific standards. In the case of Bayesian predictions, we observed a clear trend in favour of significantly higher accuracy and reliability levels when using non-geographic-specific standards. One of the main advantage of Bayesian predictions over maximum likelihood methods of estimation is an overall increase in accuracy with high levels of reliability on a fraction of the test sample and, importantly, across all age categories (contrary to methods based on regression analysis). Importantly, in the case of Bayesian non-adult predictions, and contrary to age estimation techniques based on regression, a better quality does not depend on age.