Beyond the white line: a multicentre retrospective analysis of radiographic and clinical predictors of inferior alveolar and lingual nerve injury following mandibular third molar surgery.

BACKGROUND: Inferior alveolar nerve (IAN) and lingual nerve (LN) injury remain among the most significant complications of mandibular third molar surgery. Earlier studies relied primarily on panoramic predictors such as interruption of the white line (IoWL), often without three dimensional validation. This multicentre retrospective analytical study aimed to identify independent predictors of neurosensory injury and to develop and internally validate a composite risk prediction model. METHODS: Records of 2,850 third molar extractions performed between January 2019 and December 2024 across three tertiary centres were reviewed. After exclusions, 2,330 cases were analysed, including 710 with pre-operative Cone beam computed tomography (CBCT) performed selectively in patients considered at higher risk based on clinical assessment and panoramic radiographic findings rather than as routine imaging. Radiographic and clinical parameters were evaluated using standardized criteria. IAN and LN function were assessed using identical standardized neurosensory tests rather than subjective symptom reporting alone. Univariate and multivariable logistic regression analyses identified predictors of IAN and LN injury, and model performance was assessed using receiver operating characteristic analysis. RESULTS: IAN injury occurred in 4.1% and LN injury in 1.9% of cases, with most resolving within three months. IAN injury was associated predominantly with radiographic and operative difficulty markers, whereas LN injury was primarily related to surgical technique factors, particularly lingual flap retraction. Independent predictors of IAN injury included IoWL (OR = 3.41; p < 0.001), loss of cortication (OR = 2.76; p < 0.001), inter radicular canal position (OR = 4.12; p = 0.002), extensive ostectomy (OR = 2.23; p = 0.011), and surgery duration > 30 min (OR = 1.96; p = 0.032). The regression model demonstrated excellent discrimination (AUC = 0.87; 95% CI 0.83–0.90) and good calibration (Hosmer–Lemeshow p = 0.47). The panoramic-only model demonstrated an accuracy of 71.8% in the full cohort, while the combined panoramic and CBCT model achieved an accuracy of 88.4% within the CBCT subset. These values are not directly comparable due to differences in the underlying study populations. CONCLUSION: Panoramic signs, particularly IoWL, remain clinically valuable, but risk stratification may be enhanced when interpreted alongside selectively obtained CBCT findings and surgical parameters in higher-risk cases. The composite model may assist risk stratification and selective imaging but requires independent external validation before clinical implementation.