Nomogram prediction of the lymph-vascular space invasion in cervical cancer: comparison of 2009 and 2018 staging systems

TL;DR

This study developed a nomogram to predict lymph-vascular space invasion in cervical cancer patients using clinical indicators, showing similar performance across two staging systems.

Contribution

A novel nomogram model for predicting LVSI in cervical cancer that performs consistently across 2009 and 2018 FIGO staging systems.

Findings

Five clinical variables (age, pathology, DSI, SCC-Ag, LDH) were selected for the LVSI prediction model.

Adding FIGO staging did not significantly improve the model's predictive performance.

The nomogram effectively stratified patients into high-risk groups with higher LVSI positivity.

Abstract

Lymph-vascular space invasion (LVSI) is a crucial prognostic factor in cervical cancer (CC), and its assessment is essential for developing personalized treatment strategies.

The primary objective of this study was to focused on constructing LVSI predictive model based on clinical indicators and evaluating its predictive performance across different FIGO staging cohorts.

We included 691 patients, with 348 patients having 2009 FIGO stage IB1-IIA2 CC assigned to Cohort 1, and 343 patients with 2018 FIGO stage IB1-IIIC1r CC assigned to Cohort 2. In Cohort 1, univariable and multivariable regression analyses, along with Mallows’ Cp, R squared-R, and LASSO, were used to select variables forming Model 1. Model 2 included the FIGO stage. We compared the contribution of different FIGO stages to the LVSI prediction model in both cohorts. The final LVSI prediction model for the entire cohort was constructed using selected variables and risk stratification was established. The models were evaluated through internal validations using ROC curves, C-index, Clinical Impact Curve (CIC), and Decision Curve.

Five variables were incorporated into Model 1: age, Pathology, Depth of Stromal Invasion (DSI), SCC-Ag, and Lactate Dehydrogenase (LDH). Model 2 was established by incorporating the FIGO staging system. Compared with the two models, there was no significant difference in ROC, ΔC-index and ΔNRI. Adding FIGO clinical staging did not significantly improve predictive value. Model 1’s variable were included in the nomogram for the combined cohort. The AUC for the model-development cohort and validation cohort was 0.754 (95% CI: 0.711, 0.798) and 0.789 (95% CI: 0.727, 0.852), respectively. In both cohorts, risk stratification effectively distinguished the high-risk group, which had a significantly higher proportion of positive cases compared to the low and middle-risk groups (p < 0.01).

Our nomogram predictive model demonstrates robust LVSI prediction performance across different staging systems.

Figures