Diagnostic Accuracy of Frozen Sections for Intraoperative Detection of Spread through Air Spaces in Early-Stage Lung Cancer: Systematic Review and Meta-Analysis

Document Type

Conference Proceeding

Publication Date

5-27-2026

Publication Title

J Clin Oncol

Keywords

paraffin, cancer mortality, conference abstract, diagnosis, diagnostic accuracy, diagnostic test accuracy study, frozen section, human, lung cancer, meta analysis, metastasis, null result, receiver operating characteristic, recurrent disease, sensitivity analysis, surgery, systematic review, univariate analysis

Abstract

Background: Tumor spread through air spaces (STAS) is an independent predictor of recurrence and lung cancer–specific mortality in early-stage lung cancer. Intraoperative detection of STAS using frozen section analysis may aid surgical decision-making by guiding appropriate selection between lobectomy and sub-lobar resection. Methods: We systematically searched five databases from inception to October 2025 to identify PICO-concordant studies evaluating the diagnostic accuracy of intraoperative frozen section analysis for detecting STAS, using permanent paraffin sections as the reference standard. Pooled sensitivity, specificity, SROC curves, and diagnostic odds ratios were estimated using Bayesian bivariate hierarchical and random-effects models. Heterogeneity, sensitivity analyses, and publication bias were assessed in R (v2024.12.0+467) using the meta and mada packages. Results: Using a Bayesian bivariate hierarchical model, frozen section analysis demonstrated a pooled sensitivity of 0.66 (95% credible interval [CrI], 0.51–0.79) and specificity of 0.87 (95% CrI, 0.68–0.95), with a diagnostic odds ratio (DOR) of 12.74 (95% CrI, 1.86–23.63). Substantial heterogeneity was observed (I²E sensitivity: 0.82 [0.74–0.89]; I²E specificity: 0.72 [0.47–0.96]), with between-study variance estimates of 0.74 for logit sensitivity and 2.80 for logit specificity. In univariate analyses, pooled sensitivity was 0.60 (95% CI, 0.49–0.70; I² = 65.6%) and specificity was 0.87 (95% CI, 0.65–0.96; I² = 93.5%), yielding a DOR of 13.90 (95% CI, 5.95–32.49; I² = 88.6%). Heterogeneity remained stable on leave-one-out and subgroup analyses. Deeks’ funnel plot showed no evidence of publication bias (p = 0.62). Conclusions: Frozen section analysis demonstrates relatively high specificity but moderate sensitivity for STAS detection, making it more suitable as a rule-in rather than a rule-out test. Positive intraoperative findings may support escalation to lobectomy, whereas negative results should be interpreted with caution.

Volume

44

Issue

16

First Page

e20062

Share

COinS