Pretreatment predictors of orthodontic extraction-planning decisions: development and internal validation of an interpretable clinical prediction model

Background Orthodontic extraction planning remains a complex and experience-dependent aspect of treatment planning. This study aimed to develop and internally validate an interpretable clinical prediction model for documented extraction-planning decisions using routinely available pretreatment variables. Methods This single-center retrospective study with internal validation included 763 orthodontic patients, who were divided into training and hold-out validation cohorts using stratified random sampling according to the extraction-planning outcome. Candidate pretreatment predictors included demographic, dental, cephalometric, soft-tissue, and periodontal variables. A multivariable logistic regression model was developed in the training cohort and evaluated in the hold-out validation cohort for split-sample internal validation, while bootstrap resampling within the training cohort was used to estimate optimism in model discrimination. Model performance was evaluated using discrimination, calibration, Brier score, decision curve analysis, bootstrap internal validation, and subgroup analysis in borderline and non-borderline cases. Sensitivity analyses were performed by adding periodontal risk score and additional conventional cephalometric variables. The model was intended to estimate the probability of an extraction-planned treatment strategy rather than an objective clinical indication for extraction. Results The final model retained five pretreatment predictors: upper arch crowding, A-point–nasion–B-point (ANB) angle, upper lip to the esthetic line (E-line) distance, lower lip to E-line distance, and overjet. The model showed acceptable discrimination, with an area under the curve of 0.779 in the training cohort and 0.794 in the validation cohort. In the validation cohort, the Brier score was 0.182, the calibration intercept was 0.104, and the calibration slope was 1.001, indicating good agreement between predicted and observed extraction probabilities. Bootstrap internal validation estimated an optimism-corrected area under the curve of 0.772. In subgroup analysis, discrimination was higher in non-borderline cases than in borderline cases, with area under the curve values of 0.839 and 0.704, respectively. Adding periodontal risk score did not improve validation performance, and the extended nine-variable model provided only minimal improvement in discrimination with increased model complexity. Conclusions This study developed and internally validated an interpretable five-variable clinical prediction model for estimating the probability of an extraction-planned treatment strategy. The model demonstrated acceptable discrimination and good calibration and may provide decision-support information for orthodontic treatment planning, although discrimination was lower in borderline cases. It should not be interpreted as an independent determinant of the objective need for extraction. External validation in independent and multicenter cohorts is required before clinical implementation.