Predictive model for impaction of lower third molars.

The study was carried out to create and test a model for predicting impaction of lower third molars on the basis of radiographic findings at age 20 years. Fifty-six initially unerupted lower third molars were followed up for 6 years. Five radiographic findings in panoramic tomograms at age 20 were taken as variables. Clinical status at age 26 was taken as response. The radiographic features studied were angulation of tooth, development of root, state of impaction, depth in bone, and relation of the tooth to the ramus of the mandible and the second molar. With the use of logistic regression, univariate and bivariate analyses, and clustering techniques, a decision tree was constructed that indicated accuracies of prediction on the basis of single variables or pairs of variables. The most important predictor was the type of impaction. The model predictions and test teeth findings were in agreement in 94% of instances. It was concluded that the model is suitable for predicting lower third molar impaction at age 20.