This systematic review and meta-analysis evaluated the diagnostic accuracy of intraoperative frozen section for detecting malignancy in thyroid nodules with preoperative Bethesda III cytology. The clinical question addresses whether frozen-section reporting during surgery meaningfully discriminates malignant from non-malignant lesions in this cytologically indeterminate category and whether results can guide the extent of thyroid surgery.
The study protocol was prospectively registered in PROSPERO (CRD420261416683). The investigators searched PubMed, Embase, Web of Science, Europe PMC, and the Cochrane Library for studies that reported a separable Bethesda III cohort, intraoperative frozen-section findings, and final surgical histopathology. Eligible reports were required to provide data enabling construction of study-level 2 × 2 diagnostic accuracy tables for the Bethesda III subgroup.
The authors defined a frozen-section result as positive only when malignancy was explicitly reported. Results reported as benign, suspicious, indeterminate, deferred, inconclusive, or describing a follicular-pattern were classified as non-malignant for the analysis.
Study-level 2 × 2 tables were synthesized using random-effects logit models. Because every included study reported zero false-positive frozen-section results, a full bivariate model with freely estimated covariance was not identifiable. To accommodate this, the authors applied a pseudo-bivariate hierarchical summary receiver operating characteristic (HSROC) approximation. The QUADAS-2 tool was used for risk-of-bias assessment.
Ten studies met inclusion criteria, together comprising 1,069 Bethesda III patients or nodules. The dataset therefore reflects a pooled experience across multiple centres and study designs, as reported in the included manuscripts. The individual study characteristics, patient selection criteria, and local frozen-section reporting practices were those reported by the original studies.
The pooled sensitivity of intraoperative frozen section for malignancy in Bethesda III nodules was calculated as 43.1% (95% CI, 21.3–67.9). Sensitivity varied substantially between studies.
The pooled specificity was 98.8% (95% CI, 97.0–99.5). Specificity showed negligible between-study variance; notably, all included studies reported zero false-positive frozen-section results in the Bethesda III cohorts.
The approximate HSROC area under the curve (AUC) was reported as 0.89 when including all studies and 0.87 when excluding a single influential study (Mao 2023), but the authors note that these AUC estimates were driven largely by uniformly high specificity rather than by consistently high sensitivity.
Sensitivity across studies was highly heterogeneous (Q = 83.45, I2 > 90%, T2 = 2.15). Specificity showed minimal heterogeneity. Excluding the study by Mao 2023 reduced pooled sensitivity to 35.6% (95% CI, 23.8–49.5) and lowered sensitivity heterogeneity to moderate levels (Q = 12.44, I2 = 36%, T2 = 0.25), while specificity remained effectively unchanged at 98.7% (95% CI, 96.7–99.5).
The pooled results indicate that intraoperative frozen section in Bethesda III nodules exhibits excellent specificity but limited and inconsistent sensitivity. In practical terms, a malignant frozen-section result is highly likely to reflect true malignancy (making the test useful as a rule-in tool). However, non-malignant, indeterminate, deferred, or inconclusive frozen-section results should not be interpreted as reliably excluding cancer and therefore should not be used alone to determine the extent of thyroid resection.
These findings imply that when a frozen section yields an unequivocal malignant diagnosis in a Bethesda III nodule, conversion to a more extensive procedure may be supported by the test result. Conversely, a non-malignant frozen-section report does not provide sufficient negative predictive value to justify limiting surgery without additional data.
This work is reported as a preprint and has not undergone peer review; the authors explicitly note that it should not be used to guide clinical practice until validated. The analysis was constrained by available published data: all included studies reported zero false-positive frozen-section results, which limited the modelling options and required a pseudo-bivariate HSROC approximation. Study-level heterogeneity in sensitivity was high, reflecting variability in patient selection, pathology practice, and reporting categories across included studies. Details on individual study methods and potential sources of bias are those reported in the original articles; QUADAS-2 was applied but specific domain judgments were reported in the manuscript.
Across 10 studies and 1,069 Bethesda III nodules, intraoperative frozen section demonstrated very high specificity (~98.8%) but limited and variable sensitivity (pooled ~43.1%, lower when excluding an influential study). The evidence supports use of frozen section as a rule-in diagnostic adjunct when it reports malignancy, but a non-malignant or non-definitive frozen-section result should not be relied upon to exclude cancer or to determine the extent of thyroid surgery by itself. The study is a preprint and conclusions should be interpreted in that context.