Integrating Clinical and Transcriptomic Profiles Associated with Vitamin D to Enhance Disease-Free Survival in Cervical Cancer Recurrence Using the CatBoost Algorithm

• Published in: Diagnostics (Basel) Vol. 15; no. 13; p. 1579
• Main Authors: Senthilkumar, Geeitha, Pitchaimuthu, Renuka, Dhanasekaran, Seshathiri, Panneerselvam, Prabu Sankar
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 21.06.2025; MDPI
• Subjects: Alfacalcidol; Algorithms; Biomarkers; Calcifediol; Cancer; Cancer therapies; Cervical cancer; cervical cancer recurrence; Chemotherapy; Datasets; Development and progression; Disease; disease-free survival; DNA methylation; Forecasts and trends; Gene expression; Genetic aspects; Human papillomavirus; Infections; long non-coding RNA; Machine learning; machine learning predictors; Medical prognosis; Medical screening; Metabolism; Metastasis; Nutrition research; Oncology, Experimental; Prevention; Radiation therapy; Relapse; Vitamin D; Vitamin deficiency; Womens health; machine learning predictors; long non-coding RNA; disease-free survival; cervical cancer recurrence; vitamin D
• ISSN: 2075-4418; 2075-4418
• DOI: 10.3390/diagnostics15131579

Background/Objectives: Cervical cancer is a leading cancer-related cause of death among women, with recurrence being a serious clinical issue. Recent evidence demonstrates that long non-coding RNAs (lncRNAs) affect cancer recurrence. This research investigates vitamin D’s regulatory actions in the recurrence of cervical cancer, centering on the involvement of lncRNA. Clinical data on 738 patients shows that greater serum vitamin D levels are linked to reduced recurrence rates and enhanced disease-free survival (DFS). Methods: A transcriptomic analysis of CaSki cervical cancer cells using data from the GEO dataset GSE267715 identified that vitamin D controls genes that prevent cervical cancer recurrence. Machine learning predictors CatBoost, LightGBM, Extra Trees, and Logistic Regression and feature selection methods such as ANOVA F-test, mutual information, Chi-squared test, and Recursive Feature Elimination (RFE) are used to identify predictors of recurrence, evaluating model performance using accuracy, precision, recall, ROC AUC, confusion matrices, and ROC curves. Result: CatBoost performs the best overall, producing an accuracy of 95.27%. CatBoost provided an ROC AUC of 0.9930, a precision of 0.9296, and a recall of 0.9706, and this implies a significant trade-off between the ability to detect metastatic cases correctly. Conclusions: These data identify the therapeutic potential of vitamin D as a regulatory compound and lncRNA as a potential therapeutic target in the recurrence of cervical cancer.