Prognostic impact of adenomyosis in cervical cancer: insights from machine learning-driven survival analysis.

OBJECTIVE

The aim of this study was to determine whether adenomyosis is an independent prognostic factor in cervical cancer using integrated survival analysis and machine-learning models.

METHODS

This retrospective cohort study included 131 patients with early-stage cervical cancer treated surgically between 2008 and 2020. Patients were stratified by the presence (n=28) or absence (n=103) of adenomyosis based on final histopathology. Kaplan-Meier curves and log-rank tests assessed overall survival. Independent prognostic factors were identified through multivariate Cox regression and logistic regression analyses, supplemented by machine-learning decision tree modeling to evaluate variable importance and model performance.

RESULTS

Women with adenomyosis had no significant differences in tumor size, histology, depth of stromal invasion, lymphovascular space invasion, parametrial involvement, lymph node metastasis, or International Federation of Gynecology and Obstetrics 2018 stage. Kaplan-Meier analysis demonstrated shorter overall survival in the adenomyosis group (median overall survival 31.3 vs. 64.8 months, log-rank p=0.045). Multivariate Cox regression identified age, tumor size, International Federation of Gynecology and Obstetrics stage III, histologic grade, lymphovascular space invasion, parametrial infiltration, and vaginal involvement as independent determinants of overall survival (all p<0.05). Adenomyosis status did not retain prognostic significance after adjustment (HR 0.91, p=0.818). Decision tree models corroborated these findings, with International Federation of Gynecology and Obstetrics stage and tumor size emerging as the most influential predictors of survival and recurrence.

CONCLUSION

Although adenomyosis was associated with shorter overall survival in univariate analysis, it did not function as an independent prognostic factor after adjustment for established clinicopathological variables in both multivariate Cox regression and machine-learning decision tree models. These findings indicate that prognostic stratification in cervical cancer should remain guided by tumor burden, stage, and histopathological risk factors.