Applicability of Willems model for dental age estimations in Brazilian children

Several studies described tooth development as a reliable pathway for age estimations. Depending on the considered life span, the dental age indicators vary. In children, combinations of developing teeth provide the best information about age. In sub adults third molar mineralization is almost exclusively considered. The aim of this study was, firstly, to verify the Willems model in a Brazilian sample. Secondly, to observe differences between the Willems model and a new developed Brazilian model. Thirdly, the information of permanent teeth (PM) and third molar (TM), development was combined for age estimation in children. A sample of 1357 panoramic radiographs of Brazilian males (M) and females (F), with age between 5 and 23 years was collected. The technique of Gleiser and Hunt modified by Kohler (1955) [34] was applied for third molar staging in the entire sample. The Demirjian staging technique was used on the mandibular left permanent teeth (except third molars) of all individuals from 5 to 15 years. Kappa and weighted Kappa statistics were performed to verify inter- and intra-observer reliabilities. Based on the obtained Demirjian scores the Willems model was verified. Next the data were split to develop a new Brazilian model based on the Willems method and to verify the established model. The accuracy in age prediction between the Willems model and the new Brazilian model was compared. Additionally, regression models including PM, TM and PM plus TM information were compared. The Kappa and weighted Kappa statistics revealed high agreement between observers (0.88 Kappa; 0.93 weighted Kappa). The differences between predicted and chronological age for the verified Willems model were expressed in mean errors of -0.17 and -0.38 year for F and M respectively. The differences in mean error between the new developed Brazilian model and the Willems model were 0.02 (F) and 0.20 (M) year. The regression models combining PT and TM information provided only in the age range between 14 and 15.99 years a small decrease in root mean squared error (0.2 year) in females. The new developed Brazilian model provided similar age predicting performances as the Willems et al. model. Added TM was only providing more accurate age estimations in the ages of 14 and 15 year in F.