Related Experiment Video
Updated: May 31, 2026
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
Published on: April 18, 2025
A three-gene expression signature predicts lymph node metastasis in cervical squamous cell carcinoma: development and
Yachen Duan1, Jiajia Li1, Xiaohua Wu1
1Department of Gynecological Oncology, Fudan University Shanghai Cancer Center, Shanghai, China.
Background:
Lymph node metastasis (LNM) is a major prognostic determinant in early-stage cervical squamous cell carcinoma (SCC); however, conventional preoperative imaging has demonstrated sensitivities below 60% for sub-centimeter metastases, resulting in treatment misallocation for approximately 20-30% of patients.
Purpose:
This study aimed to develop and validate a minimal gene expression signature that can be performed on routine biopsy specimens to predict preoperative LNM in cervical SCC.
Methods:
This retrospective biomarker discovery and validation study had a two-phase design. In the discovery phase, we analyzed transcriptomic data from 116 The Cancer Genome Atlas (TCGA) cervical SCC samples and identified differentially expressed genes (DEGs) using Benjamini-Hochberg (BH) false discovery rate correction (FDR < 0.10). Then, we refined them using the least absolute shrinkage and selection operator (LASSO) regression analysis and multivariate logistic regression analysis. The locked signature was independently validated in a prospectively collected cohort of 202 patients (101 LNM-positive patients and 101 LNM-negative patients) from the Fudan University Shanghai Cancer Center using quantitative reverse transcription-polymerase chain reaction (qRT-PCR), with histopathological lymphadenectomy confirmation as the reference standard. Performance was assessed based on area under the curve (AUC), sensitivity, specificity, predictive values, and bootstrap internal validation. Decision curve analysis and a combined molecular-clinical model were also evaluated.
Results:
Of the 231 DE genes, a three-gene signature (LOC494141, GLOD5, and GML) was identified by sequential filtering. In the independent validation cohort, the signature achieved an AUC of 0.745 (95% CI: 0.676-0.814), with a sensitivity of 62.38%, a specificity of 64.36%, a positive predictive value of 63.64%, and a negative predictive value of 63.11%. Bootstrap validation confirmed model robustness (optimism-corrected AUC: 0.722; calibration slope: 0.913; Hosmer-Lemeshow p = 0.387). A combined model integrating the signature with tumor size and lymphovascular space invasion achieved an AUC of 0.789 (95% CI: 0.724-0.854), with a significant incremental value [net reclassification improvement (NRI) = 0.42, p = 0.001; integrated discrimination improvement (IDI) = 0.065, p < 0.001]. Decision curve analysis demonstrated net clinical benefit across threshold probabilities of 20-70%. At a 20% population prevalence, the adjusted negative predictive value reached 87.3%.
Conclusion:
This three-gene expression signature provides clinically informative preoperative risk stratification for LNM in cervical SCC. Intended as a complementary tool within integrated clinical assessment frameworks rather than a standalone diagnostic tool, this affordable qRT-PCR-based assay holds particular promise for resource-limited settings, pending prospective multicenter validation.
