Diagnostic accuracy of intraoperative frozen section in thyroid nodules with Bethesda III cytology: systematic review and meta-analysis

Background Bethesda III thyroid nodules remain diagnostically indeterminate, and intraoperative frozen section is used selectively to support surgical decision-making. Its value in this specific cytological category is uncertain because a malignant result may be highly specific while non-malignant and non-definitive results may fail to exclude cancer. Objective To estimate the sensitivity and specificity of intraoperative frozen section for detecting malignancy in thyroid nodules with preoperative Bethesda III cytology. Methods The protocol was prospectively registered in PROSPERO (CRD420261416683). A systematic review was conducted in PubMed, Embase, Web of Science, Europe PMC, and the Cochrane Library. Studies were eligible when they reported a separable Bethesda III cohort, intraoperative frozen-section findings, and final histopathology. Frozen section was classified as positive only when malignancy was reported. Benign, suspicious, indeterminate, deferred, inconclusive, and follicular-pattern results were classified as non-malignant. Study-level 2 x 2 tables were synthesised with random-effects logit models. Because all studies reported zero false-positive results, a full bivariate model with freely estimated covariance was not identifiable; a pseudo-bivariate HSROC approximation was therefore used. QUADAS-2 was used for risk-of-bias assessment. Results Ten studies comprising 1,069 Bethesda III patients or nodules were included. The pooled sensitivity was 43.1% (95% CI, 21.3-67.9) and the pooled specificity was 98.8% (95% CI, 97.0-99.5). Sensitivity was highly heterogeneous (Q = 83.45, I2 >90%, T2= 2.15), whereas specificity showed negligible between-study variance. Excluding Mao 2023 reduced pooled sensitivity to 35.6% (95% CI, 23.8-49.5) and reduced sensitivity heterogeneity to moderate levels (Q = 12.44, I2 = 36%, T2 = 0.25); specificity remained 98.7% (95% CI, 96.7-99.5). The approximate HSROC AUC was 0.89 including Mao 2023 and 0.87 excluding Mao 2023, but these values were driven largely by the uniformly high specificity. Conclusions Intraoperative frozen section in Bethesda III nodules has excellent specificity but limited and variable sensitivity. It is more suitable as a rule-in test than as a rule-out test. A non-malignant or non-definitive result should not be used alone to exclude malignancy or determine the extent of thyroid surgery.