Class-Stratified, Axis-Resolved Inter-Rater Agreement of Cephalometric Landmark Identification and Six Sagittal Discrepancy Measurements: A Secondary Analysis

Background/Objectives: Cephalometric analysis remains the principal tool for diagnosing sagittal jaw discrepancies, and its reproducibility depends on the precision with which landmarks are identified along both coordinate axes. This study is a secondary, class-stratified analysis of a cephalometric dataset whose pooled analysis has been published previously. Its aim was to establish whether inter-rater agreement for 12 landmarks and for six sagittal measurements (ANB, Wits, Tau, Yen, Sar and W) differs between skeletal Classes I, II and III. Methods: Twenty-four lateral cephalograms (n = 8 per skeletal class) were assessed twice, seven days apart, by 15 orthodontists trained in a 5-h calibration course. No new measurements were collected for the present work. Agreement was quantified using the two-way random-effects intraclass correlation coefficient ICC(2,1) for absolute agreement. Results: Horizontal coordinates showed excellent agreement in every class (ICC 0.90–0.96), whereas vertical coordinates were markedly more variable (ICC 0.52–0.94). Among the sagittal measurements, Wits and Yen showed the most class-independent agreement (ICC 0.87–0.90 and 0.76–0.86); Tau, Sar and W reached their highest values in Class III (0.93–0.95) and ANB its lowest in Class I (0.55, 95% CI 0.32–0.84). The methodologically important observation is that the ICC and the standard error of measurement (SEM) need not order measurements in the same way: a measurement that ranks patients consistently (high ICC) does not necessarily locate an individual patient precisely. For a threshold-based decision such as skeletal-class assignment, absolute precision is the relevant property, yet published reliability tables rarely report it. Conclusions: Inter-rater agreement for the six measurements was not uniform across skeletal classes in this dataset. Because every class-specific estimate rests on eight cephalograms with wide and overlapping confidence intervals, and because ANB and Wits both contributed to the class assignment itself, these findings are exploratory and hypothesis-generating.