To compare the survival difference between 2018 International Federation of Gynecology and Obstetrics (FIGO) stage IIB cervical cancer (CC) patients with unilateral parametrial invasion (UL) and bilateral parametrial invasion (BL) disease, and explore the significant role of parametrial invasion (PI) in prognosis prediction.
A total of 506 stage IIB CC patients were identified from the multi-center study, and patients were divided into UL and BL groups according to gynecological and radiological examination. Survival outcomes were estimated and compared between 2 groups before and after propensity scoring matching (PSM). The role of upper 2/3 vaginal invasion (VI) in impacting survival probability was also assessed. The random forest (RF) model was constructed and validated to select important features related to survival outcomes and predict prognosis for these patients. The SHapley Additive exPlanation (SHAP) was further introduced to provide a better understanding toward the findings from the RF model.
Significant better 5-year overall survival (OS) was observed among patients with UL disease whether before (BL: 61.7% [95% CI: 57.0%-66.4%]; UL: 84.8% [95% CI: 82.4%-87.2%]; HR = 2.83, 95% CI: 1.90-4.20, P < 0.001) or after PSM (BL: 61.3% [95% CI: 56.6%-66.0%]; UL: 81.2% [95% CI: 77.3%-85.1%]; HR = 2.51, 95% CI: 1.56-4.04, P < 0.001). Similar findings could also be observed in terms of progression-free survival (PFS). The presence of VI didn’t significantly impair the survival probability, whether in the UL or BL group (all P > 0.05). RF model was constructed, which possessed decent predictive ability both in the training (area under the receiver operating characteristic curve [AUC] = 0.893; 95% CI: 0.874-0.912) and validation cohort (AUC = 0.879; 95% CI: 0.801-0.957). PI was identified to be the paramount feature in affecting the survival outcomes for stage IIB CC patients through the Beeswarm summary plot and bar chart in SHAP analysis.
Our findings demonstrated that 2018 FIGO stage IIB CC patients with BL disease had a worse prognosis than those with UL disease, and PI was the most significant feature in prognosis prediction for these patients.