A gene-based predictive model for lymph node metastasis in cervical cancer: superior performance over imaging techniques.

Objective: Lymph node metastasis (LNM) critically impacts the prognosis and treatment decisions of cervical cancer patients. The accuracy and sensitivity of current imaging techniques, such as CT and MRI, are limited in assessing lymph node status. This study aims to develop a more accurate and efficient method for predicting LNM.

Methods: Three independent cohorts were merged and divided into training and internal validation groups, with our cohort and those from other centers serving as external validation. A predictive model for LNM in cervical cancer was established using the LASSO regression and multivariate logistic regression. The diagnostic performance of the predictive model was compared with that of CT/MRI in terms of accuracy, sensitivity, specificity, and AUC.

Results: Using RNA-seq data, four independent predictive genes (MAPT, EPB41L1, ACSL5, and PRPF4B) were identified through LASSO regression and multivariate logistic regression, and a predictive model was constructed to calculate the LNM risk score. Compared with CT/MRI, the model demonstrated higher diagnostic efficiency, with an accuracy of 0.840 and sensitivity of 0.804, compared to CT/MRI's accuracy of 0.713 and sensitivity of 0.587. The predictive model corrected 81% of misdiagnoses by CT/MRI, demonstrating significant improvements in accuracy and sensitivity.

Conclusion: The predictive model developed in this study, based on gene expression data, significantly improves the preoperative assessment accuracy of LNM in cervical cancer. Compared to traditional imaging techniques, this model shows superior sensitivity and accuracy. This study provides a robust foundation for developing precise diagnostic tools, paving the way for future clinical applications in individualized treatment planning.