The Use of Readily Obtained Patient and Parent Data in Mandibular Growth Prediction.

Abstract

The ability to predict a patient's mandibular growth would be a powerful tool for the orthodontic practitioner. Despite several past attempts, a simple, accurate and reliable method still does not exist for mandibular growth prediction. Using only data readily obtained in a routine orthodontic clinical exam, I created several unique predictive models over an average T1-T2 interval of 2.77 years for 447 patients between the ages of 9 and 16 at T1 using a best-fit least squares linear method. Evaluating all possible combinations of 14 different non-age-related variables and three different age-related variables, 65,536 unique models were constructed to predict either the change in articulare-pogonion (∆ArPg) or the difference between the actual ∆ArPg and the expected ∆ArPg as predicted by a historical population average from T1 to T2 for three groups: males, females, and a subset of the females whose menarchal status is known. I assessed the mean absolute error (MAE) of each model using a leave-one-out cross-validation approach on half of the data (model-building sets) for each group. I then fit the most accurate models to the model-building sets, tested their accuracies (MAE) at predicting ∆ArPg on the other half of the data (validation sets), and compared these accuracies to those obtained when using historical population averages to predict ∆ArPg. The population averages were represented by polynomial growth curves (PGCs) which were fit to the pooled data of three published longitudinal growth studies (Michigan, London, and Philadelphia). Only three models proved to be more accurate than the PGCs when applied to the validation sets, but not to a clinically significant degree, being merely 0.07, 0.06, and 0.01mm more accurate. The variables used in this study provide no clinically significant insight into a patient's mandibular growth rate over a typical orthodontic treatment interval beyond what can be obtained from a historical population average.