Are the results of the Wits appraisal clinically interchangeable when evaluating 2D and 3D images?

Abstract Purpose To assess whether the results of the Wits appraisal derived from two- (2D) and three-dimensional (3D) images of the same cone-beam computed tomography (CBCT) acquisition are numerically interchangeable and to quantify diagnostic reclassification when conventional 2D thresholds are applied to 3D values. Methods This retrospective method-comparison study analyzed 233 CBCT datasets with a field-of-view adequate for evaluating the Wits appraisal. A CBCT-derived lateral cephalogram was generated for each scan as a digitally reconstructed radiograph (DRR), and 2D Wits (AO–BO) was computed by projecting Points A and B onto an occlusal line (interincisal–MVC6). The 3D Wits was computed on the volume by orthogonally projecting Points A and B onto an occlusal plane (interincisal, 1.6CF, 2.6CF) and measuring the signed in-plane anteroposterior distance (AO 3D –BO 3D ). The linear association between 2D and 3D measurements was assessed using Pearson’s correlation coefficient (r). Agreement was evaluated using Bland–Altman bias and limits of agreement (LoA), Lin’s concordance correlation coefficient (CCC), κ statistics, and equivalence testing (two one-sided tests [TOST]; δ = ±1 and ±2 mm). Diagnostic categories were defined using clinically used ±2 mm thresholds (class II > 2; class I −2 to +2; class III < −2 mm). Exploratory subgroup analyses assessed whether Δ = 3D − 2D Wits differed across 3D skeletal classes and vertical skeletal patterns. Results The 3D method yielded systematically lower values than the 2D (Δ = 3D − 2D: −1.97 ± 1.39 mm; p < 0.001) with LoA from −4.69 to +0.76 mm. Statistical correlation was strong (r = 0.935), but concordance was incomplete (CCC = 0.814). Using ±2 mm class cut-offs, three-class agreement was 60.1% (κ = 0.380), and 39.5% of the cases shifted toward class III on the 3D analysis. TOST did not demonstrate equivalence for δ = ±1 mm or δ = ±2 mm. In the available-case exploratory vertical-pattern analysis ( n = 225), the difference between 3D skeletal classes differed modestly across hypodivergent, normodivergent, and hyperdivergent groups, but the continuous SN^MP angle was not significantly associated with 3D − 2D Δ. Conclusion Wits values computed in 2D on CBCT-derived DRRs and in 3D on the corresponding CBCT volume were strongly associated but not numerically interchangeable in this dataset; therefore, conventional 2D thresholds should not be transferred directly to 3D values without 3D-specific validation.