Diagnostic Accuracy of Panoramic Radiography for Assessing Maxillary Posterior Root Protrusion into the Maxillary Sinus: A Cluster-Adjusted Prediction Model Using CBCT as Reference
TL;DR
This model has the potential to serve as a clinical decision-support tool to help prioritize the need for CBCT in patients pre-evaluated with panoramic radiography, however, its performance reflects apparent, in-sample estimates, and external validation is required before it should be considered for routine clinical use.
Abstract
Background/Objectives: The proximity of the maxillary sinus floor to the tooth apices is critical for dental procedures, so preoperative evaluation is crucial for preventing complications. This study aims to examine the accuracy of panoramic radiographs in determining the relationship between the maxillary sinus and teeth, in comparison with cone-beam computed tomography (CBCT), which serves as the reference standard, and to develop a cluster-adjusted clinical prediction model to assist clinicians in identifying patients who may benefit from CBCT imaging. Methods: A total of 3436 teeth were evaluated using CBCT and panoramic radiographs for the relationship between root tips and sinuses. The McNemar–Bowker test and the R program were used to assess how accurate panoramic radiographs are by comparing them to CBCT imaging. The performance metrics of panoramic radiography were calculated, and the generalized estimating equation (GEE) logistic regression model and the interactive risk calculator were developed. Results: The McNemar–Bowker test indicated a statistically significant systematic difference between the two modalities (χ2 = 190.01, p < 0.001); the overall three-category agreement was 82.2% (cluster-corrected 95% CI: 80.4–84.0%), corresponding to a dichotomized (protrusion vs. non-protrusion) accuracy of 87.6%. A cluster-corrected GEE logistic regression model, including panoramic classification, tooth type, age, and sex and developed and internally evaluated in this single-center cohort, predicted root protrusion identified using CBCT with high apparent discrimination (AUC = 0.948; cluster-based bootstrap 95% CI: 0.941–0.955, 1000 resamples; optimism-corrected AUC = 0.948) and satisfactory apparent calibration. Conclusions: This model, developed and evaluated internally within a single-center cohort, has the potential to serve as a clinical decision-support tool to help prioritize the need for CBCT in patients pre-evaluated with panoramic radiography. However, its performance reflects apparent, in-sample estimates, and external validation in independent populations is required before it should be considered for routine clinical use.