Can Childhood Longitudinal Cephalograms Predict Adult Facial Soft‐Tissue Dimensions? A Bayesian Modelling Study

Abstract

ABSTRACT Introduction Predicting facial soft‐tissue growth is essential for treatment planning and long‐term outcome forecasting in orthodontics, orthognathic surgery and facial aesthetic surgery. This study evaluated whether longitudinal childhood cephalograms can predict adult facial soft‐tissue dimensions using Bayesian growth models. Methods Serial lateral cephalograms from four studies in the AAOF Craniofacial Growth Legacy Collection (Bolton‐Brush, Denver, Iowa, Oregon) were analysed. The sample comprised 96 subjects (39 females, 57 males), over an age range of 6–29 years. Eight soft‐tissue linear measurements were analysed: upper and lower facial heights, upper and lower lip lengths, upper and lower lip thicknesses and soft‐tissue chin thickness at pogonion and menton. Population‐level and individual‐level Bayesian models were fitted for each variable. Predictive accuracy was assessed using five‐fold cross‐validation, with a childhood cutoff age of 14 years and prediction of withheld adult observations at age > 20 years. Predictive accuracy was quantified using the mean absolute error, bias and 95% prediction interval coverage. Results Mean absolute error ranged from 0.73 mm (lower lip thickness) to 2.85 mm (lower facial height), with 95% prediction interval coverage of 94.1%–99.0%. Bias was close to zero for all variables. Residual standard deviation reduction from the individual‐level model ranged from 3% (upper lip length) to 42% (lower lip length). Prediction accuracy was generally higher in females than males. Conclusion Longitudinal childhood cephalometric records contain predictive information regarding adult facial soft‐tissue dimensions beyond population‐based growth standards. Prediction accuracy and inter‐individual variation both differed by soft‐tissue variable and sex.