Two-dimensional cephalometry is widely used for monitoring orthodontic treatments and for quantifying the outcome of maxillofacial surgery. Despite careful use of a cephalostat, successive radiographs might differ due to slight differences in patient posture. This study evaluates the reliability of lateral cephalometric measurements and estimates the impact of patient positioning on this reliability. We studied cephalograms of 104 patients; 31 of them had two radiographs because the first was deemed unsuitable for cephalometric analysis. Using AudaxCeph 3.0 (Audax, Ljubljana, Slovenia), two observers traced each cephalogram twice, one month apart. We evaluated intra- and interobserver agreement via Bland-Altman analysis, intraclass correlation coefficient (ICC), standard error of measurement, and smallest detectable difference (SDD). First, we studied the reliability of the hard tissue part of the Tweed-Merrifield analysis for 73 single cephalograms and for the better ones of patients with two exposures. Then, we studied 31 unsatisfactory cephalograms, and the ones recorded at improved patient posture. Although intraobserver bias was less than 0.5° or 0.3 mm, interobserver bias was significant for most measurements. Intraobserver reliability was high (ICC > 0.9), whereas interobserver reliability was good (ICC > 0.83) except for FMPA, FMIA and OP. Head rotations and inclinations had little impact on reliability (e.g., interobserver SDD decreased for 3 of 11 measurements). We conclude that averaging the positions of bilateral structures enables a reliable cephalometric analysis in spite of imprecise patient posture. Retaking cephalograms is ethically questionable in such cases.
Keywords: Interobserver agreement; Intraobserver agreement; Lateral cephalograms; Subject posture.